EPA-450/3-78-014a
March 1978
GROWTH EFFECTS OF MAJOR
LAND USE PROJECTS
(WASTEWATER FACILITIES)
Volume I: Model Specification
and Causal Analysis
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air and Waste Management
Office of Air Quality Planning and Standards
Research Triangle Park, North Carolina 27711
-------
EPA-450/3-78-014a
GROWTH EFFECTS
OF MAJOR LAND USE PROJECTS
(WASTEWATER FACILITIES)
Volume I: Model Specification
and Causal Analysis
by
Peter H. Guldberg, Ralph B. D'Agostino, and Richard D. Cunningham
Walden Division of Abcor, Inc.
850 Main Street
Wilmington, MA 01887
Contract No. 68-02-2594
EPA Project Officer: Thomas McCurdy
Prepared for
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air and Waste Management
Office of Air Quality Planning and Standards
Research Triangle Park, North Carolina 27711
March 1978
-------
This report is issued by the Environmental Protection Agency to report
technical data of fnterest to a limited number of readers. Copies are
available free of charge to Federal employees, current contractors and
grantees, and nonprofit organizations - as supplies permit - from the
Land Use Planning Office, Office of Air Quality Planning and Standards,
Environmental Protection Agency, Research Triangle Park, North Carolina
27711; or, for a nominal fee, from the National Technical Information
Service, 5285 Port Royal Road, Springfield, Virginia 22161.
This report was furnished to the Environmental Protection Agency by Walden
Division of Abcor, Inc., Wilmington, Massachusetts 01887, in fulfillment
of Contract No. 68-02-2594. This report has been reviewed by the Land Use
Planning Office, EPA and approved for publication. Approval does not
signify that the contents necessarily reflect the views and policies of
the Environmental Protection Agency, nor does mention of trade names or
commercial products constitute endorsement or recommendation for use.
, .
I
Publication No. EPA-450/3-78-014a
i i
-------
ACKNOWLEDGEMENTS
Special appreciation goes to Mr. Thomas McCurdy, the U.S. Environmental
Protection Agency project officer for this study, whose extensive assistance
and advice was ind1spensable in the conduct of the study. Appreciation is
also extended to all EPA technical committee members for guidance given
throughout the study. The technical committee included representatives of
the Office of Transportation and Land Use Policy, the Municipal Construction
Division (Facilities Requirements Branch) and the Control Program Develop-
ment Division.
In addition, we wish to thank the more than one hundred individuals in
city/county/regional planning agencies and transportation departments nation-
I
wide who cooperated with us during the data collection task and helped provide
the data on which this study is based. Their time and cooperation was in-
va 1 uable.
Urban Systems Research & Engineering, Inc. (USRE) of Cambridge, MA were
subcontractors to Walden on this study, assisting in the definition of basic
model concepts, infrastructure relationships and exogenous variables, and in
the testing and refinement of the causal and predictive models. The coopera-
tion of Dr. James F. Hudson, who guided the USRE technical effort, is deeply
appreciated.
Finally, recognition is
difficult task of field data
Mr. Michael Geraghty and Mr.
due the Walden staff who performed the long and
collection - Mr. Mahesh Shah, Ms. Diane Gilbert,
Kei th Kennedy.
i i i
I
-------
Secti on
I
II
III
TABLE OF CONTENTS
Ti t1e
I NTRODUCTI ON . . . . . . . . . . . . . . . . . . . . . .
A. Study Overview. . . . . . . . . . . . . . . . . . .
B. Technical Background. . . . . . . . . . . . . . . .
1. Federal Wastewater Funding Programs. . . . . . .
2. Wastewater Facilities. . . . . . . . . . . . . .
3. Land Use Impacts. . . . . . . . . . . . . . . .
C. Problem Statement. . . . . . . . . . . . . . . . . .
D. General Approach. . . . . . . . . . . . . . . . . .
E. Glossary of Terms. . . . . . . . . . . . . . . . . .
f
PHASE I - INITIAL MODEL SPECIFICATION. . . . . . . . . .
A. Perform Literature Searches. . . . '. . . . . . . . .
B. Aerial Photographic Interpretation. . . . . . . . .
C. Define Modeling Approach. . . . . ~ . . . . . . . .
1. Theory of Induced Development. . . . . . . . . .
2. Approach to Theory Testing,. . . . . . . . . . .
3. Model Form and Characteristics. . . . . . . . .
4. Technique Selection for Path Analysis. . . . . .
D. Define Induced Development. . . . . . . . . . . . .
E. Define Major Project. . . . . . . . . . . . . . . .
F. Define Area of Analysis. . . . . . . . . . . . . . .
G. Initial Choice of Variables. . . . . . . . . . . . .
H. Specify Initial Path Analytic Land Use Model. . . .
1. RES Equation. . . . . . . . . . . . . . . . . .
2. COMM Equation. . . . . . . . . . . . . . . . . .
3. OFFICE Equation. . . . . . . . . . . . . . . . .
4. MANF Eq ua ti on . . . . . . . . . . . . . . . . . .
5. HIWAYS Equation. . . . . . . . . . . . . . . . .
6. EDUC Equation. . . . . . . . . . . . . . . " . .
7. REC Equation. . . . . . . . . . . . . . . . . .
8. WHOLE Equation. . . . . . . . . . . . . . . . .
9. OTHER Equation. . . . . . . . . . . . . . . . .
Page
1-1
1-1 '
1-3
1-3
1-4
1-8
1-13
1-14
1-17
2-1
2-1
2-2,
2-2
2-2'
2-3
2-5
2-6
2-11
2-17
2-23
2-29
2-44
2-58
2-60
2-61
2-62
2-63
2-63
2-64
2-65
2-65
PHASE II - DATA COLLECTION. . . . . . . . . . . . . .. 3-1
A. Case Study Selection. . . . . . . . . . . . . . .. 3-1
1. Selection Criteria. . . . . . . . . . . . . .. 3-1
2. Information Sources. . . . . . . . . . . . . .. 3-3
3. Final Selection. . . . . . . . . . . . . . . .. 3-9,
B. Data Collection. . . . . . . . . . . . . . . . . . .' 3-10
1. Transportation Data. . . . . . . . . . . . . .. 3-10
2. Survey Forms. . . . . . . . . . . . . . . . .. 3-18
3. Field Surveys. . . . . . . . . . . . . . . . .. 3-29
4. Create Computer Data File. . . . . . . . . . .. 3-35
iv
-------
TABLE OF CONTENTS (CONTINUEDL
Section
Ti t1 e
IV
PHASE III - CAUSAL ANALYSIS. . . . . . . . . . . . . . .
A. Path Analysis. . . . . . . . . . . . . . . . . . . .
1. Preselection of Exogenous Variables. . . . . . .
2. Identification.................
3. Multicollinearity. . . . . . . . . . . . . . . .
4. Suppressor Variables. . . . . . . . . . . . . .
5. Theory Trimmi ng . . . . . . . . . . . . . . . . .
6. Comparison of Model Forms. . . . . . . . . . . .
B. Coefficient Stability Analysis. . . . . . . . . . .
1. Extreme Value Index. . . .'. . . . . . . . . . .
2. Coefficient of Variation Index. . . . . . . . .
C. Determine Net Causal Effects. . . . . . . . . . . .
1 . Approac h . . . . . . . . . . . . . . . . . . . .
2. Results and Conclusions. . . . . . . . . . . . .
D. VMT Model Validation. . . . . . . . . . . . . . . .
1. Develop Test Criteria. . . . . . . . . . . . . .
2. Model Validation, . . . . . . . . . . . . . . . .
V
SUMMARY OF RESULTS AND CONCLUSIONS. .
. . . . .
. . . ..
VI
REFERENCES. . . . . . . .
.. .. . . ..
..........
APPENDIX A - Path Analysis
. .. . . .
.. . . ..
. .. .. .. .. .
APPENDIX B - Causal Analysis in GEMLUP Projects
.. . . .. .
APPENDI X C - Aeri a 1 Photographi c InterpretaUon .
. .. . ..
APPENDIX D - Bibliography. .
.........
.....
APPENDIX E - Case Study Data*
.. .. .. .. .
.. .. .. ..
. . .. .. ..
APPENDIX F - Condescriptive Output* .
..........
APPENDIX G - Simple Correlation Output* .
.. .. .. .. .. .. .. ..
APPENDIX H - Statistical Output of the Final Path
Ana1ysis*. . . . . . . . . . . . . .
. .. .. ..
APPENDIX I - One Way Frequency Distributions of Model
Coefficient Values from the Jackknifing
Ana 1 ys 1 s *.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..
*
Appendices E through I are not included in this volume. They may be
ordered from the Strategies and Air Standards Division, U.S. Environmental
Protection Agency MD-12, Research Triangle Park, NC 27711.
v
Page
4-1
4-1
4-2
4-3
4-3
4-7
4-8
4-24
4-27
4-30
4-34
4-37
4- 37
4-38
4-51
4-51
4-56
5-1
6-1
A-1
B-1
C-1
D-1
E-1
F-1
G-1
H-1
1-1
-------
Table
2-1
2-2
2-3
2-4
2-5
2-6
2-7
2-8
2-9
2-10
2-11
2-12
2-13
2-14
3-1
3-2
3-3
. 3-4
3-5
3-6
3-7
4-la
4-lb
4-2
LIST OF TABLES
Ti tle
Page
Initial List of Endogenous Variables ~ . . .. . . . . . . 2~13
Point Estimates of the Construction Period of Interceptor
Sewer Projects. . . . . . . . . . . . . . . . . . . . . . 2-21
Construction Grant Information on a Sample of Current
Wastewater Projects in Federal Regions II, IV, and VI 2-24
Definition of Endogenous Model Variables. . . . . . . .. 2-30
Definition of Exogenous Model Variables. . . . . . . . . 2-33
Independent Variables for the RES Model Equation. . . . . 2-49
Independent Variables for the COMM Model Equation. . .. 2-,50
Independent Vari abl es for the OFFICE Model Equation 2-51
Independent Variables for the MANF Model Equation. . . . 2-52
Independent Variables for the HIWAYS Model Equation. . . 2-53
Independent Variables for the EDUC Model Equation. . . . 2-54 .
Independent Variables for the REC Model Equation. . . . . 2-55
Independent Variables for the WHOLE Model Equation. . .. 2-56
Independent Variables for the OTHER Model Equation. . . . 2-57
Summary of Survey Questionnaire Returns. . . . . . . .. 3-5
Sample Survey Questionnaire. . . . . . . . . . . . . .. 3-6
Case Study Major Projects. . . . . . . . . . . . . . .. 3-7
Selected Variables Found Significant In Urban
Transportation Planning. . . . . . . . . . . . . . . . . 3-14
u.S. Geological Survey Land Use and Land Cover
Cl ass i fi ca ti on Sys tern. . . . . . . . . . . . . . . . .. 3-31
Compatibility Between the U.S. Geological Survey and
GEMLUP Land Use Classification Systems. . . . . . . . .. 3-32
Summary of GEMLUP Case Study Transporta ti on Data. . . .. 3-34
Instrumental Variables Used In The Initial Two-Stage
Least Squares Analysis. . . . . . . . . . . . . . . . .. 4-4
Instrumental Variables Used In The Final Two-Stage
Least Squares Analysis. . . . . . . . . . . . . . . . .. 4-4
Confirmation of Sufficient Identification In The Initial
Two-Stage Least Squares Analysis. . . . . . . . . . . .. 4-5
vi
-------
LIST OF TABLES (CONTINUED)
Table
Title
4-3
Independent Variables Not Significantly Correlated With
Dependent Variables In The Simultaneous Block of
Equations. . . . . . . . . . . . . . . . . . . . . . . .
4-4
Extreme Values Which Accounted for a Substantial Portion
of Unexplained Variance in the Path Analysis and Produced
La rge Re s i dua 1 s . . . . . . . . . . . . . . . . . . . . .
Comparison of Coefficients of Variation for a Pure Rates
and a Mixed Totals-Rates Model. . . . . . . . . . . . .
Comparison of the Number of Standardized Residuals
Greater than 1.0 in Absolute Value For Two Model Forms
Comparison of Path Regression Coefficients from the
Final Path Analysis and the Jackknifing Stability
An a 1 y s is. . . . . . . . . . . . . . . . . . . . . . . .
Extreme Value Index of Stability for all Path Regression
Coeffi cients in the Causal Model Equations. . . . . . .
Stability Index for all Path Regression Coefficients in
the Causal Model Equations Based on the Effect of Varia-
tion in the Coefficients on the Dependent Variable
Summary of Direct and Indirect Causal Effects on the
Va ria b 1 e RE S . . . . . . . . . . . . . . . . . . . . . .
Summary of Direct and Indirect Causal Effects on the
Variable COMM .....................
Summary of Direct and Indirect Causal Effects on the
Variable OFFICE. . . . . . . . . . . . . . . . . . . . .
Summary of Direct and Indirect Causal Effects on the
Va ri ab 1 e MAN F ......................
Summary of Direct and Indirect Causal Effects on the
Variable HIWAYS . . . . . . . . . . . . . . . . . . . . .
Summary of Direct and Indirect Causal Effects on the
Variable EDUC . . . . . . . . . . . . . . . . . . . . . .
Summary of Direct and Indirect Causal Effects on the
Va ri ab 1 e REC . . . . . . . . . . . . . . . . . . . . . .
Summary of Direct and Indirect Causal Effects on the
Variable WHOLE. . . . . . . . . . . . . . . . . . . . .
Summary of Direct and Indirect Causal Effects on the
Variable OTHER. . . . . . . . . . . . . . . . . . . . .
4-5
4-6
4-7
4-8
4-9
4-10
4-11
4-12
4-13
4-14
4-15
4-16
4-17
4-18
vii
Page
4-9
4-13
4-25
4-26
4-28
4-32
4':'35
4-39
4-40
4-41
4-42
4-43
4-44
4-45
4-46
4-47
-------
LIST OF TABLES (CONTINUED)
Table
Titl e
Page
4-1-9
Compilation of Nationwide Data on the Relationship
Between Trip Generation Rates and Quantities of
Land Use. . . . . . . . . . . . . . . . . . . . . . . . . 4-53
Validation Results for Trip Production Equations Developed
for Pittsburgh, PA ................... 4-54
Suggested Validation Criteria for Trip Generation Rates. 4-55
Trip Length Data Distributions Compiled by the Federal
Highway Administration. . . . . . . . . . . . . . . . . . 4-57
Actual and Predicted Trip Length Data in Miles. . . . . . 4-59
Comparison of Actual and Predicted Trip Generation Rates
by Land Use. . . . . . . . . . . . . . . . . . . . . . . 4-65
Comparison of Actual and Predicted Daily Trips in the
Area of Ana lys is by Land Use and Purpose. . . . . . . . . 4-67
Comparison of Actual and Predicted Total VMT for the Area
of Ana 1 ys is. . . . . . . . . . . . . . . . . . . . . . . 4-68
4-20
4-21
4-22
4-23
4-24
4-25
4-26
J
"
viii
-------
Figure.
1-1
1-2
2-1
2-2
2-3
2-4
2-5
4-1
4-2
4-3
4-4
4-5
4-6
4-7
Exhi bit
3-1
lIST OF FIGURES AND EXHIBITS
Titl e
Typical Elements of a Sewage Collection, Treatment &
Disposal System. . . . . . . . . . . . . . . . . . . . .
Technical Approach to the Development of a Statistical
Mode 1 for Predi cti ng the Growth Effects of Major.
Wastewater Projects. . . . . . . . . . . . . . . . . .
Example of the Displaced Growth Phenomenon Within a
Hypothetical Drainage Basin. . . . . . . . . . . . . . .
Hypothetical legal Service Area of a Wastewater Major
Project Illustrating the Locational Effects of the
Project's Collection Network. . . . . . . . . . . . . .
Initial Specification of the Path Analytic land Use
Mode 1 . . . . . . . . . . . . . . . . . . . . . . . . . .
Seven Equation Simultaneous Block in the Initial Path
Analytic land Use Model. . . . . . . . . . . . . . . . .
Two Equation Recursive Block in the Initial Path
Analytic Land Use Model. . . . . . . . . . . . . . . . .
Final Path Analytic Model. . . . . . . . . . . . . . . .
Scattergram of Actual Versus Predicted Work Trip
Len 9 th s . .. .. . . . . . . . . . . . . . . . . . . . . . . .
Scattergram of Actual Versus Predicted Other Trip
lengths. . . . . . . . . . . . . . . . . . . . . . . . .
Scattergram of Actual Work Trip Lengths Versus Model
Re sid ua 1 s . . . . . . . . . . . . . . . . . . . . . . . .
Scattergram of Actual Other Trip Lengths Versus Model
Re s ; d ua 1 s . . . . . . . . . . . . . . . . . . . . . . . .
Scattergram of Actual Versus Predicted VMT . . . . . . .
Scattergram of Actual VMT Versus Model Residuals. . . .
Titl e
Regional Planning Agency Survey
Form. . . . . . . . . .
ix
Page
1-5
1-16
2-18
2-26
2-46
2-47
2-48
4-12
4-60
4-61
4-62
4-63
4-70
4-71
Page
3-19
-------
types, large residential developments and large industrial/office parks,
has been performed previously and is reported on in .the three volume set of
GEMLUP-I final reports [17-19]. The secondary air quality impacts of
wastewater projects are determined by the land use growth induced by such
projects. That is, the air impacts are emissions from (1) residential
complexes that appear at the end of and along new sewer lines, (2) other
service-oriented land uses (commercial, industrial, office, government)
that relate to residential development, and"(3) motor vehicles used as
tran.sportation between development areas. Thus, the key to understanding
secondary air quality impacts is to first understand the growth effects of
a wastewater facility on land use in a region. The objectives of the study
effort were: (1) to develop and test a path-analytic causal model that
represents the induced land use ten years after construction of a wastewater
major project, (2) to develop and validate a simplified predictive model
of induced development, (3) to test and correct the GEMLUP-I VMT model [19],
and (4) to develop worksheets that can be used to predict induced land use
and associated air pollution emissions. For these purposes, data were
collected from forty (40) case study wastewater projects nationwide.
The study project was divided into four major phases:
I.
II.
III.
IV.
Definition of basic concepts and initial model specification,
Data collection,
Causal analysis of the land use model using path analysis,
Development of predictive equations for the land use model
and worksheet procedures.
Two separate technical reports are to be prepared. This is the first
volume report and covers phases I-III of the study. The remainder of this
chapter examines the technical background to induced development, gives a
statement of the problem, an outline of the general study approach, and a
glossary of terms used in this report. Chapters II-IV summarize the
technical performance on phases I through III, respectively, while Chapter V
summarizes the results and conclusions for these first three phases.
1-2
-------
1.
INTRODUCTION
A.
STUDY OVERVIEW*
Pursuant to 40 CFR 51.12(d)-(h), State Implementation Plans must
contain provisions to prevent any national ambient air quality standards
from being exceeded. These provisions are called Air Quality Maintenance
Area (AQMA) plans, and estimating the air quality impact of major land
use and urban development projects is a necessary part of AQMA planning
[42-47J. In addition, the National Environmental Policy Act [48J and the
Council on Environmental Quality (CEQ) "Guidelines on the Preparation of
Environmental Impact Statements II [49]** require the considerati.on of
secondary impacts from major projects. CEQ states that:
"Many federal actions, in particular those that involve the
construction or licensing of infrastructure investments (e.g.,
highways, airports, sewer systems, water resource projects,
etc.), stimulate or induce secondary effects in the form of
associated investments and changed patterns of social and
economic activities. Such secondary effects through their
impacts on existing community facilities and activities, or
through changes in natural conditions, may often be more
substantial than the primary effects of the original action
itself [49J."
This has been a particular concern for wastewater systems, since their
primary impacts tend to be small, i.e., sewers and treatment plants
generally improve water quality, but they may lead to significant
negative secondary impacts. The probable large indirect impacts (redi-
recting growth and inducing development) of a new or expanded regional
sewage treatment facility on ambient air quality, and the need for some
procedure to ascertain what its impact will be, is recognized in the AQMA
Guideline series ([42J:A-7ff, [45J:21ff). To date, EPA has not developed
a model to estimate what the ambient impacts will be for use in AQMA
planning. Thus, it was the purpose of this study to develop such a model.
This study is entitled the Growth Effects of Major Land Use
Projects (GEMLUP-II), and it addresses the induced growth effects of
wastewater major projects. Similar research on two other major project
* .
This section is partially based on material in reference [21J.
**
CEQ is in the process of promulgating regulations for the Preparation of
EISs. . . .
1-1
-------
B.
TECHNICAL BACKGROUND
1.
Federal Wastewater Funding Programs
\ .
Understanding the past history of federally funded wastewater collec-
tion and treatment projects is important to the current study since the identi-
fication of wastewater case study projects will depend to a large extent upon a
review of published federal reports, as well as state and water/sewer district
reports. Federal financial aid in the construction of municipal sewage treat-
ment works was first authorized in 1948 (PL 80-845). This loan program was
never implemented because necessary funds were not appropriated by Congress.
The Federal Water Pollution Control Act of 1956 (PL 84-660) included the first
authorization for federal grants to assist in the construction of waste treat-
ment works. The Act authorized an 'appropriation of $50 million per year for
such grants to be allocated to the States on the basis of relative population
..
and per capita income. The grants were limited to 30% of the eligible project
cost to a maximum of $250,000. Appropriations were increased during the early
1960's, and major amendments to PL 84-660 occurred in 1966. At that time,
appropriation authorizations were increased, th~ maximum dollar limitation on
grants was dropped, and the federal share was increased to a maximum of 55%.
Most recently, the Federal Water Pollution Control Act Amendments of 1972
(PL 92-500) set up a three-phased program for the elimination by 19S5 of pol-
lutant discharges into the nation's navigable waters. 'First, an extensive
matching grant program was instituted for the construction of municipal waste-
water facilities (Sections 201-207). Under this program, the federal share has
increased to 75% of eligible costs and annual expenditures have risen above
$1 billion. Secondly, the National Pollution Discharg~ Elimination System has
been adopted to control point source discharges through a permit program
(Section 402). Finally, to coordinate the activities of the first two programs,
an interrelated group of water quality planning processes are being implemented,
viz. facility, area-wide, and state-wide planning (Sections 201,208,211, and
303). This coordinated effort of wastewater planning represents the first
attempt by Congress to integrate land use planning with the control of water
pollutant discharges.
1-3
-------
Administration of this rapidly expanding federal program has passed
through many different agencies over the years. Much of the earliest federal
wastewater funding came from the U.S. Public Health Service (under HEW) and
the Community Facilities Administration (the predecessor agency of HUD).
From 1962 to 1965, federal funding of sewer projects was accomplished under
the Accelerated Public Works Act of 1962 (PL 87-658) for which the Area
Redevelopment Administration (ARA) of the Department of Commerce was the lead
agency. In 1965, the Economic Development Administration took over these
duties from the ARA, and, during the late 1960s the Appalachian Regional Com-
mission also participated ln the funding process. Finally, in 1970, EPA
inherited most of the responsibility for overseeing all federal wastewater
funding and planning.
. 2.
Wastewater Facilities
a.
Basic Elements
Sewerage systems are required to collect and treat the materials
entering them. The basic elements of wastewater management systems, shown
in Figure 1-1, are the treatment plants which handle wastes delivered to
them, and the various types of sewers carrying the material. The first
stage of a sewer is the service connection to a residence, which transports
waste from the household to small collector or lateral sewers serving
streets. These feed into trunk sewers serving whole subdivisions. The
larger sewers which receive wastes from a number of trunk lines are known'
generically as "interceptors", though this term originally meant only
lines which "intercepted" untreated discharges to streams. The service
area of a wastewater system is usually defined by gravity to include the
watershed surrounding the lines, though use of pumping stations and force
mains can allow the wastes to traverse watershed boundaries, and therefore
extend the service area. A full discussion of the concept of legal service
area is included iIT Section II.
Community wastewater facilities have generally been constructed
in piecemeal fashion in the past. Some typical examples follow:
1-4
-------
WA~1E COLLECTION
~ TREATMENT
SERYltE
CONNKT"N ()'~tHAA6ER& )
, ~~~
nUNKS ~
r"L, ~ 0 C MAINS
] D 'NYWf,Pr,IS
.....
I
U1
II
FIGURE- 1-1
TYPICAL ELEMENTS OF A
$EWAGE cou£tT'ON,
TREA1h\ENt; ~ DI~PO&Al
~Y~TEtI\.
Source: [76J
SOLIDS
PIS,OSAL
I I
GREATMENT 0
~\) 0 8
OUTFALL
R£'lfVIN.
STREAN
~
W l
If
~
\~
J.
~
~
\
-------
. A town puts in sewers in the early 1900s that drain to nearby
streams or lakes. Interceptor lines are put in around 1960 to
pickup discharges to the streams and transport them to a newly
built treatment plant. The potential for induced land develop-
ment associated with interceptor construction in this case may
be small since only the means of treatment is changing, not the
availability of sewer connections. An example is the Oity of
Los Altos, CA [50]. In 1957, an extensive sewer system was
built in Los Altos to pickup the waste load from septic tanks
that were polluting the local water supply. Since the majority
of land within the city was developed prior to the construction
of the sewer system, no significant expansion has occurred
subsequently.
. Several towns in a region put in sewers in the 1920s that drain
. to separate primary treatment p1 ants. By 1960, the plants are.
used beyond capacity and equipment is in need of repair. Instead
of constructing a new treatment plant in each town, one large
regional treatment plant is bui 1 t and interceptor 1 ines (perhaps
with pumping stations) are built out to adjoining towns to pick
up each town.s waste stream. In the process, the interceptor
lines traverse unsewered, rural areas between towns and open
this land up to development pressures. An example is the City
of Phoenix, AZ [51]. In 1965, work was completed on the 91st Avenue
treatment pl ant. This 45 MGD* faci 1i ty picked up the waste
streams of the nearby cities of Scottsdale, Tempe, and Mesa (in
addition to Phoenix), allowing them to abandon their outdated
plants. In the process, the interceptor lines crossed developable,
vacant land in and around Phoenix, opening these lands up to
intense development.
. A rapidly growing urban center updates its wastewater facilities
by building a new treatment plant in 1960. Although sewer
service has been available in the urban core for decades, outlying
areas have relied on septic disposal systems. Due to the poor
,'percolation characteristics of the local soil, however, these
septic systems have been causing both health and water quality
problems in the community. Therefore a large amount of excess
capacity is designed into the new treatment plant to allow urban
fringe and suburban areas access to the wastewater facilities.
The extension of large interceptor sewers into predominantly
rural areas ()pens.. up vast areas of land to intensive development
pressures. An example is Met~opolitan Washington, D.C. Between
1957 and 1966, 16 different wastewater projects involving the
construction or extension of interceptor sewers in Prince George's
and Montgomery Counties in Maryland were completed by the Washington
Suburban Sanitary Commission at a cost of over $13 million [52].
More recently, between 1967 and 1973, 23 more interceptor projects
were completed in the same area at a cost of over $170 million.
As a result, Metropolitan Washington experienced the highest rate
of growth during the 1960's of any large metropolitan area in the
*
Million gallons per day.
1-6
-------
U.S. [53]. Be~we:n 19.60 and 1972, metropolit~n popul&tion grew
from 2 to 3 mllhon persons. In gener&L th,s new development
followed the interceptor lines into predominantly rural areas
outside Washington, D.C. spreading suburban sprawl.
b.
Sys tem Capacity
The amount of sewage generated within a community can be simply
defined as water consumption minus losses (e.g., lawn sprinkling) plus addi-
tions (e.g., infiltration). Sewage flow varies quite widely, showing seasonal,
daily, and hourly fluctuations. For health reasons, wastewater collection
systems are designed to accomodate the flow of the heaviest hour of the year
for the design population. Therefore, interceptor lines are always built with
"excess capacity" - that is, capacity beyond that needed to handle the peak
flow from the present population of the area served. This excess capacity is
available to developers to serve future projects that might connect to the lines.
The proportion of excess capacity provided may vary from a low of 10-20% up to
200% or more, depending upon the growth implied by the design population fore-
cast and the estimates of average flow and peaking factors. EPA recently reduced
its design factors for sewers considerably;because of the costs of overdesign.
c.
Package Plants
In the 1950s, in response to the demand for treatment facilities
to serve isolated subdivisions and institutions, manufacturers developed
miniaturized sewage treatment plants. Package plants can be regarded as
filling the gap between individual septic tanks and large custom-designed
municipal treatment plants [54], in the 1000 to 100,000 gallon per day range.
For example, ~mrris County, NJ had 15 separate package plants in existence in
1965 to serve developing, suburban areas [55]. As interceptor lines have become
available in this and many other areas, the waste load is diverted to sewers
and the package plants removed.
d.
Wastewater Districts
Although the traditional organization to provide sanitary services
has been the local government (municipal or county), regional commissions have.
1-7
-------
also been formed for this purpose. Special districts, which have been created
to operate a specific service and are largely financially independent, can be
found today in all metropolitan areas immediately outside the central city and
at numerous other locations within the U.S. serving small localities. The
U.S. Census of Governments reports the existence of 1,411 special sewer
districts in the U.S. in 1972 [56]. The direct reasons for their emergence
has been the past refusal by the administrators of the existing central
(municipal) systems to extend their service, such as sewer lines, very far
into other political units, and the inability of the new suburban munici-
palities to finance a complete set of utilities. Federal incentives for.
area-wide approaches (under the Water Pollution Control Act Amendments of
1956 and 1972) have also increased the use of regional systems.
3.
Land Use Impacts
The secondary effects of infrastructure investments are indirect
long-term changes that occur as a result of operation or use of new facilities.
. Several recent studies have utilized an approach that analyzed the secondary
impacts of wastewater facilities. A study examining the growth of northeast
Philadelphia from 1945 to 1962 found a correlation between high density zoning
and the enormous rise in real estate values of land with public sewer connections
[57]. In metropolitan Washington D.C., Rabe and Hudson [58] report positive
correlations between highway and sewer investments, and residential construction
from 1960 to 1970, while Promise and Leiserson r53] have found statistically
significant relationships between population growth, sewer capacity, and dis-
tance to interceptor lines. Shapiro and Tabors [59] developed estimates of
sewer impact on land values for Clay, New York, also showing significant effects.
A study of the influence of interceptor sewer policy on development in Fairfax
County, Virginia, found that interceptor sewers affected the location of new
sprawl of new subdivisions [60]. Based upon "self-fulfilling prophecies of
inflated population projections" by county sewer authorities, sewer investme~ts
created "intense financial pressures for residential development". The
rationale of the county authorities was that these major capital investments
had to be repaid through "hook-up .and service charge, revenues on the service
line" [61].
1-8
-------
The CEQ has concluded that the secondary impacts of interceptor sewers
upon undeveloped land at the urban fringe is the most critical element of waste
treatment facility planning as related to land use planning [62]:
. "Interceptor sewers are defined as the major lines that run from the
collector sewers to the treatment plant. Because the location of a
new interceptor signtficantly increases the number of buildable lots
along its right of way, a key issue is its capacity. There is a
general tendency for s:uch lines to be oversized in order to assure
the necessary capacity for future development, but the oversizing
itself can contribute to the extent of development that occurs. II...
. "A related land use impact caused by large in'terceptor sewers is
their tendency to be designed to run for long distances between
existing towns before reaching the treatment plant. Such lines open
up large areas of what may have been previously undeveloped land
between the towns."...
. "Another phenomenon related to the construction of large interceptors
is the tendency for developers to move immediately to the end of the
new line in order to take advantage of both the available sewer service
and the low land costs on the far urban fringe. The result is a costly
leapfrog and fill-in development pattern, which increases the diffi-
culty of properly planning the time and size of other public facilities
and spreads the urban area out in a pattern that is wasteful of land
and energy resources. II
The significant relationship between extension of interceptor sewers into a com-
munity and the resultant urban sprawl patterns has been researched and confirmed
in several .CEQ-commissioned reports [61,63-67]. While PL 92-500 and EPA regu-
lations do not allow the use of federal funds for collector lines in sewer
developments built after 1972 (Section 211), excess capacity for growth has been
commonly included in the more major investments.
The costs associated with sewage disposal systems are not always real-
ized at the time of installation. Builders prefer public sewered land for their
developments since (1) the sewer connection fee is, in general, lower in cost
to them than constructing individual septic systems, cesspools. or neighborhood
package plants, and (2) they are not restricted by soil, drainage, or water
quality restrictions in their development of the land. However, the true costs
of a large wastewater collection and treatment system are borne by the resi-
dents of the area, not the developer, through increased property taxes to retire
1-9
-------
the large debt 'incurred. Thus', in The Costs of Sprawl [67], sprawl was de-
termined to be the IImost expensive form of residential development in terms
of economic costs, environmental costs, natural resource consumption, and
many types of personal cost. II
In another report commissioned by CEQ [66], interceptor extensions
were described as creating a direct incentive to the land developer to build
at the urban fringe: IINew sewers increase the density of possible development
and thus the potential economic Irentl (and development profit) per unit of
land to the owner or developer. II For example, a study of secondary impacts
in New Jersey [80] concluded that:
1I...Sewers are the critical ingredient and the guiding force for
growth in New Jersey. As the cost of land and construction rises,
more townhouses and multi-family units will be built in proportion
to single family homes. Sewers are esse~tial for this higher
density construction. As a result, the role of sewers as a growth
determinant will become even stronger in the future. II
In certain locations, sewers make medium and high density development feasible
where previously, due to soil conditions and septic tank limitations, only
single family homes on large lots could be built. Thus, adding sewers removes
a development constraint. Their absence often prevents intense development of
an area, although in the absence of laws prohibiting on-site disposal systems,
not all development is precluded. Many times builders will develop an un-
s~wered area, installing individual septic tanks without regard to their
appropriateness, effectiveness, or water quality impacts. If these septic
tanks do not function properly because of adverse soil conditiions, munici-
palities must rescue them by installing interceptor sewers. Thus, every
homeowner pays double for sewer service. In addition, such improvement pro-
jects are far more costl'y than building sewers through open land due to the
obstructions of existing structures, roadways, etc.
Attempts to control the land development associated with wastewater
facilities "have involved both delaying building permits and outright sewer
moratoria. But recent studies of communities in which these policies are in
1-10
-------
use have shown that far from giving impetus to reconcentration of development
in urban core areas, the moratoria can encourage sprawl by sending builders
to jurisdictions farther out where there are no restrictions and land is cheap
[68]. And even when such controls are effective, they can cause other problems
in the areas of water quality and real estate values. An example community is
San Diego, CA. This city currently exceeds the standards for several air pol-
lutants, and like most of southern California, has a severe oxidant problem.
San Diego County. is also a high growth area due to its climate and large tracts
of vacant, developable land. As a result, the region has had to rapidly
expand wastewater facilities numerous times in the past. In the past few
years, EPA has consciously delayed funding on construction of new wastewater
facilities in the area due to the potential for their interfering (indirectly)
with the attainment of air standards [77]. The effect of delays in construction
funds has been widespread moratoria on connection~ to existing sewer lines and,
thus, de facto moratoria on new residential construction. The net effects have
been both good and bad. Uncontrolled growth has been halted for a while in
San Diego, and the Comprehensive Planning Organization has been given signifi-
cant powers for planning future development in the County. However, real
estate values have skyrocketed to unrealistic values, speculation is rampant,
. and inadequate treatment of existing wastes, resulting from the lack of new
treatment facilities, is causing serious water quality problems.
The above discussion of induced development represents a view which
has also been stated as follows:
.,. . - -..-
liThe comprehensive plan for an area represents a policy decision
which is logically followed by implementation programs, one of
which is the planning and construction of utilities. Precisely
because there are no effective means to control overall "develop-
ment" consistent with public goals, it is necessary and responsible
to at least conform public spending for capital construction with
adopted policy. Sewerage plans become 'self-fulfilling prophecies'
by opening new areas for development, and therefore, often 'cause'
the sprawl that functional planners claim to be simply serving
after the fact" [69J.
1-11
-------
. ....., -- ,_. -..
- .._-- --. --- --.. .'.... -- . -.-..._-_. - --
- ....._- .~--- .. -.
--'--."--' ..'. -.-----
- ~- ------.
The planning community, however, is not unanimous in this view. For example,
the director of New York's clean water program has the opposite viewpoint:
"The granting of federal funds for pOllution abatement facilities is not the
cause of expanding land use, but is the cure for the effects of land use
created by other sociological forces" [70]. Indeed, a 1965 survey of sewer
service in the New York City metropolitan area found that, in general, sewers
were lagging behind development in suburban areas, not causing it [71].
Package plants and septic ~ystems were the principal means of wastewater
treatment in the tri-state area in the 1950s and 1960s. Also, in Metropolitan
Seattle, studies have shown that the implied cause-effect sequence (sewers
"induce" developemnt) can run both ways. The Puget Sound Council of Govern-
ments has raised many alternative reasons for the perceived sewer-sprawl
relationship [69]:
. "To what extent does sprawl merely in fill behind the 1956 Federal
Aid to Highways Act? Sewerage construction is funded on a high
priority basis as the population withdraws from the central city.
Is central city deterioration a cause of shifting "demand," or
the consequence of construction-oriented categorical grants? Does
the provision of utilities reinforce "demand" as well as merely
responding to demand? What part of "demand" is due to constitu-
tionally protected rights, and what part is due to real estate tax
exemptions, capital gains benefits, average cost pricing of utilities,
deductibility of taxes and interest paid on borrowed money, roads
funded ironically on the basis of distance traveled (gas tax), etc."
Finally, a recent study for the National Water Commission [72] ~oncluded that:
. "Federal investment in water resource development has not had
noticeable effects on the growth or decline of communities. This
assistance has for the most part come after a need is recognized,
not before. While Federally supported water-related facilities
may permit growth to occur, they do not cause its occurrence...
...Fundamental economic and location factors determine whether a
community will grow or decline, and the availability of water-
related facilities and services plays a minor role... (since)
...the costs for sewage collection and treatment represent a
relatively small proportion of community intrastructure and
service costs."
-""-----'-
1-12
-------
In summary, although land use patterns, and particularly the phenome-
non of suburban sprawl, are the result ofa complex set of historical, economic,
social, and political interactions, the role of wastewater facilities in
inducing development can be a major one. Thus, the growth effects of such
projects should be recognized and dealt with to ensure the attainment of
planning goals for a region, including the attainment and maintenance of
air quality standards.
C.
PROBLEM STATEMENT
Extensive research has been done to design models for simulating the
physical characteristics (demand, flow) of storm and wastewater systems. As a
result, many dynamic models for designing entire networks are available
(e.g., EPAls Storm Water Management Model (SWMM), Illinois' Storm Water Sewer
Simulation Model (ISS), and the Systems Operational Analysis Model (SOAM)).
Although research has also been undertaken to examine the secondary impacts
of wastewater facilities (as discussed in previous sections), similar model
development has not occurred. Predictive models of secondary development are
required to answer questions concerning the impact, a priori, -of infrastructure
investments. One model that includes infrastructure as a determinant of land
use growth is the EMPIRIC model [73] (or in its more recent form, PLUM), an
"activity allocation model" that apportions population and employment pro-
jections to areas based on measures, in part, of sewer service availability.
However, EMPIRIC does not simulate the more indirect long-term impacts of in-
frastructure investments, nor does it attempt to include any causal specifica-
tion. Recently, research has been done to develop methodologies for secondary
air quality impacts [78,79]. However, these approaches have proven to be
less than innovative, since they generally use the design population projection
for a sewage treatment plant as the basis for simple growth effects assessments.
Land use patterns and their attendant activities have a significant
impact upon the type and amount of air pollution generated over a region.
To the extent that land use can be associated with the discharge of pollu-
tants, it is necessary to plan future land use and transportation that is
compatible with acceptable levels of air quality. Comprehensive planning has
in the past been relatively insensitive to air quality considerations. New
1-13
-------
'-
I
air quality management procedures require that the planning community generate
and'use analytical tools to incorporate air quality constraints into the
p1anni,ng process [42-47J. A recent review of the state-of-the-art in models
relating land use planning to air qual ity considerations [44] concluded that
existing techniques are limited in application by the area-specific data base
on which they are developed and their inability to disaggregate projected
pollutant emissions by land use category. Thus, there is a need for a
generaZized analytical tool that can (1) predict land use development induced
by major projects (such as wastewater facilities) by category type, and (2)
convert such detailed land use projects into pollutant emissions for use in
air quality management analyses. It was for these reasons that the GEMlUP-I
model [17-l9J was developed and the current study was undertaken to develop a
model of the induced development associated with major wastewater collection
and treatment systems. GEMlUP relates to a number of EPA programs [74J,
including AQMA planning, Environmental Impact Statement (EIS) review, the
prevention of significant air quality deterioration or non-degradation, and
the recent emissions offset policy promulgated for growth in non-attainment
.areas. Explicit or implicit in these programs is an evaluation of air quality
impacts of land use plans or project developments. To perform such assess-
ments, EPA has recently ruled that a portion of Section 208 grant money for
waste quality management planning may be used for "air quality assessments of
existing or projected development to be served by wastewater treatment facili-
ties" [75J. Most recently**, Congress has amended the Clean Air Act, Section 316
of which now empowers the EPA Administrator to deny construction funds for
wastewater facilities if an area is not covered by an SIP which "expressly
quantifies and provides for" emissions related to secondary growth effects.
D.
GENERAL APPROACH
The principal objectives of this study were to formulate a statis-
tical methodology to predict air pollutant emissions from:
. Induced land development associated with the construction and
operation of wastewater facilities in a community.*
*
The causal model of induced development used in this study does not take
into account the effects of mitigating measures. Variables measuring restric-
tion on on-site disposal and hook-ups to existing interceptor lines were in-
cluded in the development of the model, however, see Section II.
**
The EPA office of Transportation and land Use Policy, Washington, DC, is in
the process of preparing Guidelines for reviewing and mitigating secondary
imp~cts of wastewater treatment facilities. These Guidelines will be
available in early 1979.
1-14
-------
. Motor vehicular traffic associated with the induced land
development.
The ability to accurately predict the secondary development induced by major
wastewater collection and treatment projects is dependent on understanding
the complex interrelationships inherent in such a model. Thus, an important
objective was to formulate and test a causal theory of induced development
using path analysis. Additional objectives were to test and correct the motor
vehicle traffic model developed as part of the previous GEMLUP-I study [19]
for use in the air pollutant emissions projection procedure, and to integrate
the predictive land use model, the traffic model, and land use based emission'
factors in a set of easy-to-use worksheets.
The approach to fulfilling these objectives involved
execution of four separate project phases shown schematically
and summarized below.
the sequenti a 1
in Figure 1-2
In Phase I, our primary interest was to define the basic concepts
on which this study is based. The first step in this process was to
determine the mOdeling approach. Next, the infrastructure causal relationships
of wastewater facilities in communities were studied and the knowledge used to
define "induced development" in the model. The concepts of a "wastewater major
project" and the "area of analysis" for induced development were also studied
and defined. Knowing the important causal relationships enabled us to de-
velop a list of model variables representative of the relevant factors involved
(e.g., the major project, land use, regional growth). Finally, specific
causal relationships between ,the variables were hypothesized and formalized
in an initial causal model.
The principal objective of Phase II was to collect a sufficiently
large. diverse, and thereby representative, cross-sectional data base on which
to develop the model. Due to the critical importance of the quality of this
data base and the large manpower effort required to collect such data, the
first task required a careful selection of 40 case study wastewater major
projects distributed on a nationwide basis, which had the potential, upon
construction, to induce a significant quantity of land development in their
communities and for which all the requisite data were available. To this end,
1-15
-------
PHASE I
INITIAL MODEL SPECIFICATION
I
0'1
. DEFINE
MODELING
APPROACH
IDENTIFY
INFRASTRUCTURE
RELATIONSHIPS
-----,
I
DEFINE MAJOR
PROJECT AND
AREA OF
ANALYSIS
INITIAL CHOICE
OF VARIABLES
PHASE II
DATA COLLECTION
PHASE III
CAUSAL ANALYSIS
!
I PRELIMINARY
I CASE STUDY
SURVEY
r .n -- -'
COEFFICIENT
STABILITY
ANALYSIS
G--------,
DETERMINE
-- NET CAus:JL
EFFECTS
----r
I
,
i
0---- '---
PHASE I V
DEVELOPMENT OF PREDICTIVE EQUATIONS
l MODEL
VALI DATION
----~
i
CASE STUDY
SELECTION
SPECI FY
PRELIMINARY
MODEL
\
\-ll~
I
I
, I
i VALIDATE r-
PATH. GEMLUP
ANALYSIS i TRAFFIC
I MODEL
--~~ - -~-~~~--=L---
DATA
COLLECTION
THEORY
TRIMMING
FI GURE 1-2
TECHNICAL APPROACH TO THE DEVELOPMENT OF A STATISTICAL MODEL
FOR PREDICTING THE GROWTH EFFECTS OF MAdOR WASTEWATER PROJECTS
STEPWISE
REGRESSION
ANALYSIS
j
DEVELOP
EMI SSIONS
PROJECTION
WORKSHEETS
-------
a case study mail survey based upon the screening of over 15,000 federally
funded wastewater collection and treatment projects nationwide was performed.
Final selection of the 40 case study projects was made, a data collection
training course was conducted for field personnel, followed by site visits to
the case study areas to collect the required data.
The objectives of Phase III were to develop the final causal model
and validate the GEMLUP-I VMT model [19]. The initial causal model was tested
using the set of case study data and the statistical techniques of path analy-
sis. This approach verified whi.ch of the hypothesized causal relationships
were significant, trimmed those that were not, and determined model parameters
for the final causal model. Tasks were also performed to trace the direct and
indirect effects the model variables have on one another and to validate and
correct the GEMLUP-I VMT model.
In Phase IV, predictive equations of induced land use associated with
wastewater major projects were developed and validated. These predictive
equations, the GEMLUP traffic model, and GEMLUP land use based emission factors
[18] were then used as the basis for an emissions projection procedure. The
procedure was developed in workbook form to serve as a general i zed analytical
tool for use by planners and environmental engineers in predicting the induced
land use and air pollutant impacts associated with major wastewater collections
and treatment systems.
E.
GLOSSARY OF TERMS
A dependent variable is the output variable of a mathematical
function, which has as its arguments one or more input or independent variables.
Given a set of interrelated variables, an exogenous variable is one
whose value is determined by forces external to or outside the system of
relationships. An endogenous variable is an internal variable of the system
which is completely determined (i .e., caused) by one or more exogenous and/or
endogenous variables. The relationships of anyone endogenous variable can
be translated into mathematical form by designating it as the dependent
variable of a function whose independent variables are all related causal
factors, whether they be endogenous or exogenous variables.
Instrumental variables are exogenous variables which are introduced
into nonrecursive systems of causal relationships (i.e., those involving
1-17
-------
interactions between endogenous variables) to allow the estimation of structural
coefficients. A detail list of conditions associated with instrumental varia-
bles is given in Heise [15].
A wastewater major project is defined as the construction or extension
of interceptor or collector sewer lines during the period 1958 to 1963 in a
community in the United States. If construction date information is not
available, then a grant funding date in the period 1956-1960 is acceptable.
The project had to affect an increase in absolute system collection capacity
of 1 MGD or more, and had to cost a minimum of $200,000 to construct. Phased
projects are considered if the first phase of construction on the collection
network was complete within 1958-1963 and the last phase of construction was
complete by 1965.
Induced development is defined as urban land use associated with,
caused, stimulated, or allowed by, or located because of the construction and
operation of a major project.
Secondar~ development is induced development which has a direct
causal relationshlp with a major project.
Tertiary development is induced development which has a direct
causal relationship with secondary development, and so is only indirectly
related to a major project.
area Whic~h~r~~~~~a6~ ~~~~~t~: ~ow:~~e~~i~~ ~~o~~~tc~~l~~i~~~dn~~w~~~ ~~nd
the major project. In the case of essentially flat terrain (e.g., river
deltas), this area is restricted to the locus of points no greater than
1,000 feet from any point along the interceptor line of the major project.
The legal service area of a wastewater major project is defined as
the drainage basin of the major project plus any additional areas connected
to the collection network by pumping stations and force mains.
The area of anal~sis is defined as the legal service area of a
wastewater major project ln the base year. It must be a minimum size of
5,000 acres (or approximately 8 square miles), and contain significant
amounts of vacant developable land, some of which must be more than 5,000
feet from the nearest interceptor line in the base year.
"201" refers to Section 201 of the Federal Water Pollution
Act Amendments of 1972 (PL 92-500). Section 201 calls for detailed
for the wastewater treatment facilities needed to achieve the goals
Act.
Control
planning
of the
"208" refers to Section 208 of the Federal Water Poll ution Control
Act Amendments of 1972 (PL 92-500). Section 208 provides for the designa-
tion of state and areawide agencies for the purpose of developing effective
water quality management plans for areas that, because of "urban-industrial
concentrations" or other factors, have "substantial water quality control
problems. II The approach is aimed at integrating controls over municipal and
industrial wastewater, storm sewer runoff, nonpoint source pollutants and
land use.
1-18
-------
II.
PHASE I - INITIAL MODEL SPECIFICATION
The first phase of the GEMLUP-II study consisted of a series of initial
analyses to define basic concepts, specify the initial causal model, and identify
data requirements. Thus, this phase served as a general guide and basic frame-
work for the entire study.
A.
PERFORM LITERATURE SEARCHES
As the first step in Phase I of this study, literature searches were
performed to identify all past research material on results and technical ap-
proaches to:
. Causal analyses and predictive modeling of induced development,
in general, and of wastewater facilities specifically
. Simplified VMT predictive modeling procedures
. Land use identification and categorization from aerial photography,
and the accuracy and precision of such techniques
In addition, searches were made to identify potential case studies from techni-
cal reports and wastewater trade journal publications. This material was needed
at different points throughout the study.
The key to a successful literature search is the "search strategy"
employed. This term refers to the application of an accumulated knowledge of:
(1) what data sources to use, such as published directories, bibliographies,
indexes, computer data bases, and (2) how to use them, including the manipula-
tion of Boolean logic in developing complex keyword descriptors for search pur-
poses. Specific, effective search.strategies were developed for the above pur-
poses with the assistance of information specialists in Abcor's Technical Infor-
mation Center (TIC), and these were applied both manually and using computerized
retrieval systems. The data bases searched on are the following:
. Air Pollution Abstracts (APTIC)
. Engineering Index (COMPENDEX)
. Envi ronment Abstracts (ENVIROLINE)
2-1
-------
Geo Abstracts F: Urban and Regional Planning
Journal Indexes:
. ASCE Journal of Surveying and Mapping
ASCE Journal of Urban Planning and Development
Environment and Behavior
Environment and Planning
Transportation Engineering
Monthly Catalog of U.S. Government Publications
. National Technical Information Service (NTIS)
. Pollution Abstracts (POLLUTION)
. Science Citation Index (SCISEARCH)
. Smithsonian Science Information Exchange (SSIE)
. Urban Affairs Abstracts
. Water Resources Abstracts (WRSIC)
(MOCAT)
All relevant citations obtained are listed in Appendix D by search
objectives, along with references obtained from past work in these technical
areas. Thus, Appendix D serves as a comprehensive bibliography for this study.
B. AERIAL PHOTOGRAPHIC INFORMATION
Since aerial photographs were the principal source for land use data
in this study, a task was performed to determine the. typical errors associated
with land use data estimated in this manner. Appendix C provides a complete
summary of this task.
c.
DEFINE MODELING APPROACH
1.
Theory of Induced Development*
In order to develop a causal model of the induced development as-
sociated with a wastewater facility, it was necessary to first hypothesize a
theory of induced development. Of the many types of scientific theory dis-
cussed in the literature [22,23,81-83], the most rigorous type of theory that
we could ascribe to is "factor theory". Factor theory involves a selective,
explicit enumeration of all factors thought to influence a given phenomenon
(in this case, land use), is characterized by narrow and non-overlapping gener-
alizations, and utilizes empirically defined variables to represent the factors
*
This section and the next are based in part on material written by Thomas
McCurdy, Project Officer, in February 1976.
2-2
-------
involved. While almost every effort at causal explanation involves factor
theory, it is limited theory because it does not readily suggest other general-
izations due to its relative narrow focus. Consequently, the factor theory was
operationalized in the form of a model.
The preliminary hypothesis of induced land use development was an
elaboration of the following theory:*
. Construction of a regional wastewater treatment facility enables
interceptor sewer lines to be extended into previously unsewered
areas. The availability of public-financed sewer connections on
vacant land is one factor that encourages large industrial and/or
residential developments; industrial complexes generate jobs
which result in the construction of additional nearby residential
development, and all of these induce retail establishments to
locate near them and generate demand for community facilities.
Increased development leads to the construction of streets and
highways to improve accessibility to the area. Better access
fosters continued urban development, particularly highway-
oriented industrial, office, and commercial land uses. Addi-
tional sources of employment spur on another round of residential
development, and so forth.
2.
Approach to Theory Testing
Selection of an approach for testing the theory of induced develop-
ment was guided by the study objectives and constrained by resource limitations.
Since the theory was only a preliminary hypothesis of the true causal relation-
ships, a deterministic approach** was inappropriate due to its dependence on
the specification, a priori, of the correct model structure. That is, use of
a deterministic model would not allow one to test or refine the theory of in-
duced development. Also, since it was desired to have a model which would rep-
resent the average cause and effect response between variables, it was necessary
* The theory is not new. Most of the urban development models referenced
later in the text are based upon the same general relationships posited in
our theory; they are usually less explicit than the GEMLUP model, however.
**For example, see Lowry [84], Forrester [85], Hill [73], Seidman [86], and
Center for Real Estate and Urban Economics [87]. In addition to these
general models there are many single sector models that have been developed
since the early 1960's. See references [44] and [88] for a review of urban
development models.
2-3
-------
that the approach chosen be capable of dealing with the random errors present
in case study data. A deterministic model does not recognize such errors;
therefore, it was decided to use a statistical approach.
Limitations of time and resources made a dynamic modeling approach
infeasible, because of the effort involved in obtaining sufficient longitudinal
data to incorporate time into the system. A difference approach would involve
solving a system of partial differential equations for the change in develop-
ment patterns over time. If used, however, the solution would be deterministic
and therefore inappropriate, as previously discussed. Since the study objective
of quantifying causal relationships required the use of a technique which could
represent the direction of causal action between an ordered set of model vari-
ables, a static approach using path analytic techniques (based on regression
analysis) was selected to test the theory of induced development. This approach
involved formalizing the theory in a causal path diagram and then testing it
using cross-sectional data and the statistical techniques of path analysis.
Mathematically, this is equivalent to specifying and solving a system of simul-
taneous, linear equations in which some of the sources of variations (i.e.,
unobserved variables) may be unspecified and where the number of observations
exceeds the number of variables. Three problems associated with unobserved
variables can arise with this type of approach. First, there may be important
causal variables which are left out of the model. Second, the variables
chosen may not be true measures of the effects being researched. Third, there
will always be random effects (residuals) in any system which cannot be
modeled. In response to the first problem, we specified as complete an initial
model hypothesis as possible to avoid leaving out any significant variables.
As a result, we assumed all other unobserved effects to be random in nature.
We did not address the second problem because the sample size placed severe
constraints on the depth of our analysis into the model structure. The third
problem is dealt with through the assumption that we are dealing with ~andom
, effects (residuals) which are independent with each other, and with the exo-
genous variables in the model. Thus, their covariances are not needed or
included in the model. Path analysis is not restricted by this assumption; it
is adopted only heuristically to keep the model form simple enough ,for solution.
The assumption is justified to the extent that all major system variables have
been specified explicitly'in the model [15].
2-4
-------
The static approach to testing the inherently change-oriented
theory was justified by three factors: (1) the theoretical assumption that
induced development follows a single basic causal structu~ for all cases or
observations, (2) the use of a cross-sectional data for variables observed in
a static state and the assumption that input variables are initialized at time
zero and held constant long enough so that all the causal consequences in the
system are realized, and (3) the use of time-lagged exogenous variables in the
system. The conceptual usefulness of these factors in testing causal theory
is well described in Heise [15] and Blalock [89].
3.
Model Form and Characteristics
A thorough review of the technical literature did not provide
enough information on induced development associated with wastewater facilities
to be able to pre-define the form of relationships in the system. The form,
therefore, was assumed to be linear. This is not a bad a priori approximation
in most social science applications, and it allows the use of well-developed
statistical techniques to test for causality. There is accumulating evidence
that many social systems can be approximated by a linear function as long as
operating conditions remain fairly stable [15]. Even complex nonlinear rela-
tionships can often be approximated by a constant relation in discrete sub-
regions of the relationship [90]. Also, if a relationship is thought to be
nonlinear on theoretical grounds (e.g., multiplicative, exponential), the data
can be transformed prior to entering it in the linear analysis.*
A time period of 10 years, specifically 1960 to 1970, was used in
the land use model. The selection of this time period is based on two factors.
First, from a preliminary review of the technical literature related to this
study, it was judged that a time period of 5-10 years was reasonable for al-
lowing the secondary and tertiary impacts associated witr a wastewater major
project to be realized. Second, the principal source for the comprehensive
demographic and economic data required in the model was the U..S. Census, taken
*
As part of the testing of the path analytic model, nonlinear forms were
considered.
2-5
-------
at the beginning of each decade. Although some states undertake a census at
mid-decade, it was determined that these efforts are limited in scope and
sporadic, and do not provide either the detail or quality of data available
from the U.S. Bureau of Census [91J.
The possibility of using the GEMLUP-I model [19J as a portion of
the GEMLUP-II model was examined. It was found, however, that such a Hpiggy-
backed" model would be unworkable since the GEMLUP-I models were developed for
a fixed area of analysis while the GEMLUP-II model will use a variable area of
analysis.
In summary, the GEMLUP causal approach consisted of testing a
cross-sectional, path analytic (simultaneous equations) model that explains
total land use development in the vicinity of a wastewater major project 10
'years after its construction.
4.
Technique Selection for Path Analysis
Two statistical techniques were used to verify the hypothesized
path analytic model of induced land use and to determine model parameters:
two-stage least squares and ordinary least squares multiple regression. The
first technique was used to solve for path coefficients in the model equations
affected by simultaneity, i.e., those containing feedback causality between
variables. Ordinary least squares regression was used to solve for path coef-
ficients in recursive (not interconnected*) equations. There are many other
multivariate linear statistical techniques available for data analysis purposes.
Presented below are brief discussions of why these other techniques were not
used in the current study.
a.
Factor Analysis and Canonical Correlation
Unlike forms of regression analysis, these techniques do not
make a distinction between endogenous and exogenous variables [92J. Thus, the
direction of causal action, an important element of this study, cannot be
specified. Also, if a simultaneous treatment of all causal factors is to be
*
And assuming independent errors.
2-6
-------
considered with these techniques, then the individual relationships between
specific endogenous and sets of exogenous variables, which are desired, cannot
be obtained. Finally, the factors (e:g., principal components) and canonical
variates output by these procedures can defy description in other than mathematical
terms, i.e., they cannot be related to measurable, physical quantities. Thus,
a theory of causality cannot be verified with these techniques.
b. Three-Stage Least Squares and Maximum Likelihood
Estimation Techniques
Three-stage least squares is an alternative path analytic
technique which involves essentially performing a two-stage least square
analysis, obtaining the residual errors, and then simuZtaneousZy re-estimating
the model coefficients in all model equations. Three-stage least squares is
called a fuZZ-info~ation method because of the last stage in which all the
structural coefficients are determined at one time. The advantage of this
approach is that it produces more efficient (i.e., precise) model coefficients,
although not necessarily more accurate ones. It is a tool for fine tuning the
coefficients in a model. The principal disadvantage of any full-information
method is that it incorporates model specification errors of anyone equation
into the calculations for all the model equations [15]. Thus, if but one of
the equations is mis-specified, the errors involved may cascade throughout the
entire model. For this reason, these methods are most appropriately used when
there is a high level of confidence in the theoretical specification of causal
linkages. A second disadvantage to full-information methods is the reduction
in the available degrees of freedom for statistical fit due to the simultaneous
consideration of all model variables in one estimation step. Due to the dis-
tinct possibility for error propagation in using three-stage least squares,
and the problems associated with small sample size, this technique was judged
inappropriate in the current study.
The maximum likelihood technique is an alternative method for
estimating the coefficients of the path analytic model, which differs from two-
stage least squares in that it offers the ability to specify the form of the
relationships between unobserved variables (not in the model) and those included
2-7
-------
in the model structure. The price for being able to make these new assump-
tions is increased model complexity and statistical constraints. The latter
is particularly troublesome when the sample size is small, and in some instances
it may be impossible to use such techniques without first discarding several
model variables to free up more degrees of freedom in the statistical analysis.
This technique represents a very general approach that allows one to specify a
more complex model structure, perhaps involving correlations between error terms
or causal relationships between observed and unobservable variables. This
technique is most appropriate when the causes and effects being researched are
not directly measurable and it is necessary to represent them by related
indicator variables which can be observed. In the current study, indicator
variables are not a problem since the principal model outputs, the endogenous
land use variables, are specifically what we wish to study (and ultimately
predict). Also, there is no basis for assuming a more complex model structure
I .
in the current study that would justify the use of these techniques. And if
there were, it is doubtful the limited sample size could accommodate such
complexity. In addition, maximum likelihood techniques require the assumption
of multivariate normality (which two-stage least squares does not). Finally,
these techniques are full-information methodi, and so suffer from the ~ame
disadvantages as three-stage least squares. For the above reasons, these
techniques were judged inappropriate for the current study.
c.
Robust Regress ion
Robust regression is a class of experimental statistical
techniques that differ from ordinary least squares in certain aspects [93J.
For example, robust regression may involve regression on the median rather
than the mean of the data, or it may involve a weighted least squares where
observations far from the mean are discounted. Other approaches are to elimi-
nate so-called outlyers completely in the analysis or substitute in a fixed
value. The principal advantage of robust regression is in the detection and
modification of outlying data points. Implicit in its use is the assumption
that such outlyers are undesirable and are introducing errors in the estimation
of structural coefficients. In the current study, there is no basis for making
2-8
-------
such an assumption, and thus no basis for weighting observati.ons. Due to the
small sample size employed, an outlying data point may simply be the only
value available in a particular portion of a variable's range. For this reason,
robust regression was not used in this study. .
d.
Ridge Regression
This technique is a modification of ordinary least squares in
which a dampening factor "k" is included in the process of estimating model
parameters [94-96]. The advantage of ridge regression is that it produces more
,stable estimates of the regression coefficients than ordinary least squares.
In order to do this, the traditional assumption of unbiased estimates has to
, be discarded. Ideally, the ridge estimates will be closer to the true model
parameters because the average instability error (imprecision) of ordinary
least square estimates is greater than the bias (inaccuracy) of the ridge esti-
mates. Thi s will be the case if the "correct" value of k is chosen. Unfortu-
nately, there is not established technique for choosing the best value of k,
and in general, a trial and error approach is required.
Due to its capabilities, the ridge regression is designed
to handle many of the problems inherent in the GEMLUP path analysis, such as
unstable coefficients, small sample size and multicollinearity. For this reason,
the technique was seriously considered for use in this study. Since it is not
a replacement for two-stage least squares, but only a modification of ordinary
least squares, it would have to :be incorporated into both steps of two-stage
least squares. A standard computer program for ridge regression does not
exist*; therefore, extensive software development would have been required.
Also, since the algebraic solution to two-stage least squares would no longer
hold when the k factor was introduced, that analysis would have to be done as
, two ordinary least square analyses, with an interim manipulation of residuals.
This would greatly increase data analysis costs. Finally, each equation would
have to be solved numerous times, each with a different k value, the results
plotted, and an optimum k chosen from the graphs. Limitations of time and
*
One is currently being developed as part of the SPSS [92] system, however,
there is no projected date of availability.
2-9
-------
resources precluded such a significant amount of additional work, and therefore
ridge regression was not used in this study. This is unfortunate, since the
problems associated with small sample size are the most serious limitations
imposed on the path ana)ysis by its use in the GEMLUPstudy.
e.
Modification of Two-Stage Least Squares for Undersized Samples
One major problem encountered in the first GEMLUP study was
the problem of having more exogenous variables than case study samples. When
this is the case, the sample is said to be undersized. The approach used
previously was to retain for the analysis only those variables which theory
indicated were important and those that correlated highly with the endogenous
variables of the system. Thiel [97J suggests a modification of the two-stage
least square technique that permits the retention of all the exogenous variables
in the analysis. The technique is based on the asymptotic (large sample)
properties of least squares analysis. While the technique does not appear to
have been employed often it does have great promise and would be extremely
useful in this study. However, since a standard computer program for Thiel's
method does not exist, extensive software development would have been required.
Limitations of time and resources unfortunately precluded the associated ad-
di ti ona 1 work.
f.
Other Considerations
In addition to the above techniques, several others were
suggested to deal with problems that sometimes arise in statistical data
analysis. The first of these problems is heteroskedasticity, which involves
unequal variances in the model equation residuals. This problem was addres~ed
in the path analytic work by normalization of all model variables, as described
on pp. '2-14 and 2-15.
2-10
-------
A second problem that was identified is intrinsic non-linearity
in the causal relationships. Many techniques exist for dealing with this
problem, however, they were generally not applicable to the current problem
since we found no basis (through a review of the technical literature) for
assuming non-linearity or a particular non-linear form. Thus, a linear form
for the path analytic model was assumed, as previously discussed. . The validity
of this assumption was checked by considering other, non-linear forms in
testing the model and comparing the results.
The related problems of autocorrelation and serial correlation
in the case study data were mentioned, along with Box-Jenkins techniques. It
was judged that these problems do not exist in this study since the independent
variables employed in the model are defined for a single point in time, the
base year of a major project, and so temporal correlation cannot occur between
variables.
The problem of discontinuous dependent variables was raised
and the techniques of logit analysis and discriminant analysis mentioned
as possible solutions. It was judged that this issue was not a problem in
the current study, since all of the dependent (land use) variables are by
nature continuous in their function.
D.
DEFINE INDUCED DEVELOPMENT
Induced development is defined as urban land use associated with,
caused, stimulated, or allowed by, or located because of the construction and
operation of a wastewater facility. This definition provided a basis for the
selection of endogenous variables for the causal analysis. The approach used
to selecting these variables involved a review of the technical literature to
identify what types of land use are principally induced by a wastewater major
project (i.e., what are the infrastructure relationships of wastewater facili-
ties in a community), and which are measurable in an objective way, for inclusion
in the causal and predictive models. The initial choice of model variables
was also constrained by the fact that model outputs must be compatible with
the land use based emission factors [18] developed in the previous GEMLUP
study, which were used in Phase IV of this study to produce an emissions pro-
jection procedure.
2-11
-------
Several recent studies have examined the secondary impacts of waste-
water facilities, and reached the following similar conclusions. Wastewater
facilities have their greatest growth-inducing impacts on single family housing
[58,60,61,64,66J. They also encourage medium to high density housing [57,58,61,
64,66J by removing the development constraint associated with soil conditions
and septic tank limitations. Other important land uses for which a wastewater
infrastructure acts as a stimulus are industrial (manufacturing), commercial,
educational, transportation and recreational developments [53,61,66J. Due to
the large effect of wastewater major projects on residential and industrial
, development, the situation occurs in which these secondary developments then
induce other tertiary land uses, namely commercial, office-professional ser-
vices, manufacturing, wholesale-warehousing, non-expressway highways, hotel-
motel, cultural-entertainment, religious, educational and recreational de-
velopments [17J. Note that the implied temporal ordering here is not repre-
sented explicitly in the GEMLUP model; rather, the objective is to illustrate
how a wastewater major project can induce growth of many different land use
types. The initial list of endogenous variables, given in Table 2-1, reflects
all of the land uses discussed above, which are thought to be ultimately induced
by the construction and operation of wastewater projects. In addition, it
includes a disaggregation of the residential and commercial land use categories
into more specific sub-categories based on size. These detailed data are re-
quired to accurately estimate associated air pollutant emissions, through
application of the GEMLUP land use based emission factors [18J. Also, this
disaggregation will allow the user of the GEMLUP model to introduce a specific
density distribution for residential and commercial land use that reflects
local and future development trends, rather than the density patterns of the
period 1960-1970. The endogenous variables listed in Table 2-1 will represent,
in the GEMLUP model, predicted land use in the area of analysis 10 years after
the construction of a wastewater major project.
Certain land use categories are not included in the initial list of
endogenous variables. Some of these are independent of wastewater facilities,
such as passive open space, agriculture and mining. Others, though requiring
some form of wastewater services, are usually the result of decisions made by
regional, state or federal governments, and so their construction and location
is generally not determined by the existence of a wastewater major project.
2-12
-------
TABLE 2-1
INITIAL LIST OF ENDOGENOUS VARIABLES
Land Use Type
Units
1.
Res i denti a 1
Owe 11 i ng Units
1,000 ft2*
1,000 ft2*
1,000 ft2*
1 ,000 ft2*
Lane Miles
1,000 ft2*
Acres
1,000 ft2*
a. Single family detached (1 unit/structure). .
b. Single family attached (2 units/structure).
c. Mobile homes.
d. Multifamily low rise (3-4 units/structure).
e. Multifamily high rise(5+ units/structure).
2.
Commerci a 1
a. <50,000 sq. ft. of gross leasable area (GLA)
b. 50,000-100,000 sq. ft. of GLA
c. >100,000 sq. ft. of GLA
3.
Office
4.
Manufacturi ng
5.
Wholesale-Warehousing
6.
Non-expressway highway
7.
Educati anal
8.
Active, Outdoor Recreation
9.
Other Urban Land Uses (hospitals, hotels
and motels, cultural facilities, churches)
* Of total floor area.
2-13
-------
Examples of the latter are correctional institutions, military installations,
universities, airports and sport stadiums or arenas. Package treatment plants
are often used for these facilities, allowing practically any location where
vacant land is available.
Certain land uses are combined in the initial list of endogenous vari-
ables. The category of "other urban land uses" includes variables which are
only indirectly related to wastewater major projects, and which by themselves,
involve very small amounts of land. Thus, on an individual basis, these cate-
gories are very difficult to predict. So, they are aggregated.
Another issue associated with definition of induced development was
determining whether to model total land use, or change in land use, at the
end of the 10-year period of analysis. It was decided to model total land
use for three reasons. First, data were not available to efficiently deter-
mine changes in land use. 1970 land use was extracted from aerial photographs,
USGS quadrangle topographical and land use/land cover maps, and local land
use maps for that time period. All of these sources were generally less
readily available for 1960. Second, the air quality emissions projection
procedure required predictions of total land use as input variables. If only
change in land use was predicted, a step would have to be included in the
worksheet procedure to add in base year land use quantities. Since emission
projections are the principal output of the GEMLUP tool, it is important to
keep the computational procedure as simple as possible. finally, if 'change
in land use is desired, it can be obtained by subtracting base year amounts
from model predictions. This information is less likely to be desired than
estimates of total air pollutant emissions.
The endogenous variables listed in Table 2-1 were normalized prior
to their use in the path analytic model. This requirement was due to the
selection of a variable area of analysis which exhibited wide variability
between case study major projects.* For example, consider the form of the
GEMLUP model for a recursive equation:
Y = B + B1Xl + ..... + B X + E
o P P
(1)
*
Exogenous model variables were specified on a density or percent basis, i.e.,
normalized as well, due to the variability in area of analysis. See page 2-31.
2-14
-------
--.-. - '.'"
-_.-- .
- ~- -. -...
..-. "'.' ~ "",.-..,. .
. -...-.-. n___.--.._..._._..._..~ 0
where Y is the endogenous variable; Xl"""'XP are the exogenous variables;
So' Sl"""'Sp are the model coefficients; and E is the error term. In
estimating the 8's, it can be shown that if the standard deviation of E is
constant for all case study samples, then ordinary least squares will pro-
duce good estimates of the B1s. However, if this is not the case, then ad-
justments are required before employing the least square technique. One
such example of when an adjustment is needed arises in this study due to
the wide variability in areas of analysis in which the endogenous variable Y
is measured. As the total area increases, so will the magnitude of Y, in
general, and thus so will the standard deviatio~ pf E. That is, la~ge proj-
ects are apt to be more variable (i .e., larger standard deviations) than smaner
ones regardless of the values of the Xis. If this is the case, than a more
appropriate model than (1) is
-M = 80 + 81X, +... TPpXp + 11
(2 )
w~ere f(A) rep_r~sents a~_appr?p.T!a~~_.tr~nsformati~n of the ar~a o~ analysis A
so as to produce an error term 11 with constant standard deviation. The func-
tion f(A) is determined as follows. For a fixed set of Xis, if the standard
deviation of Y (cry) is found to be proportional to A, then f(A) = A; if cry
is found to be proportional to VA, then f(A) =VA, and so forth [12].
Since a very large number of case study samples would be required to analyze
the variability of (cry) with A, for a fixed set of Xis, the technical litera-
ture was examined. No references could be found related to this question,
however. Thus, the intuitive judgment was made that the standard deviation
of Y is proportional to A directly, and so f(A) = A was used.
....~ . .- --.-
A separate problem associated with defining the induced development
is the approach used to investigate the phenomenon of displaced growth. A
wastewater major project will induce changes in land uses within its region
by two mechanisms. First, the'project will affect the growth of the region
as a whole, and second, the project will produce changes in the pattern of
2-15
-------
-'---~----"" --.---. -'- _.n_. --.--.-.-.. ~_. --. .-. -- -
~-----_. "----'-"._'--~--'----- --~_...._--
.----- -----------.-- - ---
growth within the region by a substitution or displacement effect. This
latter effect is characterized by growth that is not "new" or directly in-
duced by the project itself, but rather is a relocation within the region
to the area of analysis as a result of the project. At a particular loca-
tion in the region, the substitution effect of the project may be either
positive or negative - the project may act to attract growth to the immediate
area that would not have occurred there without the project, or lead to a
relocation of the activity. to another area in the region. Likewise, the
major land use project could make the immediate area less attractive to
certain land uses and, therefore, make other locations comparatively more
attractive. . To summarize, a major wastewater project induces changes (simul-
taneously) in both the total land use and its location within a region.
One way to categorize types of induced development is the following:
Development that is new to a region, attracted by the major
project from other regions
Development that is indigenous to a region, but located in
the specific area of analysis due to the major project
Development that is "pent up" in the area of analysis and
allowed to occur by the major project.
Another approach is that any development can be viewed as using up or displacing
a development potential that might have been used elsewhere. Thus, develop-
ment is never lIindigenous" to, or "pent Upll in an area, but rather develop-
ment potential is directed to, or located in, areas most favorable to it. A
more fundamental viewpoint reflects the fact that development is rarely
caused by one key factor (wastewater facilities), but rather a combination
of variables. Since it is virtually impossible to determine the exact source
of development potential which manifests itself in the case study project
areas. it was decided that the key to understanding and defining the induced
2-16
-------
.--- ...
growth of a major wastewater project is a methodology to measure. and analyze
displaced growth effects. Or stated differently, the need to address the
issue of displaced growth is due to the fact that, even though the model of
induced development produced by this study is project-specific, the ef-
fects of a wastewater collection system are often more regional in nature,
attracting (i.e., displacing) growth from land areas surrounding the service
area in which public sewer connections are most easily available. An illus-
tration ~f this is given in Figure 2-1, where a hypothetical drainage basin is
shown in which an urbanized core has been sewered for many years, although
only primary treatment was applied. The construction of a new regional
secondary treatment plant in a nearby large city allows an interceptor line
to be built into the area to pick up its entire waste load. Assuming excess
capacity has been built into this line, a new (publicly financed) collection
system rapidly spreads through a section of the vacant land of the community,
providing obvious development potential. This example illustrates that it
is necessary to predict not only the total amount of future land use in a
community, but where development will occur, e.g., in sewered or:unsewere~
areas. New development will obviously not occur exclusively in vacant,
sewered areas since other locational factors (e.g., transportation facili-
ties, availability of markets, zoning) can be equally or more important.
However the existence of sewer connections can be a deciding influence and,
therefore, a cause of displaced growth.
A technique for estimating displaced growth effects will be outlined
in the second volume technical report.
E.
DEFINE MAJOR PROJECT
For the purpose of model specification and case study selection, an
explicit definition of what constitutes a major wastewater project was de-
termined. ~his definition was bas.ed on the premise that we are only inter-
ested in wastewater projects which have the potential for inducing and/or
redirecting land development, i.e., those that have the effect of increasing
or expanding the number of sewage connections in a service area. Examples
of projects that do not have the potential for significant induced growth
2-17
-------
SEWER
TRUNK LINES
NEWLY SEWERED
VACANT LAND
TO REGIONAL
FACI LITY
URBANIZED SEWERED AREA - BASE YEAR
FIGURE .2-1
EXAMPLE OF THE DISPLACED GROWTH PHENOMENON WITHIN A HYPOTHETICAL
DRAI NAGE BAS IN
2-18
-------
------. -~.~.. ...-.--'-'---'."--- . ---_._-.~ --. '--~--.. -~-~- ..-- - ----- .--'- ---'-.- -.
effects are: (1) those to simply upgrade treatment technology (e.g., from
primary to secondary, or from direct discharge of raw sewage to any level
of treatment), or (2) those that increase plant capacity only to more nearly
match collection network capacity, e.g., to handle periods of high flow dur-
ing wet weather from combined storm-sanitary sewers (preventing overflow
conditions). The principal criteria which describe the definition of major
projects can be categorized as type, size, and timing requirements and are
discussed separately below.
Wastewater projects can consist of either new construction of, or ex-
tensions to: (1) treatment plants, (2) interceptor lines and collection
networks, (3) outfall sewers (discharge lines from the treatment plant, if
it exists, to the receiving body of water), or (4) waste stabilization ponds
and appurtenances. We are not interested in the last two project types,
since these involve the treatment and discharge of wastes, and do not affect
the availability of sewer connections. Project type (1), expansion of treat-
ment plant capacity, is also, by itself, not an element that will likely in-
crease land development. When collection network capacity exceeds treatment
capacity, rarely are additional sewer connections prohibited [60]. And if
restrictions are enacted, such moratoria are generally difficult to maintain,
even in cases of treatment plant overloading. Courts are unlikely to allow
such moratoria to stand more than a year or two since Sewers are considered
a public asset [64J. Thus, collection network capacity is a greater growth
constraint than treatment plant capacity. Consequently, an increase in the
tteatment capacity of a wastewater project generally does not increase its
growth-inducing effects, unless the collection network is concurrently or
subsequently expanded as well. Project type (2), expansion of the collection
network, is the wastewater project type we are primarily interested in for
the purposes of defining a major project. Past research has shown that the
major secondary impacts associated with wastewater facilities "result from
the placement, sizing and staging of interceptor sewers, and the provision
of reserve capacity in these sewers" [98J. Usually, a project involves the
construction or extension of various combinations of the four project types.
Thus, although our principal interest is in the expansion of wastewater col-
lection systems, changes in treatment and disposal of wastes may also be in-
volved.
2-19
-------
A minimum size requirement was necessary to ensure a wastewater project
is "major", i.e., large enough to have had the potential, upon construction,
for inducing a "significant quantity" of land development in its legal service
area. Based upon discussions with EPA personnel from the Municipal
Construction Division and the Office of Transportation and Land Use Policy,
the conclusion was reached that wastewater projects with a collection network
capacity below 1 MGD (million gallons per day) are too small to induce a
significant quantity of land development. Thus, this figure was chosen as
one minimum size requirement. Since sewage flow typicab1y varies between
90 and 200 gallons per capita per day, the upper limit is generally used in
wastewater system design [99,100]. Thus, a capacity of 1 MGD will fulfill
the wastewater needs of approximately 5,000 persons. A second size require-
ment was that the wastewater project had to cost at least $200,000. Inter-
ceptor sewer projects for the period 1960-1970 cost typically in the range
of $100,000 to $10,000,000 to construct [101]. The objective of the minimum
cost requirement was again to eliminate small projects which could not include
a detectable amount of land development.
Timing requirements were set to ensure that project impacts had been
realized by 1970, the year in which final land use in the area of influence
was measured. These requirements were specified as a 2-year time period
centered around the base year of 1960 during which construction on the major
project had to be essentially completed, i.e., from 1958 to 1962. Although
choosing a wider time period would increase the availability of case study
major projects, it would also increase the unexplainable variance in the
model. This criteria was left open to revision, subject to the availability
of sufficient case study projects which could satisfy it. Since construction
completion dates may not always be available from information sources used
to select case study major projects, an alternative timing requirement was
set. The project had to receive matching federal grants (if allocated) be-
tween 1956 to 1960. The offset of 2 years from the construction period require-
ment of 1958 to 1962 is to account for two effects. First, there ;s usually a
few months delay between the time a grant is made and construction begins
(although this lag can vary from 1 month to 2 years). Second, the average
construction period for a project costing over $200,000 is 1-2 years, see
Table 2-2 (although it can range anywhere from 10 months to 3 1/2 years).
2-20
-------
TABLE 2-2
POINT ESTIMATES OF THE CONSTRUCTION PERIOD
OF INTERCEPTOR SEWER PROJECTS
Total Project Cost Months to Complete Construction
1967-69 $ Lower limit Average Upper limit
$ 25,000 4.7 7. 1 10.8
50,000 6. 1 9.3 14. 1
109,000 7.7 11.7 17.9
250,000 10. 1 15.4 23.5
500,000 12.1 18.5 28.1
1,000,000 14.3 21.8 33. 1
2,000,000 16.5 25.2 38.3
3,000,000 ,17.9 27.3 41.5
Source: [101 ]
2-21
-------
.. ~_.,_"__.P .-.' .-...--
~ ........-------..
..--'..-.---'.W' .
Another timing problem which was encountered is related to the fact
that wastewater construction projects are often constructed in several
phases, several years apart. This project phasing is quite common and is
caused by: (1) the large costs of such projects (despite federal matching
funds), (2) the reluctance of communities to finance a large debt all at
once, and (3) projected demand over time. For example, the City of Denton,
Texas (1970 population of 40,000) completed four different wastewater pro-
jects within the period of 1960-1970 [52J. . In 1960, construction began on
a $1.2 million project to extend interceptor sewer lines and construct a
new city treatment plant. In 1963, the collection network was again expanded
at a cost of $230,000. In 1966, a $260,000 project constructed additional
sewer outfa11s and in 1968 a $1.3 million addition was made to the sewage
treatment plant. For our purposes, only the first two projects are of
interest since they alone affected the availability of sewer connections.
To eliminate the 1960 project in Denton from consideration as a case study
because of phasing would be a mistake, however, since initiation of the
second project could have been due to extensive development pressures in
the conmunity (population increased 49% from 1960 to 1970). Thus,
conmunities in which most of the development potential of a wastewater
major project is rapidly used may experience project phasing more often than
other conmunities. To exclude such phasing might bias the resultant model.
Therefore, we extended the definition of wastewater major project to allow
phased projects on a limited basis, by considering projects in which a second
phase of collection network construction was completed by 1965. In such
...----- ..."..-.'W .-.-..-. .-.
cases, the system ca~acity was defined as that present after the second
phase of construction was completed. This approach treats the construction
as really one planned project, spread over time. The definition also allows
5 years for the induced growth effects associated with the second phase to
be rea 1 i zed.
2-22
-------
In summary, the definition determined was:
A wastewater major project is considered to involve principally
the construction or extension of interceptor or collector sewer
lines during the period 1958 to 1962 in a community in the United
States. If construction date information is not available, then
a grant funding date in the period 1956-1960 is acceptable. The
project had to affect an increase in absolute system collection
capacity of 1 MGD or more, and had to cost a minimum of $200,000
to construct. Phased projects are considered if the first phase
of construction on the collection network was complete within
1958-1962 and the last phase of construction was complete by
1 96 5 .
F.
DEFINE AREA OF ANALYSIS
Careful study was made of the size and characteristics desired for
the area of analysis to be used in this study to measure the induced land
use impacts of major wastewater projects. In the GEMLUP-I project [17J, a
fixed area of 10,000 acres was used because it "captured" most secondary
development (75-80%) associated with the project types investigated. In the
current study, however, the "size" of a wastewater major project relates to
its legal service area, which can vary from few hundred to several hundred
thousand acres (see Table 2-3). Thus, it was necessary to use a variable
area of analysis to accommodate such wide variability in project size.
For a given major project, selecting the size of the area of
analysis required cognizance of two potential problems. First, if the area
selected was too large, the influence of the major project on the end state
land use would be small compared to the influence of regional growth factors.
Thus, although the model might make accurate predictions, it would not be
simulating principally the induced growth effects of the major project. On
the other hand, if the area selected was too small, the model might not repre-
sent all of the associated induced development and again it would be under-
estimating the major project's influence. The approach employed in defining
the area of analysis involved determining first the area in which most secon-
dary land use occurs, and second, the area in which most tertiary land use,
induced by secondary development, occurs and whether this extends beyond the
boundaries of the secondary development. Note that the partitioning of induced
2-23
-------
TABLE 2-3
CONSTRUCTION GRANT INFORMATION ON A SAMPLE OF CURRENT WASTEWATER
PROJECTS IN FEDERAL REGIONS II, IV AND VI
Servi ce Area
(acres)
Percent Initial Design
Developable Population Population
. Number of Observations 49 50 51 56
Mean .~; 20, 118 60 26,894 65,459
Standard Deviation 33,267 22 40,764 91,470
Range 200 to 0 to 96 0 to 1,470 to
149,000 1 73 , 500 402,000
Distribution 18% <1,500
18% 1,500-3,000
23% 3,000-10,000
25% 10,000-50,000
8% 50,000-100,000
10% >100,000
Source:
[1 02J .
2-24
-------
development into secondary and tertiary types is done here only for the purpose
of determining the areal extent of growth impacts from a wastewater major
project. The implied temporal ordering is not represented explicitly in the
GEMLUP model.
All secondary land use, induced directly by the wastewater major
project, will occur within what is called the "legal service area" of that
project. Since it is, in general, not possible to deny a developer access
to a publicly-financed sewer with excess capacity [64J, the "legal service
area" is defined as the watershed boundaries of the area that drains, by
gravity, to any point along the interceptor line, plus any areas outside
the watershed connected to the collection network by pumping stations and
force mains. Service of this entire area is not implausible since widely
used sewer design standards require that interceptor pipes be sized to accom-
modate the full development of the drainage basin served [13J. In the case
of essentially flat terrain (e.g., river deltas), the vertical inclination
of pipes provides the necessary gravity flow for the system. Due to the costs
of excavating, in these cases, the legal service area is usually restricted
to an area no greater than 1,000 feet from any point along the interceptor
line. The legal service area is, in general, not easy to map since watershed
boundaries pursue a course independent of municipal boundaries. A single
watershed may contain many towns, and a single town may overlay portions of
several watersheds (e.g., the City of Lexington, KY has 7 different water-
(
sheds [104J). Also, any watershed can be partitioned into sub-drainage basfns.
Where an interceptor sewer only penetrates one end of a watershed, its legal
service area may contain only a few of the sub-drainage areas in the watershed.
The legal service area defines the area in which secondary develop-
ment is possible. There is a smaller area, contained within the legal ser-
vice area, however, in which such development is most probable. An example of
this is illustrated in Figure 2-2 where the legal service case is large and
only the "ske1 eton" of a co1l ection network is constructed. Speci fically,
the extent of the publicly constructed wastewater collection network is a
single interceptor line placed parallel to the principal tributary (river)
down the middle of the watershed, with only a few cross-connecting collector
2-25
I,
-------
.,-
r~OST PROBABLE DEVELOP~1ENT AREA
NETWORK COi~STRUCTED IlII." I
t
1
I
MILE
FIGURE 2~2
HYPOTHETICAL LEGAL SERVICE AREA OF A WASTEWATER MAJOR PROJECT
ILLUSTRATING THE LOCATIONAL EFFECTS OF THE
PROJECT'S COLLECTION NETWORK
2-26
-------
lines. In this case, the area closest to the river (including its flood plain)
has the maximum potential for development due to the easy access to the col-
lector network. Conversely, the area farthest from the interceptor line, near
the watershed boundary, has the lowest development potential due to the cost
a developer must incur to construct a wastewater line down to the existing
network. This example illustrates the idea that the quantity of induced de-
velopment is some declining function of the distance from the wastewater
co"iection network. Promise and Leiserson [53] have studied this effect
through a statistical analysis of housing data for the period 1963 to 1970
in suburban Washington, D.C. (Montgomery County, MD). The analysis considered
the relationship between residential development and the distances to the
nearest interceptor line and showed that the majority of development occurred
within 5,000 feet (or 1 mile) of the line, with none beyond 10,000 feet (or
2 miles) distance. Work related to other land use types (e.g., industrial)
was not found in the literature, and so these results were assumed to be
representati ve of all secondary development.
The task of determining the size of the area in which tertiary land
use occurs can be compared to the "area of influence" investigation that was
performed in the GEMLUP-I study [17]. The factor inducing development is
presumably newly constructed residential housing or industrial plants. In
the GEMLUP-I study, it was assumed that a radius of 2 miles from the center
of a Residential or Industrial major project encompassed most induced develop-
ment. The issue here, however, is more complex. Unlike the previous study,
it is generally not possible to define a center for the secondary develop-
ment (the cause of tertiary development) since this can occur anywhere in a
project's legal service area, but will most likely occur within 1 mile of
an interceptor. This effect spreads out tertiary development over a wider
area, at most, encompassed by the locus of points within 3 miles of an in-
terceptor sewer line. However, the presence of more readily available
sewer connections in the secondary development area simultaneously reduces
the aerial extent of tertiary deve~opment. This, in conjunction with the
fact that most tertiary development locates near its secondary development,
led to the judgment that the composite of the secondary and tertiary land use
2-27
-------
areas generally defines an "area of analysis" no larger than the wastewater
legal service area. Also, an expansion of the analysis beyond the legal ser-
vice area to directly analyze the contribution of secondary growth to all
tertiary uses would require a thorough analysis of all growth patterns in the
region. For such an effort, a full regional land use model would have been
required, along with extensive data for perhaps an hour's travel around the
service area [105J. This was judged to be beyond both the scope of the study
and the state-of-the-art (particularly in joint travel-location forecasting).
The wide variance in area of analysis size for major projects, shown
in Table 2-3, indicates that small and large areas are likely to be mixed in
the selection of case study projects. This causes a potential problem since
small areas are less likely to include as much tertiary development as large
areas. The interactions among induced land use types come in two forms, of
which only one is significant for this project. On a large scale, inter-
actions are determined by joint locational economies. For example, commer-
cial uses need markets -- either households or businesses, and households
need both housing and employment opportunities. These are the important
interactions for this study. On a more detailed, local level, however, dis-
. .
ruptions in these interactions can occur. Parcels of land often develop as
all residential or industrial because of zoning and localized scale econo-
mies. Land is also commonly sold in large units, as agricultural activity
gives way to urban development [106]. This leads to a market with local
heterogenities and, when combined with on-site environmental impacts (par-
ticularly from industry and highways), to diseconomies of joint location on
a local scale [61]. These two forces -- small scale homogeneity and large
scale diversity -- confound each other in the analysis of small sites, and
can lead to indeterminate effects. Since the GEMLUP model assumes a single
causal hypothesis, a lower limit on area of analysis size of 5,000 acres wa~
set to avoid such inconsistencies. This lower limit was based on the pro-
fessional judgment of the staff after review of several wastewater major
projects [102].
Another criteria related to the area of analysis involves the poten-
. tial for induced development from a major project. Areas which are extensively
2-28
-------
urbanized in the base year wi.ll contain practically no vacant, developable
land. In this case the relati.ve growth effect of a major project will be
imperceptible. Thus, the decision was made to exclude such cases by requir-
ing the area of analysis to include significant amounts of vacant developable
land. Also, some of this land must serve as a control area for measuring
the displaced growth effects of the major project and so must be located more
than 5,000 feet from the nearest interceptor sewer line.
Therefore, the following definition was used in this study:
The area of analysis is defined as the legal service area
of a wastewater major project in the base year. It must be a
minimum size of 5,000 acres (or approximately 8 square miles),
and contain significant amounts of vacant developable land,
some of which must be more than 5,000 feet from the nearest
interceptor line in the base year.
G.
INITIAL CHOICE OF VARIABLES
The objectives of this task were to: (1) define the initial list of
endogenous land use variables in terms of standard Land Use Codes (LUC) [109],
and (2) choose an initial set of exogenous model variables. Defining the
endogenous variables in terms of an established classification system re-
quired a systematic review of each LUC category and an assessment of its suita- .
bi1ity as a part of each GEMLUP land use variable. The results of this analysis
are given in Table 2-4. Note that the "Office" variable name has been appended.
to reflect the existence of "Professional Services" in this variable, and that
"""" . .
hospi'ta1 land use has been moved to the "Office-Professional Services" variable.
Since hospital land use was initially placed in a miscellaneous category, it
was judged more appropriate to include it in a functionally related endogenous
variable. Finally, as discussed in the section "Define Induced Deve10pmentll,
the endogenous variables exclude the following land uses:
. LUC 4: Transportation, communication, and utilities
(except LUC 45: Highway and street right-of-way)
. LUC 67: Governmental Services
. LUC 682 and 683: Universities and Trade schools
. LUC 72: Public assembly
. LUC 76: Parks
. LUC 8 and 9: Resource production and undeveloped
land/water areas.
2-29
-------
TABLE 2-4
DEFINITIOU OF ENDOGENOUS MODEL VARIABLES
Name"
Description
Land Use Codes [106J
RES
Number of dwelling units per 10,000 acres of area
of analysis in 1970.
COMM
Commercial land use" per 10,000 acres of area of
analysis in 1970 in 1,000 square
feet.
OFFICE
Office-Professional services land
. use "per 10,000 acres of area of analysis
1n 1970 in 1,000 square feet.
N
I
(.oJ
a
MANF
Manufacturing land use per 10,000 acres
of area of ahalysis in 1970 in 1,000 sq. ft.
Wholesale-warehousing land use per 10,000
acres of area of anaTysis in 1970 in
10,000 square feet.
i'~on-expressHay highway lane m"iles per 10,000
acres of area of analysis in'1970.
Educational land use per 10,000 acres of area
of analysis in 1970 in 1,000 square feet.
WHOLE
HIWAYS
muc
11
12
13
14
19
Household units
Grou p quarters
Residential hotels
Mobile home parks or
Other residential
courts
Reta 11 trade
Personal services
Repair services
Contract construction services
Finance,insurance, and real
estate services
631-636 Business services (excludes
638,639 warehousing)
65 Professional services
692 Welfare and Charitable
servi ces
699 Other services
Manufacturing
Wholesale trade
Warehousing and storage services
45
Highway and street right-of-way
52:-59
62
64
66
61
2,3
51
637
681
Nursery, primary, and secondary
educat ion
-------
TABLE- 2-4 (CONTINUED)
DEFINITION OF ENDOGENOUS MODEL VARIABLES
Name
Description
Land Use Codes [106]
RIT
Active, outdoor recreational land use per 10,000
acres of area of analysis in 1970 in acres.
73 Amusements
74 Recreational activities
75 Resorts and group camps
15 Transient lodgings
691 Religions activities
71 Cultural activities and
nature exhibitions
79 Other cultural, entertain-
ment, and recreational
mHffi
Other urban land uses per 10,000 acres of area of
analysis in 1970 in 1~000 square feet.
rv
I
~
---
-------
'---~_.~-.--~'-." ~---_._.
. ------.
Next, a .list of exogenous variables thought to be causally related
to the endogenous land use variables was prepar~d. The types of exogenous
variables that were required can be generally categorized as representing:
. Service area characteristics in
. Regional growth factors for the
(1960-1970)
. The development potential of the wastewater major project
(1960) .
the base year (1960)
time period of analysis
The base year characteristics of the wastewater service area include measures
..
of socioeconomic variables, accessibility, and land use constraints, and serve
I'
a~ surrogates for detailed estimates of base year land use, which could not be
entered explicitly in the model due to the unavailability of data. Land use
-~_._. ~, -- ...-.
growth in the area of analysis is due to both' reg;.onol activities and the de-
velopment potential of the major project. Thus, exogenous variable categories
have been specified for these major effects. The identification of exogenous
variables in each of these categories was based on literature review and
previous experience (i.e., professional judgement) of the staff in secondary
impacts of infrastructure investments *. The exogenous variables se-
lected for each of the three general categories are given in Table 2-5. Due to
the wide variability in case study project area size, all exogenous variables,
with the exception of AREA, are normalized to a percent or per acre basis.
AREA has been included in the list of exogenous model variables to control
. '
for any variation in the general structure of causal relationships related
to the variation in area of analysis scale.
The time period for the majority of exogenous variables is the year
(t+O) when the wastewater major project was completed and became operational.
Due to data constraints, however, published data sources (e.g., the U.S.
Census) had to be used as the basis for several exogenous variables. In
*
See references [17, 19, 53, 57-61, 63-66, 80, 107J.
2-32
-------
TABLE 2-p
DEFINITION OF EXOGENOUS MODEL VARIABLES
Name - Description
Data Source
Service Area Base Year Characteristics:. Socioeconomic Variables
DUACRE = Dwelling units per mile2 in area of analysis in 1960.
= (100*du6Q)/acre
where: du60 = 1960 census tracts* housing units i2 100s
acre = 1960 area of census tracts in miles
VACHSE = Percent vacant available dwelling units in area of
ana 1 ys is in 1960.
INCOME = Relative medium income of fanrilies and unrelated
individuals in area of analysis compared to county
income levels in 1960.
= (lO*inc)/median
where: i nc
= 1960 median income for families in
census tracts in $lOs
= 1960 median income for families in
the county**
VACOFF = Vacancy rate of offi ce bui 1 dings in area of ana lys i s
i n 1960.
median
UNEMP = Unemployment rate in area of analysis (census tracts)
in 1960.
OFFJOB = Office employment per mile2 in area of analysis in 1960.
= (100* smoff )/acre
where: smoff = 1960 office employment in census
tracts in 1 OOs***.
WWJOB = Warehouse and wholesale employment per mile2 in area
of analysis in 1960.
= (lOO*wwemp)/acre
where: wwemp = 1960 employment in wholesale tradet in
cens us tracts in 1 OOs.
Census
Census
Census
Census
Census
BOMA/Planning
Agency
Census
Census
Census
* Census tracts refers to those tracts most closely approximating the
area of analysis in areal extent.
** County refers to the county containing most of the legal service area.
*** FIRE, Business Services, Public Administration, and Repair Services.
+ Trucking, Warehousing, and Wholesale Trade.
2-33
-------
TABLE 2 -5 (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
Name - Oescri pti on
Data Source
MANJOB = Manufacturing employrrent per mile2 in area of analysis
in 1960. .
= (lOO*manemp}/acre .
where: manemp = 1960 manufacturing employment in
census tracts in 100s.
JOBS = Total employment per mile2 in area of analysis in 1960.
(lOO*totemp)/acre
where: totemp = 1960 total civilian employrrent in
census tracts in 100s.
NONHH = Non-household population per mile2 in area of analysis
i n 1960.
= (100*nonh60)/acre
where: nonh60 = 1960 population in group quarters in
census tracts in 100s.
= Percent of total families with income below $3,000
in area of analysis (census tracts) in 1960.
RENTS* = Percent of total hous; ng un; ts that are renter occu-
pied in area of analysis in 1960.
POOR
VALUE = Median value of housing units in area of analysis
(census tracts) in 1960.
ROOMS = Median number of rooms in housing units in area of
analysis (census tracts) in 1960.
KIDS
= School age children per 100 households in area of
analysis in 1960.
= 100*sch60/du60
where: sch60 = 1960 population 0-14 years of age
in census tracts in 100s.
Census
Census
Cens us
Census
Census
Census
Census
Census
* Used only to predict the disaggregation percentage variables of SFDET,
MF, and SFATT.
2-34
-------
TABLE 2-5 (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
Name - Description
Data Source
UNIV
= Categorical variable to indicate the presence or
absence of a college or university in the area of
analysis in 1960. .
where: 1 = a college or university existed in census
tracts in 1960.
o = none existed~
RELJOB = Relative employment density of the. area of analysis
in 1960.
= JOBS/(lOO*rempl/areall
where: remp1 = 1960 total employment in the SMSA
in 100s. . 2
areal = 1960 SMSA land area in mile.
RELWW = Relative warehouse and wholesale employment density
of the area of analysis in 1960.
= WWJOB/(lOO*rwwemp/areal)
where: rwwemp = 1960 employment in warehouse and
wholesale trade in the SMSA in 100s.
RELOFF = Relative office employment density in the area of
analysis in 19.60..
= OFFJOB/(lOO*roffl/areal)
where: roff1 = 1960 SMSA office employment in 100s.
RELMAN = Relative manufacturing employment density of the area
of analysis in 1960.
= MANJOB/(100*rempl*manper/areal)
where: manper = percent of 1960 total SMSA employment
in manufacturing.
GOVT = Total county government expenditures in 1962 in 106 $.
LAND
= Price of vacant land in area of analysis in 1960
relative to median regional income.
= price/median
where: price = median price of one qcre of residential
vacant land ($) in area of analysis
in 1960. '
Planning
!\gency
Census
Census
Census
Census
Census
Census
Planning
Agency
2-35
-------
.
TABLE 2-5 (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
Name - Description
Data Source
POPDEN
SFDET
MF
= Population density of the area
in persons per mi le2 . "
= (1 OO*pop )/acre
where: pop = total population
1960 in 100s.
of analysis in 1960
of census tracts in
= Percent of housing that is single family
area of analysis in 1970.
= unitl/hse70
where: unitl =
detached in
hse70=
number of single family detached
units in census tracts in 1970 in 100s.
total number of housing units in
census tracts in 1970 in 100s.
= Percent of housing that is multifamily in area of
analysis in 1970.
= (unit34+unit5+unit50)/hse70
where: unit34 = number of 3 and 4 family housing
units in census tracts in 1970 in
100s.
unit5 = number of 5-49 family housing units
in census tracts in 1970 in 100s.
unit50 = number of 50+ family housing units
in:census tracts in 1970 in 100s.
SFATT = ((hse70-unitl)/hse70)-MF
= Percent of housing that is single family attached
in area of analysis in 1970.
PCOMMl
= Percent of commercial development less than
50,000 ft2 in floor area.
= comml/(comml + comm2 + comm3)
where: comml = total commercial floor space in
1,000 ft2 in area of analysis in
1970 for buildings with less than
50,000 ft2. .
comm2 = total commercial floor space in
1,000 ft2 in 1970 in area of analysis
for buildings with less than 100,000 ft2
but greater than 50,000 ft2.
comm3 = total commercial floor space in 1,000
ft2 in 1970 in area of analysis for
buildings with greater than 100,000 ft2.
2-36
Census
Ce n s u s
Census
Census
Census
Planning
Agency
-------
TABLE 2-5 (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
Name - Description
Da ta Source
PCOMM2 = Percent of commercial development with
floor area between 50,000 and 100,000 ft2.
= ~omm2/(comml + comm2 + comm3)
PCOMM3 = Percent of commercial development with
floor area greater than 100,000 ft2.
= comm3/(comml + comm2 + comm3)
Service Area Base Year Characteristics: Accessibility
DISCBD = Distance in miles from centroid of area of analysis
to centroid of nearest CBD* in year (t + 0)**
RRMILE = Railroad mileage per mile2 in the area of analysis
analysis in year (t + 0)
= (rail*640)/AREA
where: rail = railroad mileage in area of analysis
in year (t + 0)
= Limited-access highway interchanges per mile2 in
area of analysis in year (t + 5)
= (intchg*640)/AREA
where: intchg = number of limited access inter-
changes in area of analysis in
year (t + 5)
ACCESS
I NTDEN
= Relative limited-access highway interchange density
of the area of analysis in year (t + 5)
= ACCESS/(ctyacc*640/county)
where: ctyacc = number of limited access inter-
changes in the county in year (t +5)
county = area of county in mile2
TRANS = Number of transit stops (bus..and commuter rail) in
the area of analysis in year (t + 0)
AIRPRT = Distance in miles from centroid of area of analysis
to centroid of nearest commercial airport in the
year (t + 0)
USGS topo-
graphic map
USGS topo-
graphic map
USGS topo-
graphic map/
Planning
Age n cy
USGS topo-
graphic map/
Planning
Agency
Planning
Agency
USGS topo-
graphic map
* Central Business District, defined as the center of the nearest urban
area with population exceeding 100,000.
**t b tile year tne wastewater major project was completed and became
"operational.' . -' ". ...'.. .
2-37
I '
-------
TABLE 2 -5 : (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
~,
Data Source'
Name - Descripti on
Service Area Base Year Characteristics:
Land Use Constraints
AREA
= Area of analysis in acres
VACANT = Percent vacant developable acreage in area of analysis
in year (t + 0)
= vacdev/(AREA-vacund)
where: vacdev = vacant developable acreage in area of
analysis in year (t + 0)
vacund = vacant undevelopable acreage in area
of analysis in year (t + O)
RlONED = Percent of total acreage zoned for res'idential use in
the area of analysis in year (t + 0)
= rzonel AREA
where: rzone = acres of land zoned for residential use
in the area of analysis in year (t + 0)
ClONED = Percent of total acreage zoned for commercial use in the
area of analysis in year (t + 0)
= czone/AREA
where: czone = acres of land zoned for commercial use
in the area of analysis in year (t + 0)
OlONED = Percent of total acreage zoned for office use in the
area of analysis in year (t + 0) ,
= ozone/AREA
where: ozone = acres of land zoned for office use in
the 'area of analysis in year (t + 0)
IlONED = Percent of total acreage zoned for industrial use in
the area of analysis in year (t + 0)
= i zone/AREA .
where: izone = acres of land zoned for industrial use
in the area of analysis in year (t + 0)
SOILS
= Percentage of total area of analysis hav~ng a "severe"
soil type classification for urban developw~nt suita-
bil ity ,
= s 011 / AREA
where: soil =
Acreage in area of analysis with a
"severe" soil type classification with
regard to suitability for urban devel-
opment.
Project Data
Planning
Agency
Planning
Agency
Planning
Agen cy
Planning
Agency
Planning
Agency
Planning
Agency
P"l anning
Agencyj'stJi 1
Conservati on
Servi ce (SCS)
2-38
-------
TABLE 2-5 (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
Name - Description
Data Source
ONLOT
= Percentage of total area of analysis having a "severe"
soil type classification for on-lot sewage disposal.
suitabi 1 i ty.
= onsite/AREA
where: onsite = Acreage in area of analysis with a
"severe" soi 1 type classification for
on-lot sewage disposal suitability.
= Categorical variable to indicate the severity of gov-
ernmental restrictions on on-lot sewage disposal during
the period (t + 0) to 1970.
where: 4 = on-lot disposal is prohibited entirely.
3 = on-lot disposal is prohibited except on
1 arge lots.
2 = on-lot disposal is permitted but percola-
ti on tes ts are requi red.
1 = on-lot disposal is permitted but package
plants are prohibited.
o = no restrictions.
LI MITS
Categorical variable to indicate the presence or
absence of governmental policies designed to limit the
number of hookups to the sewerage system in area of
analysis anytime during the period (t + 0) to 1970.
Examples are sewer moratoria and rationed connections.
where: 1 = policies on hookup limitations existed
o = policies did not exist.
TLIMIT = The number of years during the period of analysis that
on-site sewage disposal was limited
= 1970-yl i mit
where: ylimit = the year on-site disposal limitations
went into effect (or 1970 if no
res tri cti ons) .
POLICY =
TPOLIC = The number of years during the period of analysis that
sewer system hookup limitations were in effect
= 1970-ypoli c
where: ypolic =
the year limitations on sewage system
hookups went into effect (or 1970 if
no restrictions)
Planning
Agency/SCS
Planning
Agency /
Local
Government
Planning
Agency /
Loca 1
Government
Planning
Agency /
Local
Government
Planning
Agency /
Local
Government
2-39
-------
TABLE 2-5 (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
Narre - Descripti on .
Da ta Source
Regional Growth Factors
ENERGY = Relative electrical energy cost factor in municipality
compared to average U.S. commercial rate in 1960.
= encost/$51.59
where: encost=
cost of energy for commercial users
in 1960 in units of dollars per
1500 KWh.
DRIVE = 100s of workers who drive per mile2 in the county in 1960.
= (autoco)/county
where: autoco = workers who drive to work in county
in 1960 in 1 OOs .
RIDET = WQrkers who ride mass transit per mile2 in the county
i n 1960.
:::: massco/county
where: massco = workers who use mass transit in county
in 1960 in 1 OOs . .
PERCHG = Percent change in county population 1960-1970
DENCHG = Population density change in the county 1960-1970.
= (copop2/county - copopl/county)
where: copop2 = county population in 1970.
copopl = county population in 1970.
----.--------.. 2
I JOBCHG = Change in total regional employment per mile 1960~1970.
I = 100*(remp2 - rempl)/SMArea .
I where: rempl = 1960 total employment in SMSA in lOOse
remp2 = 1970 total employment2in SMSA in lOOse
SMArea= area of SMSA in miles.
HSECHG = Percent change in total regional housing units 1960-1970.
= (rhse2 - rhsel)/rhsel
where: rhsel = total housing units in SMSA in 1960
in 1 OOs .
rhse2 = total housing units in SMSA in 1970
i n 100s.
Census
Census
Census
Cens us
Census
Census
Census
Census
Ce ns us
Census
Census
2-40
-------
TABLE 2-5 (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
Name ~ Description
, COMCHG = Percent change in total regional retail trade* employment
1960-1970.
= (rcom2 - rcoml)/rcoml
where: rcoml = total retail trade employment in SMSA
in 1960 in lOOse .
rcom2 = total retail trade employment in SMSA
in 1970 in lOOse
POPDIF = Percent change in regional population 1960-1970.
MANCHG = Percent change in regional manufacturing employment'
1960-1970.
= (perman*remp2-manper*rempl)/(manper*rempl)
where: perman = % of 1970 total SMSA employment in
manufacturing
manper = % of 1960 total SMSA employment in
manufacturing
SERCHG = Percent change in regional services** employment 1960-
1970.
= (rser2
where:
- rserl )/rserl
rserl = 1960 SMSA service employment in lOOse
rser2 = 1970 SMSA service employment in lOOse
HOSCHG = Percent change in regional hospital
= (rhosp2 - rhosp1)/rhospl
where: rhosp1 = 1960 SMSA hospital
rhosp2 = 1970 SMSA hospital
employment 1960-1970.
employment in lOOse
employment in 100s.
EDUCHG = Percent change in regional educational employment
1960-1970.
= (red2 - redl )/red1
where: red1 = 1960 SMSA educational employment in 100s.
red2 = 1970 SMSA educational employment in lOOse
OFFCHG = Percent change in regional office*** employment 1960-
1970.
= (roff2 - roffl)/roff1
where: roff1 = 1960 SMSA office employment in 100s.
roff2 = 1970 SMSA office employment in 100s.
,Data Source
Ce n s us
Census
Census
Census
Census
Census
Census
Census
Census
Census
Census
Census
Census
* Food. and Dairy, Eating and Drinking, and Other Retail.
** Professional and Related Services, Other Personal Services, Entertainment,
and Welfare and Fraternal Organizations.
*** FIRE, (Finance. Insurance, and Real. Estate) Business Services. Public
Adm~nistration, and Repair Services.
2-41
-------
TABLE 2 -5 (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
. . ..' .. .. .. .
Name - Description
Data Source
PERCAP = Median fam11y income 1n SMSA in 1970.
EMPOP = Regional (SMSA) employment to population ratio in 1970.
= remp2/rpop2
where: rpop2 = 1970 SMSA population in 100s.
CAPCHG = Change in median family SMSA income 1960-1970.
= PERCAP - per60
where: per60 = median family income in SMSA in 1960.
STAY
= Index of mobility - % of 1960 families who were in the
same house in 1955.
Wastewater Treatment Major Project Characteristics
= Number of years available for secondary growth to occur.
= phyear-yea r. .
where: phyear = the year aerial photographs were taken
from which land use data was extracted.
year = the year construction was completed on
the major project on initial phase.
PHASE = Categorical variable to indicate the presence or
absence of phasing of the major project.
where: 1 = phasing has occurred.
o = phasing has not occurred.
TIME
COST
= Normalized per capita local costs ($) of the major
project in area of analysis in year (t + 0).
= 1,000* {totcst/pi1 - fedcst/pi2)/popcom
where: totcst = total major project construction
cost (1,000 $)
fedcst = federally funded share of major
project cost (l,OOO $)
= consumer price index for "year"
= consumer price index for the year of
federally funding of the major project
popcorn = population served by major project
facility in year (t + 0)
pil
pi2
,
LENGTH = Running length of interceptor sewer 1 ines in miles
going through relatively undeveloped land Co:;, 1 du per
acre) in area of analysis in year (t + 0).
Census
Census
Census
Census
Census
Planning
Agency
Project
Data
Project
Data
Project
Data
Project.
Data
Project
Data
Project
Data
2-42
-------
TABLE 2-5 (CONTINUED)
DEFINITION OF EXOGENOUS MODEL VARIABLES
. '" .."" . ., .. .
Name - Description
Da ta Source
CROSS = Index of available undeveloped land in the area of
analysis through ~hich interceptor sewers go through
in t + O.
= 640*LENGTH/AREA
SERVED = Percent of the area of analysis easily served by the
major project in year (t + 0)
= area5k/AREA
where: area5k = acres of land within .5,000 feet of the
major project interceptor sewer in
area of analysis in year (t + 0)
= Total hydraulic design capacity of the major project
wastewater collection system in million gallons per
day (mgd) in year (t + 0) (or in 1965 if a phased
project)
CAPACl
CAPAC2 = Total hydraulic design capacity of the major project
wastewater treatment plant in mgd in year (t + 0)
PEAK
= Actual Peak flow in the major project wastewater system
in mgd in year (t + 0)
RECAP = Reserve capacity of the wastewater major project
collection system in year (t + 0)
= CAPACl - PEAK
RECAP 1 = Percent reserve capacity of the wastewater major
project collection system in year (t + 0)
= 100* RECAP/PEAK
RECAP2 = Percent reserve capacity of the wastewater major
project treatment plant in year (t + 0)
= 100* (CAPAC2 - PEAK)/PEAK
Project
Data
Project
Data
Project
Data
Project
Data
2-43
-------
- ..... ~.. .., ~ . .
these cases, the time period for which the variables defined is 1960, the
average value of (t+O). In addition, the variable ACCESS is defined for
the year (t+5) since the construction of new limited-access highway inter-
changes is commonly known five years in advance. Note that regional growth
variables are based on standard Metropolitan Statistical Areas (SMSAs) which
are sufficiently large so that measures of regional growth processes are
essentially independent of the induced growth effects of major projects [102J.
The variable PEAK required to define the exogenous variables RECAP,
RECAP1 and RECAP2 may not always be available directly from project data
files. In case it is not, the actual peak flow in gallons per day in the
major project wastewater collection system in 1960 may be estimated by first
computing the average flow as follows:
Average flow =
~ (industrial flows) +
50 * (1960 population in service area) +
300 * ~ (pipe 1engthi * pipe diameteri)
1 .
The factor of 50 gallons per day is an average per capita,use [108J. The
.. . ... .' .
last term computes infiltration flow, using a factor of 300 gallons per
day [108J per inch diameter per mile of pipe, summing over all major sewer
pipes. P:ak f1~w is then ~b~ained b,y applying the scaling factor given in
fi g~re 2-6 of reference [100]. ' .
H.
SPECIFY INITIAL PATH ANALYTIC LAND USE MODEL
. ~~..-.-.,. ....
The objective of this task was to formalize the factor theory of
..--..- ._.._~ -
;ndu~ed d~~~lopme~t-i; a 'causal--path dia-g~a~':'-' Path d1agrams--a-re a schema'tic
representation of the hypothesized causal relations among variables. Causa1~
ity is shown by a single-headed arrow (or path) and simultaneity (reciprocal caus-
ation) between two variables is shown by a double-headed arrow. Use of this
technique requires two assumptions about causality in the system: (1) a "weak"
causal order exists among the variables and it is known, and (2) relationships
among variables are causally closed. Weak causal ordering exists in a two
2-44
-------
variable set, X. and X., if it is known on logical, empirical, or theoretical
1 J "
grounds that X. (the exogenous variable) may affect X. (the endogenous
1 "J
variable) and that X. cannot affect X.. Causal closure is simply the concept
J 1
that, given a weak causal ordering (X. ~ X.), the observed covariation between
1 J
the two variables must be due only to the causal dependence of X. on X.,
J 1
their mutual dependence on some outside variable(s), or a combination of these
factors [92J. Since path analysis is only capable of testing theories, rather
than deducing them, the most comprehensive model possible was initially
hypothesized. This ensures that the final path analytic model will represent
the causal structure closest approximating the real world.
The approach to this task was to take the list of initial "model
variables previously compiled, and using our knowledge of the infrastructure
relationships involved, to examine all possible causal relationships between
I. .
each endogenous variable and all other exogenous and endogenous variables.
For each variable pair, we hypothesized:
. If a relationship is plausible and if so, why,
. The direction of causal action, and
. Whether the hypothesized relationship is "weak"
or "strong" (subjectively).
Where a causal relationship was perceived, it was coded into the path
diagram shown in Figure 2-3, and the direction and strength of causal action
"are denoted by the "+" and "_" signs placed on the paths. Thus, Figure 2-3
represents the initial path analytic land use model. The specification of
the model resulted in 9 structural equations of which 7 are simultaneous.
These 7 equations, related to each other by interactions, are summarized in
Tables 2-6 through 2-12 and the path diagram representing tnese equations is
shown in Figure 2-4. The remaining 2 recursive equations are summarized in
Tables 2-13 and 2-14 and Figure 2-5. The complete initial model contains 9
. " . . . ; . .' .. .. ~
., .. . . . .
endogenous and 32 exogenous variables. An additional 34" ex6genous variabl~s
serve as alternatives to those specified in the initial model (see Tables 2-6
through 2-14). The choice between prime and alternative exogenous variables
was made based on the results of a zero-order correlation analysis for all
model "variables (see Section IV). This analysis also provided the necessary
data to check for the problems of multicollinearity and suppressor variables.
2-45
-------
N
I
..l="
01
r::\
+~
+
+,+
OTHE R
OFFICE
+
+
B
+
+,+
+,+
+
-,-,+
COMM
++"
+ -,+ +
+
+
REC RES ++ HIWAYS
+ + -
++ +,+,- +
+
+
++
+
+,+,-
+
+
MANF
EDUC
+,-
Note: Some Variables Appear Twice for Ease of Presentation.
Path Coefficients Separated by Commas Refer to Multiple Exogenous Variables, Respectively.
Double "+" Signs Denote Strong Causal Relationships.
FIGURE 2-3
INITIAL SPECIFICATIOn OF THE PATH ANALYTIC LMm USE MODEL
-------
OFFICE
+
, u.
B
'.' , G
+ + I .
COM~'
I'\)
I
~
......
+ +
8
--,
+'
+,+
+,+,+
++
s
I ---
G
+ -,+
+
+
+
RES
HHIAYS
REC
+
+
MANF
EDUC
+
+ -
,
Note: So~e Variables Appear Twice for Ease of Presentation.
Path Coefficients Separated by Commas Refer to Multiple Exogenous Variables, Respectively.
Double U+U Signs Denote Strong .Causal Relationships.
. FIGURE 2-4
: SEVEN EQUATION SJMULTANEOUS BLOCK IN THE INITIAL PATH ANALYTIC LAND .USE MODEL
-------
+
-,-,+
G
+,+
+,+
OTHE R
OFFICE
COMI1
I
~
---
8
--- ------
".-- -- ---.-
RES
N
I
~
ex:>
+
+,+,-
MANF ++ WHOLE
+
Nate: Some Variables Appear Twice for Ease of Presentation.
- Path Coefficients Separated by Commas Refer to Multiple Exogenous Variables, Respectively.
Double "+" Signs Denote Strong Causal. Relationships.
FIGURE 2...5
TWO EQUATION RECURSIVE BLOCK IN THE INITIAL PATH ANALYTIC LAND USE MODEL
-------
TABLE 2-6
INDEPENDENT VARIABLES FOR THE
RES MODEL EQUATION
Descri ptor
Alternate(s}
Prime
Endogenous
Endogenous
Endogenous
Endogenous
MP* sewer
capaci ty
MP sewer service
MP timing
On-si te di sposal
res tri cti ons
Popul ati on
density
Regi ona 1
i nfl uences
I n come
Emp 1 oyment
characteri sti cs
Housing
characteri sti cs
Land cos ts and
taxes
Mobility
Developable land
Zoning
Access i b il i ty
+MANF
+OFFICE
+HIWAYS
tREC
+RECAPl
+ LENGTH
+ TI ME
-LIMITS
+DUACRE
+HSECHG
+INCOME
+JOBS
+VALUE
-LAND
-STAY
+VACANT
+RZONED
+ACCESS
+CAPAC1, +CAPAC2, +RECAP, +RECAP2, +PEAK
+CROSS, +SERVED, +COST
+PHASE
-POLICY, -TLIMIT, - TPOLIC
+POPDEN
+PERCHG, +POPDIF, +JOBCHG
+PERCAP, +CAPCHG, -POOR,
+RELJOB, -UNEMP, -POOR
-VACHSE, +ROOMS, -NONHH
- GO VT
+AREA
* Major Project
+INTDEN, +TRANS, +DRIVE, +RIDET
i '
2-49
-------
TABLE 2-7
INDEPENDENT VARIABLES FOR THE
COMM MODEL EQUATION
Descriptor
Alternative(s} .
Prime
Endogenous
Endogenous
Endogenous
,
Income
Employment
characteristi cs
Populati on
characteri s ti cs
Accessibility
Zoning
MP timing
Regi ona 1
i nfl uences
+RES
+HIWAYS
+OFFICE
+INCOME
+OFFJOB
+POPDEN
+ACCESS
. +CZONED
+ TIME
+COMCHG
+PERCAP, +CAPCHG, -UNEMP, -POOR
+JOBS, RELOFF, +RELJOB
+UNIV, +NONHH
+INTDEN, +TRANS, +DRIVE, +DISCBD, +RIDET
+OZONED
+OFFCHG, +POPDIF
2-50
-------
TABLE 2-8
INDEPENDENT VARIABLES FOR THE
OFFICE MODEL EQUATION
Descriptor Prftre A 1 ternati ve(s} , ','
Endogenous +COMM
Endogenous +RES
Endogenous +MANF
Endogenous +HIWAYS
Income +INCOME +PERCAP, +CAPCHG
Employment +oFFJOB +RELOFF, +RELJOB, +JOBS
character; sti cs
Popul ati on +POP DEN +NONHH
character; s t; cs
Access; bi 11 ty +ACCESS +INTDEN, +TRANS, +DRIVE, -DISCBD, +RIDET
Zoning +OZONED +CZONED
MP timing + TI ME
Reg; onal +OFFCHG +JOBCHG, +POPDIF, HOSCHG
;nfl uences
Development +VACANT -VACOFF, -GOVT
constra1 nts
2-51
-------
TABLE 2-9
INDEPENDENT VARIABLES FOR THE
MANF MODEL EQUATION
Descriptor Prime . Alternative(s)
Endogenous +HIWAYS
MP sewer +RECAPl +CAPAC1, +CAPAC2, +RECAP2, +RECAP, +PEAK
capaci ty
MP sewer +LENGTH +CROSS, +SERVED
service
Emp 1 oyment +MANJOB +JOBS, +RELJOB, RELMAN, +EMPOP, +UNEMP
characteri s ti cs
Rai 1 +RRMILE
Accessibility
Road and Ai r +ACCESS + I NTDEN, '-AI RPRT
Accessibility
De ve 1 op ab 1 e +VACANT +AREA
land
Zoning +IZONED
On-site disposal -LIMITS - TLIMIT
restri cti ons
MP timing +TIME
land costs -lAND -GOVT
and taxes
Regi ona 1 +MANCHG +JOBCHG, +POPDIF
influences
2-52
-------
TABLE 2-10
INDEPENDENT VARIABLES FOR THE
HIWAYS MODEL EQUATION
Descriptor Prime
. Endogenous +RES
Endogenous +COMM
Endogenous +MANF
Endogenous +OFFICE
Population +DUACRE
characteristics
Employment +JOBS
characteristics
Motor vehicle +DRIVE
densi ty
Distance to -DISCBD
trip generators
MP timing + TIME
Limited access -INTDEN
highways
Alternative(s)
+POPDEN, +POPDIF, +UNIV
+RELJOB, +MANCHG,+JOBCHG, +OFFCHG, +EMPOP
+RIDET
-AIRPRT
-ACCESS
2-53
-------
TABLE 2-11
INDEPENDENT VARIABLES FOR THE
EDUC MODEL EQUATION
Descriptor Prime Alternative(s)
Endogenous +RES
Population
characteristics +DUACRE -NONHH, +UNIV, +POPDEN
Housing
characteristics +VALUE -VACHSE
School Age
children +KIDS
Zoning +RZONED
MP timing + TI ME
Mobility -STAY
Regional
influences +HSECHG +PtRCHG, +POPDIF, + EDUCHG
2-54
-------
TABLE 2-12
INDEPENDENT VARIABLES FOR THE
REC MODEL EQUATION
Descriptor Prime A1ternative(s)
Endogenous +RES
Endogenous +EDUC
Income +INCOME +VALUE, -POOR, +GOVT
Population +KIDS -POPDEN, +UNIV
characteristics
Land constraints -LAND +VACANT, +AREA
Zoning +RZONED
Regional +POPDIF +PERCHG
influences
2-55
-------
TABLE 2-13
INDEPENDENT VARIABLES FOR THE
WHOLE MODEL EQUATION
Descriptor
P ri me
Al ternati ve (s)
Endogenous
Endogenous
Endogenous
Employment
characteri s ti cs
Land costs
Popul a ti on
characteri st i cs
Accessibility
Deve 1 opab 1 e
1 and
Zoning
MP t i mi n g
Regional
in f1 uences
+MANF
+HIWAYS
+COMM
+WWJOB
+MANJOB, +RELJOB, RELWW, +RELMAN, +EMPOP
-LAND
+POPDEN
+ACCESS
+VACANT
+RRMILE, +INTDEN, -AIRPRT, -DISCBD
+AREA
+IZONED
+ TIME
+COMCHG
+POPDIF, +MANCHG
2-56
-------
TABLE 2-'14
INDEPENDENT VARIABLES FOR THE
OTHER MODEL EQUATION
Descriptor Prime A1ternative(s)
Endogenous +RES
Endogenous +OFFICE
Endogenous +MANF
Endogenous +REC
Income +INCOME +PERCAP, +CAPCHG, -POOR
Employment +JOBS +RELJOB, +EMPOP
characteristics
Population +UNIV +POPDEN
characteristics
Land constraints -LAND +VACANT, +AREA
Accessibi 1 i ty +ACCESS +n~TDEN, +TRANS, +DRIVE, +RIDET
Distance to -DISCBD
population center
MP timing + TIME
Regional +SERCHG +PERCHG, +POPDIF, +JOBCHG
influences
\
2-57
-------
The maximum number of variables for anyone equation in the re-
cursive block is 12. The total number of exogenous variables in the simul-
taneous block is 29. Thus, the total number of instrumental variables for
the two-stage least squares solution of the simultaneous block approaches
the study sample size of 40. The result is that that the available degrees
of freedom for model testing of anyone simultaneous equation are 10
(=40-29-1). This exceeds the minimum of 6 degrees of freedom thought to be
necessary to obtain meaningful results in the initial path analysis.
Note that the variables used in the initial model specification
exclude four exogenous variables for which data could not be obtained (see
SectionIII).These variables are DENCHG, ONLOT, SOILS, and ENERGY. Although
these variables were not originally excluded from the initial model, it is
instructive to include these changes at this point.
The theoretical basis for the initial model is discussed, by endo-
genous variable, in the following sections. Note that in these discussions
the implicit assumption is made that all of the variables in a given equation
are held constant while considering the effects of a particular independent
variable. Also, although only the prime independent variable is discussed,
the basis for corresponding alternative variables is identical.
1.
RES Equati on
RES = MANF + OFFICE + HIWAYS + REC
+ RECAPl + LENGTH + TIME
- LIMITS + DUACRE + HSECHG
+ INCOME + JOBS + VALUE
- LAND - STAY + VACANT
+ RZONED + ACCESS
2-58
-------
.. -. ..- -.._0. ~ . .
Residential developrrent (RES) is hypothesized to 'respond to in-
creased manufacturing (MAN F) and office-professional services (OFFICE) activity
due to employrrent opportunities, and to highway lane miles (HI WAYS) , as such
development (particularly high density residential) locates in areas well
served by transportation networks. In addition, active, outdoor recreational
development (REC) is included as a causal factor since vacation homes often
locate near such facilities.
Three different wastewater major project variables (RECAP1,
LENGTH, TIME) are included since wastewater facilities have strong growth-
inducing impacts on residential development (both single family and high
dens ity) .
Growth in regional housing stock (HSECHG) was included to control
for the influence of regional growth on residential development in the area of
analysis. The index of mobility (STAY) was included to control for regional
shifts in population.
, Five base year socia-economic indicators are included in the RES
equation. Residential density (DUACRE) and employment density (JOBS) are
both expected to influence total future residential development. New residential
construction is attracted to areas with relatively high income (INCOME) and
existing residential property values (VALUE), and low vacant land prices (LAND).
Three separate variables for representing base year land use constraints are
included in the RES equation. A constraint on residential development (par-
ticularly high density) is the existence of on,..site..wastewate,r"disposal re-
strictions (LIMITS).
--._-----_.. ...
"-'-~.' .,-...... --. .
- .- -- _#' ..---. ----_.. . ---_. .
Since development is controlled to some extent by the zoning classifications
set in an area, it is expected that the amount of land zoned for residential
(RZONED) is causally related to residential development. In addition, a
prime factor encouraging new residential development is the availability of
vacant, developable land (VACANT). The density of limited-access highway
interchanges (ACCESS) is a measure of an area's accessibility, and so can
have an effect on the location of new residential construction, particularly
on the fringe of 1 arge metropol itan areas.
2-59
-------
..-. .. .~-
. 2.
COMM Equation
COMM = RES + OFFICE + HIWAYS + COMCHG .
+ INCOME + OFFJOB + POPDEN
+ ClONED + ACCESS + TIME
Commercial development (COMM) is hypothesized to respond to resi-
dential (RES) dev~lopment and business activities (OFFICE), as both of these
stimulate demand for commercial services. The existence of a transportation
network (HIWAYS) is also important in servicing demand, particularly with
larger commercial developments, such as regional shopping malls.
included to
ment in the
Growth in regional retail-wholesale trade employment (COMCHG) was
control for the influence of regional growth on commercial develop-
area of analysis.
Three base year socio-economic measures are included in the COMM
equation. Relative median income (INCOME) generally determines the amount of
commercial expenditures and, hence, the amount of commercial development.
Densities of population (POPDEN) and office employees (OFFJOB) are indicators
of the potential market for commercial services.
The density of limited-access highways interchanges (ACCESS) is a
measure of an area's accessibility, and so has a strong effect on the location
of commercial development.
...- - - .__h_.__.
.-----. .--_. -----
Since development ;s controlled to some extent by the zoning
classifications set in an area, it is expected that the amount of land zoned
for commercial use (ClONED) is a determinant of commercial development.
Finally, the amount of commercial development is thought to be
related to the number of years available for secondary growth (TIME).
2-60
-------
3. OFFICE Equation
OFFICE = RES + COMM + MANF + HIWAYS
+ INCOME + OFFJOB + POP DEN
+ ACCESS + OZONED + TIME
+ OFFCHG + VACANT
Office-Professional Services development (OFFICE) is hypothe-
sized to respond to residential (RES) and manufacturing (MANF) development
as these create the principal market for Office-Professional Services. The
proximi ty to commerci a 1 development (COMM) and to transportation networks
(HIWAYS) are strong locational factors in office building location.
Growth in regional office employment (OFFCHG) was included to
control for the influence of regional growth on office-professional services
development in the area of analysis.
Three base year socio-economic measures were included in the
OFFICE equation. Office employment density (OFFJOB) is expected to increase
total future office development. Densities of population (POPDEN) and office
employees (OFFJOB) are indicators of the demand and personnel available for
office-professional services.
The density of limited-access highway interchanges (ACCESS) is
a measure of an area'~ accessibility, and so has a strong effect on the loca-
tion of office-professional services development.
Since development is controlled to some extent by the zoning
classification set in an area, it is expected that the amount of land zoned
2-61
, '
, .
I
I,
-------
for office use (OZONED) is a determinant of office-professional services
development. In addition, a prime factor encouraging new office development
is the availability of vacant, developable land (VACANT).
Finally, the amount of office development is thought to be
related to the number of years available for secondary growth (TIME).
4.
MANF Equation
MANF = HIWAYS + RECAPl + LENGTH + MANJOB
+ RRMILE + ACCESS + VACANT + IZONED
-LIMITS + TIME - LAND + MANCHG
Prime location requirements for manufacturing industries (MANF)
include proximity to good highways (HIWAYS, ACCESS) and availability of
railroad access (RRMILE).
Three different wastewater major project variables are included
since the placement (LENGTH), sizing (RECAP1), and timing (TIME) of waste-
water facilities effects the location of industrial facilities (MANF).
Growth in regional manufacturing employment (MANCHG) was included
to control for the influence of regional growth on manufacturing development
in the area of analysis.
Two base year socio-economic measures are included in the MANF
equation. Manufacturing employment density (MANJOS) is expected to increase
total future manufacturing development, while vacant land costs are expected
to decrease such development.
Three separate variables for representing land use constraints
are included in the MANF equation. A constraint on industrial development
is the existence of qn-site wastewater disposal restrictions (LIMITS). Since
development is controlled to some extent by the zoning classifications set in
an area, it is expected that the amount of land zoned for industrial use (IZONED)
is causally related to manufacturing development. In addition, a factor en-
couraging new manufacturing development is the availability of vacant, devel-
opable land (VACANT).
2-62
-------
5.
HIWAYS Equation
HIWAYS = RES + COMM + OFFlfE + MANF
+ DUACRE + JOBS + DRIVE
- DISCBD + TIME - INTDEN
Non-expressway highway lane mile development (HIWAYS) is hypothe~
sized to respond to residential (RES), commercial (COMM), office-professional
services (OFFICE), and manufacturing (MANF) developments since each of these
generate substantial vehicular traffic and highway facilities service this
transportation need.
The relative limited access highway interchange density of the
area of analysis (INTDEN) is expected to restrict the development of non-
expressway highway development since these two types of highway service many
of the same needs. Highway facilities provide access to population and em-
ployment centers, such as central business districts. Thus, an inverse
relationship between the distance to the CBD (DISCBD) and highway development
is expected. .
Two base year socio-economic indicators are included in the
HIWAYS equation. Residential density (DUACRE) and employment density (JOB)
are both expected to increase future highway construction.
The regional density of motor vehicle drivers (DRIVE) i-s con-
sidered to be a further indicator of the demand for highway facilities.
Finally, the amount of highway development is thought to be
related to the number of years for secondary growth (TIME).
6. EDUC Equation
EDUC = RES + DUACRE + VALUE + KIDS + RZONED + TIME
- STAY + HSECHG
Educational development (EDUC) is hypothesized to respond to
residential development (RES) since this housing stock generally determines
the market for educational facilities.
2-63
-------
Growth in regional housing stock (HSECHG) was included to control
for the influence of regional growth on educational development in the area
of analysis. An index of mobility (STAY) was included to control for regional
shifts in population.
Three base year socio-economic measures are included in the EDUC
equation. Public school enrollment density (KIDS) is expected to increase
the ~otal future educational facility development. Residential density (DUACRE)
and the value of existing residential property (VALUE) are both expected to
increase the amount of total future educational facilities.
Since development is controlled to some extent by zoning classi-
fications set in an area, it is expected that the amount of land zoned for
residential (RZONED) is causally related to future educational development.
Finally, EDUC is thought to be related to the number of years
available for secondary growth (TIME).
7.
REC Equati on
REC = RES + EDUC + INCOME + KIDS
- LAND + RZONED + POPDIF
Active, outdoor, recreational development (REC) is hypothesized
to respond to residential (RES) development due to the demand it creates for
recreational opportunities. Educational facilities (EDUC), due to their
youth orientation, also generate significant demand for active, outdoor
recreational facilities.
. Growth in total regional
control for the influence of regional
the area of analysis.
population (POPDIF) was included to
growth on recreationa'l facilities in
Three base year socio-economic measures are included in the REC
equation. Relative median income (INCOME) and public school enrollment
density (KIDS) are determinants of the amount of active, outdoor recreation in a
2-64
-------
community. To a
size and numbers
dential (RZONED)
lesser extent, the cost of vacant land (LAND) determines the
of recreational areas. The amount of land zoned for resi-
also gives an indicator of the future demand for recreational
areas.
8. WHOLE Equation
WHOLE = COMM + MANF + HIWAYS + ACCESS
+ WWJOB - LAND + POPDEN
+ VACANT + IZONED + TIME + COMCHG
Wholesale-warehousing development (WHOLE) is hypothesized to
respond to related manufacturing (MANF) and commercial (COMM) developments.
A locational factor which attracts wholesale-warehousing is accessibility
to highways (HIWAYS, ACCESS).
Growth in regional retail-wholesale trade employment (COMCHG)
was included to control for the influence of regional growth on wholesale-
warehousing development in the area of analysis.
Three base year socio-economic indicators are included in the
WHOLE equation. Densities of population (POPDEN) and wholesale/warehousing
employees (WWJOB) are indicators of the market and personnel for new
wholesale/warehousing development. Due to the large land area of such
developments, they are very sensitive to the cost of vacant land (LAND).
Prime factors encouraging new wholesale/warehousing are the
amount of land zoned for industrial use (IZONED) and the amount of vacant
developable land (VACANT).
Finally, WHOLE is thought to be related to the number of years
available for secondary growth (TIME).
9. OTHER Equation
OTHER = RES + OFFICE + MANF + REC
+ SERCHG + INCOME + JOBS + UNIV
- LAND + ACCESS - DISCBD + TIME
2-65
-------
Other urban development types (OTHER) include religious,
cultural-entertainment, and hotel-motel activities. These are hypothesized
to respond to residential development (RES) as this is the principal source
of demand for the services of other urban development types. In addition,
office-professional services (OFFICE), manufacturing (MANF) and active,
outdoor recreational (REC) developments are causal factors since each can
generate overnight visitors and, hence, demand for hotel-motel services.
Growth in regional services employment (SERCHG) is included to
control for the influence of regional growth on other urban development
types in the area of analysis.
Three base year socio-economic measures are included in the
OTHER equation. Since other urban development includes hotel-motel land
uses and, since employment opportunities provide a market for hotel-motel
services, total employment density (JOBS) is expected to increase total
future other urban development. The existence of a university or college
(UNIV) in the area is expected to increase the development of cultural
-activities to service the academic community and to increase hotel-motel
development through accommodation requirements for professionals and
student's families. All of these are elements of the variable OTHER. The
types of development including OTHER are also closely related to relative
income levels (INCOME).
Since cultural, religious and hotel-motel activities all
service, and hence locate,~ear population and employment centers, an inverse
relationship between the distance to the nearest central business district
(DISCBD) and other urban development is expected. The density of limited-
access highway interchanges (ACCESS) is a measure of an area's accessibility,
and so can have an effect on the location of new cultural-entertainment and
hotel-motel activities.
The cost of vacant land (LAND) is often a prime factor in the
location of OTHER developments.
Finally, OTHER is thought to be related to the number of years
available for secondary growth (TIME).
2-66
-------
III.
PHASE II - DATA COLLECTION
The principal objective of the second phase was to collect a sufficiently
large and diverse cross-sectional data base on which to develop the causal and
predictive model.
A.
CASE STUDY SELECTION
Due to the critical importance of data quality and the large man-
power effort required to collect the data base for this study, the first task
was to carefully select a set of 40 projects for case studies. These case
studies were wastewater major projects, distributed nationwide, which had the
potential, upon construction, for inducing a significant quantity of land
development in their communities. Identification of the case study projects
was based primarily upon an extensive mail survey of municipalities and re-
gional planning agencies, supplemented by data obtained from a review of sec-
ondary (published) data sources.
The approach to the selection of case study major projects involved
three steps. First, a list of selection criteria were developed. Second,
these criteria were applied to information derived from primary and secondary
sources to obtain a preliminary list of 46 potential case studies. Finally,
detailed inquiries were made regarding each potential case study to verify
its compliance with everyone of the selection criteria, and a final set of
forty case study major projects compiled. These efforts are reported on in
the sections below.
1.
Selection Criteria
The selection criteria were designed to standardize certain as-
pects of the model data in order to avoid difficulties in the analysis and
to minimize the overall effort of the data collection program. They are
summarized as follows:
a.
Geographical Distribution Requirements
These were specified to ensure a generalized model applica-
In summary:
b1e nationwide.
3-1
-------
. All cases had to be located in the continental
United States.
. At least two cases had to come from each federal
region.
. No more than five cases could come from anyone
state.
. No more than two cases could come from anyone
SMSA*, and
. The area of analysis of any two cases could not be
contiguous or overlapping.
b.
Size, Type and Timing Requirements
These were the specified by the definition of a wastewater
major project, as repeated below:
. A wastewater major project is considered to involve
principally the construction or extension of inter-
ceptor or collector sewer lines during the period.
1958 to 1962+ in a community in the United States.
If construction date information is not available,
then a grant funding date in the period 1956-1960
is acceptable. The project had to affect an in-
crease in absolute system collection capacity of
1 MGD or more, and had to cost a minimum of $200,000
to construct. Phased projects are considered if
the first phase of construction on the collection
network was complete within 1958-1962+ and the last
phase of construction was complete by 1965.
c.
Potential for Induced Growth Requirements
These were specified
could have occurred in connection with
by the definition of area of analysis,
to ensure that induced development
a wastewater project, and are given
as repeated below:
The area of analysis is defined as the legal service
area of a wastewater major project in the base year.
It must be a minimum size of 5,000 acres (or approxi-
matelyeight square miles), and contain significant
amounts of vacant developable land, some of which
must be more than 5,000 feet from the nearest inter-
ceptor line in the base year.
*Standard Metropolitan Statistical Area.
+It was necessary to extend this time period to 1963 in order to obtain 40
case study wastewater projects.
3-2
-------
d.
Availability of Data Requirements
These ensured that the municipality, regional planning
agency or water district associated with each case study project had sufficient
records for the project to estimate values for important exogenous model varia-
bles. Specifically, the following data were required:
. A map of the legal service area showing the extent
of the wastewater collection network in the base
year
. Base year population in the
. Actual peak wastewater flow
tern in the base year
. Total hydraulic design capacity of the major pro-
ject (both collection network and treatment plant).
legal service area
in the wastewater sys-
2.
Information Sources
Case study major projects were identified from both primary and
secondary data sources, as discussed below.
a.
Primary Data Sources
The primary source of information for case study selection
was a nationwide mail survey. Since the base year of interest is 1960, it
is expected that many of the sewer project records, which are almost two
decades old, may be lost or destroyed. For this reason, the survey effort
was directed at two possible sources of information: (1) those municipalities
in which the project is located, and (2) regional planning agencies.
The first survey consisted of regional planning agencies,
the best source of case study projects for GEMLUP-I. The 1977 Directory of
such agencies, obtained from the National Association of Regional Councils
[118], was used to identify the agencies. Due to the large number of such
groups (669) we could not survey them all. Therefore, within each federal
region, one or two representative states were chosen for the survey, and
only regional planning agencies which were also A95 Clearinghouses (the local
3-3
-------
agency responsible for overseeing federally funded projects) were
The total number of agencies surveyed is 213; relevant statistics
with the survey are summarized by federal region in Table 3-1.
contacted.
associated
The second survey was of municipal department of public
works in the selected States. EPAls Project Register [52J of over 15,000
wastewater projects nationwide receiving matching federal funds between 1956
and 1973 was screened using the definition of major project to obtain the
survey list. A sample of the mail survey questionnaire is shown in Table 3-2.
The total number of municipalities surveyed was 291. Relevant statistics as-
sociated with this survey are summarized by federal region in Table 3-1.
The results of the case study mail survey (see Table 3-1)
show a total of 39 case study wastewater projects identified by this approach.
Note that it was necessary to extend'the time period for construction ,of a
major project to 1963 to obtain a sufficient number of case studies. Of the
39 selected, 9 or 23% finished construction in 1963. A list of the selected
case study major projects is given in Tabl~:'3-3.
b.
Secondary Data Sources
As part of the literature search performed in Phase I of
this study, an exhaustive search for published case studies was undertaken.
A review of secondary data with regard to the case study selection criteria
produced only seven additional case study major projects, these are numbers
27,36-40, and 46 in Table 3-3 [101, 110-112J. Most of the potential projects
identified in secondary sources were inappropriate due to noncompliance with
one or more of the selection criteria. The remainder did not present suffi-
cient data to judge the appropriateness of a project. This does not mean
that the secondary data sources were unessential to the selection process.
In fact, they were invaluable in providing data on the wastewater projects
identified through the mail survey which allowed the selection of the case
studies. Sources that proved to be particularly useful were the following:
3-4
-------
TABLE 3-1
SUMMARY OF SURVEY QUESTIONNAIRE RETURNS
Federa 1 Regi on
Number
Ma i 1 ed
Number
Returned
N umbe r
I denti fyi ng
Projects
Number of
Case Study
Projects
Regional Planning
Councils
I 26 4 1 2
II 3 1 1 1
III 26 4 2 1
IV 27 2 1 0
V 30 7 2 2
VI 24 4 0 0
VII 20 1 0 0
VIII 9 0 0 0
IX 18 1 0 0
X 30 4 1 0
Subtota 1 213 28 8 6
Municipalities
I 39 10 9 4
II 21 3 3 1
III 31 6 4 2
IV 52 12 8 7
V 39 " 10 9 2
VI 34 8 6 5
VII 20 2 2 2
VIII 9 1 1 1
IX 35 19 14 6
X 11 5 4 3
Subtota 1 291 76 60 33
Tota 1 504; 104 68 39
Return rate = 21%
% of returns identifying projects = 65%
% of total identifying projects = 14%
3-5
-------
TABLE 3-2
SAMPLE SURVEY QUESTIONNAIRE
Name of Respondent:
Organization:
Address:
Name of Project:
Start/Completion Dates:
location of Project
Municipality - Name:
Drainage Basin
or Sewershed - Uame:
Size:
Size:
Approximate Size of
Interceptor Service Area:
Design Excess Capacity of Interceptor Main:
%
Available Data Sources in Area for Planning Information
3C Planning Agency:
Detailed land Use Inventory or Aerial Photographs
for r1unicipality circa 1970:
Telephone #:
Project #:
Circle One
Acres
or
Sq. Miles
Acres or
Sq. Miles
Please attach a map of any scale of the area encompassing the interceptor
project. Feel free to ~~ke any pertinent comments concerning the project
below. If you have any questions, call Peter Guldberg at 617-657-4250 Xl19.
Thanks for your assistance.
3-6
-------
.
.
.
.
"
. .
TABLE 3-3
CASE STUDY MAJOR PROJECTS
ID ID
Number Project Name Location Number Project Name Location
1 Will i manti c SS* Wi11imantic, CT 19 Lick Creek Interceptor Da n v i 11 e, I L
2 Leeds and Ryan Road Northampton, MA 20 Roseville Interceptor Rosevi 11 e, MN
Interceptors 21 Denton SS Denton, TX
3 Woodhaven and Vinebrook Lexi ngton, MA 22 Lamesa SS Lamesa, TX
Sewers
4 Hudson SS Hudson, MA 23 Garl and SS Garl and, TX
5 Mi 1 ford SS Mil ford, MA 24 Co 1 dwa ter Creek Interceptor St. Louis County,
MO
6 Eatontown SS Eatontown, NJ 25 Liberty West Si de Sewers Liberty, "MO
7 Southside Interceptor Plainfield, NJ 26 Florence SS" Fl orence, CO
8 Cox Creek SS Anne Arundel County, MD 27 Boul der SS Boul der, CO
w 9 Division 0-22nd Street Richmond, VA
I 28 Casa Grande SS Casa Grande, AZ
. -...J Interceptor
10 Western Branch SS Portsmouth, VA 29 South Phoenix Interceptor S. Ph oe nix, AZ
11 Crane Creek SS Me 1 bourne, FL 30 South Interceptor Vallejo, CA
12 Lakeland SS Lakeland, FL 31 Demair Interceptor Turlock, CA
13 Southeast Interceptor Clearwater, FL 32 Vacavi 11e SS Vaca vi 11 e, CA
14 Tar River Interceptor Greenville, NC 33 Vancouve r SS Vancouver, WA
15 Upper Walnut Creek Ra lei gh, NC 34 North and South Interceptors Auburn, WA
Interceptor 35 Everett SS Everett, WA
16 Asheboro SS Asheboro, NC 36 Bennington SS Bennington, VT
17 McMullen Creek SS Charlotte, NC 37 Jamestown SS Jamestown, NY
18 E ffi ngham SS Effingham, IL 38 Farmington SS Fa rmi ngton, NM
* Sewe r Sys tem
-------
TABLE 3-3 (Continued)
CASE STUDY MAJOR PROJECTS
10 10
N umbe r Project Name Location Number Project Name Location
39 Vista/Carlsbad SS Vi s ta, CA 43* Oran ge SS . . Orange, TX
40 Mos cow SS Moscow, 10 44* Texarkana SS Texarkana, TX
41* South Amherst Amherst, MA 45* Northern Interceptor Livermore, CA
Interceptor 46* Greensboro SS Greensboro, AL
42* Ri chfi e 1 dlEdi na Ri chfie 1 d/Edi na,
Interceptor MN
* Not selected as one of the final 40 case study major projects
W
I
0:>
-------
. The Accelerated'Public Works Directory for 1964 [111]
and the EPA'Wastewater Contract Awards computer data
file, maintained by the EPA Municipal Construction
Division [101], both of which provided wastewater com-
pletion dates for all types of wastewater projects
funded under PL-87-658 and PL-84-660.
. The EPA 1968 Inventory, Municipal Waste Facilities [112]
which tabulates all municipal waste facilities in-
stalled by 1968, including non-federally funded pro-
jects. This publication provided data on completion
date, drainage basin, and plant design capacity and
initial average wastewater flow.
In addition to published data sources, the EPA Municipal
Construction ~ivision supplied information [113] on two wastewater facilities
that might be potential case study projects. The first of these involved three
sewage treatment plants for Lancaster, PA. Of these three projects, only one
was constructed in the time period of interest and this one unfortunately in-
volved only the upgrading of the treatment plant and not an extension of the
collection network. The second set of facilities resides in Chattanooga, TN
and involves two sewage treatment plants. Unfortunately, neither of these
projects was associated with an expanded collection network until 1964 when
construction on interceptor extension began. Thus, no case study projects
were selected from these data.
3.
Final Selection
Table 3-3 lists data for wastewater exogenous model variables
which had to be known to determine whether a given project satisfied the se-
lection criteria for a case study. Since the preliminary list of 46 projects
drawn from primary and secondary sources did not provide all such data, a
telephone survey of municipal departments of public works, regional planning
agencies and state soil conservation service offices was carried out to com-
plete the data shown in Table 3-3. In many instances values for the variables
PEAK and CAPACl could not be obtained and so estimates were used instead. In
these instances, the peak flow in a major project's collection system was es-
timated by scaling a known average flow by the population-dependent peaking
3-9
-------
factors given in Figure 2-6 of reference [100]. When the hydraulic design
capacity of the collection network had to be estimated, one of two different
techniques was employed. The most common estimation procedure was to calculate
CAPACl as follows:
CAPACl = PEAK * (1 + RECAP1/100)
The second method used was to sum estimated hydraulic flow rate for each major
pipe in the collection system. Hydraulic flow was determined from nomographs
of Manning's formula [100] which required pipe diameter and slope. In the
absence of data for the latter parameter, average values for hydraulic flow
based on pipe diameter were used [114].
, .
The telephone survey revealed that three major projects should
be dropped from consideration as case studies, since they no longer satisfied
the definition of major project. The first of these, #41 in Amherst, MA, only
has a design excess capacity in the base year of 5%, i.e., there was essentially
no excess capacity in the collection system. The second, #42 in Richfield, MN,
was completed in 1954, not 1961 as previously thought. The last project was
#46 in Greensboro, AL. Since this area has never been surveyed by the U.S.
Geological Survey, no maps were available and it would have been very difficult
to define an area of analysis and collect data for this case. At the
Project Officer1s direction, three additional case study projects were elimin-
ated (#45 in Livermore, CA; #43 in Orange, TX; and #44 in Texarkana, TX) to
form the final set of 40 case study major projects.
B.
DATA COLLECTION
1.
Transportation Data
Prior to the data collection effort, the GEMLUP-I VMT model [19]
was reviewed to identify what data were required to test and validate the
model. Thus, the purpose of this task was to determine the most appropriate
3-10
-------
spatial extent of data sources; the best available agency sources of data;
and to initially select variables for which data were to be collected for
GEMLUP-II. In performing this task, an initial choice of VMT case studies
was made.
a.
Spatial Extent of Data Sources
The basic question to be dealt with concerns the most appropriate
transportation planning geographic unit for analyzing and aggregating data.
The practice in transportation planning is to use either traffic districts
or traffic zones. Traffic districts are relatively large spatial units nor-
mally utilized in projecting regional travel. Traffic zones, on the other
hand, are spatial units roughly corresponding with census tracts. In examin-
ing each of these traffic data units, it was kept in mind that the range in
size of GEMLUP-II study areas would average from 20,000 to 40,000 acres with
some study areas as small as 5,000 acres. A second factor which was considered
was that it would necessarily be easier to aggregate data from smaller traffic
data units that it would be to disaggregate data from larger units; this con-
sideration is important in maintaining flexibility in the general VMT model.
In considering the use of data from traffic districts versus those
from traffic zones, it was concluded that traffic zones would best serve the
purposes of GEMLUP-II. First, traffic zones are more to scal~ with the selec-
ted study areas. Second, traffic zone population and economic data are more
detailed and correspond quite closely with Census Tracts. Third, the number
of traffic zones per study area is manageable with regard to the volume and
type of data they would generate. (For example, a recent study of a 38.8
square mile Southern California city of 24,832 acres contained 86 traffic
zones [115]). Lastly, it is generally agreed that some reliable trans-
portation models are built upon zonal data [116].
All of these factors clearly point to the conclusion that for
GEMLUP-II, the utilization of traffic zone data are far superior to the gener-
alized--and often incorrect--data that could be obta'ined from traffic districts.
3-11
-------
b.
Agency Sources of Data
The next basic questions associated with approaches to the VMT model
validation was the best agency source of data. It was initially suggested
that agencies associated with the so-called 3-C process be utilized. These
are transportation agencies operating under the Federal-Aid Highway Act of
1962 which seeks to develop transportation programs based on a IIcontinuing,
comprehensive transportation planning process carried on cooperatively in the
state and local communities [44].11 There are, however, several glaring
problems associated with using data from 3-C agencies. First, the process
is applicable only to areas with a population of 50,000 inhabitants or more;
the majority of VMT case studies initially selected have a population below
this figure. Second, unnecessary difficulty would be encountered in disag-
gregating informmation from 3-C agencies as to be applicable to sub-areas
covered by the case studies; this is verified by virtue of the fact that most
3-C agencies are State Highway Departments and their data commonly relate to
fairly large regional areas. Finally, experience has shown that 3-C agency
data vary widely in its accuracy and detail; no two 3-C agencies studies
follow identical methods or procedures; and the process outlined in the Fed-
eral Highway Administration (FHWA) Policy and Procedure Memoranda, PPM 50-9
are not always rigorously followed [117].
A more appropriate source of data was Regional Planning Agencies
covering areas in which the case studies are located. Of the case studies
initially selected for transportation data collection, all were within an
area covered by a Regional Planning Agency. These planning agencies were
founded at a point in time where it is highly likely that they would have
developed data under the appropriate study period of 1970 [118]. These Re-
gional Planning Agencies were also likely to have a finer level of data deal-
ing with transportation, population, and socia-economic characteristics.
It should be noted, however, that a given Regional Planning Agency
could not have all the data needed for validation of the VMT model. In such
a case, additional sources had to be contacted for data. These included a
3-12
-------
City or County Traffic Engineer for data dealing with hierarchy of streets,
traffic volume and design capacity, and vehicle speed. Also, County and City
Planning Departments had information on population, socio-economics, and land
use which were particularly useful in addressing trip productions and attrac-
tions on a zonal basis.
c.
Initial Choice of VMT Case Studies
Of the GEMLUP-II case studies listed in Table 3-3, 11 were selected
for transportation data collection. Data from these 11 areas were used in the
validation of the VMT model. They are:
VMT Case Studies
Lexington, Massachusetts
Richmond, Virginia
Charlotte, North Carolina
Clearwater, Florida
Roseville, Minnesota
St. Louis, Missouri
Boulder, Colorado
S. Phoenix, Arizona
Vallejo, California
Auburn, Washington
Bennington, Vermont
(test case)
The criteria associated with the selection of these VMT case studies were geo-
graphic representativeness, urban development pattern, socio-economic charac-
teristics, and the existence of a Regional Planning Agency under the study
period. These case studies range in population (popcom) from 1,000 to
201,564. In this way, a full range of cases were available for testing the
VMT model and determining the overall validity of it as a generalized model.
d.
Initial Choice of Variables
On the basis of the literature search and analysis of that litera-
ture, there were numerous variables which respective authors found useful in
their specific transportation studies. These variables differed according to
the particular purpose of their studies [116,119-123]. For example, Table
3-4 lists and ranks those variables which the U.S. Department of Transportation
3- 13
-------
TABLE 3-4
SELECTED VARIABLES FOUND SIGNIFICANT IN URBAN TRANSPORTATION PLANNING
A.
VARIABLES FOUND SIGNIFICANT IN ZONAL TRIP GENERATION
1.
2.
3.
Demographic Data
a.
b.
c.
d.
e.
Total Population
Age, Sex, Race, etc.
'Number of Household Units
School Enrollment
Family Life Cycle
Economi c Data
a.
b.
c.
d.
e.
f.
g.
h.
i.
j.
k.
1.
Total Employment (1 )**
Selected Employment (2)
Employment by Industry (3)
Employment by Residence (4)
Labor Force (5)
Labor Force by Occupation and Industry (6)
Median Income
Income Stratified
Automobile Ownership
Dwelling Units without Automobiles
Reta i 1 Sales
Average Home Value
Land Use Data
a.
b.
Specific Activities
Selected Categories
B. VARIABLES FOUND SIGNIFICANT IN DWELLING UNIT TRIP GENERATION
a.
b.
c.
d.
e.
f.
g.
h.
i.
j.
k.
1.
Ca r Owners hi p
Family Size
Number of Persons 5 Years old and dver in Household
Length of Residency
Fami 1 y Inc ome
Number of Persons 16 years old and over
Number of Persons 16 years old and over who drive
Age of the Head of Household
Distance,from CBD
Stage in Family Life Cycle (7)
Occupation of Head of Household
Structure Type
1*
3
1
2
3
1
1
3
1
3
3
1
3
1
2
2
3
3
1
1
1
1
3
2
1
1
2
3
1
1
1
* Key to Weights:
Data.
** Number referenced to notes
Source: [116J
1 = Essential Data, 2 = Desirable Data, 3 = Useful
appearing on the next page.
3-14
-------
TABLE 3-4 (CONTINUED)
SELECTED VARIABLES FOUND SIGNIFICANT IN URBAN TRANSPORTATION PLANNING
(1) Total Employment. All persons employed in the study
area, or employment by place of employment.
(2) Selected Employment. An employment grouping for the
study area that classifies workers by occupation,
industry, or class-of-workers. .
(3) Employment by Industry. All employment in the study
grouped by SIC, or employment by place of employment
or industry.
(4) Employees by Residence. All persons living in the
study area who are employed someplace.
(5) Labor Force. All persons living in the study area
who are either employed or unemployed.
(6) Labor Force by Industry. All persons living in the
study area who are either employed or unemployed, and
grouped according to industry of employment or experience
of unemployment.
(7) Stage in the Family Life Cycle. Based on age of the
head of the household, marital stamus, and age and
number of children.
f .
3-15
\
-------
found significant in transportation planning. In a study performed by the
Mi chi gan Department of State Hi ghways, the fo 11 owing vari ab 1 eswere determi ned
to be significant in predicting trips at the zonal level [123]:
1. Total Population
2. Number of !Me 11 ing Untts
3. Autos Available
4. Resident Labor Force
5. Total Employment
6. Manufacturing Employment
7. Government Emp 1 oyment
8. Retail and Wholesale Employment
9. Service Employment
10. Selected Employment
11. Retail Sales
Finally, a nationwide survey [116] ranked the importance of seven household
characteristics for trip generation as follows:
Variable
Rank
Regress i on Beta
Coeffi cient
Fami 1y Si ze
Car Ownership
Income
Stage in Family Life Cycle
Occupation of Head of Household
Density of Neighborhood
Distance from CBD
1
2
3
4
5
6
7
.29
.23
. 14
. 13
.11
. 10
Insi gni fi cant
As is self-evident from the above, the choice of variables differs from stuqy
to study. These lists and tables reflect the "biases" of individual studies.
In selecting variables for validating the overall GEMLUP VMT model, as well
as trip generation and trip length characteristics, the following three inter-
related factors must be considered:
1.
2.
3.
Pattern of land use (location and intensity);
Social and economic characteristics of the study
The type and extent of transportation facilities
study area.
area; and
in the
In examining these factors within the context of the overall GEMLUP VMT model,
the following variables were selected as being most appropriate for the
validation exercises:
3-16
-------
1.
2.
Pattern of Land Use
a.
Selected Categories of Land Use (acres, square feet)
by zone (tabulation} and by map
Social and Economic Characteristics
a.
b.
Total Population by Zone
Total Employment by Zone
Selected Employment by Occupation, Industry, or
Class-of-Workers by Zone
Number of Households (Dwelling Units Density) by Zone
Median Income by Zone
Stratified Income by Zone
Automobile Ownership by Zone (Registration and Fleet
Mix--car, truck, bus)
c.
d.
e.
f.
g.
3. Transportation Facilities
a. Map of Transportation Network (local, collector, arterial,
expressway)
b. Traffic Volume (ADTs) and Design Capacity for Each
Major Roadway
c. Average Trip Lengths for Each Major Roadway
d. Number of Interzonal Trips ~y Purpose per Day (home,
work, shopping, social-recreation, school, personal
business)
Vehicle Speed for Each Major Roadway
Zonal Trip Generation Rates by Land Use Activity.
Zonal VMT
e.
f.
g.
These variables are specifically identified as an assurance that all necessary
information for VMT model validation will be obtained. The rationale behind
this approach is that the data, methods, and procedures employed in areas
throughout the country vary considerably; for example, one cannot be confident
that information on trip generation rates are necessarily available nor that
this information was calculated with accuracy. Thus, it may be necessary to
estimate trip rates for a given purpose by dividing the number of interzonal
trips (3d) by zonal land use (la). Similarly, zonal VMT can be estimated by
summing over all major roadways the product of ADT (3b) and roadway length (3a).
The initial choice of VMT variables, therefore, may represent data that will
not be needed for each VMT model cas~ study, but it will ensure that enough data
are available to guarantee overall consistency in validation. In addition, this
3-17
-------
data will be invaluable in modifying the existing GEMLUP VMT model to incor-
porate more detailed considerations if the validation indicates a need for
such revision.
2.
Survey Forms
The list of data requirements developed in Phase I, the list of
data requirements for the GEMLUP-I VMT model validation, and the list of 40
case study wastewater major projects defined the data collection program.
These requirements were formalized in a set of survey forms. Survey forms
for the collection of data from all sources but U.S. Census publications are
reproduced in this section as Exhibit 3-1. Census data collection was per-
formed by the Project Officer and coded directly onto keypunch forms. Ques-
tions on soil suitability for buildings and septic tanks were included on
the regional planning agency form after discussions with the u.S. Soil Con-
servation Service revealed that their published soil data are available for
only a small fraction of the land area in the U.S. Planning agencies were
expected to have access to a larger set of unpublished soil data.
3-18
-------
OMB Clearance
No. 158-576007
EXHIBIT 3-1
REGIONAL PLANNING AGENCY SURVEY FORM
APPOINTMENT SCHEDULE
Dates
Time
Name
Organization
Address
Telephone Number
3-19
-------
EXHIBIT 3-1
REGIONAL PLANNING AGENCY SURVEY FORM
Name of Project
Year Completed (t)
Comnuni ty Served
County
Ri ver Basi n
Area of Analysis
acre s
USGS MAP AND ROAD ATLAS DATA (Try to compute these before visit to RPA)
1. Distance in miles from centroid of area of analysis
to centroid of nearest CBD (majority over 100,000
population) in year t
2. Railroad mileage in area of ' analysis in year t I I
3. Distance in miles from centroid of area of analysis I I
to centroid of nearest commercial airport
*4. Number of limited-access highway interchanges in I I
area of analysis in year (t+5)
*5. Number of limited-access highway interchanges in I I
the county in year (t+5)
*6. Non-limited-access highway lane miles in area of I I
analysis in 1970 (use worksheet)
SPECIFIC QUESTIONS (Show them map of area of analysis)
1.
Did a college or university exist in the area of
analysis in 1960?
What was the median price ($) of residential vacant
land in the area of analysis in 1960?
(Alt. source: Call local realtors, real estate
appraisers, visit county recorder for real estate
sale prices)
*
Verify with RPA the year of construction of interstate highways shown on
USGS maps or road atlas and count only those interchanges in service in
year (t+5) and those lane miles in service in 1970.
3-20
2.
3a.
How many passenger commuter rail stops were there
in the area of analysis in year t?
(Alt. source: Call local transit authority)
How many local bus stops (transit system) were there
in the area of analysis in year t?
(A 1 t. source: Ca 11 1 oca 1 bus company)
b.
c.
Sum (3a) and (3b)
-------
EXHIBIT 3-1
4a. How much vacant developable acreage existed in area
of analysis in year t?
b. . How much vacant undevelopable acreage existed in
area of analysis in year t?
(Alt. source: Land use map for 1960)
5.
How many acres in the area of analysis were zoned
for the following uses in year t?
Residential
Commerci a1
Offi ce
Industrial
6a. Which of the following is true with regard to
restrictions on residential on-lot sewage disposal
for the time period t to 19701 (Note: If more
than one applies, code largest number)
4 = on-lot disP9sa1 prohibited entirely
3 = on-lot disposal prohibited except on large lots
2 = on-lot disposal permitted but percolation tests
are required
1 = on-lot disposal permitted but package plants
prohi bi ted
o = no restrictions
b. What year did on-lot disposal restrictions go into
effect?
(Note: If no restrictions, code 1970; if year is
less than 1958, code 1958)
(Alt. source: Municipal health dept., DPW or sewer
authori ty
During the period t to 1970, were any government
policies in effect to limit the number of hookups
to the sewage system in the area of analysis?
7a.
1 = Yes
o = No
b. What year did such policies take effect?
(Note: If no policies, code 1970; if year is less
than 1958, code 1958)
(Alt. source: Municipal DPW or sewer authority)
3-21
-------
EXHIBIT 3-1
8. How much active, outdoor recreational acreage was there
in the area of analysis in 19701 This is defined to
include:
o Golf courses, driving ranges, miniature golf
o Tennis courts
o Ska ti ng ri nks
o Riding stables
o Ski and tobaggan slopes
o Athletic fields
o Play lots and playgrounds
o Swimming beaches and community pools
. Yacht clubs, marinas and boating access sites
. Dude ranches
This is defined to exclude passive open space portions
of large reservations, arboretums, parks~and conserva-
tion areas such as state/national forests, wildlife
refuges. .
9.
The following questions are required only if part of
the area of analysis is lunb10cked" (i .eo, not covered
by census tracts or blocks). In such cases, answer
these questions only for the unblocked portions of the
area of analysis.
a. Number of dwelling units in 1970 - take off of aerial
photos
b. What is the proportion of multifamily dwelling units
(3 or more households) in low rise and high rise
structures?
c. What is the proportion of single family dwelling units
in detached or attached (duplex) structures?
10.
Soil suitability data for area of analysis-acreage
with a "severe" soil type classification with
rega rd to"
a. Urban development (building sites)
b. On-lot sewage disposal (septic tanks)
(Alt. source: State Office of U.S. Soi1 Conservation
Service - see attached list).
3-22
% I Low
% I High
% I Detached
% I Attached
I
I
-------
EXHIBIT 3-1
HIWAYS WORK SHEET
Link Name
Length (miles) # of Lanes
3-23
Lane Mi les
-------
EXHIBIT 3-1
IN-HOUSE DATA COLLECTION FORM
Community Served by Major
Year of Completion (t)
Year of Federal Funding
Project
USGS MAPS
1.
Using available watershed boundary and wastewater collection network
maps for the major project, IIdrawll the boundaries of the area of
analysis on USGS 1: 24,000 quadrangle maps of the area. The
following definitions apply:
The area of analysis is defined by the watershed boundaries
of the area that drains, by gravity, to any point of the waste-
water collection network, plus any areas outside the watershed
connected by pumping stations and force mains. In the case of
essentially flat terrain (e.g., river deltas), the area of analysis
is restricted to the locus of points within 1,000 feet of any
interceptor or trunk sewer line.
2.
Compute the area of analysis in acres (refer to
list of case study projects as the variable AREA
has already been computed for some projects).
3.
Enter the name of the river basin which makes up
most of the area of analysis.
Compute the length of interceptor sewer lines
(in miles) transversing relatively undeveloped
land « 1 du per acre) in the area of analysis
in year t.
4.
5.
Compute the acreage of land within 5,000 feet of
any interceptor sewer line in the area of
analysis in year t.
SOMA DATA
1.
Enter the vacancy rate for office buildings in
the area of analysis in 1960.
%l
CONSUMER PRICE INDEX
1.
Enter the consumer prlce index for the
community served and year t.
2.
Enter the consumer price index for the
community served and the year of federal funding.
3-24
-------
EXHIBIT 3-1
SURVEY FORM OF LAND USE FROM AERIAL PHOTOGRAPHS
Community Served by Major Project
Aerial photographs for area of analysis (1970)
USGS 1:24,000 topographic map with boundaries of area of
analysis shown (1970 .
Land use map of area (1970) - included with some project
materials, otherwise ask regional planning agency if available
USGS 1:250,000 land use/land cover map - included with some
project materials
American Hospital Association Guide, 1973
Interview with regional planning agency.
1. Check scale of photograph by measuring known distance on USGS topographic
map and photograph. Indicate scale: 1 inch = feet
Resources:
2.
3.
Indicate year of coverage:
the number of grid squares equivalent to the
For the scale, determine
following areas:
1,000 ft2 =
50,000 ft2 =
100,000 ft2 =
4.
For the scale, determine the reference shadow length of a known l-story
s truc ture = inches
5.
For each structure, record the number of grid squares of roof area and
the number of stories (estimated from shadow length) on the worksheet in
the appropriate land use category. Apportion unknown uses equally among
probable categories as you are working and, separately, keep track of the
amount you apportioned in the miscellaneous column on the worksheet.
Note, you should ignore buildings in the following use categories:
. Resi denti a 1
. Recreational - Parks
. Transportation - Utilities
. Government - Public Assembly
. Universities - Trade Schools
6. Compute total building floor area in 1,000 ft2 for each land use category
and record on summary sheet.
3-25
-------
EXHIBIT 3-1
1,000 Ft2 Floor Area
Category
Defi nition
Conmerci a1
<50 K
50-100 K
> 1 00 K
Office-Professional
w I Manufacturi ng
I
N L
0\
Wholesale-Warehousing
I Public and Private
Schools
I Other
LUC*
52-59
62
64
66
61
63 (-637)
65
692, 699
2, 3
51
637
681
15
691
71, 79
Reta i 1 trade
Personal services
Repai r servi ces
Contract construction
services
Finance, insurance, real estate
Business services
Professional services (includes
Welfare - charitable services
hospota1s)
Manufacturi ng
Wholesale trade
Warehousing and storage services
Nursery, primary and secondary schools
Hote 1 s-mote 1 s
Churches
Cultural facilities
* Standard Land Use Codes, U.S. Urban Renewal Adm. and Bureau of Public Roads, 1965.
-------
EXHIBIT 3-1
W
I
I'\)
'"
COlTm!rci al Public and
<50 K 50-100 K >100 K Office-Professional Manufacturi ng Wholesale-Warehousing Private Schools Other
-------
EXHIBIT 3-1
TRANSPORTATION DATA COLLECTION FORM
Data is collected for only 10 of the 40 case studies.
For the a~ea of analysis circa 1970 determine if zonal transportation
studies have been done. If so, obta~n a map showing the traffic zones in
the area of analysis or mark those on your USGS map. If zonal data are
not available, gather the following data by the smallest geographical
unit available and obtain a map showing the boundaries of such units.
Collect data for the following variables. For each variable, record .the
values by zone on separate sheets of paper or Xerox the data directly from
reports and attach. No structured data form has been provided since the
number of zones may vary from 10 to 100 or more. .
1.
Pattern of Land Use
2
a. Land use (acres, ft ) by zone (tabular or map) for the follow-
ing categories: residential ~ conrnercial, industrial, wholesalei warehousing,
schools, office, recreation.
2.
3.
Social-Economic Characteristics
a.
b.
c.
d.
e.
f.
g.
Total population by zone
Total employment by zone
Emplo~ent by occupation by zone
Dwelling units per zone
Median income by zone
Strati fi ed income by zone
Automobile ownership by zone (registration and fleet mix -
car, truck, bus)
Transportation Facilities
a.
Map of transportation network (local, collection, arterial,
expressway)
Traffic volumes (ADT ) and design capacity for each major
roadway
Average trip length by purpose
Number of interzonal trips by purpose per day (home, work,
shopping, social-recreation, school, personal business)
Average vehicle speeds for each major roadway
Zonal trip generation rates by land use activity (same cate-
gories as la).
Zona 1 VMT
b.
c.
d.
e.
f.
Alt. Sources:
g.
3a, b, and e
3g
la, 2a-f
County or city traffic engineer
State department of motor vehicles
. County or city planning departments
3-28
-------
3.
Field Surveys
a.
Data Sources
The survey forms in the previous section were used by the Walden
staff in personal visits to each major project site to gather data from local/
regional planning agencies, municipal department of public works/sewer com-
missions/chamber of commerce, regional river basin commissions/water resources
councils, and state transportation (3C) planning agencies. The data collected
on-site consisted principally of information on the major project, land use in the
area of analysis (from aerial photographs, published reports, and special agency
unpublished reports), various regional parameters, watershed boundaries, and
VMT data for the GEMLUP VMT model validation task. We collected field data for
36 of the 40 cases; the Project Officer was responsible for the remaining four
cases. Transportation data were collected for 11 of the 40 case studies.
One key data item needed to define the area of analysis for each
case study was a watershed map for the area. There is no single source for
watershed maps nationwide. Thus, the combined sources listed below provided
adequate data:
The U.S. Geological Survey (USGS) has published a series
of "State Hydrologic Unit Maps" showing the drainage
basin boundaries, rivers and streams of all river basins
exceeding 700 square miles in area. These maps have
been prepared at a scale of 1:500,000 and were available
for 24 eastern states* and the District of Columbia.
Although at too small a scale to always identify the
drainage basins for case study major projects directly,
these maps provided guidance in this effort.
. Although the USGS does not print watershed boundaries
on their quadrangle maps, these were deduced from
elevations. Also, most USGS district offices
had local drainage area maps which could be reviewed
on-site, but not purchased.
. In some states, Regional River Basin Commissions, Water
Resources Councils or Water Districts had done water-
shed mapping as part of their river basin studies.
*Alabama, Florida, Georgia, Illinois, Indiana, Maine,
Massachusetts, Rhode Island, Connecticut, Minnesota,
Vermont, New Jersey, New York, North Carolina, Ohio,
Carolina, Virginia, West Virginia, Wisconsin.
Maryland, Delaware,
Michigan, New Hampshire,
Pennsylvania, South
3-29
-------
Supplementing the aerial photographs used to estimate land use
in the areas of arialysis were USGS land use/l~nd cover maps at a scale of
1:250,000 for certain portions of the country. Overlay maps with hydro-
logical units and/or census tract boundaries were also obtained. Each of
these maps required a USGS topographical map for reference purposes. Al- .
though these land use maps could not be used as a direct source of land use
data due to the small map scale and the fact that the roof area of urban
structures is not shown, with overlays of hydrological unit information,
they provided guidance in identifying the type of urban land use for parcels
in the area of analysis of a major project. In addition, an overlay of cen-
sus tract boundaries on land use helped provide an inventory of all census
tracts contained in the area of analysis of a major project. This informa-
tion was needed for the census data collection effort.
The land use/land cover maps published by the USGS are based on
remote sensing data, viz. LANDSAT satellite images and high-altitude, color-
infrared aerial photographs. Using these sources, data are presented at the
two levels of detail shown in Table 3-5. It is noteworthy that the USGS
classification system {s resource oriented (i.e., it emphasizes land cover),
while, by contrast, the Standard Land Use Coding (LUC) Manual [109] is people
oriented (i.e., it emphasizes land use). Still, there is a degree of compat-
bility between the first level categories of the Standard Land Use Coding
Manual and the second level categories under "Urban or Built-up Land" of the
USGS classification system. Since the GEMLUP endogenous land use variables
are defi ned in terms of LUC categori es, the correspondence between the" USGS
and GEMLUP classification systems can be determined. This comparison is
given in Table 3-6 which shows that USGS categories 11,13, and 14 can in-
dicate the dominance of one specific GEMLUP land use in a given area. The
other USGS categories, although less informative because they indicate the
dominance of anyone of several possible GEMLUP land use types, still pro-
vided some guidance in the land use identification process.
3-30
-------
TABLE 3-5
US GEOLOGICAL SURVEY LAND USE AND LAND COVER CLASSIFICATION SYSTEM
Code
Level I
Description
. Code
Urban or Bu i 1 d-u p Land
11
12
13
14
15
16
17
2
Agricultural Land
21
22
3
23
24
31
32
Rangeland
4
33
41
42
43
51
52
53
54
61
62
71
72
73
74
75
76
77
81
82
83
84
85
91
92
Forest Land
5
Water
6
7
Wetland
Ba rren Land
8
Tundra
9
Perennial Snow or Ice
Leve 1 II
Description
Residential
Commercial and Services
Industrial
Transportation, Communi-
cations, and Utilities
Industrial and Commercial
Compl exes
Mixed Urban or Build-up
Land
Other Urban or Build-up
Land
Cropland and Pasture
Orchards, Groves, Vine-
yards, Nurseries, and
Ornamental Horticultural
Area s
Confined Feeding Opera-
tions
Other Agricultural Land
Herbaceous Rangeland
Shrub and Brush Range-
land
Mixed Rangel and
Deciduous Forest Land
Evergreen Forest Land
Mixed Forest Land
Streams and Canals
Lakes
Reservoirs
Bays and Estuaries
Forested Wetland
Nonforest Wetland
Dry Salt Flats
Beaches
Sandy Areas other than Beaches
Bare Exposed Rock
Strip Mines, Quarries, and
Gravel Pi ts
Transitional Areas
Mixed Barren Land
Shrub and Brush Tundra
Herbaceous Tundra
Bare Ground Tundra
Wet Tundra
Mi xed Tundr.a
Perennial Snowfields
Glaciers
Source: [124].
3-31
-------
TABLE 3-6
COMPATIBILITY BETWEEN THE U.S. GEOLOGICAL SURVEY AND GEMLUP LAND USE CLASSIFICATION SYSTEMS
U.S. Geological Survey
Code Description
GEMLUP
Description
11
Residential
=
Res ident ia 1
12*t
Commercial and services
=
Commerc ia 1
Office-Professional Services
Wholesale-Warehousing
Educational
Active, Outdoor Recreation
Other Urban Land Uses
Manufacturing
Non-expressway highway
. 13
14**
W
I
W
N
15*
Indu stria 1
Transportation, communications,
a nd uti 1 it i e s
Industrial and commercial
compl exes
=
=
=
Commercial
Office-Professional Services
Wholesale-Warehousing
Manufacturing
Residential
Commerc i a 1
Office-Professional Services
Wholesale-Warehousing
Manufacturing
Educat ion
Active, Outdoor Recreation
Other Urban Land Use
16*
Mixed urban or built-up land
=
17t
Other urban or built-up land
=
* USGS codes 12, 15 and 16 also include LUCs 67 (Governmental Services), 682 (Colleges) and 683 (Trade
Schools), not in GEMLUP
** USGS code 14 includes all of LUC 4, not just 45 (Highway and street right-of-way)
t USGS codes 12 and 17 also include LUC 76 (Parks), not in GEMLUP
Source: [}24]
. .
r:J
c'"
-------
b.
Test-Training Case
As a prelude to the actual data collection, a training-test case
was undertaken involving the Project Officer, the Walden program manager, and
all field personnel. The objectives of the training case were to:
. Collect data for one of
. Test the feasibility of
data,
. Verify the time estimates developed for the data
collection, task,
. Provide training to the field personnel, and
. Isolate any potential problem areas for solution.
the 40 case studies,
collecting the requisite
The actual data collection process was then initiated and required in excess
of 20 man-weeks of effort to complete.
c.
Results
Early on in the data collection, at the Project Officer's direc-
tion, data collection for soil suitability (item #10 on the Regional Planning
Agency Survey Form) was discontinued. Field staff observed that urban
development in the period 1960-1970 occurred on land having a Iisevere" soil
type classification without regard to suitability for urban development. Since
such classifications did not really restrict urban development in the study
period, and since soil restrictions are probably better in explaing why a
sewage treatment plant is constructed rather than the development that
follows, SOILS and ONLOT were not appropriate exogenous variables. for the
model. Thus, the data needed to represent these variables was not collected.
Additional reasons for not collecting the data were difficulty in obtaining
data for just the area of analysis, lack of U.S. Soil Conservation Service
surveys in most regions and lack and comparability of soil classifications
in different regions.
The transportation data collected for 11 of the 40 case studtes
was examined and the data available (and thus obtained) is summarized in
Table 3-7. The data are slightly less consistent than shown since variable
definitions often varied between case studies, in terms of categories and
uni ts .
3-33
-------
TABLE 3-7
SUMMARY OF GEMLUP CASE STUDY TRANSPORTATION DATA
Variables for Which Data Were Available and Collected (See Page 3-17)
Transportation 1 2 3 Zonal
Case Studies a b c d e f g a b( 1) b(2) c d e f g Map
1
Lexington, MA .; .; .; .; .;
2
Benn ington, VT '; .; .; .; .; .;
6
Ri chmond, VA .; .; .; .; .; .; .; .; .; .; .; .; .;
3
Charlotte, NC .; .; .; .; .; .; .; .; .;
5
Clearwater, FL .; .; .; .; .; .; .; .; .; .; .; .; .; .; .;
Rosev;l1e, MN .; .; .; .; .; .; .; .; .; .; .; .;
It It 5
St. Louis, MO .; .; .; .; .; .; .; .; .; .; .; .; .; .;
v..> 5
I Boulder, CO .; .; .; .; .; .; .; .; .; .; .;
I~ 5 8
S. Phoenix, AZ .; .; .; .; .; .; .; .; .; .; .; .;
I Vallejo, CA .; .; .; .; .; .; .; .; .; .; .;
I 7 8
Auburn, WA .; .; .; .; .; .; .; .; .; .; .;
.1 Substitution for Hudson, MA. 5 For area of analysis as a whole.
2 Extra transportation case study. 6 Units are in elapsed time, not distance.
3 Work trips only. 7 For all trips as a whole.
It Major arterials only. 8 Trip generation equations for area as a whole by
tri p purpose, not land use.
-------
Missing data from U.S. Census publications caused four separate
problems. First, the lack of data for encost eliminated ENERGY from the
model. Second, the lack of data for copop1 eliminated DENCHG. However, a
similar variable named PERCHG was introduced as a replacement. Third, 9 of
40 values for wwemp were unavailable. Lack of complete data would have elim-
inated two important variable WWJOB and RELWW from the mode1*. Thus, estimates
for these nine values were obtained from SMSA data as follows:
wwemp = rwwemp (acre/areal)
Finally, missing data was a problem in 1970 census tract housing data by
family unit size (variables unit34, unit 5, and unit50). These missing data
did not effect the path analysis, but reduced the sample size for the varia-
bles MF and SFATT used in the Phase IV disaggregation analysis of the
predictive equations.
4.
Create Computer Data File
This task involved error checking and loading of the GEMLUP-II data
files onto our computer system. Keypunched cards of data were loaded onto
disk files and a program was written and executed to transform these raw
data into a file of the variables chosen for the initial path model. Next
each variable was analyzed using the CONDESCRIPTIVE command of SPSS and the
data statistics (e.g. variable means and extremes) examined for errors.
Several errors in the coding forms and the transformation program were dis-
covered. All errors detected in this quality assurance check were corrected.
The final GEMLUP-II data file is reproduced in Appendix E and the output of
the final CONDESCRIPTIVE run is given in Appendix F. In order to test the
initial path model in a non-linear, multiplicative form, a second data
file was created. This file is identical to the first except that the
natural logarithm of every data value was computed and stored.
*
Although SPSS has the capability to handle missing data in statistical
analyses, TSP does not and TSP is eventually used to analyze each land use
category.
3-35
-------
IV.
PHASE III - CAUSAL ANALYSIS
The principal objective of the third phase of this study was to test
the initial causal model using the set of case study data and the statistical
techniques of path analysis. The approach involved determining which of the
hypothesized causal relationships were significant and selecting the parameters
for the final causal model. The final path analytic model for induced land
use was also used to trace the direct and indirect effects that variables
have on each other and summarize the total net effects of such relationships.
The final task was to validate the GEMLUP-I VMT model [19] using data col-
lected from 11 of the c~se study major projects.
A.
PATH ANALYSIS
The objective of this task was to test the initial causal model
developed in Phase I with the statistical techniques of path analysis. A
full discussion of the concepts, assumptions, and problems associated with
path analysis and its applicability to the current study is contained in
Appendices A and B. Two statistical techniques were used to verify the
hypothesized path models of induced land use and to determine model param-
eters: two-stage least squares and ordinary least squares multiple re-
gression. The first technique was used to solve for path coefficients in
the system of non-recursive equations connected by feedback loops. To
solve for path coefficients in the other model equations that are not inter-
connected, ordinary least squares regression techniques was employed. All
variables were standardized (to a mean of 0 and standar~deviation of 1)
prior to the path analysis. Thus, the analysis yielded standardized path
coefficients (8 weights) as the model parameters. This approach was selected
since standardized path coefficients (8 weights) provide an effective way of
comparing the relative importance of various causal variables, independent of
different units and scales. Thus, path coefficients can be used to judge
the significance of model paths in theory trimming, the second task of
Phase I I 1.
4-1
-------
Two different computer software packages were employed for the
statistical analysis in this task. The first of these is the Time Series
Processor (TSP) developed at Harvard University and the Massachusetts
Institute of Technology (MIT) [125]. TSP'is a programming language oriented
towards the statistical analysis of time series, with specific applications
to econometric research. TSP was used for all two stage least squares
analysis in the GEMLUP study. The Statistical Package for the Social
Sciences (SPSS) [92] was also used in the path analysis due to the compre-
hensive system of programs it provides for statistical data analysis.
1, .
Preselection of Exogenous Variables
The initial path analytic model consists of 9 structural
equations, of which 7 are simultaneous and 2 are recursive. The complete
model contains 9 endogenous and 32 exogenous variables. An additional 34
exogenous variables serve as alternatives to those specified in the initial
model. Before testing the initial model, it was necessary to choose between
the prime and alternative independent variables in each equation. The selec-
tion was based on the results of a zero-order correlation analysis for all
model variables (see Appendix G). The selection process used was as follows:
. First, eliminate those independent variables whose correlation
with the dependent variable is opposite in sign to that
hypothesized. If all signs were incorrect, the prime variable
was retai ned.
. Then, compare correlation values of the remaining independent
variables and select the variable with the highest absolute
value. If two or more variables have similar correlation
values, the one thought to have the strongest causal relation-
ship was chosen.
The effects of this task were to substitute many alternative variables for
the prime independent variables in the initial model. A complete discussion
of variables in model equations is given in section IV.A.5.c.
4-2
-------
2.
Identifi cation
In order to obtain consistent estimates of the model coefficients
in the two-stage least squares analysis, it was necessary that each equation
satisfy the condition of identification which ensures an adequate number of
instrumental variables. This condition is stated as:
m > q- 1
o -
where:
m = the number of instrumental variables, i.e., the
o number of exogenous variables which do not appear as
independent variables in an equation
q = the number of independent, endogenous variables
in an equation
Due to the substitutions of independent variables, the total set of instrumental
variables in the simultaneous equation block increased to 32 (see Table 4-la).
This increase reduced the available degrees of freedom for model testing of
anyone simultaneous equation to 7 (=40-32-1). This still exceeds the mini-
mum of 6 degrees of freedom thought to be necessary to obtain meaningful
results in the initial path analysis. As variables were trimmed, the set of
instrumental variables eventually was reduced to 12 (see Table 4-lb).
Application of the identification condition to each initial
simultaneous equation indicates all are properly identified (see Table 4-2).
The identification condition was reapplied at each step of the theory trimming
task (see Section IV.A.5).
3.
Multicollinearity
To identify potential problems of multicollinearity, the zero-
order correlation~ between independent variables in each simultaneous
equation were examined. Values of Irl ~ 0.80 usually indicate a problem.
4-3
-------
TABLE 4-1a
INSTRUMENTAL VARIABLES USED IN THE INITIAL
TWO-STAGE LEAST SQUARES ANALYSIS
CAPACl
LENGTH
PHASE
TLIMIT
PERCHG
PERCAP
POOR
NONHH
LAND
STAY
AREA
RZONE D
DRIVE
JOBS
UNIV
TRANS
OZONED
TIME
POPDIF
POP DEN
VACOFF
PEAK
r~ANJOB
RRMILE
ACCESS
I ZONED
LIMITS
JOBCHG
EMPOP
DISCBD
VALUE.
KIDS
TABLE 4-1b
INSTRUMENTAL VARIABLES USED IN THE FINAL
TWO-STAGE LEAST SQUARES ANALYSIS
RECAPl
NONHH
POP DEN
PERCHG
LAND
RZONED
DRIVE
ACCESS
CZONED
JOBS
INCOME
OZONED
4-4
-------
TABLE 4-2
CONFIRMATION OF SUFFICIENT IDENTIFICATION IN THE INITIAL
TWO-STAGE LEAST SQUARES ANALYSIS
Dependent
Variable
Number of
Ins trumenta 1
Variables (mo)
Number of Independent
Endogenous Variables
(q)
Suff1 c1 ency
Condition
m > q-l
o -
RES 18 4
COMM 25 3
OFFICE 24 4
MANF 21 1
HIWAYS 26 4
EDUC 25 1
REC 27 2
18 ~ 3
25 ~ 2
24 ~ 3
21 ~ 0
26 ~ 3
25 ~ 0
27 ~ 1
4-5
-------
Examination of the correlation output shown in Appendix G revealed that most
values of Irl are less than 0.30, with two instances where it approached or
exceeded 0.80. The first of these occurred between MANF and COMM (r = 0.90).
This pair of independent variables appears in three equations (OFFICE, HIWAYS
and WHOLE). The second case of multicollinearity occurred between RES and
EDUC (r=0.78), both independent variables in the REC equation.
There are two solutions to a problem of multicollinearity.
The first is to form a single composite variable which represents both
variables (e.g., using factor analysis), while the second is to simply
eliminate one of the pair. The first approach, if used, would not allow a
complete statement of causality since composite variables are often hard to
interpret physically. Thus, the second approach was used, i.e., one of the
two intercorrelated variables was removed, prior to the first path analysis.
In the OFFICE equation, the following simple correlations
are relevant:
rOFFICE, MANF = 0.69
rOFFICE, COMM = 0.71 rRES, COMM =
rHIWAYS, MANF = 0.30
rHIWAYS, COMM = 0.39
rRES MANF = 0.51
, .
0.50
These statistics reveal that while COMM is more highly associated with
OFFICE than MANF; COMM is also associated with slightly more redundancy* in
the equation. Causally, MANF is secondary to RES as a market generator for
Office-Professional services, while COMM is the primary locational factor.
Thus, MANF was eliminated from the equation.
relevant.
In the HIWAYS equation, the above listed correlations are
These statistics reveal that COMM is more highly associated with
*
As exhibited by correlation with other independent variables, such as RES
and HIWAYS in this example.
4-6
-------
HIWAYS than MANF, and that both are associated with about equal redundancy
in the equation. Causally, both. vartables represent generators of motor
vehicle traffic, with the relationship Between COMM and HIWAYS thought to be
slightly stronger. Thus, MANF was eliminated from the equation.
In the WHOLE equation, the following correlations are
relevant:
rWHOLE, MANF = 0.42
rWHOLE, COMM = 0.25
rHIWAYS, COMM = 0.39
rHIWAYS, MANF = 0.30
These statistics reveal that MANF is more highly associated with WHOLE than
COMM, and that COMM is associated with more redundancy in the equation.
Causally, both represent a portion of the same market cycle in which
wholesale-warehousing development occurs. Thus, eOMM was eliminated from
the equation.
In the REC equation, the following correlations are
re levant:
rREe, RES = 0.42
rREC, EDUC = 0.34
rRES, POOR = 0.26
rEDUC, POOR = 0.53
These statics indicate that RES is more highly associated with REG than EDUG,
and that EDUC is associated with more redundancy in the equation. Causally,
EDUe is secondary to RES as a generator of demand for recreational space.
Thus, EDUe was eliminated from the equation.
Testing for multicollinearity was also performed after
the initial path analysis by identifying any unusually large (in absolute
value) beta weights and standard errors of the model coefficients.
4.
Suppressor Variables
Suppressor variable problems generally occur in simultaneous
systems when an independent variable is included which ;s causally unrelated
4-7
-------
.-. -~ ~.~--_._--- "-~ -~--~
to the dependent variable of the equation in which it appears. To identify
potential problems, the zero-order correlations between dependent and inde-
pendent variables were examined. Correlations not significant at the 20%
level (i .e., Irl < 0.14) were assumed to indicate a potential problem. A
list of the potential problem variables is given in Table 4-3. It is not de-
sirable to trim these variables from the model using the simple correlations
as the only guide; intercorrelations may be masking significant relationships
between variables. Thus, at this point, it was sufficient to note where
problems may occur so that they could be addressed in the theory trimming of
the mode 1 .
5.
Theory Trimmi-ng
The objecti ves
results of the initial path
nificant and redundant, and
of this task were. to examine the statistical
analysis to determine which paths are insig-
to trim these paths from the model.
a.
Approach
Our approach to the execution of this task was as
follows. First, a set of statistical criteria were developed to determine
which paths should be trimmed from the structural equations. Next these
criteria were applied to the results of the initial path analysis and the
model refined. Additional path analyses were then performed until a final
path model was determined. Thus, the trimming procedure was an iterative
process. For example, at each step the remaining path coefficients were
examined to see how the deletion of a path affected the ability of the model
to reproduce the original observed correlations; this is particularly
critical in two-stage least squares. .
In deciding what paths to drop, we did not choose one
fixed significance level (e.g., five percent) against which to compare the
model t and F statistics. In the two stage least squares analysis, the
values calculated for a t-statistic can only approximately* be interpreted
as being distributed as t, due to the first stage estimation of all feed-
back loop variables. Thus, it was necessary to use a more general rule in
*
Asymptoti cally
4-8
-------
TABLE 4-3
INDEPENDENT VARIABLES NOT SIGNIFICANTLY
CORRELATED WITH DEPENDENT VARIABLES.
IN THE SIMULTANEOUS BLOCK OF EQUATIONS
Dependent Variable
Independent Variables
RES
COMM
OFF! CE
MANF
HIWAYS
EDUC
REC
WHOLE
OTHER
STAY, LENGTH, PHASE
POPDIF, TIME, UNIV, OZONED
TIME, JOBCHG, VACOFF
LAND, LIMITS, LENGTH, TIME
ACCESS, DRIVE, DISCBD, UNIV, EMPOP, TIME
PERCHG, NONHH, TIME
UNIV
LAND, COMCHG, TIME
TRANS, 1NCOME, EMPOP, UNIV, TIME
4-9
-------
interpreting these statistics. In ordinary least squares analysis~ an ac-
cepted guidel~ne is that if t orF* is greater than 1.0~ the variable being
tested raises the adjusted R2 of the equation~ i.e.~ it explains a signifi-
cant additional amount of the variance of the dependent variable. Con-
versely~ a't or F less than 1.0 indicates that the variable in question
lowers the adjusted R2 of the equation.
The adjusted R2 is defined as:
R2 = R2 - (k-l) (1-R2)
a N-k
where:
k = number of independent variables
N = number of data samples.
The conventional R2 statistic can yield deceptive results when the signifi- '
cance of independent variables isin question. For example~ simply adding
a variable to any regression equation~ whether it is at all correlated with
the dependent variable or not~ will raise the R2 and indicate additional
variance has been expla ined. The R2 a gives a more conservative~ unbiased
estimate of the amount of variance explained in the dependent variable
through the regression equation.
The path coefficient (6) indicates the expected stan-
dard deviation change in the dependent variable given a one standard devia-
tion change in a particular independent variable~ all other effects held
constant. Comparing the 6's for different variables allows one to evaluate
the relative importance of the various independent variables because of the
standardization. In general~ 6 < 0.1 indicates a variable is not contri~
buting importantly to the regression equation.'
In trimming the path analysis model then~ the follow-
ing tests were applied:
* 2
F(1~N-k-1) = t (N-k-l)
4-10
-------
. A test to ensure statistical significance.
I tl of F ~ 1. 0 .
. A test to ensure independent variables
make a significant contribution, 8 ~ 0.1.
In addition to the above, the following criteria were also applied:
. A test to ensure a maximum value for the
model parameters, 8 < 1.0, indicating no
redundant paths through excessive inde-
pendent variable correlation.
. A test to ensure the sign of the path co-
efficient (8) is correct and not counter
to the original model hypothesis.
. A test to ensure a path being trimmed does
not totally eliminate an exogenous or in-
strumental variable from the model which
is deemed a priori to be of substantive
causa 1 importance.
b.
Results
The results of the final path analysis are shown in the
form of the final causal path diagram in Figure 4-1. The path coefficients
. 2 . h -...
(8) are shown on each path and the R statistic for each equation displayed
in the box of the associated dependent variable. The results indicate that
the final model explains the more than half of variance in case study data
with R2 values ranging from 0.27 to 0.82 and averaging 0.54. The residuals
of the final regressions do not exhibit any trends or patterns. indicating
the remaining unexplained variance (1-R2) is not due to poor specification
of the model, but rather due to wide variance in the case study data (i.e.
the problem of trying to develop one generalized model for a broad range.
of situations). Several of the equations do have one or two very large
residuals which account for 30-50% of the unexplained variance by them-
selves (see Table 4-4). An examination of these data points did not uncover
any errors in input data, but it did reveal that most large residuals cor-
respond to the highest or second-highest values for the dependent variable
and that these are all more than three standard deviations away from the
mean, i.e. they are extreme values. An examination of the distributions
4-11
-------
J
0.15
0.45
OTHE R
0.51
0.12
OFFICE
0.72
0.49
0.50
COMM
0.63
0.18
0.42
0.21
0.51
REC 0.25 RES 0.21 0.27 HIWAYS -0.12
.l::" 0.30 0.82 0.41
I
'-" 0.33 0.48 0.32
N
0.23
0.27
EDUC
0.70
0.34
0.26
MANF
0.54
0.22
0.31
0.32
WHOLE
0.27
I
I>
FIGURE q-l
FINAL PATH ANALYTIC MODEL
-------
TABLE 4-4
EXTREME VALUES WHICH ACCOUNTED FOR
A SUBSTANTIAL PORTION OF UNEXPLAINED VARIANCE
IN THE PATH ANALYSIS AND PRODUCED LARGE RESIDUALS
Dependent Case Study Number Percent of Unexplained
Variable of Extreme Value Variance Caused by Extreme Value
C
-------
for the dependent variables involved (all but RES and OFFICE) showed that
they were skewed by these extreme values. Thus, there is a statistical
basis for excluding these points from the model. Such exclusion would,
however, restrict the model's applicability to a much narrower range of cir-
cumstances. Since it is the objective of this study to develop a generalized
model, it was decided not to exclude the extreme value data. The complete
statistical output of the final path analysis, including residual plots,
is given in Appendix H.
In the course of trimming individual equations, only
one instance of multicollinearity (i.e., Isl~l.O) was observed. This in-
volved the independent variables POPDEN and OFFJOB which appeared together
in the COMM equation in the third path analysis. The intercorrelation of
these two variables was strong (R = 0.93) and caused instabilities in the
COMM equation. Of the two, POPDEN was retained because of its stronger
causal and statistical relationship with COMMa
Of the potential suppressor variables uncovered
through examination of the zero-order correlations, only five were not
eliminated during the normal course of theory trimming. To test whether
these five variables might be causing errors in the coefficients of other
model variables, the equations in which they appeared were re-estimated,
excluding them, to see if such changes did occur. In the case of three
variables, STAY in the RES equation, POPDIF in the COMM equation and
LENGTH in the MANF equation, very large changes did occur and so these sup-
pressor variables were deleted (as discussed below). The exclusion of the
other two variables, ACCESS in the HIWAYS equation and INCOME"in the OTHER
equation, produced changes in other model coefficients of 9% or less. Thus,
it was concluded that a suppressor variable problem did not exist with
these two and so they were retained.
c.
Discussion of Individual Equations
The actual path analysis was carried out using variables
defined in English units. The path regression coefficients shown in the
4-14
-------
printout in Appendix H reflect this fact. The path coefficients (S) shown
in Figure 4-1 are independent of the units chosen. The summaries of individual
equations in this section give the path regression coefficient for each
final model equation in both English and metric units.* The conversion to
metric involved only the dimension of distance. The following conversion
factors were used:
mi1e2 + km2, multiply by 2.589
1,000 'fl + m2 , multiply by 92.90
acres + m2 , multi ply-' by 4,047
miles + km , multiply by 1.609
acre-1 + m-2 divide by 4,047
mile-l + km-1, divide by 1.609
In addition to equation coefficients, the following discussions also list
the R, R 2 and coefficient of variation** for each final equation. The latter
statistic is a measure of the precision, of each model equation.
(1) RES Equation. In the first path analysis, REC,
LENGTH, POOR and AREA were trimmed (low S,t statistics or incorrect sign).
The sign for OFFICE was negative, counter to the original hypothesis.
OFFICE was kept however because the statistical relationship with RES is
very strong and can be explained if it is assumed that when a large amount
of office development occurs, it sets the character of the area for further
office/inddstrial development and not housing. The sign for LAND was positive,
counter to the original hypothesis. LAND was kept ,though because the statis-
tical relationship is strong and it is plausible that vacant developable
land, which is on the edge of development, will have already risen in value
in comparison to other parcels of land. Thus, residential development will
more likely occur in areas where land prices are higher. The variable
VACANT was added to replace AREA, representing the important causal factor
*
A list of model variables giving both
in the following page. .
**
Standard error of the regression as a
mean.
English and metric units is given
percentage of the dependent variable
4-15
-------
------,-------- -
LIST OF CAUSAL MODEL VARIABLES AND
CORRESPONDING ENGLISH AND METRIC UNITS
Variable
Name
English
Units
Metric
Units
. RES
COMM
OFFICE
MANF
HIWAYS
EDUC
REC
WHOLE
OTHER
RECAPl
NONHH
POPDEN
PERCHG
LAND
RZONED
DRIVE
ACCESS
CZONED
JOBS
INCOME
OZONED
RRMILE
MANJOB
PEAK
JOBCHG
GOVT
KIDS
STAY
VALUE
POOR
POPDIF
WWJOB
Dwelling units per 104 acres of land
103 ft2 floor area per 104 acres
324
10 ft floor area per 10 acres
103 ft2 floor area per 104 acres
Lane mi 1 es per 104 acres
324
10 ft floor area per 10 acres
Acres recreational land per 104
acres
103 ft2 floor area per 104 acres
103 ft2 floor area per 104 acres
Percent** collection sys reserve cap.
People per mile2 of land
People per mile2 of land
Percent + change in county pop.
Unitless (land cost/regional income)
percent+ of area zoned residential
100s of people per mile2 of land
Interchanges per mile2 of land
percent+ of area zoned commercial
People per mile2 of land People
Unitless (tract family inc./county fam. inc.)
Percent + of area zoned offi ce
Miles of track per mile2 of land
People per mile2 of land
Million gallons per day (mgd)
People per mile2 of land
Millions of dollars (106 $)
People per 100 dwelling units
Perc.ent + of fami 1 i e~ 1 n. ,same house
$ (median value of housing)
percent+ of families with low income
percent+ change {n regional pop.
People per mile2 of land
Dwelling units per 104 m2 of land
242
m floor area per 10 m of land
242
m floor area per 10 m of land
2 4 2
m floor area per 10 m of land
Lane km per 104 m2 of land
242
m floor area per 10 m of land
m2 recreational land per 104 m2
of land
2 4 2
m floor area per 10 m of land
m2 floor area per 104 m2 of land
*
2
People per km of land
People per km2 of land
*
*
*
100s of people per km2 of land
2
Interchanges per km of land
*
. 2
per km of land
*
*
km of track per km2 of land
People per km2 of land
*
People per km2 of land
*
*
*
*
*
*
People per km2 of land
*
Same as English units.
**
A value of 10% is inserted as 10.
tA value of 10% is inserted as 0.1.
4-16
-------
of land constraints. The trimming of REC made the EDUC and REC equations
recursive in subsequent analyses.
In the second path analysis, CAPAC1, TLIMIT, and VACANT
were trimmed (low s,t statistics or incorrect sign). The variables
RECAP1 and LIMITS were added as replacements, respectively. Also, the varia-
bles JOBS and SERVED were added to represent the important causal factors
of employment characteristics and major project sewer services.
In the third path analysis, LIMITS, PHASE and SERVED were
trimmed (low B,t statistics). In the fourth path analysis, MANF, OFFICE,
PERCAP and JOBS were trimmed (low s,t statistics). CAPCHG and UNEMP were
added to replace the latter two. Also, STAY was trimmed because of suppressor
variable problems, i.e. its simple correlation with RES is insignificant
and its inclusion in the equation causes instabilities in other model co-
efficients (e.g. betas change by more than 100%). The trimming of MANF
makes the MANF equation recursivp..
In the fifth path analysis, CAPCHG and UNEMP were trimmed
(incorrect sign and low 8,t statistics) and the final equation became:
Metric
Unstandardized
English RES = 40.36 HIWAYS + 12.89 RECAP1 - 7.244 NONHH
+ 2.007 POPDEN + 10,419 PERCHG + 2416 LAND
+ 2798 RZONED + 892.8 DRIVE - 5,444
RES = 25.08 HIWAYS + 0.003185 RECAP1 - 0.004634 NONHH
+ 0.001284 POPDEN + 2.574 PERCHG + 0.5970 LAND
+ 0.6914 RZONED + 0.5712 DRIVE - 1.345
Standardized
RES = 0.21 HIWAYS + 0.32 RECAP1 - 0.19 NONHH
+ 0.48 POPDEN + 0.40 PERCHG + 0.22 LAND
+ 0.13 RZONED + 0.41 DRIVE
R = 0.91, R2 = 0.82, Coefficient of Variation = 0.32
4-17
-------
(2) COMM Equation. In the first path analysis, RES,
PERCAP, JOBS, UNIV, TRANS and TIME were all trimmed (low S,t statistics and/
or incorrect sign). Three new variables were selected from the list of al-
ternatives to replace trimmed variables representing the important causal
factors of income, accessibility and zoning. These variables are INCOME,
ACCESS and ClONED.
In the second path analysis, INCOME was trimmed (incorrect
sign and low s,t statistics) and replaced by CAPCHG. Also, POPDEN and
OFFJOB were added to represent the important causal factors of population
and employment characteristics.
In the third path analysis, CAPCHG was trimmed (incorrect
sign). Also, POPDIF was trimmed because of suppressor variable problems,
i.e. its simple correlation with COMM is insignificant and its addition to
the equation causes instabilities in other model coefficients (e.g. betas
change by 100% with values approaching 1.0). In addition, OFFJOB was
trimmed because of a multicollinearity problem. The final equation became:
Unstandard i zed
English
COMM = 2.097 OFFICE + 13.98HIWAYS +
+ 0.1206 POPDEN - 649.5
COMM = 2.097 OFFICE + 807.2 HIWAYS +
+ 0.007167 POPDEN - 14.91
3,082 ACCESS + 8,249 ClONED
Metric
183.2 ACCESS + 189.4 ClONED
Sta nda rd i zed
COMM = 0.50 OFFICE + 0.21 HIWAYS + 0.18 ACCESS + 0.16 ClONED
+ 0.08 POPDEN
R = 0.80, R2 = 0.63, Coefficient of Variation = 0.75
(3) OFFICE Equation. In the first path analysis, RES,
TRANS, POPDEN, TIME, JOBCHG and VACOFF were all trimmed (low S,t statistics
and/or incorrect sign). Three new variables, ACCESS, OFFCHG and VACANT were
added to replace trimmed variables representing the important causal factors
of accessibility, regional influences and land constraints.
4-18
-------
/
In the second path a~alysis, PERCAP and VACANT were trimmed
(low B,t statistics) and replaced by CAPCHG and GOVT.
~ .
In the third path analysis, CAPCHG, OFFCHG and GOVT were
trimmed (low S,t statistics). INCOME was added as a replacement for CAPCHG.
I . .
I ~
In the fourth path analysis, ACCESS was trimmed (low s,t
statistics) and the final equation became:
Unstandardized
English OFFICE = 0.1149 COMM + 2.904 HIWAYS + 0.1853 JOBS
+ 265.8 INCOME + 22,133 OZONED - 390.9
OFFICE = 0.1149 COMM + 167.7 HIWAYS + 0.01101 JOBS
+ 6.102 INCOME + 508.1 OZONED - 8.973
Metric
Standardized
OFFICE + 0.49 COMM + 0.18 HIWAYS + 0.20 JOBS
+ 0.12 INCOME + 0.35 OZONED
R = 0.85, R2 = 0.72, Coefficient of Variation = 0.56
(4) MANF Equation. In the first path analysis, AREA,
IZONED, LIMITS, LAND and JOBCHG were all trimmed (low 6,t statistics).
Two new variables GOVT and MANCHG were added to represent the important
causal factors of tax costs and regional influences.
In the second path analysis, TIME and MANCHG were trimmed
(low s,t statistics) and JOBCHG was added to replace the latter.
4-19
-------
In the third path analysis, LENGTH was dropped because of
suppressor variable problems (i.e. change in other betas of up to 100% re-
sult from its inclusion in the equation).
In the fourth path analysis, JOBCHG and GOVT had marginally
low Band t statistics, but were not trimmed because of their causal impor-
tance. The final equation became:
Unstandardized
English MANF = 1,984 RRMILE + 3.754 MANJOB + 111.3 PEAK
+ 16.38 HIWAYS + 3,627 ACCESS + 5.969 JOBCHG
- 2.201 GOVT - 1,647
MANF = 73.28 RRMILE + 0.2231 MANJOB + 2.555 PEAK
+ 945.7 HIWAYS + 215.6 ACCESS '+ 3.547 JOBCHG
- 0.05052 GOVT - 37.Rl
Metric
Standardized
MANF = 0.32 RRMILE + 0.31 MANJOB + 0.22 PEAK
+ 0.21 HIWAYS + 0.18 ACCESS + 0.09 JOBCHG
- 0.08 GOVT
R = 0.74, R2 = 0.55, Coefficient of Variation = 1.00
(5) HIWAYS Equation. In the first path analysis, DISCBD,
UNIV,EMPOP, DRIVE, and TIME were all trimmed (incorrect sign). The variable:
POPDEN was added to replace UNIV, representing the important causal factor
of population characteristics.
In the second path analysis, OFFICE and POPDEN were
trimmed (low B, t statistics) and DUACRE was added to replace the latter.
Also, the variables JOBS and AIRPRT were added to replace EMPOP and DISCBD.
4-20
-------
In the third path analysis, AIRPRT, JOBS and DUACRE were
trimmed (low s,t statistics and/or incorrect sign). The final equation be-
came:
Unstandard ized
English HIWAYS = 0.001452 RES + 0.007517 COMM
- 29.75 ACCESS + 19.54
HIWAYS = 0.002336 RES + 0.01209 COMM
- 0.03062 ACCESS + 0.007769
Metric
Standard ized
HIWAYS = 0.27 RES + 0.51 COMM
- 0.12 ACCESS
R = 0.64, R2 = 0.41, Coefficient of Variation = 0.58
(6) EDUC Equation. In the first path analysis, NONHH,
RZONED and PERCHG were trimmed (low B,t statistics or incorrect sign). The
sign of STAY was positive, counter to the original hypotheses. STAY was
kept however because the causal relationship is strong and can be explained
as a secondary effect (after RES) where less mobile residential areas (i.e.
more stable areas) are more likely to encourage educational programs and
hence building development. The variable EDUCHG was added to replace
PERCHG, representing the important causal factor of regional influences.
The variable KIDS was redefined as school age children per 100 households
in 1960 to allow a stronger causal relationship.
In the second path analysis, EDUCHG was trimmed (incorrect
sign and low B,t statistics). The variables HSECHG and POPDIF were.
added as replacements, and POPDEN was added to represent population
characteristics.
In the third path analysis, HSECHG, TIME, and POPDIF were
trimmed (incorrect sign and/or low B,t statistics). The final equation
became:
4-21
-------
- . ..'- "-
Metric
Unstandardized
English EDUC = 0.03587 RES + 8.059 KIDS + 1447 STAY
+ 0.03636 VALUE + 0.07166 POPDEN - 1713
EDUC = 3.332 RES + 0.1850 KIDS + 33.22 STAY
+ 0.0008347 VALUE + 0.004259 POPDEN - 39.32
Standa rd i zed
EDUC = 0.48 RES + 0.43 KIDS + 0.34 STAY
+ 0.26 VALUE + 0.23 POPDEN
R = 0.84, R2 = 0.70, Coefficient of Variation = 0.40
(7). REC Equation.
RZONED were trimmed (incorrect sign).
represent the important causal factor
In the first path analysis, AREA and
A new variable, VACANT, was added to
of land constraints.
In the second path analysis, UNIV, PERCHG a,nd VACANT were
trimmed (incorrect sign and/or low S,t statistics). The variables POPDEN,
POPDIF and LAND were added as replacements, respectively.
In the third path analysis, LAND was trimmed (incorrect sign
and low S,t statistics). The final equation became:
Unstanda rd i zed
English
REC = 0.007844 RES - 805.0 POOR + 0.04382 POPDEN
+ 229.2 POPDIF + 127
Metric
REC = 31.74 RES - 805.0 POOR + 0.1134 POPDEN
+ 229.2 POPDIF + 127
Standard i zed
REC = 0.25 RES - 0.27 POOR + 0.33 POPDEN + 0.23 POPDIF
R = 0.55, R2 = 0.30, Coefficient-of Variation = 0.81
The low R2 is possibly due to the fact that large, active recreational areas
are usually dependent on governmental decisions and the availability of
appropriate natural resources. Neither of these effects could be quantified
for the causal model.
4-22
-------
---"-- -
(8) WHOLE Equation. In the first path analysis, MANF,
LAND, POPDEN, AREA, IZONED, TIME and COMCHG were all trimmed (low 8,t statis-
tics). The variables VACANT and MANCHG were added to represent the important
causal factors of land constraints and regional influences.
In the second path analysis, MANCHG and VACANT were trimmed
(low 8,t statistics). The variable POPDIF was added to represent regional
influences.
In the thir.'d path analysis, POPDIF was trimmed (incorrect:
sign) and the final equation became:
Unstanda rd i zed
English WHOLE = 9.022 WWJOB
- 94.28
WHOLE = 0.5362 WWJOB
- 2. 164
+ 1,422 ACCESS + 5.151 HIWAYS
Metric
+ 84.51 ACCESS + 297.4 HIWAYS
Standardized
WHOLE = 0.41 WWJOB + 0.29 ACCESS + 0.27 HIWAYS
R = 0.52, R2 = 0.27, Coefficient,of Variation = 1.12
(9) OTHER Equation. In the first path analysis, all varia-
ble except REC, MANF and INCOME were trimmed (low 8,t statistics and/or in-
correct sign). The variables ACCESS, POPDEN and SERCHG were added to
represent the important causal factors of accessibility, population charac-
teristics and regional influences.
In the second path analysis, POPDEN, ACCESS and SERCHG
were trimmed (low 8,t statistics and incorrect sign). The variables
INTDEN, DRIVE and RIDET were added to represent accessibility, and the
variables POPDIF and JOBCHG were added to represent regional influences.
JOBS was added to represent employment characteristics.
In the third path analysis, JOBCHG, DRIVE, POPDIF, INTDEN
and JOBS were trimmed (incorrect sign and low 8,t statistics). In addition,
RIDET was trimmed (low B,t statistics). The final equation became:
4-23
-------
Unstandardi zed
English OTHER = 1.407 REC + 0.1176 MANF + 430.1 INCOME - 527.0
Metric OTHER = 0.03230 REC + 0.1176 MANF + 9.873 INCOME - 12.10
Standard i zed
OTHER = 0.42 REC + 0.45 MANF + 0.15 INCOME
R = 0.72, R2 = 0.52, Coefficient of Variation = 1.02
6. Comparison of Model Forms
The first alternative model form considered was a mixed rates
and totals model (viz., the form used in GEMLUP-I) in which the endogenous
land use variables are not normalized by case study area. The initial path
analysis was rerun using the mixed rates/totals form, and the results com-
pared with that from the initial path analysis of the pure rates model.
Comparisons were made using a coefficient of variation statistic (the stan-
dard error as a percent of the dependent variable mean) and are shown in
Table 4-5. The results indicate a lower coefficient of variation. in six out.
of nine equations using a pure rate model.
model form was rejected.
Thus, the first alternative
The second alternative form considered was a multiplicative
model of the form:
b c d
Yl = a Y2 Xl X2 .....
where Yl' Y2 are endogenous variables, Xl and X2 are exogenous variables,
and a,b,c and d are model parameters. The initial path analysis was rerun
using the multiplicative form (by taking the natural logarithm of all varia-
bles), and the results compared with that from the initial path analysis of
the additive model. Due to the non-linearities in the multiplicative form,
it was inappropriate to compare standard statistical measures. Instead,
the residuals for each form were examined to determine how many occurred
outside the standardized bounds of -1.0 or +1.0, see Table 4-6. This compari-
son of residuals can be thought of as a goodness of fit test, similar to the
4-24
-------
TABLE 4-5
COMPARISON OF COEFFICIENTS OF VARIATION
FOR A PURE RATES AND A MIXED TOTALS-RATES MODEL
Dependent Variable Pure Rates Model Mixed Totals-Rates Model
RES 0.44 0.55.
Ca-1M 0.78 1.03
OFF ICE 0.60 0.80
MANF 1.08 1.20
HIWAYS 0.64 0.58
EDUC 0.42 0.54
REC 0.84 1.10
WHOLE 1.13 1.02
OTHER 1.03 0.67
*
Endogenous variables as totals, all exogenous variables as rates.
4-25
\
)
-------
TABLE 4-6
COMPARISON OF THE NUMBER OF STANDARDIZED
RESIDUALS GREATER THAN 1.0 IN ABSOLUTE VALUE
FOR TWO MODEL FORMS
DEPENDENT
VARIABLE
ADD ITIVE MODEL
<-1.0 >+1.0 Total
MUL TIPLICATIVE
<-1.0 >+1.0 Total
RES 6 4 10 5 9 14
COMM 3 4 7 6 6 12
OFFICE 3 4 7 5 5 10
MANF 4 3 7 8 6 14
HIWA YS 6 5 11 5 6 11
EDUC 4 1 5 4 6 10
REC 4 6 10 7 7 14
WHOLE 3 4 7 3 1 4
OTHER 1 3 4 2 3 5
4-26
-------
R2 statistic. The results indicate a lower number of large residuals in
eight out of nine equations using the additive form. In addition, some of
the residuals were plotted versus their dependent variables to check for
any patterns. No distinct trends were apparent, though the residuals from
the additive model did appear to be more randomly distributed. Thus, the
multiplicative form was rejected.
B.
COEFFICIENT STABILITY ANALYSIS
The usefulness of any path analytic model is dependent on the
stability of the path estimates. In the current study, where only a rela-
tively small data sample (40) was available for estimating model parameters,
the question of stability was therefore important. The statistical technique
chosen to investigate stability was jackknifing. This procedure involved 40
separate analyses of each model equation using only 39 data samples. For
each analysis, a different sample from the total set of 40 was excluded.
This approach provided 40 separate estimates of each of the model coefficients,
from which mean and standard deviations, and one-way frequency distributions
were prepared. Two separate indices of coefficient stability were then de-
veloped for interpreting the data. The jackknifing analysis was applied to
all nine equations in the model. In the jackknifing analysis, only the path
regression coefficients (b) were used. Path coefficients (8) do not provide
a valid basis for comparisons of systems operating on different data popula-
tions since the variance of the independent variables is included in the path
coefficients and variance can change dramatically from sample to sample.
The means and standard deviations computed from the set of 40
values for each model coefficient are compared on Table 4-7 with, the coefficient
values from the final path analysis. The results show the jackknifing re-
sults and final path analysis path regression coefficients to be practically
identical in most cases, with the exception of the COMM equation where dif-
ferences are appirent. The standard deviation of the jackknifing coefficients
is, on the average, only l~ percent of the coefficient mean value, indicat-
ing low variance in the coefficient values and thus no significant instabil-
ities in general.
4-27
-------
TABLE 4-7
COMPARISON OF PATH REGRESSION COEFFICIENTS FROM THE FINAL
PATH ANALYSIS AND THE JACKKNIFING STABILITY ANALYSIS
Path Regression Coefficients*
Hodel Independent Final Path Jackknifing
Equat ion Variable Analysis Mean Standard Deviation
RES NONHH -7.244 -7.311 0.505
HIWAYS 40.36 39.27 15.31
RECAPl 12.89 12.89 0.965
POPDEN 2.007 2.009 0.166
PERCHG 10,419 10 ,389 840. 1
LAND 2,416 2,443 315.2
RlONED 2,798 2,814 510.4
DRIVE 892.8 888. 1 85.68
COMM OFF IC E 2.097 1.362 0.369
HIWAYS 13.98 10.78 4.482
ACC ES S 3,082 4, 170 351.2
ClON ED 8,249 9 , 261 1 ,287
POPDEN 0.1206 0.2480 0.0490
OFF IC E COMM 0.1149 0.1420 0.0140
HIWAYS 2.904 2.155 1 .131
JOBS 0.1853 O. 1660 0.0170
INCOME 265.8 273.5 41.28
OZONED 22, 133 21,561 2,390
MAr~F HI WAYS 16.38 16.05 3.893
PEAK 111.3 112.1 22.89
MANJOB 3.754 3.759 0.2830
RRM I LE ",984 1,978 334.4
ACCESS 3,627 3,585 919.0
GOVT -2.201 - 2. 1 86 O. 7200
JOBCHG 5.969 6.033 2.283
HIWAYS RES 0.001452 0 . 001 0 < .0001
COMM 0.007517 0.0080 0 . 001 0
ACCESS -29.75 -29 .82 8.215
EDUC RES 0.03587 0.0360 0.0020
KIDS 8.059 8.060 0.3130
STAY 1 , 44 7 1 ,448 87.82
VALUE 0.03636 0.0360 0.0050
POPDEN 0.07166 0.0720 0.0060
*Shown to four significant digits, wherever.'possible.
4-28
-------
TABLE 4-7 (Continued)
COMPARISON OF PATH REGRESSION COEFFICIENTS FROM THE FINAL
PATH ANALYSIS AND THE JACKKNIFING STABILITY ANALYSIS
Path Regression Coefficients*
Model Independent Final Path Jackknifing
Equati on Variable Analysis Mean Standard Deviation
REC RES 0.007844 0.0080 0 .001 0
POOR -805.0 -805.0 68. 13
POPDEN 0.04382 0.0440 0.0040
POPD I F 229.2 228.2 22.75
WHOLE WWJOB 9.022 9.015 0.8550
ACCESS 1,422 1 ,406 289.1
HIWAYS 5.151 5.211 0.8290
OTHER REC 1 .407 1 . 394 0.1690
MANF 0.1176 0.1170 0.0200
INCOME 430.1 430.6 49. 13
*Shown to four significant digits, wherever possible.
4-29
-------
The percent of instability in each coefficient can be compared
with the average imprecision introduced by errors inherent in the land use
data on which the model is based. A previous analysis of the errors as-
sociated with estimating land use from aerial photographs concluded the
average estimate to be 87% accurate (with actual values varying widely de-
pending on the interpreter's skill, the photograph's scale and the classifica-
tion system used). Introducing this error estimate (f13%) for the land use
variables into the model equations produces an average imprecision of 13%
for all non-endogenous variable coefficients. Comparison of this value with
the 15% total variance estimated by the jackknifing indicates that aerial
photo errors may be one of the principal causes of instabilities in the model
coefficients.
One-way frequency distributions of the coefficient values are
graphed in Appendix I for the nine model equations. To more accurately
quantify the stability of the path regression coefficients and to allow com-
parisons, between equations, two separate indices of stability were developed.
1.
Extreme Value Index
The first. stability index that was used is a statistical
test for extreme values in a sample developed by Dixon [126J. The null hypothe-
sis assumes the value farthest from the mean does not differ significantly
from the range of values in the data set, i.e., is not an extreme value. As-
sumptions are made that the data are independent observations and normally
distributed. The coefficient values that were tested in the current applica-
tion, however, are a dependent set. The violation of this assumption, however,
does not invalidate our use of the test statistic as a guide for judging
stability. The 40 values for each model coefficient were ordered from
lowest to highest Xl
-------
The critical rl level tested for was 0.414, corresponding to the one percent
significance level. If rl exceeds 0.414, then given an independent sample,
there is a 99 percent probability that the deviation of the lowest value Xl
from the range of values in the data set is significant and that Xl is an
extreme value. Again as a guide, if rl exceeds 0.414 ill our samples, we will
refer to th~ corresponding Xl as an extreme value significant at the .01
level. A separate test on the highest data value X40 was also performed. A
second ratio to test the "highest data point was computed as follows:
Xk - Xk-2
r' =
40 X k - X 3
k = 40
The values of rl and r40 for each model coefficient are shown in Table 4-8.
Where the ratio was found to be significant at the one percent level, the
case number of the data sample excluded from the path analysis corresponding
to the extreme value is indicated.
When an extreme value occurs in the data set of path regres-
sion coefficient values, it indicates that the exclusion of data from a part-
icular case number is causing a significant change in the value of the co-
efficient. Normally one would test for a change in value when certain data
were included, however, it was impractical to compute coefficients for all
possible data combinations, and thus, detection of changes was accomplished
by excluding each sample, one at a time. If exclusion of.one sample from a
least squares analysis causes a significant change in a coefficient's values,
then that data sample, when included, is exerting a greater effect on the
analysis results than other samples.
The repetition of case study numbers in a given model
equation in Table 4-8 (e.g. 18 appears 3 times in the OTHER equation), and the
correspondence between many of the case study numbers with those in Table 4-4
indicates that skewness in the dependent variable distributions is the
principal cause of model coefficient instabilities. Thus, the extreme value
indices are limited to assessing the overall stability of an equation, not
its individual coefficients. In this light, only the COMM, MANF, WHOLE and
OTHER equations can be classified as having appreciable instability in their
4-31
-------
TABLE 4-8
EXTREME VALUE INDEX OF STABILITY FOR
ALL PATH REGRESSION COEFFICIENTS IN THE CAUSAL MODEL EQUATIONS
Extreme Value Case Number Exc1 uded
Model Independent Ratio if Extreme Va1ue*
Equation Variable r1 r40 Xl X40
RES HIWAYS 0.338 0.190
RECAP1 0.541 0.064 34
NOi~HH 0.551 0.246 40
POPDEN 0.525 0.255 7
PERCHG 0.525 0.212 11
LAND 0.128 0.113
RZONED 0.504 0 . 277 37
DRIVE 0.761 0 . 260 7
COMM OF FI C E 0.662 0.709 35 3
HIWAYS 0.221 0.339
ACCESS 0.366 0.442 20
CZOi~EU 0.578 0.432 11 29
POPDEN 0.616 0.332 3
OFF ICE COMM 0.184 0.498 35
HIWAYS 0.342 0.399
JOBS 0.354 0.087
INCOME 0.684 0 . 185 17
OZONED 0.800 0.415 3 8
MANF HIWAYS 0.642 0.305 18
PEAK 0.688 0.560 35 9
MANJOB O. 219 0.492 30
RRMILE 0.545 0.677 30 34
ACCESS 0.207 0.632 30
GOVT 0.270 0.592 34
JOBCHG 0.464 0.658 30 18
HIWAYS RES 0.385 0.196
COMM 0.076 0.353
ACCESS 0.292 0.429
EDUC RES 0.446 0.354 11
KIDS 0.277 0.493 29
STAY 0.471 0.300 3
V ALU E 0.796 0.708 3 17
POPDEN 0.402 0.183
*At the 1 percent significance level.
4-32
-------
TABLE 4-8 (Continued)
EXTREME VALUE INDEX OF STABILITY FOR
ALL PATH REGRESSION COEFFICIENTS IN THE CAUSAL MODEL EQUATIONS
Extreme Value Case Number Excluded
Mod e 1 Independent .Rati 0 if Extreme Va1ue*
Equat ion Variable r1 r40 Xl X40
REC RES 0.204 0.446 7
POOR 0.476 0.395 3
POPDEN 0.305 0.241
POPDIF 0.200 0.099
WHOLE WWJ OB 0.480 0.616 14 17
A CC ESS 0.888 0.484 20 18
HIWAYS 0.528 0.716 14 18
OTHER REC 0.853 0.404 18
MANF' 0.860 0.852 18 35
Ii~COME 0.659 0.641 18 17
*At the 1 percent significance level.
4-33
-------
model estimates. These are the only equations having greater than 50 per-
cent extreme values. Since multiple occurrences .of the same case number
in a given equation is not as strong an indication of instability as the
occurrence of the same number of different case numbers, the COMM and WHOLE
equations can be classified as having the greatest instabilities.
2.
Coefficient of Variation Index
A secondary stability index was developed to quantify the effect
of the total variation in each model coefficient on the predicted values for
the dependent variable. The variation in each path regression coefficient
was expressed as half of the difference between the maximum (bm ) and mini-
ax
mum (bmin) values obtained in the jackknifing analysis. This difference was
then multiplied by the mean value of the associated independent variable
(Xi) to obtain the total effect on the dependent variable. This quantity was
then normalized by the mean value of the associated dependent variable (V) to
obtain the percentage change:
j (b b . )x.
P. = I max - ml n 1
- 1 --
-' 2Y
The computed Pi values are summarized in Table 4-9.
Of the 43 values of Pi listed in Table 4-9, only 5 of them (or
12%) exceed 30% and only 2 of them (or 5%) exceed 40%. The average value of
Pi is 0.16 (16%). Thus, no widespread instability problem exists and the
large instabilities that do occur are restricted to the COMM and OTHER equa-
tions.
4-34
-------
TABLE 4-9
STABILITY INDEX FOR ALL PATH REGRESSION COEFFICIENTS IN THE
CAUSAL MODEL EQUAHONS BASED ON THE EFFECT OF VARIATION
IN THE COEFFICIENTS ON THE DEPENDENT VARIABLE
Model I nd epend en t Percentage Change in
Equati on Variable Dependent Variable
RES HIWAYS 0.21
RECAPl 0.06
NONHH 0.02
POPDEN 0.13
PERCHG 0.07
LAND 0.05
RlONED 0.06
DRIVE 0.06
COMM OFFICE 0.41
HIWAYS 0.30
ACCESS 0.05
ClONED 0.08
POPDEN 0.19
OFF ICE COMM 0.16
HIWAYS 0.28
JOBS 0.07
INCOME 0.28
OlON ED 0.09
MANF HIWAYS 0.23
PEAK 0.22
MANJ OB 0.07
RRMILE 0.23
ACCESS 0.12
GOVT 0.07
JOBCHG 0.13
HIWAYS RES 0.10
COMM 0.15
ACC ESS 0.06
EDUC RES 0.08
KIDS 0.15
STAY 0.19
VALUE 0.34
POPDEN 0.05
4-35
-------
TABLE 4-9 (Continued)
STABILITY INDEX FOR ALL PATH REGRESSION COEFFICIENTS IN THE
CAUSAL MODEL EQUATIONS BASED ON THE EFFECT OF VARIATION
IN THE COEFFICIENTS ON THE DEPENDENT VARIABLE
Model
Equation
I ndepend ent
Vari abl e
Percentage Change in
Dependent Variable
REC
RES
POOR
POPDEN
POPDIF
WWJ OB
ACC ESS
HIWAYS
O. 15
0.17
0.11
0.05
0.17
0.19
0.27
WHOLE
OTHER
REC
MANF
INCOME
0.27
0.34
0.49
4-36
-------
C.
DETERMINE NET CAUSAL EFFECTS
1. Approach
The objectives of this task were to quantify direct and
indirect causal effects between model variables, and to summarize the
total net relationships. This causal tracing analysis was performed
for every variable interaction in the final path analytic model.
I
The total net effects obtained in this manner correspond to the path coef-
ficients of the IIreduced-form" equations [15]. An introduction to causal
tracing is given in Section 8 of Appendix A.
In general, the total effect (T) from an input variable (X) to a
dependent variable (Y) is determined by:
T =[(E1 + E2 + .....) (l-L1)(l-L2){l-L3)""']*
( 1 - L 1 ) ( 1 -L 2) ( 1 - L 3) . . . . .
..... are distinct open paths from X to Y
..... are the return effects for all distinct
loops' .provi di ng relevant feedback
* is a special operation in which the multiplications
in the numerator and in the denominator are carried
out before division, terms are deleted if they multiply
the effects of touching paths, and division is carried
out only after such terms have been deleted.
A more thorough description of the procedure is given by Heise [15]. An ex-
ample of this technique is illustrated below for the input variable OFFICE
and the dependent variable OTHER. Referring to Figure 4-1, it can be seen
there are two open paths:
where:
E" E2'
L" L2'
E, : OFFICE +COMM +HIWAYS +RES +REC +OTHER
E1 = 0.50 * 0.51 * 0.21 * 0.25 * 0.42 = 0.00562
E2 : OFFICE +COMM +HIWAYS +MANF +OTHER
E2 = 0.50 * 0.51 * 0.21 * 0.45 = 0.0241
These open paths touch four feedback loops:
L, : OFFICE ++ COMM
L, = 0.49 * 0.50 = 0.245
4-37
-------
L2 .: COMM ++ HIWAYS
L2 ~ 0.21 * 0.51 ~ 0.107
L3 : HIWAYS ++ RES
L3 = 0.27 * 0.21 ~ 0.0567
L4 : c:-~~~~~~(~(~~~~(~(~~~~~~()
L4 = 0.50 * 0.51 * 0.18 = 0.0459
Thus, the total net effect is:
T = [(El + E2)(1-Ll)(1-L2)(1-L3)(1-L4)]*
(1-Ll)(1-L2)(1-L3)(1-L4)
Carrying out the multiplication in the denominator, all of the cross terms
are deleted except for L1L3 since these loops do not touch. Carrying out the
multiplication in the numerator, all of the E and L cross terms are deleted
since every feedback loop touches El and E2. The resultant expression is:
T =
El + E2
(1-Ll-L2-L3-L4+L1L3)
= 0.05
2.
. ...
Results and Conclusions
The direct, indirect and total causal effects between the 9
endogenous land use variables and all input variables are listed in Table 4-10.
through 4-18. Conclusions, for each model equation, are given below by summari-
zing the 5 most important causal factors, in descending order, with a prefix
sign indicating the direction of causal action. Note that a rliscussionof the
reasons for including each variable that appears in the model is given in
Section II .H.
(1)
RES Equation
The most important causal factors for residential develop-
ment are measures of (+) population characteristics, (+) accessibility,
(+) regional growth, (+) reserve collection capacity of the wastewater
4-38
-------
TABLE 4-10
SUMMARY OF DIRECT AND INDIRECT CAUSAL EFFECTS
ON THE VARIABLE RES
Causal Factor Path Coefficients
Type Variable Direct Indi rect Total
COMM O. 19 0.19
OFFICE 0.10 0.10
Land Use MANF
Variables HIWAYS 0.21 0.02 0.23
EDUC
REC
WHOLE
OTHER
Population POPDEN 0.48 0.05 0.53
Characteristics NONHH -0.19 -0.01 -0.20
KIDS
Employment JOBS 0.02 0.02
Characteri sti cs MANJOB
WWJOB
Accessi bil ity ACCESS 0.0005 0.0005
DRIVE 0.41 0.03 0.44
RRMILE
Regi ona 1 POPDIF
Inf1 uences PE RCHG 0.40 0.03. 0.43
JOBCHG
Incorre I N COME 0.01 0.01
POOR
Major RECAPl 0.32 0.02 0.34
Project PEAK
Zon ing RZONED 0.13 0.01 0.14
CZONED 0.03 0.03
o ZO NED 0.04 0.04
Mobility STAY
Housing VALUE
Cha rac te ri s ti cs
Land Costs, LAND 0.22 0.02 0.24
Taxes GOVT
4-39
-------
TABLE 4-11
SUMMARY OF DIRECT AND INDIRECT CAUSAL EFFECTS
ON THE VARIABLE COMM
Causal Facto r Path Coefficients
Type Variable Di rect Indi rect Total
RES 0.06 0.06
OFFICE 0.50 0.34 0.84
Land Use MANF
Variables HIWAYS 0.21 0.14 0.35
EDUC
REC
WHOLE
OTHER
Population POPDEN 0.08 0.12 0.20
Characteristics NONHH -0.02 -0.02
KIDS
Employment JOBS 0.17 0.17
Cha ra cte ri s ti cs MANJOB
WWJOB
Access i bil i ty ACCESS 0.18 0.08 0.26
DRIVE 0.06 0.06
RRMILE
Regional POPDIF
Infl uences PERCHG 0.06 0.06
JOBCHG
Income INCOME 0.10 0.10
POOR
Major RECAPl 0.05 0.05
Project PEAK
Zoning RZONED 0.02 0.02
CZONED 0.16 0.11 0.27
OZONED 0.30 0.30
Mobil ity STAY
Housing VALUE
Characteristics
Land Cos ts, LAND 0.03 0.03
Taxes GOVT
4-40
-------
TABLE 4-12
SUMMARY OF DIRECT AND INDIRECT CAUSAL EFFECTS
ON THE VARIABLE OFFICE
Causal Factor Path Coefficients
Type Variable Di rect Indirect Tofa 1
RES 0.14. 0.14
COMM .. 0.49 0.41 0.90
Land Use MANF
Variables HIWAYS 0.18 0.30 0.48
EDUC
REC
WHOLE
OTHER
Population POPDEN 0.14 0.14
Cha racte ri s tics NONHH -0.03 -0.03
KIDS
Emp 1 oymen t JOBS 0.20 0.10 0.30
Characteri s ti cs MANJOB
WWJOB
Access i bi 1 i ty ACCESS 0.12 0.12
DRIVE 0.06 0.06
RRMILE
Regional POPDIF
Inf1 uences PERCHG 0.06 0.06.
JOBCHG --
Income INCOME 0.12 0.06 O. 18
POOR
Major RECAPl 0.05 0.05
Project PEAK
Zoning RZONED 0.02 0.02
CZONE!} 0.16 0.16
OZONtD 0.35 0.17 0.52
Mobi1 ity STAY
Housing VALUE
Characteristics
Land Cos ts, LAND 0.03 0.03
Taxes GOVT
4-41
-------
TABLE 4-13
SUMMARY OF DIRECT AND INDIRECT CAUSAL EFFECTS
ON THE VARIABLE MANF
Causal Factor Path Coefficients
Type Variable Di rect Indirect Total
RES 0.Q8 Q.08
COMM 0.14 0.14
Land Use OFfIcE 0.10 0.10
Variables HIWAYS 0.21 0.21
EDUC
REC
WHOLE
OTHER
Population POPDEN 0.05 0.05
Character; s ti cs NONHH -0.01 -0.01
KIDS
Employment JOBS 0.02 0.02
Character; sti cs MANJOB 0.31 0.31
WWJOB
Access i bi 1 i ty ACCESS 0.18 0.0005 0.18
DRIVE 0.03 0.03
RRMILE 0.32 ~.. 0.32
Regional POPDIF
Infl uences PERCHG 0.02 0.02
JOBCHG 0.09 0.09
Income INCOME 0.01 0.01
POOR
Major RECAP 1 0.02 0.02
Project PEAK 0.22 0.22
Zoning RZONED 0.01 0.01
CZONED 0.03 0.03
OZONED 0.03 0.03
Mobil ity STAY
Housing VALUE
Character; s ti cs
Land Cas ts, LAND . 0.01 0.01
Taxes GOVT -0.08 -0.08
4-42
-------
TABLE 4-14
SUMMARY OF DIRECT AND INDIRECT CAUSAL EFFECTS
ON THE VARIABLE HIWAYS
Causal Factor Path Coefficients
Type Variable Di rect Indi rect Total
RES 0.27 0.09 0.36
COMM 0.51 0.18 0.69
Land Use QFfI.CE 0.46 0.46
Variables MANF"
EDUC
REC
WHOLE
OTHER
Population POPDEN 0.25 0.25
Characteri s ti cs NONHH -0.07 -0.07
KIDS --
Employment JOBS 0.09 0.09
Characteristics MANJOB
WWJOB
Access i bi 1 i ty 'ACCESS -0.12 0.12 0.002
DRIVE 0.15 0.15
. RRMILE
Regional POPDIF
Infl uences PERCHG 0.14 0.14
JOBCHG
Income INCOME 0.06 0.06
POOR
Maj or RECAP1 0.12 0.12
Project PEAK
Zoning RZONE D 0.05 0.05
CZONED 0.15 0.15
OZONED 0.16 0.16
Mobil ity STAY
Housing VALUE \
Characteri s ti cs
Land Cos ts, LAND 0.08 0.08
Taxes GOVT
4-43
-------
TABLE 4-15
SUMMARY OF DIRECT AND INDIRECT CAUSAL EFFECTS
ON THE VARIABLE EDUC
Causal Factor Path Coefficients
Type Variable Di rect Indi rect Total
RES 0.48 0.48
COMM .0.07 0.07
Land Use OFFICE . . 0.05 0.05
Variables MANF
HIWAYS 0.11 0.11
REC
WHOLE
OTHER
Population POPDEN 0.23 0.26 0.49
Cha racte ri s tics NONHH -0.09 -0.09
KIDS 0.43 0.43
Emp 1 oymen t JOBS 0.01 0.01
Characteristics MANJOB
WWJOB
Access i bi 1 i ty ACCESS 0.0002 0.0002
DRIVE 0.20 0.20
RRMILE
Regional POPDIF
Infl uences PERCHG 0.19 0.19
JOBCHG
Income INCOME 0.006 0.006
POOR
Major RECAPl 00115 0.15
Project PEAK
Zoning RZONED 0.06 0.06
CZONE!} 0.01 0.01
OZONtD 0.02 0.02
Mobil ity STAY 0.34 0.34
Housing VALUE 0.26 0.26
Characteri s ti cs
land Cos ts, LAND 0.11 O. 11
Taxes GOVT
4-:-44
-------
TABLE 4-16
SUMMARY OF DIRECT AND INDIRECT CAUSAL EFFECTS
ON THE VARIABLE REC
Causal Factor Path Coefficients
Type Variable Di rect Indi rect Total
RES 0.25 0.25
COMM 0.04 0.04
Land Use OFFrCE 0.02 0.02
Variables MANF
HIWAYS 0.06 0.06
EDUC
WHOLE
OTHER
Population POP DEN 0.33 0.13 0.46
Characteristics NONHH -0.05 -0.05
KIDS
Employment JOBS 0.005 0.005
Characteri s ti cs MANJOB
WWJOB
Access i bi 1 i ty ACCESS 0.0001 0.0001
DRIVE 0.11 0.11
RRMILE
Regional POP DI F 0.23 0.23
Infl uences PERCHG 0.11 0.11
JOBCHG
Income INCOME 0.002 0.002
POOR -0.27 -0.27
Major RE CAP 1 0.09 0.09
Project PEAK
Zoning RZONE D 0.03 0.03
CZONED 0.008 0.008
OZONED 0.014 . 0.014
Mob il ity STAY
Housing VALUE
Characteri s ti cs
Land Cos ts, LAND 0.06 0.06
Taxes GOVT
4-45
-------
TABLE 4-17
SUMMARY OF DIRECT AND INDIRECT CAUSAL EFFECTS
ON THE VARIABLE WHOLE
Causal Factor Path Coefficients
Type Variable Di rect Indi rect Total
RES 0.10 0.10
COMM 0.19 0.19
Land Use OFF! CE 0.12 0.12
Variables MANF
HIWAYS 0.27 0.27
EDUe
RE.C
OTHER
Population POPDEN 0.07 0.07
Cha racte ri s ti cs NONHH -0.02 -0.02
KIDS
Emp 1 oymen t JOBS O.O~ 0.02
Cha racte ri s ti cs MANJOB
WWJOB 0.41 0.41
Ac ce s sib i 1 i ty 'ACCESS 0.29 0.0006 0.29
DRIVE 0.04 0.04
' RRMILE
Regional POPDIF
Infl uences PERCHG 0.04 0.04
JOBCHG
Income INCOME 0.01 0.01
POOR
Major RECAPl 0.03 0.03
Project PEAK
Zoning RZONED 0.01 0.01
CZONED 0.04 0.04
OZONED 0.04 0.04
Mobil ity STAY
Housing VALUE
Characteri s ti cs
Land Cos ts, LAND 0.02 0.02
Taxe s GOVT
4-46
-------
TABLE 4-18
SUMMARY OF DIRECT AND INDIRECT CAUSAL EFFECTS
ON THE .VARIABLE OTHE.R
Causal Factor Path Coefficients
Type Variable Di rect Indi rect Total
RES 0.14 0.14
Land Use COMM 0.07 0.07
Variables OFFICE 0.05 0.05.
MANF 0.45 0.45
HHJAYS 0.12 0.12
EDUC
REC 0.42 0.42
WHOLE
Population POPDEN 0.22 0.22
Characteristics NONHH -0.03 -0.03
KIDS
Emp 1 oymen t JOBS 0.01 0.01
Characteri s ti cs MANJOB 0.14 0.14
WWJOB
- ~
Access i bi 1 i ty ACCESS 0.08 0.08
DRIVE 0.06 0.06
RRMILE 0.14 0.14
Regional POPDIF 0.10 0.10
Infl uences PERCHG 0.05 0.05
JOBCHG 0.04 0.04
Income INCOME 0.15 0.006 O. 16
POOR -0.11 -0.11
Major RECAPl 0.04 0.04
Project PEAK 0.10 0.10
Zoning RZONED 0.02 0.02
CZONED. 0.01 0.01
OZONtD 0.02 0.02
Mobil i ty STAY
Housing VALUE
Characteri s ti cs
land Cos ts, LAND 0.03 0.03
T axe s GOVT -0.04 -0.04
4-47
-------
major project and (+) land costs. It is interesting to note that variables
representing the major project treatment plant capacity were not significant
in the causal model, i.e., it is the collection network not the treatment
facilities which can be said to induce residential development. This is
consistent with the discussion of wastewater major project impacts in
Sections 1.8.3 and II.E. Other factors, which were hypothesized in the
initial model, but which turned out not to be important causal factors, are
manufacturing development, outdoor recreational development, on-site disposal
restrictions, income, employment characteristics and vacant developable
land. Vacant developable land is a prerequisite for induced development.
The reason it did not appear to be causally important is probably because
all case study areas were chosen a priori to include significant amounts of
vacant land.
(2) COMM Equation
The most important causal factors for commercial development
are measures of (+) office development, (+) non-expressway highway lane miles,
(+) office zoning, (+) commercial zoning, and (+) expressway interchange den-
sity. Major project timing was hypothesized as a causal factor, but the
analysis rejected this variable. However, collection network reserve capacity
is a small indirect causal factor (through residential development). All
other hypothesized causal factors were verified in some measure, large or
sma 11 .
(3) OFFICE Equation
. The most important causal factors for office-professional
development are measures of (+) commercial development, (+) office zoning,
(+) non-expressway highway lane miles, (+) employment density, and (+)
family income. Again, an hypothesized relationship with major project timing
was rejected by the analysis. However, collection network reserve capacity
is a small indirect causal factor (through residential development). Other
causal factors which turned out not to be important were manufacturing de-
velopment and development constraints.
4-48
-------
(4) MANF Equation
The most important causal factors for manufacturing develop-
ment are measures of (+) railroad mileage density, (+) manufacturing employ-
ment density, (+) wastewater major project peak flow, (+) non-expressway
highway lane miles, and (+) expressway interchange density. Although a
measure of wastewater system capacity was found to be significant, it is
interesting to note that measures of sewer service (area of land served)
and major project timing were rejected by the analysis. Other causal fac-
tors which turned out not to be important were on-site disposal restrictions
and vacant developable land.
(5) HIWAYS Equation
The most important causal factors for non-expressway high-
way development are (+) commercial development, (+) office development, (+)
residential development, (+) population density, and (+) office zoning. Al-
though the hypothesized relationship with major project timing was rejected
by the analysis, collection network reserve capacity is an indirect causal
factor (through residential development). Other causal factors which
turned out to be unimportant were manufacturing development and distance to
trip generators.
(6) EDUG Equation
The most important causal factors for educational develop-
ment are (+) population density, (+) residential development, (+) school-age
children per household, (+) neighborhood stability, and (+) residential real
estate values. Again, although the hypothesized relationship with major
project timing was rejected, collection network reserve capacity is an in-
direct causal factor (through residential development). All other hypothe-
sized causal factors were verified in some measure, large or small.
(7) REG Equation
The most important causal factors for active, outdoor
recreational devleopment are (+) population density, (-) percent of families
4-49
-------
below poverty level, (+) residential development, (+) growth in regional
population, and (+) growth in county population. All other hypothesized
causal factors were verified, with the exception of educational development:
(8) WHOLE Equation
The most important causal factors for wholesale-warehouse
development are (+) wholesale and warehouse employment density, (+) ex-
pressway interchange density, (+) non-expressway highway lane miles, (+)
commercial development, and (+) office development. Although the hypothe-
sized relationship with major project timing was rejected, collection net-
work reserve capacity is a small indirect causal factor (through residential
development). All other hypothesized causal factors were verified in some
measure, with the exception of manufacturing development and vacant developable
1 and.
(9) OTHER Equation
The most important causal factors for other types of de-
velopment are (+) manufacturing development, (+) recreational development,
(+) population density, (+) family income, and (+) residential development.
Although major project timing was rejected as a causal factor, wastewater
system peak flow is an indirect causal factor (through manufacturing develop-
ment. All other hypothesized causal factors were verified, with the exception
of distance to population center.
4-50
-------
D.
VMT MODEL VALIDATION
The objective of this task was to validate the GEMLUP VMT model by:
(1) testing its predictive ability with transportation data from 11 to the
40 case studies, and (2) quantifying the standard error associated with its
use in these cases. The VMT model is a methodology to predict the vehicle
miles travelled by motor vehicles in trips associated with land use within
either the region or the area of analysis. The basic technique for predict-
ing vehicle miles of travel (VMT) is simply to form the product of: (1)
predictions of land use, (2) vehicle trip generation rates, and (3) vehicle
trip lengths, and sum over all land use categories. Thus, it was our intent
not only to validate the final VMT projection, but the process of obtaining
trip generation rates and trip lengths as well (either by equation or default
values). Our approach to this task involved first developing criteria for
determining whether model performance in the validation exercise is "reason-
able", i.e., are the errors encountered of the same order as those encountered
in other simplified VMT predictive techniques and are they excessive in
comparison to typical errors introduced at other points in the GEMLUP pre-
dictive model. These acceptability criteria are discussed below.
1.
Develop Test Criteria
Two convenient measures of a model's accuracy and precision in
any validation are the standard error of estimate and correlation coefficient
values. Thus, in developing criteria, the literature was first searched for
validations of similar transportation models in which these statistics had
been quanti fied, for the three key GEMLUP parameters: (1) VMT, (2) trip
generation rates, and (3) trip lengths. The most extensive compilation of.
data on existing transportation procedures and models is given in Appendix B
of a recent publication by Comsis Corporation [127]. Although the effectivt::-
ness and limitations of 40 different procedures are discussed, these evalua-
tions are unfortunately all subjective and non-numeric. Four other publica-
tions were found that did contain significant data.
4-51
-------
The Maricopa Association of Governments has produced a summary
report of data from several hundred land use trip generation studies done
nationwide[120J. The data are arranged in linear relationships between land
use (dwelling units or 1000 square feet gross floor area) and trip genera-
tion rates. Regression analysis is then used to test the accuracy of the
concept of linear predictive equations for trip rates. Applicable data from
this publication are shown in Table 4-19. Bates [128] describes the result
of a validation analysis of trip generation equations, developed from data
on 4 southeastern U.S. cities and tested with data from 8 others. Standard
errors of estimate for VMT ranged from -21.1% to +24.5% (average i22.8%),
while for trip generation rates the errors were -53.6% to +70.0% (average
i61.8%). A study by Jeffries and Carter [129] of trip generation in small
urban areas used data from 6 states to test simple trip rate equations for
home-based trips. Correlation results ranged from 0.74 to 0.95 with an
overall coefficient value of 0.86. Finally, Ashford and Holloway [130J did
a comprehensive validation of trip production equations developed for
Pittsburgh, PA. Their results are summarized in Table 4-20.
Based on these data, it was decided to use the following as
testing criteria:
For VMT, the maximum allowable error is t23%.
For trip generation rates, the maximum allowable errors
and minimum correlations to be used are summarized in
Table 4-21. Where exact data on a given land category
were not available, overall average statistics for trip
rate prediction of 42% and R = 0.81 were used.
The existing literature on validation provides data on only
two of the three key GEMLUP parameters, viz. VMT and trip rates. For trip
lengths a different approach was required to estimate reasonable errors of
estimation. It was assumed that errors should be some percentage of the
total variation typically seen in a parameter. If errors approach 100% of
the total range, it is probable that a value above the mean could be esti-
mated as below the mean, or vice-versa. Thus, a value of 50% of the total
variation in trip length data was used as the maximum allowable error.
Only one publication was found that gave substantial data on the variation
of average trip lengths. The FHWA[1 3l] has compiled extensive data on trip
4-52
-------
TABLE 4-19
COMPILATION OF NATIONWIDE DATA ON THE RELATIONSHIP BETWEEN
TRIP GENERATION RATES AND QUANTITIES OF LAND USE
Trip Generation No. of Average
. Land Use Type Studies Standard Deviation of Average Li near
Trip Rates (%) Correlation (R)
Residential
Si ngle Fami ly 231 32% 0.88
Mul ti -Family 65 9% 0.95
All 296 21% 0.92
Commerci a 1
Sma 11 re ta i 1 58 52% 0.61
Shopping centers
Sma 11 54 50% 0.56
Medium 24 16% 0.89
Large 18 19% 0.90
~ All 144 34% o:t4
I
c.n Offices 15 36% 0.78
w
Man ufac turi ng
Sma 11 39 77% 0.26
Large 15 25% 0.87
All 54 m ~
Warehousing 16 103% 0.83
Average Statistics 49% 0.74
[
Source: [120] .
-------
TABLE 4-20
VALIDATION RESULTS FOR TRIP PRODUCTION
EQUATIONS DEVELOPED FOR PITTSBURGH, PA
Trip Type
Average Error
Corre 1 a ti on
Coefficient
Home-Based
Work
Shopping
Schoo 1
Miscellaneous
Non'-work
Other
Nonhome-Based
Average Statistics
+15.4%
-18.7%
-28.0%
+ 9.0%
- 6.5%
- 1.7%
-19.7%
:!:14.1%
0.97
0.89
0.70
0.76
0.83
Source:
[130] .
4-54
-------
TABLE 4-21
SUGGESTED VALIDATION CRITERIA FOR TRIP GENERATION RATES
Land Use Maximum Minimum
Ca te gory Error (:t%) Correlation (R)
Res i denti a 1
Single family detached 32% 0.88
Single family attached 32% 0.88
Multi-family low rise 9% 0.95
.Mu1ti-fami1y high rise 9% 0.95
Mobile homes 32% 0.88
Hotel s/Mote1 s 42% 0.81
Commerci a 1
<50K 51% 0.59
50-lOOK 16% 0.89
>100K 19% 0.90
Offi ce 36% 0.78
Manufacturi ng 51% 0.57
Wholesale/Warehousing 103% 0.83
Cultura 1 42% 0.81
Churches 42% 0.81
Hospitals 42% 0.81
Educational 28% 0.70
Recrea ti on 42% 0.81
4-55
-------
length distributions by various parameters, such as trip purpose, size of
urban area and family income. The range in these data are summarized in
Table 4-22. Average statistics extracted from these published data suggest
the following as testing criteria:
. For trip lengths, the maximum allowable errors to be used
are 44% for work trips and 40% for other trips.
The suggested maximum error criteria are not excessive in com-
parison to typical errors introduced elsewhere in the GEMLUP-I predictive
model. The average error of the land use predictive equations is !87%.[19].
The EPA emission factors used in the translation from land use to emissions
are averages for source classes and when applied to individual sources can
have errors ranging from one to several hundred percent [132]. Finally,
diffusion models, which are ultimately used to predict air quality levels,
are precise to only a factor of 2 or 3 in general [133].
2.
Mode 1 Val i da ti on
Three separate validations were performed to test the predictive
abil ity of the:
. Predictive equations for trip lengths
. Default values for trip generation rates
. VMT model as a whole using as input default values
rates and the default predictive equation for trip
for tri p
lengths.
The structure of default values and predictive equations in the existing VMT
model dictated the level of detail in the validation. Thus, trip lengths
were averaged for the area of analysis by trip purpose only (work or other).
Trip generation rates were averages for the area of analysis by land use
type and purpose, and VMT were totals for the area of analysis. A fourth
validation, involving VMT estimates based on local trip rate and trip length
data, could not be performed due to lack of sufficient, detailed data from
the transportation case studies. The validation of each of these three
elements is discussed in the sections below.
4-56
-------
TABLE 4-22
TRIP LENGTH DATA DISTRIBUTIONS COMPILED
BY THE FEDERAL HIGHWAY ADMINISTRATION
Di s tr;i buti on
Parameters
Tri p Length
Total (miles)
Range as Percentage
of Mean
Size of Urban Area
Earning a Living
Famil y Bus i ness
Civic, Educational
& Re 1 i gi ous
Social & Recreational
Fami 1y Income
. Earning a Living
Fami 1y Busi ness
Civic, Eduationa1
& Re 1 i gi ous
Social & Recreational
(excluding vacations).
. Average Statistics
Work Tri ps
Other Trips
4. 1 40%
3.2 57%
3.5 74%
8.6 66%
11.3 111%
7.9 141%
2.6 55%
14.8 113%
6.6
7.4
87%
77%
$0 urce :
[131] .
. 4-57
-------
a. - Tri p Lengths
Actual data on trip lengths were available from 8 case studies
for "work trips" and from 7 case studies for "other trips". The GEMLUP VMT
model was used to predict estimates of "work II and "other" trip lengths,
given data on network speeds and 5M5A population. These data are summarized
in Table 4-23 and scattergrams of actual versus predicted values are given in
Figures 4-2 and 4-3. Correlation coefficients are 0.49 and 0.03, respectively.
for work and other trips. Both predictive equatfonsgenerally underestimate
trip length, with an average bias of -32%. The standard errors of estimate,
are 50% and 56%, respectively, for work and other trips. These exceed the
acceptable criteria limits of 44% and 40%, respectively. Thus, it was neces-
sary to revise the predictive equations for trip length.
The technique used to revise thp. trip length equations was to
adjust the multiplicative constant in each case. The adjustment factor was
found by fitting the data to a regression line with a forced zero intercept
and computing the slope of the line. These regression lines are drawn in
Figures 4-2 and 4-3 and have slopes of 1.49 and 1.46, respectively. Thus,
the new predictive equations become:
Ll = 0.00447 * pO.20 * 511.49
L2 = 0.00219 [rO.1B * 521.40 + pO.26 * 521'25J
I
The bias of these equations, applied to the case study data, is zero. The
imprecision can be characterized by the revised standard errors of estimate.
The revised model was tested with the case study data and stand-
ard errors of estimate of 34% and 39% were obtained, respectively. As these
do not exceed the acceptable criteria limits, the revised trip length equa-
tions were judged suitable for use in the GEMLUP-II model.
The residual errors from the second validation are graphed
versus the actual trip length data in Figures 4-4 and 4-5. Examination of
these graphs show no evidence of changing variance in the error term, i.e.,
no heteroscedasti ci ty. I
I
4-58
-------
TABLE 4-23
. .
ACTUAL AND PREDICTED TRIP LENGTH DATA IN MILES
WORK TRIPS OTHER TRIPS
Transportation VMT Revised VMT Revised
Case Study Actu a 1 Model Model Actua 1 Model Model
Lex i ngton. MA ND 8.79 13.10 ND 4.43 6.47
Bennington. VT 6. 40 * 3.73 5.56 7 . 79 *. 3.86 5.64
Richmond. VA 5.70 5.05 7.52 5.70 5.10 7.45
Charlotte. NC 10.73* 7.95 11 .85 ND 5.90 8.61
Clearwater. FL 15.44 7.57 11. 28 11 .05 5.95 8.69
Roseville. MN 6.91 6.87 10.24 5.09 5.09 7.43
St. Lou; s. MO ND 11.28 16.81 ND 8.85 12.92
Boul der. CO 9.76 4.30 6.41 11.12 4.91 7.17
S. Phoenix. AZ 9.16 5.72 8.52 7.82 5.89 8.60
Vallejo. CA ND 5.72 8.52 ND 4.31 6.29
Auburn. WA 8.43 8.10 12.07 5.79 6.44 9.40
ND = No Data
*Estimated by weighted interzonal distances from origin/destination studies
ci rca 1970.
4-59
-------
r
", ",' , ,.', ..-,,:, , ,', ,
~IL~."'~UH8E~i;};tJ~~~:;~j'J~~;i;r<>ivo; ~~~ml;~~~(';;,~~~~~;MtD"WQ~k"I~IP" ~ENGTHS'
sCAtTERGRA'M:OF""; ~'CDOWN);::ACTUAL' ,.~',.':' ': ,":~'" :::,~"",.':,:"\/: :;:,>'~:'" (ACROSS)~:~ODE'{ , , '
';:,';:00; 1,f.:~:'~k~"+--;,~.:~E;:'~.__~;~~;c.;-:~,F,"-+~"; :;~ ~." +c ..- ~;~E_H - "-~ ;~.~- -+ --.,~;~~" -.- .~E;~E- -+_.~~ ;~E- -+ ~
I' .. . 1
, ..",
~,{,~:~\ 1;:';, :~,,:,"" 0, ,', ' , ;
""'"
-------
APPENDIX A
PATH ANALYSIS
A-l
-------
In this section we review briefly the concepts, assumptions, and prob-
lems associated with path analysis and its use in the GEMLUP-II study.
1.
Aim of Path Analysis
Path analysis was developed by Sewall Wright [1-4] as a technique
for studying and dealing with observed interrelated variables for which it
can be assumed that there are several "ultimate" or exogenous variables that
completely determine or cause them [5]. It is a method for studying the dir-
ect and indirect effects of variables taken as causes of other variables
considered as effects [6]. I~ is not capable of deducing causal relations
from available quantitative information (viz., correlation coefficients), but
rather it is intended to combine this quantitative information with qualita-
tive interpretation [1]. Thus, it is a technique useful in testing theories
rather than in generating the, and it can be used to study the logical con-
sequences of various hypotheses involving causal relations. In order to
implement the technique, the researcher must make explicit a theoretical
framework or model.
2.
Path Diagrams
Figure A-l presents various path diagrams which display graphically
typical patterns of causal relations among sets of variables. Figure A-la
shows the relation between seven variables, Xl' X2, X3' X4, U, V, and W,
where Xl' X2, X3, X4 are variables of the systems which are measurable.
The varlables U, V, and Ware unmeasured and outside the system, but do
affect variables in the system. These latter variables are often referred
to as disturbances or residuals. Usually they are not represented in a path
diagram and Figure A-la and Figure A-lb can be considered the same path dia-
gram. The residuals are "understood" to exist. The arrows in all the
figures indicate the causal flow. Variable Xl in Figure A-la and A-lb is
called an exogenous variable for its variability and is assumed to be deter-
mined by causes outside the system. Variables X2, X3, and X4 of Figures A-la
and A-lb are called endogenous. Their variablility is explained by exogenous
or endogenous variables in the system. Variables U, V and W of Figure A-la
can be considered exogenous. .
Sometimes there is more than one exogenous variable in the system
(see Figure A-lc and variables Xl and X2). In this case, the correlation
between the two is represented by a curved line. The relation between the
exogenous variables is usually not analyzed in path analysis.
The path diagram shown in Figure A-ld represents an interaction
between two endogenous variables X2 and X3' Analysis of this type of system
presents some serious technical problems which we will discuss later on*.
This interaction is also termed a reciprocal causation or a feedback loop.
*
See di.scussion of identifica~on in Section 9 of this Appendix.
A-2
-------
(a)
( b)
(c)
(d)
FIGURE A-l
ILLUSTRATIVE PATH DIAGRAMS
A-3
-------
3.
Approach (Structural Model)
In order to analyze the system of variables under investigation the
path diagram must be formalized mathematically. A structural model must be
developed which defines a causal structure specifying the network of causal
paths that exist between variables and identifying the parameters of causa-
tion. For Figure A-la, the mathematical model would be:
X2 = b21Xl + U
X3 = b31Xl + b32X2 + V
(A-l)
X4 = b41Xl + b42X2 + b43X3 + W
These equations are called structural equations. If the variables are stand-
ardized (i.e., have mean zero and standard deviation unity), then the b's
are called path coefficients. If they are not standardized, the b's are
called path regression coefficients or structural coefficients {there would
then also be an intercept in each equation of (A-l)). The b's are the para-
meters quantifying the causal effects.
Given a structural model, one has an explicit and quantitative
statement of theory and so it can be used to explain why variables vary
together as they do. It is possible with such a model to calculate how a
change in anyone variable in the system will affect the values of the other
variables. Also, it is possible to analyze how changing the structure of the
system would affect the character of the system [7].
4.
Connection Regression
It should be emphasized that the three equations of system (A-l)
each separately represent a multiple regression equation. The estimates
of the b's are obtained using ordinary least squares (i.e., the standard
regression technique). Up to this point, there is no difference between path
analysis and standard multiple regression. The b's are beta weights if the
Xis are standardized, otherwise they are partial regression coefficients.
5. 'Assumptions and Deviations
The transition from the path diagram of Figure A-la to the system
of structural equations (A-l) and the appropriateness of standard least
squares (i.e., multiple regression) for estimating the b's of (A-l) involve
a number of very important assumptions. We now consider these.
a.
Linearity
The change in one variable is assumed to always occur as a
linear function of changes in other variables. If it is believed or known
that the relation is nonlinear, for example, say multiplicative:
(b31Xl+ b32X2 + U)
X = e
3
(A-2)
A-4
-------
then the data must be transformed before the analysis. If there is non-
linear "causality" the usual path analysis will not detect it.
b.
Recursive
In order for the above discussion to be correct, the system
must not contain reciprocal causation or feedback loops (e.g., as in Figure
A-ld with the variables X2 and X3). The causal flow must be unidirectional.
At a given point in time, a variable cannot be both a cause and an effect of
another variable. Such a system is called recursive.
It is possible to solve a system of structural equations if
there are feedback loops by use of instrumental variables. Instrumental
variables are exogenous variables introduced into the system for this specific
purpose. They are used routinely in economic analysis [8,9]. Basically what
they do is to put enough constraints on the structural system so that by use
of estimation techniques (either two-stage least squares or limited informa-
tion least-generalized residual-variance solution) [8] the path coefficients
can be estimated. A point often not emphasized in the "traditional" sources
of path analysis is that little is known about the resulting estimates of the
path coefficients. The most one can say is that they are consistent estimates
of the true b's. This means that they converge in probability to the true b's
or, in other words, as the sample size increases they get "closer" to the
true b's. Basically, nothing is known about the estimates for small samples
(such as 40 in the current study). All the "nice" properties of the multiple
regression estimates (i.e., minimum variance, unbiased estimates) do not hold
for the new ones.
In this study we made use of the above techniques (i.e., use of
instrumental variables and two stage least squares) to solve for the bls. In
order to judge the usefulness and appropriateness of the estimated bls, four
different approaches were tried. The first three involved the use of approxi-
mate t tests (to judge if the b's are statistically significantly different
from zero), beta weights (to judge if in the standardized scale the magnitude
of the effect is "large") and the signs of the b's (to verify that the signs
of the b's are what theory would predict). The fourth approach involved the
use of a technique called "Jackknifing"* to verify that the b's were stable
(i.e., do not change radically with the inclusion or exclusion of just one
observation) .
Another way of solving the system if feedback loops exist is
to use lead-lag relations. This might be possible in longitudinal data [4].
Consider tne path diagrams of Figure A-2. In Figure A-2a, we have an inter-
action between variables X2 and X3. However, say X2 and X3 represent the
final states after a time period (say, 2 hours, for example). Further, say
it is possible to break the 2 hours into two time periods (time periods 1
and~) (see Figure A-2b). In time period 1, we may have variable X2 (called
X2l) as a cause of X3 (called X31) but not X31 as a cause of X21. Then into
time period 2 we have X3 as a cause of X2 (i.e., X3l X22). The system shown
in Figure A-2b is recursive and given sufficient data it is solvable. The
dynamics of the system over the full time period are needed. Unfortunately,
for the present study, sufficient data were not available to allow us to
implement this lead-lag device.
*The concept of "Jackknifing" is discussed further in Section 5 of this Appendix.
A-5
-------
(a)
(b)
FIGURE A-2
LEAD - LAG RELATIONSHIPS
A-6
-------
c.
Causal Priorities
The causal priorities among the variables must be "correctly"
established outside of the analysis. A major drawback of path analysis is
that it ~annot establish causal relationships using data alone. One needs
as input both data and theory. The ordering of the variables (that is, the
direction of the arrows in th~ path diagrams) must be established by the
investigator. "There is no error-check mechanism in path analysis to reject.
an incorrect ordering" [7]. It is common that the data 'can fit well two
very different orderings. The selection of the appropriate one depends on
correct theory. When one is dealing with variables ordered in time it is
often possible to know the causal priorities. Path analysis seems well
suited for this type of problem. For the present study, the ordering in
time was not available, and so a substantial effort was required to establish
causal priorities. Also, causal priorities were established by use of instru-
mental variables and two stage least squares for estimating the b's in feed-
back "loops (e.g., for the cases where it is not clear that a causal relation-
ship exists, if the estimated path coefficients (b) is zero then the analysis
indicates there is no causal relationship).
d.
Uncorrelated Residuals of Endogenous Variables
In the actual analysis the observed correlations among the
observed variables are used to estimate or identify the path coefficients.
If the residual (i.e., the disturbances) are uncorrelated, then the usual
least squares technique (i.e., multiple regression) can be used for the
estimation. If, however, the residuals are correlated, then the path coeffi-
cient cannot be identified by the ordinar~ least squares technique for there
are then too many unknowns. An implication of this assumption of uncorrelated
residuals is that all inputs must be entered explicitly into the analysis.
(It is possible to ignore a variable that affects two or more inputs as long
as it does not affect the residuals of any endogenous variables). This.
restriction of uncorrelated residuals is very important and meaningless and
confusing results may come out of the analysis if it is not met.
Because of its importance
we shall develop a bit further here the
inputs explicitly in the model. Say we
which are causal factors of a dependent
is:
to the project under consideration,
consequences of not including all
have three variables Xl' X2' and X3
variable Y. The appropriate equatlon
Y = bylXl + by2X2 + by3X3 + e
(A-3)
Here, e is the residual. Now say we do not include X3 in our system and
we "believe" the correct equation is:
Y = bylXl + by2X2 + e' (A-4)
el in (A-4) is a new residual. It is actually e+b X. It can be shown
[8] that the estimate we will obtain for b 1 will ~~tdallY be an estimate
not of byl alone, but rather of y
bYl + by3 (B31.2)
(A-5)
A-7
-------
Where b is the real path coefficient given in (A-3) and B is the
regress~~n coefficient of X3 and Xl after allowing for the ~ff~cts of X2.
T he important point here is that if we leave out an input it wi II reappear
in o'ur path coefficients. We will have no way of knowing how much of our
estimate is the real byl and how much is by3(B31.2). It is possible that
byl is pos~tive.yet.the value of (A-5) is ~egative (a compl~te.chan~e in ,
slgn). ThlS pOlnt lS not made very clear ln much of the eXlstlng llterature
on path analysis.
Another implication of this fourth assumption is that all
residuals should be uncorrelated with all other exogenous variables that
precede them in the system. If one suspects that the residuals are correlated,
then the technique of two stage least squares can be used to remove this
problem. However, i~ doing so, it introduces the previouslj discussed pro-
blem of having estimates whose statistical properties are not well known.
In the present project, for those equations not requiring two stage least
squares estimation techniques, we used a sufficient number of relevant vari-
ables in the model to minimize this problem of correlated residuals.
e.
Stability of the Estimates
In order for path analysis to be of use the resulting estimates
should be stable. They will be of little use if they change radically.
depending on the sample observations. In most applications of path analysis
this is achieved by using large samples and reliable measuring instruments.
Tukey [lOJ and Turner and Stevens [llJ with regard to this point suggested
using path regression coefficients rather than the usual path coefficients
(i.e., beta weights). Path regression coefficients tend to be more stable as
one applies the same structural model to different populations.
For the current project we had only 40 observations for use
in estimating model parameters. This is a small sample, and so there may
be a problem of obtaining stable estimates. One way of judging the stability
of the estimates obtained is to use the statistical technique of IIJackknifingli.
Instead of running one regression using 40 observations, Jackknifing involves
performing 40 regressions each containing 39 observations and then examining
the stability of the sets of path coefficients obtained. This technique
prevents the path analysis from capitalizing on chance elements in the data.
See Section IV for a full di$cussion of the application of this study.
f.
Multicollinearity
If the correlations between independent variables in a multiple
regression are extremely large in absolute value, we have what is called the
problem of multicollinearity. If present, it is usually impossible to sepa-
rate the influences of the variables involved. The effect on the path analysis
will be to produce unstable path coefficients (they will have large standard
errors) and beta weights greater than unity in value. Obviously, the purposes
of path analysis can be defeated if collinear variables are present. Notice,
these collinear variables can be both exogenous and endogenous as long as
they appear somewhere as independent variables.
A-8
-------
To handle the problem of multicollinearity we examined the zero
order (i.e., simple) correlations between variables for large correlations and
paid careful attention to the output of individual regressions looking for.
unusually large (in absolute value) beta weights and standards errors for the
regression coefficients.
g.
Suppressor Variable Problem
It
input variables.
interceptor 1 i ne
ther, say X2 arid
the equations,
is possible for a path analysis system to contain too many
For example, say X] refers to the size (diameter) of the main
and X4 is commercial land use in the area of interest. Fur-
X3 are two other variables, the structural model is given by
X2 = b21Xl + b2uU
X3 = b31Xl + b3vV
(A-6)
X4 = b41Xl + b42X2 + b43X3 + b4wW
and the simple correlation matrix is:
xl x2 x3 x4
xl 1.000 .800 .600 .000
x2 1.000 .480 .286
x3 1.000 .384
x4 1.000
Given this it can be shown that the solution for X4 is:
X4 = -Xl + .8X2 + .6X3 + .736W
(A-7)
The size of the sewer line appears to have a negative impact on commercial
land use (path coefficient is -1).
The actual situation is that a model was developed to produce
the correlation matrix and it was:
X4 = .48U + .48V + .736W
(A-8)
U,V and Ware the residuals of X2,X3 and X. X4 and Xl are actually uncor-
related. The "strange" result shown in (A~7) arises because of the following:
If we use X2 and X3 to predict X4 we will have a useful result because X2 and
X3 contain U and V which are related to X4. However, X2 and X3 also contain
Xl which is irrelevant. In (A-7) the term Xl is used to "suppress" the
irrelevant components in X2 and X3. That is, X2 and X3 do contain a portion
of Xl in them and this portion is subtracted again in the equation by adding
A-9
-------
the nega~ive term ~ith Xl.. Fig~re A-3 illus~rates the pro~lem.. Th: sy~tem
of equat10n (A-6) 1S assum1ngF1gure A-3a wh11e the true sltuat10n 1S F1gure
A-3b. The importance of this illustration is that one cannot just include
anything into a structural system and believe all will work out well. .
To handle this problem we examined the. zero order correlations
between all endogenous variables (i.e., those that are dependent variables
in some regressions) with all other variables. Those cases where zero or
near zero correlations appeared were noted. Later, if the partial relation-
ship proved to be significant, an examination was performed to see if a
suppressor variable problem existed by investigating the stability of the
model coefficients including/excluding the variable.
6.
Other Assumptions
There are other assumptions that deal solely with the fact that
path analysis is a regression technique (e.g., independence of observations).
We will not enumerate them for they are standard [12].
7.
Theory Trimming
Often the approach used in path analysis is to develop the most
elaborate system ancr then, after estimating the involved path coefficients,
attempt to refine or trim the system by dropping those paths that have coeffi-
cients close to zero. Those path coefficients near zero imply no or only a
weak causal linkage between the involved variables. The possibility for
refining or trimming a theory and thus making a theory more parsimonious is
of considerable significance.
There are a number of ways of deciding which path coefficients can
be dropped. First, there is the criterion of statistical significance e.g.,
drop all path coefficients which do not differ significantly in the statisti-
cal sense (0.05 level of significance) from zero. With large samples, this
could imply retaining some very small coefficients and with small samples
this may imply dropping some very large coefficients. Second, there is the
possibility of dropping all coefficients whose values are less than some
preassign~d fixed value (e.g., say 0.05 or 0.03) [13]. There is also the
dependence analysis procedure of Boudon [14]. In this technique, one usually
assumes that some path coefficients of the structural equations are zero. To
test this procedure, one sees how well the non-zero paths reproduce the
original correlation matrix. In general, for any trimming procedure it is
important to test how deletion of some path coefficients affects the ability
to reproduce the original observed correlations (an important property of
path analysis itself is the ability of reproducing the original observed
correlations by means of the path coefficients).
The approach used in trimming the model is outlined in Section IV
and consisted of developing a comprehensive set of statistical test criteria.
A-10
-------
(a)
(b)
FIGURE A-3
ILLUSTRATION OF SUPPRESSOR VARIABLE PROBLEM
11
A-ll
-------
8.
Quantification of Direct and Indirect Effects
An important application of path analysis is the analysis or decom-
position of a correlation into its various components [6J. For a given causal
model it is possible to determine what part of a correlation between two
variables is due to the direct causal effects and what part is due to the
indirect effects. As an example of this consider Figure A-4a. The system of
structural equations that correspond to the model of Figure A-4a is:
X2 = b2l Xl + U
X3 = b3l Xl + b32 X2 + V
(A-8)
For the sake of simplification let us assume the Xis are standardized (i.e.,
have mean zero and unit standard deviations). Suppose we wish to analyze.
the correlation between the variables Xl and X3. Let r31 represent this
correlation. Now there are two ways that Xl affects or lnfluence X3. First,
there is the direct effect of Xl and X3' represented in Figure A-4a by the
arrow going directly from the variable Xl to the variable X3. This effect
is quantified by the path coefficient (b3l) in the second equation of (A-8);
that is, a one unit change in Xl produces a (b3l) unit change in X3. Second,
Xl effects X2, which in turn effects X3. This indicates that Xl directly
effects X3 through X2. By use of standard manipulations it can be shown that
this indirect effect of Xl and X3 is quantified by the product (b32 b2l). This
represents the effect of Xl on X2 (b2l) and then the effect of X2 on X3 (b32).
Further manipulation can be done to show that the correlation r13 is equal to:
r13 = b3l + b32 b2l
(A-g)
That is, the net effect of multiple direct and indirect relationships is the
sum of the individual effects.
The above ideas can be extended so that path diagrams can be II read II
to define the net effect from one variable to another, even when the variables
are connected in complex systems involving multiple chains of causation and
intervening feedback loops. Such analyses allow one to trace the reverbera-
tions of model inputs; providing insight into how system variables will be
correlated in empirical data, and how changes in inputs would effect the
system's endogenous variables. Figure A-4b offers a more complex path analytic
model, including a feedback loop between variables X2 and X4. Suppose we wish
to analyze the total effect of Xl on X3. As before, there is a direct effect
(b3l)' but there are now numerous indirect effects. A one unit change in Xl
produces a change of (b2l) units in X2. This causes a change of (b2l b42)
units in X4. The change in X4 in turn causes another change in X2:
(b2l b4Z b24). So after one cycle, X2 is incremented by the value of a return
effect 1n the feedback loop. This effect can in turg move around the loop
again, causing still another change in X2 of (b2l b42 b2~) units, and that in
turn can move around the loop once more, ad infinitum. The net effect of
Xl on X2 is therefore:
(b2l) + (b2l b42 b24) + (b2l b4~ b2~) + ... (b2l b4~ b2~) + ...
A-12
-------
b2l
b2l
~cv
b42
(a) Simple Case'
(b) Complex Case
FIGURE A-4 ILLUSTRATION OF THE DIRECT AND INDIRECT
CAUSAL RELATIONSHIPS IN PATH DIAGRAMS
A-13
-------
This infinite geometric series can be expressed more simply by the expression:
b21
l-L
where
L = (b42 b24)' the "return effect" of the feedback loop.
The total indirect effect of Xl on X3 is therefore just:
(b21 b32)
1 - L
and the net effect of Xl on X3 is:
(b21 b32)
b31 + 1 - L
The determination of the total effect of variables on one another
in a complex path analytic system is, in general, not as simple as the pre-
ceeding example would indicate. Heise [15J has developed a generalized approach
to determining such effects, and the reader is referred to that text for a full
presentation of the techniques involved.
In the current project, we performed a decomposition analysis of
correlations into direct and indirect effects for all significant variables
contained in the final trimmed path analysis model.
9.
Identification
In this final section we briefly review the limitations imposed on
the model specification due to the statistical requirement of identification.
When there are feedback loops or reciprocal causal paths in a path
analytic model, then the residuals in the model are correlated with the endo-
genous variables and use of ordinary least square techniques will yield biased
and inconsistent estimates of the structural coefficients. For example. say
Yl and Y2 are two land use vari~bles measured in appropriate units and
say there is believed to be a feedback loop between these. Then possible
structural equations are:
Yl = al + b21Y2 + el
(A-10)
Y2 = a2 + b21Yl + e2
A-14
-------
It can be shown [16] that the residual el is correlated with Y (since Y is
a function of Yl) and the residual e2 is correlated with,Yl' 6rdinary l~ast
squares (which assumes uncorrelated residuals) applied to either of these
equations will result in biased and inconsistent estimates of the structural
parameters al' b21' a2' and b12' By biased and inconsistent estimates we
mean that the average value of the estimates will not equal the values of the
structural parameters no matter how large the sample size may be.
The approach usually followed at this point to obtain appropriate
estimates of the structural parameters is to introduce instrumental variables
and apply two stage least squares. This approach will yield consistent esti-
mates if the relevant equations are properly identified. The term lIinstru-
mental variable" merely means a variable which acts as an instrument in
making the system identifiable. Below, we review how instrumental variables
are introduced, how two stage least squares is applied and how the identifi-
ability condition arises.
Given the structural system of equation A-10, one must introduce
at least one new variable which is a IIcause" of Yl but not for Y2. Let us
call this variable X3' Similarly an X4 should be introduced for Y2' X3
and X4 are instrumental variables. There is no reason why there need be only
one instrumental variable for each endogenous variable, however there must
be at least one. To allow for more generality let us say we have a second
instrumental variable for Yl and let us call it XS' The important point is
that we need some instrumental variables which are causally related to one of
the endogenous variables, but not to the other in the system. It is possible
to have some exogenous variables (which are still often called instrumental
variables) which are causes of more than one of the endogenous variables.
For our example, let us say X6 is such a variable.
The revised structural system is now,
Yl = al + b21Y2 + b31X3 + bS1XS + b61X6 + ell
Y2 = a2 + b12Yl + b42X4 + b62X6 + e'2
(A-ll)
where ell and e'2 are not the same residuals as el and e2 of (A-10).
It can still be shown that Y2 is correlated with ell and Yl is
correlated with e'2 (i.e., we have correlated residuals), so ordinary least
squares still cannot be applied directly to obtain consistent estimates of
the structural parameter b's. However, if we manipulate the equations of
(A-ll), a new system of equations which have Yl and Y2 only on the left, viz.,
Yl = all + b'31X3 + b'41X4 + b'S1XS + b'61X6 + elll
(A-12)
Y2 = a'2+ b'32X3 + b'42X4 + b'S2XS + b'62X6 + e"2
A-1S
-------
is possible. This new set of equations gives us a way of estimating the bls
of (A-ll). Notice these equations have all the instrumental variables on the
right hand side. The equations of (A-12) are often called the reduced form
equations. Ordinary least squares may be applied to obtain estimates of the
als and bls (and in turn estimates of Yl and Y2 are possible). This is the
first stage in two stage least squares analysis.
Next, the estimates of Yl and Y2 obtained from (A-12) can b~ sub-
stit~ted into the right hand side of the equations in (A-ll). Using Yl
and Y2 to represent these estimates, the system (A-ll) with this adjustment
becomes:
A
Yl = al + b21Y2 + b31X3 + bS1XS + b61X6 + e'l
A
Y2 = a2 + b12Yl + b42X4 + b62X6 + el2
(A-13)
. A
Y2 is not correlated with e'l, since Y2 is a function of only th~ exogenous
variables X3, X4, Xs and X6 and not Yl, and for similar reasons Yl is not
correlated with e12. Tbus, ordinary least squares may now be applied to
obtain consistent estimates of structural parameters, the b's. This is the
second stage in two stage least squares analysis.
A In order to obtain consistent estimates of the b's it is necessary
that Y2 is not a linear function of X3, Xs and X6 alone.. This fs the identi-
fication problem (we want to "identify" or estimate the bls) and by bringing
X4 in the system we have this necessary condition satisfied. Recall X4 is the
instrumental variable that is causally related to Y2 but not Yl. A similar
necessary condition is needed to identify the bls of the second equation of
(A- 13) .
The necessary condition for identification can be formalized as
follows: suppose we want to estimate the b's of the first equation of the
system (A-ll). First, one counts the number of instrumental variables in
the full system. Here, there are 4 (X3, XS' X6 and X4). Next, one counts
the number of instrumental variables excluded from the equation under con-
sideration. Let us call this number mO. Here mO is 1 for variable X4.
Then we count the number of endogenous variables on the right hand side of
the equation. Let us call this number q: q for this problem is 1 for Y2.
The necessary condition for identifying the parameters of the equation under
consideration is then [16]:
mo .:::. q - 1
(A-14)
For our example, 1 > 1 - 1 = 0 so we have identification. Condition (A-14)
is necessary but not sufficient (i.e., if an equation is identified it must
satisfy (A-14), however, there are equations that satisfy (A-14) but are
not identified). The sufficient condition involves rank conditions of
various matrices [16]. Because we anticipated the existence of feedback loops
in the GEMLUP-II model, the problem of identification required careful considera-
tion. In the model specification phase of this project, care was taken to
specify exogenous variables that could serve as instrumental variables. Addition-
ally, the necessary conditions for identification (A-14) was continually checked
for compliance.
A-16
-------
11
APPENDIX B
CAUSAL ANALYSIS IN
GEMLUP PROJECTS
B-1
-------
B.
CAUSAL ANALYSIS IN GEMLUP PROJECTS*
This chapter attempts to explain why a causal analysis of secondary growth
effects is undertaken in GEMLUP projects. After the GEMLUP I reports [17-19]
were distributed the need for such an explanation became obvious because the
Project Officer received a number of questions relating to the issue of
ca usa 1 i ty. The ques ti ons went somethi ng 1 i ke thi s: IIHow do you know that
the land uses you are predicting are really secondary development induced
by the major project?1I This chapter is a long answer to that short question.
There are many subtle issues involved, and a large body of literature exists
concerning each issue. The discussion here addresses only the main issues in
as straightforward a way as possible.
1.
Prediction, Explanation, Theory, and Model
. GEMLUP projects contain tasks that directly involve prediction, expla-
nation, and mOdeling [20,21]. The projects indirectly involve theory formula-
tion and testing. The overall purpose of the projects is to develop prediction
equa ti ons to estimate the amount and type of secondary 1 and use induced by
constructing and operating a major project, such as a wastewater facility, a
large residential complex, or an office/industrial park. To ensure that the
prediction equations are formulated on a firm scientific basis, a causal analy-
sis of secondary development is undertaken to explain the structural relations
that we think exist in induced development. The explanation takes the form of
setting up and testing a model of a theory of induced development. The model
is made up of the path analytic diagrams and equations described in the first
volume of each GEMLUP final report; and it is a mathematical description of the
verbal theory hypothesized in the same volume. All of these concepts are
~laborated upon below.
a.
Prediction and Explanation
The essence of science is to explain. Usually the focus is to
explain empirically defined phenomena according to commonly accepted scientific
(definite) criteria [22]**. Explanation is formally defined as a process by
which singular events are related to other events through the use of appropriate
* This chapter was written by the EPA Project Officer, Thomas McCurdy.
**Criteria involve: A definition of what is acceptable evidence, how the
conduct of scientific inquiry should proceed, the capability to generalize
from the evidence, and the minimum standards for internal and external
validity [22]. These issues are the province of "philosophy of science. II
B-2
-------
general statements or theories [22J.+ Singular events and isolated phenomena
have no meaning: they are only facts and have no significance without generali-
zation into widely applicable laws governing the behavior of empirical events
or objects. It is these laws and generalizations that allow us to predict the
outcome of unknown events [22].
Saying the same thing less formally, we have to be concerned with
generalization and explanation in GEMLUP so that we can predict. While ex-
planation and prediction are not identical, they are obviously related. Kaplan
in the The Conduct of Inquiry says that from the viewpoint of a philosopher of
science, the ideal explanation is one that allows' prediction [23~349-350].
Predictions can be and often are made even though we are not
in a position to explain what is being predicted. This capacity
is characteristic of well-established empirical generalizations
that have not yet been transformed into theoretical laws. . . In
short, explanations provide understanding, but we can predict
without being able to understand and we can understand without
necessarily being able to predict. It remains true that if we
can predict successfully on the basis of certain explanations
we have good reason, and perhaps the best sort of reason, for
accepting the explanation.
Kerlinger and Pedhazur [6J bring the subject closer to home in
their discussion of the role of regression analyses in explaining and predict-
ing phenomena. This discussion is particularly apropos here because GEMLUP
projects use multiple regression analysis for both purposes, albeit in slightly
different forms. Stepwise multiple regression analysis is used for purely
predictive work in GEMLUP, while path analysis, which is actually multiple
regression analysis used in a particular way, is used for explanatory work.
Computationally, the techniques are very closely related.
Regression analysis can play an important role in predictive and
explanatory research framewo~ks. Prediction and explanation reflect
different research concerns and emphases. In prediction studies
the main emphasis is on practical application. On the basis of
knowledge of one or more independent variables, the researcher
wishes to develop a regression equation to be used for the prediction
of a dependent variable, usually some criterion of performance or
accomplishment. The choice of independent variables in the predictive
framework is determined primarily by their potential effectiveness
in enhancing the prediction of the criterion.
+There is a hierarchy of explanation organized according to how widely applic-
able are the findings and results of scientific endeavor. The hierarchy
follows, ranked from the least to the most generally applicable [22,24J.
1. Strict definition and description of facts, or observations.
2. Order and classification of facts in a typology.
3. Comparison and relation of facts according to a model.
4. Generalization of facts into a theory.
5. Generalization of facts into an empirical law, or a universal
generalization.
6. Generalization of facts into a fundamental generalization.
7. Generalization of facts into an axiom.
The causal analyses in GEMLUP projects involve geDeralizations of the t~pe listed
in both items three and four, as they involve developing and testing a Ifactor
theory," which is a limited generalization. This will be discussed in more de-
tail later.
B.3
-------
In an ~xplanatory framework, on the other hand, the basic emphasis
is on the explanation of the variability of a dependent variable by
using information from one or more independent variables. The choice
of independent variables is determined by theoretical formulations
and considerations. Stated differently, when the concern is ex-
planation, the emphasis is on formulation and testing explanatory
models or schemes. It is within this context that questions about
the relative importance of independent variables;(f:become particularly
meaningful. Explanatory schemes may, under certain circumstances,
be enhanced by inferences about causal relations among the variables
under study [6]. .
A concrete but simplified example would be that if a researcher
is interested only in prediction, then the focus is on the magnitude of R2, the
coefficient of determination, and its statistical significance, but if the
research is interested in explanation, the relations between the dependent
variable and its set of independent variables are important [6]. The focus
is thus on the entire regression equation and the regression coefficients,
the 81s (if standardized; bls if not standardized). Squared semi-partial
correlations,* in particular, are of prime interest because they identify
the contribution to the variance of the dependent variable that each independent
variable adds after the variance contribution of preceding variables are ac-
counted for [6~his is statistical control of independent variable effects,
analogous to control of independent variable impacts in a classic experimental
design. It is the conceptual basis for the causal analytic work done in GEMLUP,
which allows us to infer the induced secondary developmental impacts of major
land use projects. It allows us to explain secondary development caused by
large projects.+
b.
Theory and Model
As discussed above, a theory is a specific type of generalization.
There are a number of definitions of the word, but one good one is lIa plausible
or scientifically acceptable general principle or body of principles offered
to explain phenomenal! [30]. An alternative definition is a deductively re-
lated group of general statements that explain observable phenomena [22]. A
theory is an abstract and symbolic construction that relates general statements
on the basis of underlying similarities; it is not an inductive generalization
from a body of observations.
*A semi-partial correlation represents the correlation between two variables
with the influence of another variable or other variables removed from one of the
variables being correlated. The square of the semi-partial correlation coefficient
indicates the amount of variance contributed by separate independent variables in
an regression equation, as mentioned above in the text. The variance contributed,
however, is positionally dependent because it is the variance contributed after
that denoted by the preceding variables [6].
An example symbol for a semi-partial correlation isrY(2.1)'. which means the
correlation between the dependent variable y and variable 2 controlled for, or
partial led from, the correlation between variables 1 and 2.
For an extended discussion of partial and semi-partial correlation, see
pp. 84-99 in Kerlinger and Pedhazur [6].
+This point is elaborated upon in the section on causal analysis.
8-4
-------
While the discussion of theory so far seems very academic. GEMLUP
projects indirectly involve theory testing and formulation. The kind of theory
investigated in GEMLUP, however, is a special type of theory, called factor
theory. It is the most rigorous type of theory that we could ascribe to.*
Factor theory is characterized by narrow and non-overlapping
generalizations, a selective, explicit enumeration of all factors hypothesized
to influence a given phenomen, and the use of empirically defined variables to
represent factors involved in the theory [25]. The most theoretical aspects
of this type of theory lie in the selection of the general statements in the
theory and the positing of rules of interaction governing their application.
Almost every effort at causal explanation involves factor theory
because of its explicit enumeration of (all) factors leading to a particular
development. However, it is limited theory because it does not readily sug-
gest other generalizations [22J. The independent selection and arrangement
of general statements focused on a definite phenomenon in factor theory pre-
cludes such generalizations.
A model, on the other hand, depicts the structure of a theory. It
has no explanatory usefulness by itself, and there is no causal relationship,
or isomorphic correspondence, between model and theory. A model aids under-
standing by explicitly delineating structural relations in a theory, although
Meehan [22J emphasizes that extensive reliance upon a model leads often to an
overemphasis on rigor and precision and a loss of understanding. A model is
not used in an exploratory sense to discover underlying generalizations nor is
it used in the theoretical sense to explain general statements. It is used to
depict the structural bones of a system, and that is how it is used in GEMLUP.
The form of the model used to depict induced development in GEMLUP
is path ana1ysis,-a-causal relationship. This is in keeping with factor theory,
which is also a causal approach to explanation.
2.
Causal Analysis
The first topic to be addressed in this section is causality and causal
thinking. The need for, and limitatio~s of, causal relations in modeling are
also addressed. The second topic is the connection between causal thinking and
modeling and path analysis.
a.
Causa 1 i ty
Causality, or the cause-effect relationship, is an important as-
pect of science. Classical experimentation involves isolating one independent
variable from all others in a relationship, varying it, and recording the re-
sulting change (if any) on the dependent variable of interest. Change in the
*There are many kinds of theory. For instance, Meehan [22J discusses the
following kinds of theory: deductive (pure deductive and probabilistic
deductive), concatenated, instrumental, factor, and quasi-theory.
8-5
-------
dependent variable is then assumed to be caused by change in the independent
variable, as all other variables were held constant. While systems analysis
has altered this one-to-one isolated scientific approach, causality is still the
dominant philosophical principle in scientific research[26].*
liThe noti on of causal i ty appl i es when the occurrence of one event
is reason enough to expect the production of anotherll [19]. The word itself
is used in three ways: (1) as a category to denote a particular bond rela-
tionship between a cause and an effect, (2) as a principle, or general law,'
relating objects, and (3) as a doctrine which purports that causality is the
only form of determinism operating in reality [26]. We will only be concerned
in GEMLUP projects with the first use of the word.
Philosophers of science interested in causality do not believe
that the phenomenon holds 100% of the time in all circumstances. They are much
more modest in their claims. For instance, Mario Bunge in Causality states
that causal chains work best as rough approximations for relatively short
periods of time. They are valid in some contexts because causality is a rough
reflection of reality and is methodologically unavoidable: there is just no
alternative available that can be used for explanation and prediction [26].
There is a place for chance variations, random impacts, and side
reactions in causality [26]. The idea is that even if these effects do not
cancel out each other, their net effect is statistically tractable. There is
also a place for multiple causes, multiple effects, and mutual causality
(feedback). Their impact on a causal relation can be determined statistically
or, in the case of feedback, by disaggregation into asymmetrical causal stages.
What cannot be handled very well in causality is systematic interaction among
causes or effects. The reason for this is that interaction introduces non-
additivity into the system, which cannot be handled in a causal analysis. .
(Which is based upon the premise that the joint effect of many forces, or
causes, is equal to the vector sum of separate forces.) If interaction is
suspected to be a property of the system under study, it may be possible to
approximate this relation by a relatively simple transformation (e.g., a
multiplicative model, or a logarithmic transformation) so that causal modeling
can be used on the system [27]. If a transformation is not theoretically
reasonable, another form of analysis must be used for the interactive system.
Not everything can be causally modeled, in other words.
*A formal definition of causation appears in David Heise, Causal Analysis [15].
An event C, causes another event, E, if and only if .
(a) an operator exists which generates E, which responds to C, and which
is organized so that the connection between C and E can be analyzed
into a sequence of compatible components with overlapping event fields;
(b) occurrences of event C are coordinated with the presence of such an
operator: such an operator exists within the field of C;
(c) when conditions (a) and (b) are met, when the operator is isolated
from the fields of events other than C, and neither C nor E is present
to begin with, then occurrences of C invariable start before the
beginning of an occurrence of E;
(d) when conditions (a) and (b) are met, C implies E; that is, during
some time interval occurrences of C are always accompanied by
occurrences of E, though E may be present wi thout C or both events
may be absent.
B-6
-------
Additional drawbacks of causal analysis are (1) the treatment of
(causal) relations in isolation from the environment as separate events, and
(2) the downplaying of time in the system [26]. The first drawback is common
to most systematic attempts to model reality. The whole world cannot be mathe-
matically modeled, so simplications have to be made throughout the analysis.
This results in an artificial boundary being placed around a system, with the
emphasis being placed on what goes on inside the boundary rather than on the
interfaces between system and environment [28]. Not only that, but causal
analysis focuses on isolated events within the system, again oversimplifying
reality. (Bunge calls this ontologically objectionable.) It is analytic and
not synthetic, in other words.
Time is usually not explicitly handled in most causal analyses,
other than the logical time relation inherent in the definition of causality:
"an event is not caused by other events that occur after it" [15]. (This is
the so-called temporal ordering criterion of causality.) This situation is
due more to lack of knowledge and theory regarding time relations in social
systems than an inability of causal analysis to incorporate time in its methods.
In other words, causal analyses can use "higher level" techniques to get time
into the equations (finite differencing) so that meaningful conclusions can be
drawn concerning the results. We felt that GEMlUP models had to be static
and not dynamic for these reasons and one other--money. Obtaining longitudinal
data is a very expensive process and is well beyond the land Use Planning Office1s
budget. Hence, the use of static cross-sectional path anlaysis to test the
causal model [25].
b.
Path Analysis
Path analysis is a systems approach to testing causal relations in
an operationalized theory (a model). It is one of a number of partial-
correlation analysis techniques used for modeling a system of interrelated
vari ab les that ul timately depend upon outsi de (exogenous) vari ables to "dri veil
them [29]. The technique has found its greatest application in sociology and
econometrics, although it was first developed by a biologist, Sewall Wright
(in 1934). lucid explanations of path analysis are found in Van de Geer [29],
Blalock [28], Heise [15], and Benesh [17], The last citation is Volume I of
the GEMlUP reports. Chapter II of the Volume, written by Dr. Ralph D'Agostino
of Boston University, contains a good summary of the technique taken from the
references just noted; the Chapter is reproduced, in updated form, in Appendix
A of this report. ,
The only attribute of path analysis that will be discussed here
is its ability to "tease out" causal relations. This ability is the reason
why we use it in GEMlUP projects--and the reason why we can state that the
land uses identified as belng secondary development induced by constructing
and operating a major land use project are just that. It is the way in which
"disturbance" is controlled in the land use development system so that specific
cause and effect relations can be identified [15]. And this ability to tease
out causal relations via partial-correlation provides us with the answer to
the question posited earlier: i'Huw do you know that the land uses you are
predicting are really secondary development induced by the major project?" The
answer is, finally, artial-correlation anal sis as used in a s stems context
via ath anal sis allows us to infer causallt Wlt a known ro a 1 lt 0
error, or in our context, allows us to infer induced develo mente This sentence
B-7
-------
is the short answer to the short question. But we are not through yet, as
more should be said concerning partial-correlation analysis. (See also the
material on partial and semi-partial correlation presented earlier under
"Prediction, Explanation, Theory, and Model.")
Heise presents the case most clearly in a discussion of separat-
ing out the individual effects of multiple causes. The mish-mash of multiple
causation is really what the question of interest reproduced above is all about.
The question can be rephrased as: IIGiven that some development has occurred
in the area of interest after the major project was built, how do. you know it
was induced, or caused, by the major project and not something else?1I Obvious-
ly, multiple causation is the issue here, and we turn to Heise for elucidation
[15] .
Multiple causation complicates causal analysis because it creates a
situation in which the value of an effect is not determined solely by the.
cause of focal interest but by other causes as well. When trying to study
the relation between a specific cause and effect, all the other causes act
as disturbing factors that confound and frustrate analysis. To study the
relationship of interest the disturbing factors somehow must by controlled.
Three different strategies are available.
One approach, associated mainly with classical experimentation in the
physical sciences, involves isolation of the relationship of interest. Any
disturbance depends on a causal relationship, so disturbances occur because
of the existence of some operator coordinated with some source event. There-
fore, if the operator for a disturbance can be disconnected, made inoperative,
or insulated from its stimulus events, this cause-effect rela:tionship no .
longer exerts its influence. Moreover, if the causa] relations for all disturb~
ing factors are disrupted, the relation between thi remaining cause and effect
can be analyzed without worrying about disturbances. . .Employment of this
strategy is not always so simple as it may seem. It takes considerable skill
and knowledge to disrupt every conceivable operator or to control all unwanted
events. Indeed the strategy may be applicable only in problems in which a
great deal of scientific knowledge already exists to guide experimental design.
A second method of dealing with the problem of disturbances is to estab-
lish a kind of passive control over the disturbing factors by observing them
in detail to determine when one or another is creating an unwanted effect so
. that it can be discounted or otherwise adjusted. In its simplest form this
.would mean restricting analysis to periods in which disturbing factors I are not.
. creating disturbances. . . .
Third, the problem of disturbing factors can be approached from a statis-
tical perspective. It is accepted that any single observation is hopelessly
confounded by disturbing factors, but if enough instances are observed in which
the presumed source operates at certain levels it should be possible to deter-
mine whether the source has the proposed effect on the average. In averaging
over many cases, the effects of the disturbing factors hopefully will cancel
one another. . .
B-8
-------
A key requirement here is that the disturbing factors must be un-
related to the causal event of interest. If this were not true--if
one disturbing event always tended to occur with the causal event of
interest--then the effect of that disturbance would not disappear with
averaging. Furthermore, it is always necessary to obtain observations
with the presumed cause both present and absent, or set at various
levels, in order to a~certain whether the presumed source makes any
difference.
The statistical averaging idea has been elaborated in two different
directions. The most rigorous procedure (but frequently impractical
in social situations) is the statistical experiment in which a lack of
coordination between disturbing factors and the presumed cause of in-
terest is guaranteed by randomly assigning cases for observation onto
two different sets and then providing the hypothesized causal event
for just one of them. Then, because the presumed cause is uncoordi-
nated with any of the possible disturbing factors, its effect can be
assessed by finding the average magnitude when the presumed cause is
absent.
In the second approach observations of disturbing factors are used
to define groups of measurements in which each disturbance has a con-
stant value. Then the value of the effect variable in each group is
reset arithmetically to the magnitude it would have if disturbances
had not been present, whereupon the relation of interest can be ex-
amined to see whether there is correspondence between the hypothesized
cause and the adjusted effect indicating causality. In actual appli-
cations the adjustment approach really amounts to studying several
causal relationships simultaneously. The procedure for adjusting for
the impact of a second cause while studying a first requires making
an adjustment for cause one in order to determine how much to adjust
for cause two. At first, this sounds hopelessly circular and complex.
In fact, procedures exist for making this separation, given numerous
observations on all relevant variables. . . .
Heise then references two chapters in his book for elaboration of
the procedures. The first deals with the logic behind causal thinking and the
second describes methods to identify and estimate causal relations in path
analysis. The latter boils down to partial and semi-partial correlation analy-
sis. This is how statistical control is used to infer causality. This is why
we do path analytic modeling in GEMLUP.*
*If the logic is not clear here or if the reader is still not convinced, please
read Heise [15J and Kerlinger and Predhazur [6J. Blalock [27J and Van de Geer
[29J also discuss the issue, but not as directly.
B-9
-------
" _-.J,---
APPENDIX C
AERIAL PHOTOGRAPHIC INTERPRETATION
C-l
-------
'\
C. AERIAL PHOTOGRAPHIC INTERPRETATION
As part o~:the data collection for this study, it was necessary to
inventory land use--by category, in the case study areas of analysis.
Although some regions were found to maintain detailed land use
maps and inventories, these were the exception. Therefore, this information
was generally extracted from aerial photographs of the case study areas. The
experience from the GEMLUP-I project [17] indicated that this element of the
data collection had the greatest potential for measurement errors. Thus, the
objective of this task was to determine the "typical" errors associated with
estimating amounts of land use, by category, from aerial photographs. The
approach involved a comprehensive review and compilation of information in the.
technical literature.
1 .
Literature Review
A total of 33 sources were reviewed. Telephone conversations
with personnel in the U.S. Geological Survey (USGS) Geography Applications
Program supplemented these sources. Of these sources, twelve did not indicate
the level of accuracy that would result from using their recommended pro-
cedures for interpreting aerial photographs. The remaining sources offered
no consistency with regard to their method of land use interpretation, scale
of the photograph used, or photographic equipment. A summary of these sources
is indicated in Table C-l.
As noted in Table C-l, the range in photographic scale reported
in the literature sources reviewed ranged from 1:2,400 to 1 :250,000.* In
addition, the methods of interpretation varied, including ground checks by
windshield surveys of remote sensing data [31], center point checks of l-km
grid cells [32], field comparisons along linear traverses [32], the inter-
pretation of items which could only be checked through ground examinations
[33], and the direct examination of color infrared photographs [34]. Further,
the sources varied with regard to the nature of the area to be interpreted.
For example, Lindgren [34] concentrated on residential areas; Collins and
E1-Beik [35] dealt solely with a portion of the City of Leeds, England;
Branch [33] examined a minor portion of Baldwin Park, California; and Fitz-
patrick [32] covered the Central Atlantic Coastal Region stretching from the
southern border of Virginia to the southern quarter of New Jersey. In addi-
tion, the literature reveals that a considerable variety of land use classi-
fication schemes were employed. Collins and El-Beik [35] classified urban
land use according to eight major groups; Fitzpatrick [32] used a revised
USGS Level I and II classification system; and Mausel et al. [31] adopted
a system of earth features that could be delineated by machine assisted
processing of LANDSAT-l multispectral data. Finally, Simpson [36] based his
* The ratio expresses the size relationship between features on a map and the
same features on the earth's surface. Thus, 1:2,400 states that 1 inch on
the "map represents 2,400 inches (or 200 feet) on the earth's surface.
C-2
-------
TABLE .C-l
Sur.tMARY OF AERIAL PHOTOGRAPHIC ACCURACY DETEp.mNATION
("'")
I
\.N
Authors Overall Accuracy Most Accurate Category Leas t Accurate Category Comments
-
Hausel, Leivo & 85-90% Water (98%) Remote sensing data checked
Lewellen [31] by windshield survey
Harden [37] 90% Rule of thumb based on
cumulative experience
Collins and El-Beik [35] 92% Water bodies (100%) Complex urban land Stereoscope examination of
Transportation (100%) use (36.4%) 1:10,560 scale aerial photos
Open improved lands (99%) Commercial (84.5%) with ground check
Open unimproved lands (96%) Industrial (86.5%)
Residential (92%)
Fitzpatrick [41] Residential (94%) Deciduous forest (78%) USGS study of Greater Atlanta
Cropland & Pasture (93%) Region (in draft form)
Mixed Forest (91%)
Exergreen forest (90%)
Fitzpatrick (CARETS) [32] 84.9% Bays & Estuaries (89.9%) Residential (74.5%) Point check of 1:24,000 scale
Cropland & Pasture (83.6%) aeri a 1 photos
Fitzpatrick (CARETS) [32] 91% Light Crown Cover Linear Traverse Technique at
Fores t (40%) 1:100,000
Fitzpatrick (CARETS) [32] 69.5% Urban (57%) ERTS maps at 1:250,000
Fitzpatrick (CARETS) [32] 76.5% High-altitude photos
1:250,000
Lindgren [34] 99.5% ~1elling units per Color infrared photos of
residential residential areas
structure (59%) . 1 : 20 , 000
Simpson [36] 81.3% Sample test of interpreters
Branch [33] 96.5% Included items that could
only be ground checked
1 :2,400
Avery [39] 90% Rul e of thumb
--
-------
determination of overall accuracy on a test involving lithe 25 most competent
radar interpreters available nationally...llwhile Harden [37], Fitzpatrick [38],
and Avery [39] based their conclusions on a "rule of thumb" which they felt was
an acceptable level of overall accuracy to be attained by professional inter-
preters of aerial photographs.
Three sources were especially relevant. Collins and El-Beik [35]
indicated that they used 1:10,560 scale aerial photographs which were of poor
quality. Regardless, they achieved an overall level of accuracy of 92%. The
percentage of correctly identified land uses ranged from 36.4% for complex
urban land use (i.e., high density mixed uses) to near 100% for bodies of water
and transportation. More revealing, however, was the percentage over or under
estimation by land use category. Commercial areas were over-estimated by 24%,
transportation was over-estimated 22%; and complex urban land use was under-
estimated by 39.4%.
By contrast, a limited check of aerial photo data used in the
GEMLUP-I project [17] revealed that commercial land use was under-estimated
by 12%, manufacturing over-estimated by 10%, industrial over-estimated by
71% and office land use under-estimated by 40%.
In the CARETS project, Fitzpatrick [32] reported the accuracy of
determining land use by categories according to three scales of aerial photos
(see Table C-2). Figure C-l depicts Fitzpatrick's findings in the form of a
bar chart. The obvious was concluded, i.e., the more detailed the aerial
photograph, the greater the degree of accuracy per land use category. Two
exceptions, however, were noted. First, urban and built-up land was most
accurately mapped at a scale of 1:100,000. Second, nonforested wetlands
were mapped most accurately at a scale of 1:24,000 and 1 :250,000.
2.
Interpretation Factors
On the basis of the literature and professional experience, the
following factors are of significance in interpreting aerial photographs:
1.
Skill of the interpreter (Ability to decipher land
use from aerials on the basis of tone, texture, color,
shadows, slope, pattern, size, shape, and ralationship
to other features).
The quality of the photograph.
The scale of the photograph.
The season when the photograph was taken.
The equipment used for taking and interpreting the
photograph.
The extent to which the form of the photograph is
reflected in its functio~
The nature of the area of analysis.
2.
3.
4.
5.
6.
7.
C-4
-------
TABLE C-2
COMPARISON OF AREA FOR VARIOUS CATEGORIES OF LAND USE MAPPED
AT THREE SCALES FROM HIGH-ALTITUDE PHOTOGRAPHY (IN HECTARES)
Source:
[32J
.-- -~.-'
.'. -. '" '-- -...---
. -'.. .-.
. U. CODE. 1:24 000 1:100.000 1:250.000
URBAN AND BUILT-UP
Residential 11 4,069 3,441 4,950
Couunercia1 and services 12 340 464 300
Industrial 13 38 48
Extractive 14 33 112
Transportation, etc. 15 259 132 225
Institutional 16 961 1,289 875
Strip and clustered 17 60 48
Mixed 18 16
Open and other 19 409 400 325
Subtotal 1 6 169 5.950 6 675
AGRICULTURAL
Cropland and pasture 21 21,544 23,156 23,875
Orchards, etc. 22 10 404
Other 24 95 92
Subtotal 2 21 649 23.652 23.875
FOREST LAND
Heavy crown cover 41 33,740 31,550 30,950
Light crown cover 42 1.217 1.906 1.900
Subtotal 4 34.957 33.456 32.850
WATER
Streams and waterways 51 334 404 75
Lakes 52 108
Reservoirs 53 224 92 125
Bays and estuaries 54 9.316 9.150 8 850
Subtotal 5 9 874 9.754 9.050
NONFORESTED WETLAND
Vegetated 61 3 273 3.088 3.500
Subtotal 6 3.273 3.088 3.500
BARREN LAND
Sand other than beaches 72 6
Beaches 74 10
Other 75 62 100 50
Subtotal 7 78 100 50
TOTAL 76,000 76,000 76,000
C-5
-------
100
1: 24,000
1: 1 00,000
1:250,000
90
80
70
41
C)
o
-
C
41
v
...
41
~
60
>-
v
o
...
:)
v
v
c(
50
40
30
20
10
......
2
5
6
4
l eve I
Cat ego r i.e s
FIGURE C-1
A COMPARISON OF THE ACCURACY OF LEVEL I LAND USE INTERPRETATIONS
AT THREE SCALES DERIVED FROM AIRCRAFT DATA FOR EACH LAND USE
CATEGORY. PERCENTAGES DERIVED FROM FI ELD CHECK.
Source:
[32J
C-6
Em
-
Em!
-------
8. The land use classification scheme employed (Categories
should be mutually exclusive).
9. The resolution desired.
10. The availability of supplemental materials.
11. Repeatability by other interpreters.
Of these factors, the most important are (1) the skill of the interpreter,
(2) the scale of the photograph, (3) the land use classification system
employed. With regard to the first factor, Simpson [36] reported--as
indicated above--that 25 competent interpreters averaged 81.3% accuracy.
The U.S. Department of Agriculture has reported [40] two instances where
skilled soil scientists with 8 and 12 years experience achieved 82.5% and
95% accuracy, respectively. Second, the scale of the photograph is
generally significant vis-a-vis the level of accuracy. As previously
indicated, Fitzpatrick [32] reported that the accuracy of land use interpreta-
tion increased as aerial photo scale increased. By way of another example,
Lindgren [34] concentrated on residential areas usin~ photos with a scale of
1 :20,000 and attained an accuracy level of 99.5%. Finally, the land use
classification system employed should utilize categories that are mutually
exclusive so as to minimize subjective judgments. The difficulty here is
that such classification schemes will vary according to the area under
examination, the topics of particular interest to the interpreter and
limitations of the equipment used.
3.
Conclusions:
Level of Accuracy
Based on the review of aerial photographic literature, the
average overall level of accuracy--regardless of the numerous variables
involved such as scale, equipment, etc.--was 87%. This is slightly lower
than the 90% general "rule of thumb" mentionedby Harden [37], Fitzpatri ck
[38J, and Avery [39]. This "rule of thumbll, however, should be taken as an
ideal to be achieved, i.e., accuracy will typically not exceed this level.
The land use categories that typically attained the highest
level of accuracy were bodies of water (94.5%) [35]; open improved and
unimproved land (97.5%) [35]; and mixed forest (91%) [41J. The land use
categories for which the least accuracy was achieved included complex
urban use (36.4%) [35J; urban land use taken from ERTS photographs at a
scale of 1 :250,000 (57%) [32J; dwelling units per residential structure
(59%) [34J; and light crown forest cover (40%) [32]. Urban land use cate-
gories achieved the following accuracies: residential (59%-94%) [32,34,35,41J,
commercial (84.5%) [35J, Industrial (86.5%) [35] and general urban land use
(36.4% - 57%) [32,35J. Due to the wide variability in accuracy figures, by
category, the lack of sufficient literature on interpretation of urban land
use (as opposed to vegetation covers), and the numerous intervening factors,
the results of the literature review were inconclusive and IItypicalll errors
associated with each land use category could not be determined. Instead,
the average overall level of accuracy--87%--is recommended. This figure is
used in the coefficient stability analysis of the causal model (see Section IV).
C-7
-------
APPENDIX D
BI BLIOGRAPHY
D-1
,"
-------
D.
BI BLIOGRAPHY
This section summarizes all relevant citations uncovered in the literature
search task of this study. The literature search was aimed at identifying
publications in the following four technical areas:
. Causal Analysis and Predictive Modeling of Induced Development
Associated with Wastewater Facilities
. Simplified VMTPredictive Modeling Procedures
. Land Use Identification and Categorization from Aerial Photographs
. Potential Case Study Wastewater Projects
1.
Causal Analyses and Predictive Modeling of Induced Development
Associated with Wastewater Facilities.
Abt Associates, Manual for
water Treatment Facilities
Bascom, S.E., Cooper, K.G., Howell, M.D., Makcrides, A.C., and
Rabe, F.T., Secondar 1m acts of Trans ortation and Wastewater
Investments: Research Results EPA-600/5-75-013 and Review and
Bibliography (EPA-600/5-75-002), Washington, DC, 1975.
Benesh, F., Guldberg, P., and D'Agostino, R., Growth Effects of.
Major Land Use Projects: Volume I - Specification and Causal
Analysis of ~1odel, EPA Publication No. EPA-450/3-76-012a, May 1976.
Benesh, F., Gu1dberg, P., and D'Agostino, R., Growth Effects of
Major Land Use Projects: Volume III - Summary, EPA Publication
No. EPA-450/3-76-012c, Research Triangle Park, NC, September 1976.
Binkley, C., et al., Interceptor Sewers and Urban Sprawl, Lexington
Books, Lexington, MA, -]9-75.
Blalock, H., Theory Construction, Englewpod Cliffs, NJ:
Prentice-Hall, Englewood Cliffs, NJ, 1969.
Blatnik, J.A., "History of Federal Pollution Control Legis1atinl1",
Industrial Pollution Control Handbook, McGraw-Hill, New York, NY,
1971. .
Booz-Allen & Hamilton, Inc., Methodologies for the Analysis of
Secondary Air Quality Impacts of Wastewater Projects Located in
AQMAs, Environmental Impact Office, EPA Region II, New York, NY,
March 1976. .
0-2
-------
Boudon, R., "A Method of Linear Causal Analysis: Dependence
Analysis", American Sociological Review 33:365-373, 1965:..
Bureau of Regional Planning, Secondary Impact of Regional Sewerage
Systems, State of New Jersey Department of Community Affairs,
Trenton, NJ, 1975.
Computer Sciences Corporation, Infonet Times Series Processor,
Information Network Division, Los Angeles, CA, December
1974.
Council 00 Environmental Quality, The Fifth Annual Report of the
Council 0" Environmental Quality, Washington, DC, 1974.
Environmental I~pact Center, Secondary Effects of Public Invest-
ments in Hi gh\Vays and Sewers, Hashi ngton, D. C., Council on Envi r-
onmental Quality, 1974.
Environmental I~pact Center, Secondary Impacts of Infrastructure
Investments in the Denver Reoion, \lashington, D.C., Council on
EnvironITental Quality, 1974.
Fensterstock, J.C. and Speaker, D.M., Use of Environmental Analyses
on Vlaste\'/ater Facilities by Local Government, Washington, D.C.,
U.S. Environmental Protection Agency, 1974, (EPA-600/5-74-015).
Goldberger, A.S., Econometric Theory, Wiley, New York, NY, 1964.
Grava, S., Urban Planning Aspects of Water Pollution Control,
Columbia University Press, New York, Ny~g:--
Heise, D., Causal Analysis, John Wiley & Sons, New York, NY, 1976.
Heise, D. R., "Problems in Path Analysis and Causal Inference",
Sociological ~~ethodology, edited by E.F. Borgatta, Josey-Bass,
San Franci sco, CA, 1969 (pp. 38-72). .
Hill, D., "A grO\.tth allocation model for the Boston region",
Journal of the American Institute of Planners, IL(2) , 1965.
Hudson, J.F., Demand for Municipal Services: Measuring ~he Effect
of Service Quality, Ph.D. thesis, MIT Department of Civil Engineer-
ing, Cambridqe, MA, June 1975.
Johnston, J., Econometric Methods, McGraw-Hill, New York, NY. 1963.
0-3
-------
Kenney, Kenneth B., Downing, Donald A., and Hayes, Gary G.,
Urban Water Policy as an Input in Urban Growth Policy, Water
Resources Research Center, Tennessee University, Knoxville, TN,
1972 .
~erlinger, F.N., and Pedhazer, LJ., Multiple Regression in Be-
havioral Research, Holt, Rinehart and Winston~ New York, NY, 1973.
Land, K.C., "Principles of Path Analysis", Sociological Meth-
odology, Edited by E.F. Borgatta, Jossey-Bass, San Francisco,
CA, 1969. .
Milgram, G., The City Expands - A Study of the Conversion of Land
from Rural to Urban Use in Philadelphia, 1945-62, University of
Pennsylvania, Philadelphia, PA, March 1967.
Nie, N.H., Hull, C.H., Jenkins, J.G., Steinbrenner, 1<:., arid Bent,
D.H., Statistical Package for the Social Sciences (second edition),
McGraw-Hi 11, New York, NY, 1975.
Pepper, James E.. and Jorgensen, Robert E., Influences of Wastewater
Manaqcmer.t on Land Use: Tahoe Basin, Washington, DC, U.S. Envir-
onmenta 1 Protect; on Agency, 1974 (EPA-600/5-74-019; ~lTIS #PB-240-
247/7ST).
Phillips, M.B., "Developments in t~ater Quality and Land Use Planning:
Problems in the Application of the Federal Water Pollution Control
Act Amendments of 1972". Urban Law Annual, lQ.:43, 1975.
Promise, J., "Dallas and D.C. - An Insider's Comparison of '208'
Approaches to Growth Policy", 49th Annual Conference of the Water
Pollution Control Federation, Minneapol is, Mrl, October 1976.
Rabe, F. T. and Hudson, J. F., "Highway and Sewer Impacts on Urban
Development:, J. Urban Planning and Development - ASCE, 101 :217,
November 1975.
Real Estate Research Corporation, The Costs of Sprawl - Environ-
mental and Econo~ic Costs of Alternative Residential Development
Patterns at the Urban Fringe, prepared for the Council on Envir-
onmental Quality, Washington, DC, April 1974.
0-4
-------
Reid, George W. and Alquire, Robert
politan and Regional Area Water and
Bureau of Water Resources Research,
(NTIS #PB-222-262/8).
T., Systems Approach to Metro-
Sewer Planning, rlorman,OK. -
Oklahoma University, 1973
Rivkin/Carson, Inc., Population Growth in Communities in Relation
to Water Resources Policy, prepared for the National Water Com-
mission, Arlington, VA, October 1971..
Snedecor, G.W., and Cochran, W.G., Statistical Methods (6th Edition),
Iowa State College Press, Ames, lA, 1967.
Stansbury, J., "Suburban Growth - A Case Study", Population Bulletin,
28:5, February 1972. .
Stokes, E. B., "The Eval uati on of an Er-iPIRlC Model for an Urban Areall,
Environment ~ Planning A, pp. 703-715, 1973.
Tabors, R., Shapiro, 11., and Rogers, P., Land Use and the Pipe:
Planning for Sewerage, Heath, Lexington, 'MA, 1976.
Thomas, R.D., Water Problems in the Context or Ccunty Government
Decision-Makino, Water Resources Research Center Publication No. 32,
Florida Atlantlc University, Boca Raton, FL, October 1975.
Train, Russell E., IIMemo on the Consideration of Secondary Environ-
mental Effects of Construction Grantsll (Program Guidance Memorandum
#50), Washington, DC, U.S. Environmental Protection Agency, June 7,
1975.
Train, R.E., liThe EPA Programs and Land Use Planningll, Columbia
Journal of Environmental Law, 2:255, Columbia University, !"Jew York,NY
Spring 1976. - .
Turner, M.E., and Stevens, C.D., liThe Regression Analysis of Causal
Pathsll, Biometrics, ~;236-258, 19.59.
Tukey, J.W., IICausation, Regression and Path Analysis", Statistics
and t-1athematics in Biology, Edited by O. KempthornOe, T.A. Bancroft,
J.W. Gowen and J.L. Lush, Iowa State College Press, Ames, lA,
1964, pp. 35-66. ,
Urban Systems Research and Engineering, Inc., Interceptor Sewers
and Suburban Sprawl: The Impact and Constructicn Grants on
Residential Land Use, Volu~e I: Analysis, and Volume II: Case
Studies, prepared for the Council on Envi ronmenta 1 Quality,
Washington, DC, September 1974. .
Urban Systems Research & Engineering, Inc., The Distribution of
Water Pollution Control Costs, prepared for the National Commission
on Water Quality, !t!ashington, DC, 1976.'
0-5
-------
Urban Systems Research & Engineering, Inc., The Growth Shapers - The
Land Use Impacts of Infrastructure Investments, prepared for the
Council on Environmental Quality, Washington, DC, May 1976.
U.S. Congress, Federal Water Pollution Control Act (33 USC 1251
et. seq.). .
U.S. Environmental Protection Agency, "Annotated Bibliography for
Water Quality Management" (3rd Edition~) l~ater Planning Division,
October 1976.
U.S. Environmental Protection Agency, "Annotated Bibliography -
Publications and Active Projects", Comprehensive Planning and Land
Use Staff, Washington, DC, May 1976.
U.S. Environmental Protection Agency, "Land Use and Water Qual ity",
Proceedings of the EPA Pollution Control Technology Assessment
Conference, Columbus, OH, May 1974.
U.S. Environmental Protection Agency, Mitigating Secondary Impacts
from the Wastewater Facilities Program, Office of Land Use Coordina-
tion, Washington, DC, 1977.
U.S. Environmental Protection Agency, Office of Air Quality Planning
and Standards, Guidelines for Air Quality Maintenance Planning and
Analysis, Research Triangle Park, NC, 1974, Volume 3: Control.
Strategies (EPA-450/4-74-003), Volume 4: Land Use and Trans ortation
Considerations (EPA-450/4-74~004 , Volume 7: Projectin~ County
Emissions (EPA-450/4-74-008), Volume 9: Evaluating Indlrect Sources
(EPA-450/4-75-00l), Volume 13: Allocating Projected Emissions to
Sub-County Areas (EPA-450/4-74-014).
U.S. Environmental Protection Agency, Policies and Procedures for
Continuing Planning Process (40 CFR 130), reprinted in 40 FederaT
Register 55334; November 28, 1975; also reprinted in: Bureau of
National Affairs, Environmental Reporter; Federal Regulations
(131 :2673). .. '
U.S. Environmental Protection Agency, Re ulations on Pre aration
of Water Quality Management Plans (40 CFR 3 , reprinted in
40 Federal Register 55343; November 28, 1975; also reprinted in:
Bureau of National Affairs, Environmental Reporter; Federal
Regulations (131:2681).
Van de G~er, J.P., Introduction to Multivariate Anal~sis for the
Social Sciences, W.H. Freeman, San Francisco, CA, 19 1. --
Wonnacott, R.J., and Wonnacott, T.H., Econometrics, John Wiley
& Sons, Inc., New York, NY, 1970.
D~6
-------
Wright, S., IIPath Coefficients and Path Regressions Alternative or
Complementary Conceptsll~Biometrics, l£.:189, 1960.
Wright, S., liThe Interpretation of Multivariate Systemsll, Statistics
and Mathematics in Biology, Edtted by O. Kempthorne, T. Bancroft.
J. Cowen, and J. lusn, Iowa State College Press, Ames, lA, 1954
(Vol. 11-33).
Wri ght, S., liThe Method of Path Coeffi ci ents II, Annals of Mathemati ca 1
Statistics, ~:16l-2l5, 1934.
Wright, S., liThe Treatment of Reciprocal Interaction, with or without
lag, in Path Analysisll, Biometrics, ~:423, 1960.
Simplified VMT Predictive Modeling Procedures
2.
Ashford. N. and Holloway. F.M., The Permanence of Trip Generation
Equations, Florida State University, 1971, NTIS PB-204-433~
Australian Road Research Board, Proceedings of the Sixth ARRB
Conference, ~ (2): 5-22, Melbourne, Australia, 1972.
Brand, D., IITheory and Method in land Use and Travel Forecastingll,
Highway Research Record #422, pp. 10-20, 1973.
Burke. R.H.. Atkins, A.S. and Coote, G.M., Procedures for Forecasting
Vehicle Miles of Travel in National Road Planning, Australian Bureau
of Roads, Melbourne, Australia, 1974.
Cesario, F.J., IICombined Trip Generation and Distribution Model,1I
Transportation Science, ~ (3):211-223, August 1975.
Chang, H.K.. and Smith, C.l., Trips Ends Generation Research Counts,
California State Department of Transportation, July 1973, NTIS
PB-227 134/4.
Charles River Associates, A Disaggregated Behavioral Model of Urban
Travel Demand, Cambridge~ MA. March 1972 NTIS PB-2l0 515.
Chatterjee, A. and Cribbins, P.O., IIForecasting Travel on Regional
Highway Networkll, ASCE Transportation Engineering Journal, pp 209-
224, May 1972.
DeLeuw, Cather, Canada ltd., A New Procedure for Urban Transportation
Planning, Department of Highways, Ontario, Canada, September 1969.
Devlin, J., Hearne, R.. and McGuinness, P., Vehicle Miles Of Travel
For 1976, The National Institute for Physical Planning and Construction
Research, Dublin, Ireland, 1977.
Federal Highway Administration, Guide For Forecasting Traffic On The
Interstate System, U.S. Department of Transportation, Highway Planning
Program Manual #134, Washington, DC, March 1973.
D-7
-------
Federal Highway Administration, Guidelines For Trip Generation
Analysis, Washington, DC, April 1973, NTIS PB-244 925/4ST.
Federal Highway Administration, Nationwide Personal Transportation
~tudY, Report #8, "Home to Work Trips and Travel", Washington, DC,
ugust 1973, NTIS PB-242 892/8ST.
Federal Highway Administration, Nationwide Personal Transportation
~tu~t' Report #10, "Purposes of Automobile Trips and Travel",
as lngton, DC, May 1974, NTIS PB-242 894/4ST.
Federal Highway Administration, Statewide Travel Demand Forecasting,
U.S. Department of Transportation, Highway Planning Program Manual
#147, Washington, DC, November 1973. .
Federal Highway Administration, Travel Simulation For Small Cities
(Draft), U.S. Department of Transportation, Washington, DC, 1973.
Fogarty, W.J., "Trip Production Forecasting Models For Urban Areas",
ASCE Transportation Engineering Journal, pp 831-845, November 1976.
Gus tafson, R. L., "Empi ri cal Study of Factors Infl uenci ng Trip
Attraction and Trip Generation", High Speed Ground Transportation
Journal, I (3):307-321, 1973.
Highway Researcn Board, Urban Travel Demand Forecasting, Transpor-.
tation Research Board Special Report #143, 1973.
Jeffries, W.R. and Carter, E.L., Simplified Techniques For Deve10p-.
ing Transportation Plans - Trip Generation in Small Urban Areas,
West Virginia University, 1966, NTIS PB-183 217.
Kassoff, H., and Gende1l, D.S., "An Approach to Mu1 tiregional Urban
Transportation Policy Planning", Highway Research Record #348,
Highway Research Board, Washington, DC, 1971.
Keefer, L.E. and Witheford, D.K., Urban Travel Patterns for
Hospitals, Unversities, Office Buildings, and Capitols, National
Cooperative Highway Research Program Report #62, 1969.
Landis, R., The Effect of Automotive Fuel Conservation Measures on
Air Pollution, EPA Publication No. EPA-600/5-76-006, Washington,
DC, September 1976.
Mann, W.W., TRIMS - A Procedure For Quick Response Transportation
Planning, Metropolitan Washington Council of Governments, Washington,
DC, January 1975.
Maricopa Association of Governments, Trip Generation By Land Use,
Washington, DC, April 1975, NTIS PB-244 549/2ST.
D-8
-------
Meyerowitz, W., Trip Generation: Regression Analysis, Procedural
Manual, Michigan Department of State Highways, 1970, NTIS PB-198 506.
Morris, R.J., A Comparative Analysis of Trip Distribution and Traffic
Assignment Models for Transportation Planning in Developing Areas,
Stanford University, 1973, NTIS PB-232 325/1.
~;~~~~a ~a~~~~~~ T~~n~~9i ~~~r~u~~~~~~e~8~~ ~aa~ehdi n~toT:,iPDcL~"f~'.
Schneider, M., Transportation and Land Development: Unified
Theory and Prototype Model, Creighton Hamburg Inc., 1969, NTIS
PB-184941. .
Soliman, A. and Sharma, S.C., IIImpact of Land Use Changes on
Transportation Networks", Canadian Journal of Civil Engineering,
1 (3): 372-278, September 1975.
Sosslau, A.B., et a1., Travel Estimation Procedures .for uitk
to Urban Policy Issues Draft, Volume II, prepared for the Trans-
portation Research Board, Washington, DC, February 1977.
Tardiff, T.J., IITrip Generation as a Choice Processll, ASCE Transpor-
tation Engineering Journal, 103 (2): 337-348, March 1977.
Transportation Research Board, Research Record #569, Washington,
DC, 19?6, NTIS PB-256 975/4ST.
U.S. Environmental Protection Agency, Land Use Information for
Water Quality Management Planning, Washington, DC, August 1976.
Weiner, E., Kassoff, H., and Gendell, D.S., II Multi model National
Urban Transportation Policy Planning Mode111, Highway Research
Record #458, Highway Research Board, Washington, DC, 1973.
Zaryouni, M.R. and Kannel, E.J., IISynthesized Trip-Forecasting
Mode 1 for Sma 11 and Medi um-Si zed Urban Areas II .
3.
Land Use Identification and Categorization from Aerial Photographs
Adams, V.W., 1975, Earth Science Data in Urban and Regional In-
formation Systems--A Review, US Geo1. Survey Circ. 712.
Aguilar, A.M., "Area & Volume Errors in Reservoir Projectsll~ ASCE
Journal of Surveying & Mapping, pp. 287-296, November 1971. ----
American Society of Photogrammetry, Manual of Photographic
Interpretation, Washington, DC, American Society of Photogrammetry,
1960.
D-9
-------
Anderson, J.R., Hardy, E.E., and Roach, J.T., 1972, A Land-Use
Classification System for Use with Remote-Sensor'Data, U.S. Geo1.
Survey Circ. 671.
Anderson, J.R., Hardy, E.E., Roach, J.T., and Witmer, R.E.~ A'Land
Use and Land Cover Classification System for Use with Remote Sensor
Data, USGS Paper #964, Washington, DC, 1976.
Aschenbrenner, C.M., "Prob1ems in Getting Information into and
out of Air Photographs", Photogrammetric Engineering, 20 (3): 398-
401,1954.
Avery, T. Eugene, Intertretation of Aerial Photographs, Minneapolis,
MN, Burgess Publishing 0., 1968.
~~~n~~ ~v~~~ v~~ ~~e~~ it;i ~~e~~ ~n~~~1. and Aeri a 1 Information, Cambrfdge,
Branch, Melville C., Planninq Urban Environment, Dowden, Hutchinson
& Roos, Inc., 1974, Stroudsburg, PA, 1974.
Cas pan Corp., Remote Sensing for land use analysis, Houston, TX,
June 1975, NTIS N75-31938.
Collins, W.G., and E1-Beik, A.H., liThe Acquisition of Urban Land
Use Information from Aerial Photographs of the City of Leeds
(Great Britain)", Photogrammetria, ll..:71, 1971.
Colwell, R.N., "A Systematic Analysis of Some Factors Affecting
Photographic Interpretations", Photogrammetric Engineering,
20 (3) :433-454, 1954. '.
Eardley, A.J., Aerial Photographs--Their Use and Interpretation,
Harper and Brother~, Publishers, New Yorkr NY, 1942.
Environmental Research Institute of Michigan, Proceedings of the
Tenth International Symposium on Remote Sensing of Environment,
Ann Arbor, MI, October 6-10, 1975.
Fitzpatrick, K.A., Cost, Accurac~ and consistenc~ Comparisons of
Land Use Maps Made from High-Alt1tude Aircraft P otography and
ERTS Imagery, U.S. Geological Survey, Reston, VA, September 1975.
Gautum, N.C., "Aerial Photo-Interpretation Techniques for Classi-
fyi ng Urban Land Use", Photogrammetri c Engineeri ng and Remote
Se_nsing, 42(6) :815-822, 1976. . .
Hardy, Ernest E., Belcher, Donald J., and Phillips, Elmer S.,
land Use Classification with Simulated Satellite Photography,
U.S. Department of Agriculture Economic Research Service, Agri-
cultural Information Bulletin, Washington, DC, 1971.
D-10
-------
Howard, William A., and Kraht) James B., An Assessment of the
Useful ness of Small-Scale Photographi c Imagery for Acqui ring Land
Use Information Necessary to the Urban Planning Functlon, Technical
Paper No. 71-2, Denver, CO, Department of Geography, University
of Denver) September 1971, mimeo.
Lafferty, Maurice E., "Accuracy/Costs with Analysis", Photogrammetric
Engineering, 39 (5): 507-510, May 1975.
Lehman) E.J., Remote Sensing for natural resource, environmental and
regional planning, January 1975, NTIS PS-75/104.
Lindren, D.T., Dwelling Unit Estimation from Color Infrared
Photography, Dartmouth College, Department of Geography, Paper
No. 1) Hanover) NH, May 1970.
Lins, H.Fq "Land-Use Mapping from Skylab S-190B Photography",
Photogrammetric Enqineering and Remote Sensing, 1£(3):301-307,
March 1976.
Lueder) Donald R., Aerial Photographic Interpretation:Pri~ciples
and Applications, McGraw-Hill) New York, NY, 1959.
i~ausel, P .~1., Leiro, C.E., and Lewellen, M. T., II Regional Land Use
Classification Derived from Computer-Processed Satellite Datal!,
Journal of the American Institute of Planners, 42:153-164, April
1974.
Muret) Jean-Pierre, Aerial Photography and City Planning., Paris,
France, Centre de Recherche d'Urbanisme, International Federation
for Housing and Planning, 1972.
Perlman) E. and Raney) R., An Ex~eriment in the Application of
Remote Sensing to Land-Use Plannlng on the Urban Fringe, Environ-
mental Research Institute of ~1ichigan, Ann Arbor, MI, June 1975.
Ray, R.G., Aerial Photographs in Geologic Interpretation and
Mappinq, Ge010gical Survey Paper #373, Washington, DC, 19~
Simpson) R.B. and Lindgren) D.T.) Recognition of Settlement Patterns
A„ainst a Complex Background, U.S. Geological Survey, Washington, DC,
N IS N71-3430, May 1970.
Smith, H.T.U., Aerial Photographs and Their Applications, New York, NY.
App1eton-Century-Crofts) Inc.) 1943.
Soil Conservation Service, Aerial-Photo Interpretation in Classify-
ing and Mapping Soils, U.S. Dept. of Agriculture Handbook #294,
Washington, DC) October 1966.
0-11
-------
Spangle, W. and Associates; Leighton, F.B. and Associates; and
Baxter, McDonald and Company, 1976, Earth Science Information in
Land-Use Planning, U.S. Geol. Survey Circ. 721. .
Stafford, D.A., "Measuring Watershed Land-Use Changes with Air..
photosll, ASCE Transportation Enginerring Journal, pp. 117-129,
February 1976.
Stone, Kirk, "Air Photo Interpretation Procedures", Photogrammetric
Engineering, 22(1):123-132,1951.
Summerson, C. H., "A Philosophy for Photo Interpretersll, Photo-
grammetric Engineering, 20(3) :396-397, 1954. . ~--
Urban Renewal Administration, Standard land Use Coding Manual,
Washington, DC, January 1965.
U.S. Dept. of Interior, Proceedings of the National Conference on
Land Use Information and Classification, June 28-30, 1971,
Weedel, J.W. and K1ickner, R., Using remote data for land use mapping
and inventory: a user g~, Association of American Geographies,
Washington, DC, July 1974, NTIS PB-242-813.
Wellar, Barry S., Hypera1titude Photography as a Data Base in Urban
and Regional Planning, Evanston, Ill, Remote Sensing Laboratory,
Department of Geography, Northwestern University, January 1969.
Potential Case Study Wastewater Projects
4.
Abrams, P., "Plant Load Jumps Fourfold as Recreationists Come to
Town", \;Jastes Engineering, R(11):672-675, November 1961.
Barksdale, W.D'., IIModern is Word for ~!eirton Sewage P1antll,
American City,~(2):100-103, February 1961.
Bi-$tate Metropolitan Planning Commission, Metropolitan Compre-
hensive' Water Sewage, and Solid Waste Planning Study (3 Volumes),
Rock Island, IL, March 1970.
Browner, H. D., "Semi-elliptical Sewerll, Civil Engineering, ~(6):
77-79, June 1959.
0-12
-I
-------
Bureau of Regional Planning, Secondary Impact of Regional Sewerage
~~~~~~~: i~~~~on, NJ,. State of New Jersey Department of Community
"Camoufl aged Sewage Treatment", Ameri can Ci ty, 74( 5) : 1 04-1 066,
May 1959.
Camp, Dresser, & McKee, Inventory of Water and Sewer Facilities -
Eastern ,Massachusetts Regional Planning Project, Boston, MA,
May 1967.
Cawley, vJ.A., "Mill Creek Sewage Works", vlater & Sewage Works,
106(10):436-439, October 1959.
I
Central Iowa Regional Planning Commission, Des Moines Metropolitan!
Sanitary Seweraqe System Study, December 1971.
Cobb, E.8. and ~vheeler, H.R., "Progress Report on Development of
Allegheny County Sanitary Authority Facilities of Pittsburgh, Pa.",
Journal of Boston Society of Civil Engineers, 46(1):38-56, January
1959. --
Denver Regional Council of Governments, Drainage Basin Descriptions
Sub-basin Delineation and Summaries, Denver, CO, March 1972, NTIS
No. PB-219-610.
East-West Gateway Coordinating Council, St. Louis Region Water and
Sewer Report - Five-Year Program, St. Louis, MO, June 1973, NTIS No.
PB-222-723.
Graeser, H.J., "Expanded sewage I reatment for Ua lias;', Water &
Sewage Works, 106(10):411-415, October 1959.
Hahn, H.H., "Regional Hastewater Management Systems", Modes for
Environn~ntal Pollution Control, Ann Arbor Science PubTTshers,
pp. 41-60, 1973.
Hayes, G. G., Institutional Alternatives for Providing Programmed
Water and Sewer Services in Urban Growth Areas: A Case Study of
Knoxville - Knox County, Tenn., Knoxville Water Resources Research
Center, Tennessee University Report No. 18, June 1972.
Hensley - Schmidt, Inc., Chattanooga Area Regional Council of
Gover.nments Comprehensi ve ~'later and Sewer Study: Hami Hon, Wil ker
and Catoosa Counties, Chattanooga, TN, December 1971.
Houston-Galveston Regional Transportation Study, f22ulation and
Land Use 1970-1990, Houston, TX, July 1974.
Howells, D.H., proceedings of Symposium on Better Water and Sewer
Services for Small Communities in Harth Carolina, Water Resources
Research Institute, North Carolina University, December 1968.
0-13
-------
Karfman, H.L., "Mechanization Featured at New Primary Plant for
Binghamton", Wastes Engineering, E(4):190-l92, April 1961.
Kentucky Office for Local Government, Modified and Updated Compre-
hensive Water and Sewer Plan for Green River and Purchase Area
Development Districts, Frankfort, KY, 1973, NTIS Nos. PB-223-058
and PB-223-045.
Laboon, J.F., "Construction and Operation of Pittsburgh Projectll,
Journal of Water Pollution Control Federation, ~(7):758-782,
July 1961.
Lexington-Fayette County Pl anni ng Commi ssi on, A Growing Communi.!l.,
Lexington, KY, 1973.
Loucks, D.P., IIAnnual Literature Review: Administration, Systems
Analysisll, Journal Water Pollution Control Federation, 46;
1604-1611, June 1974. --
McWilliams, W., "Denver1s Sewage Disposal Prob1emll, Water & Sewage
Works, 107(7) :287-290, July 1960. .
Metro Dade County Planning Dept., Land Use Characteristics: 1960,
1970, Miami, Florida, July 1973.
Metropol itan Hashington Council of Governments, Water and Sewer
Plan Program - 1970, Washington, DC, September 1970.
Middle Flint Area Planning and Development Commission, Middle Flint
Regional Water and Sewer Systems Plan, Ellaville, GA, March '973,
N7IS ~o. PB-221-528.
Mid-Missouri Regional Planning Commission, Phase I Comprehensive
Water and Sewer Plan, Jefferson City, MO, September 1971.
Minnesota Pollution Control Agency, Wastewater Disposal Facili~
Inventory, State of Minnesota, Division of Water Quality, Roseville,
MN, July 1975.
Murray, C. E., "One Way to Get Sewage Pl ant Buil t: Prohi bit Further
Sewer Extensionsll, Wastes Engineering, 11(3):138-140, March 1960.
Oklahoma Foundation for Residential & Development Utilization, Inc..
Preliminary Listing of Municipal Wastewater Treatment Capacities,
Economic Development Administration, U.S. Department of Commerce,
September 1976 (NTIS #PB-254-430).
San Diego County Comprehensive Planning Organization, Water Dis-
tribution and Sanitary Sewerage Systems Background and Policy
Study, San Diego, CA, February 1972.
Scott, G. R., IINew Sewage Treatment Works for Boul der, COli, Ameri can
City, 74(1):129-131, January 1959.
0-14
-------
.,-
-- -."
Southwestern Illinois Metropolitan Area Planning Commission,
Comprehensive Water and Sewer Plan - Randolph County, Collinville,
IL, December 1972, NTIS No. PB-2l6-396.
Top of Alabama Regional Council of Governments, ReJLLonal Land Us~
Survey and Analyses, Huntsville, AL, May 1973.
_. .
Townsend, J.W. and Beckman, \'I.J., "Pleasant Hills Sewage Plant",
Hater.& Sewage Works, 107(10):370-376, October 1960.
Tri-State Regional Planning Commission, The Extent of Public Water
and Sewer Systems in the Tri-State Region, New York, Ny, May 1973.
i r{ I
U.S. Bureau of the' Census, Census of Population and Housing 1970,
Fourth-Count Housing Tallies.
U.S. Dept. of COTrnllerCe, Accelerated Public Works Program Directory
of Approved Projects as of July 1, 1964, Area Redevelopment Admin-
istration, Washington, DC, August 1964.
U.S. Environmental Protection Agency, 1968 Inventory, Municip~-1
Waste Facilities, Volumes I-X for each federal region, Office of
Water Programs, -Washington, DC, 1971 (EPA Publication No. OWP-l).
U.S. Environmental Protection Agency, ~roject Register - Wastewater
Treatment Construction Grants, Grants dministration DivlSlOn,
Hashington, DC, June 30, 1973.
U.S. Environmental Protection Agency, Region II, Wastewater Treat-
ment Facilities Grants for Nassau and Suffolk Counties, July 19~
U.S. Environmental Protection Agency, Region III, Bethany Beach
(Delaware) Regional Wastevlater Ti'eatment Plant, December 1972.
U.S. Environmental Protection Agency, Region IV, Cobb Cou~
(Georgia) Sewerage Improvement Proj~, July 1971.
U.S. Environmental Protection Agency, Region VI, Wastewater
Facilities, Hot Sprinqs, Arkansas, October 1972.
U.S. Environmental Protection Agency, Region VIII, Denver Regional
Environmental Statement for Wastewater Facilities and the Clean
Water Plan (Draft), Denver, CO. June 1977.
U.S. Environmental Protection Agency, Wastewater Contract Awards,
Computer data file maintained by the Municipa1~~~~truction Division.
U.S. Public Health Service, "Sewer and Water Works Construction
1960", publication number 758,1961. '
Southweste~n Illinois Metropolitan Area Planning Commission,
Comprehenslve Water and Sewer Plan - Randolph County, Collinville,
IL, December 1972, NTIS No. PB-216-396.
Top of Alabama Regional Council of Governments, Regional Land Use
Survey and Analyses, Huntsville, AL, May 1973.
0-15
-------
TECHNICAL REPORT DATA
(Please read IRUructions on the reverse before completing)
1. REPORT NO. EPA-450/3-78-014aI2. 3. RECIPIENT'S ACCESSION-NO.
4. TITLE AND SUBTITLE 5. REPORT DATE
Growth Effects of Major Land Use Projects, March 1978
Wastewater Facilities, Volume I, Model 6. PERFORMING ORGANIZATION CODE
Specification and Causal Analysis
7. AUTHOR(S) 8. PERFORMING ORGANIZATION REPORT NO.
Peter H. Guldberg, Ralph B. D'Agostino
Richard D. Cunningham C-921
9. PERFORMING ORGANIZATION NAME AND ADDRESS 10. PROGRAM ELEMENT NO.
Walden Division of Abcor, Inc.
850 Main Street 11. CONTRACT/GRANT NO.
Wilmington, MA 01887 68-02-2594 .
12. SPONSORING AGENCY NAME AND ADDRESS 13. TYPE OF REPORT AND PERIOD COVERED
Environmental Protection Agency Final
Office of Air Quality Planning and Standards 14. SPONSORING AGENCY CODE
Strategies and Air Standards Division MD-12 200/04
Research Triangle Park, NC 27711
15. SUPPLEMENTARY NOTES
16. ABSTRACT
Growth Effects of Major Land Use Projects is a research program whose goal is to
develop methodologies to predict the total air pollution emissions resulting from the
construction and operation of major land use projects. Emissions are quantified froIT
the major project, from land use induced by the major project, from secondary activi-
ty occurring off-site (e.g., electrical generating stations), and from motor vehicle
traffic associated with both the major project and its induced land uses.
This report documents the development of a causal model for the induced land 'use fron
wastewater major projects. The report discusses the theoretical basis of tha model,
the specification of an initial causal model for growth effects, sample selection
and data collection, and the testing and refinement of the causal model. A subse-
quent report will document predictive equations and worksheets for applying the
model.
17. KEY WORDS AND DOCUMENT ANAL YSIS
a. DESCRIPTORS b.IDENTIFIERS/OPEN ENDED TERMS c. COSATI Field/Group
Land Use Path Analysis
Planning Causal Analysis.
Sewage Treatment Plants Secondary Effects
Induced Land Use
18. DISTRIBUTION STATEMENT 19. SECURITY CLASS (This Report) 21. NO. OF PAGES
Unlimited Unclassified 258
20. SECURITY CLASS (This page) 22. PRICE
Unclassified
EPA Form 2220-1 (9-73)
-------
VI.
REFERENCES
.1.
Wright, S., liThe Method of Path Coefficients", Annals of Mathematical
Statistics, 5:161-215, 1934.
Wright, S., liThe Interpretation of Multivariate Systems", Statistics and
Mathematics in Biology, Edited by O. Kempthorne, 1. Bancraft, J. Cowen,
and J. Lush, Iowa State College Press, Ames, lA, 1954.
Wright, S., "Path Coefficients and Path Regressions Alternative or Com-
plementary Concepts II , Biometrics, l&.:189, 1960.
Wright, S., liThe Treatment of Reciprocal Interaction, with or without
Lag, in Path Analysis", Biometrics, l&.:423, 1960.
Van de Geer, J.P., Introduction to Multivariate Analysis for the Social
Sciencies, W.H. Freeman, San Francisco, 1971.
Kerlinger, F.N., and Pedhazer, E.J., Multiple Regression in Behavioral
Research, Holt, Rinehart and Winston, New York, NY, 1973.
Heise, D.R., "Problems in Path Analysis and Causal Interference", Socio-
logical Methodology, edited by E.F. Borgatta, Josey-Bass, San Francisco,
CA, 1969 (pp. 38-72).
Goldberger, A.S., Econometric Theory, John Wiley & Sons, New York, NY,
1964.
Johnston, J., Econometric Methods, McGraw-Hill, New York, NY, 1963.
Tukey, J.W., "Causation, Regression and Path Analysis", Statistics and
Mathematics in Biology, Edited by O. Kempthorne, T.A. Bancroft, J.W.
Gowen and J.L. Lush, Iowa State College Press, Ames, lA, 1954.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
Turner, M.L, and Stevens, C.D., liThe Regression Analysis of Causal Paths",
Biometrics, ~:236-258, 1959. . . "'"
Snedecor, G.W., and Cochran, W.G., Statistical Methods (6th Edition),
Iowa State College Press, Ames, lA, 1967.
Land, K.C., "Principles of Path Analysis", Sociological Methodology,
Edited by E.F. Borgatta, Jossey-Bass, San Francisco, 1969.
Boudon, R., "A Method of Linear Causal Analysis: Dependence Analysis"
American Sociological Review, 30:365-373, 1965.
Heise, D., Causal Analysis, John Wiley & Sons, New York, NY, 1976.
Wonnacott, R.J. and Wonnacott, T.H., Econometrics, John Wiley & Sons,
Inc., New York, NY, 1970.
Benesh, F., Guldberg, P., and D'Agostino, R., Growth Effects of Major Land
Use Projects: Volume I - Specification and Causal Analysis of Model, EPA
Publication No. EPA-450/3-76-012a, Research Triangle Park, NC, May 1976.
Benesh, F., Growth Effects of Major Land Use Projects: VOlume'II - Com-
pilation of Land Use Based Emission Factors, EPA Publication No. EPA-
450/3-76-012b, Research Triangle Park, NC, September 1976.
12.
13.
14.
15.
16.
17.
18.
6-1
-------
19.
Benesh, ~, Guldberg, P., and D1Agostino, R., Growth Effects of Major
Land Use Projects: Volume III - Summary, EPA Publication No. EPA-450/
3-76-012C, Research Triangle Park, NC, September 1976.
McCurdy, T., IIRequest for Proposal: Growth Effects of Major Land Use
Projects,1I RFP #DU-75-C181, Research Triangle Park, NC, December 23,1974.
McCurdy, T., IIRequest for Proposal: Growth Effects of Major Land Use
Projects (Wastewater Facilities), IIEPA RFP #DU-77-C007, Research Triangle
Park, NC, December 3, 1976.
Meehan, E.J., The Theory and Method of Political Analysis, The Dorsey
Press, Homewood, IL, 1965.
Abraham Kaplan, A., The Conduct of Inquiry, Chandler Publications,
San Francisco, CA, 1964. -
DeGroot, A.D., Foundations of Inference and Research in the Behavorial
Sciences, Mouton & Co., The Hague, Netherlands, 1969.
McCurdy T., Benesh, F., Guldberg, P., and D'Agostino, R., IIApplication
of Path Analysis to Delineate the Secondary Growth Effects of Major Land
Use Projects,1I in Wayne R. Ott (ed.), Environmental Modeling and Simula-
tion, U.S. Environmental Protection Agency, Washington, DC, 1976.
Bunge, M., Causality, Harvard University Press, Cambridge, MA, 1959.
Blalock, H.M., Causal Interferences in Nonexperimental Research, W.W.
Norton Press, New York, NY, 1972.
Bertalanffy, L.V., General Systems Theory, George Braziller, New York,
NY, 1968.
Van de Geer, J.P., Introduction to Multivariate Analysis for the Social
Sciences, W.H. Freeman and Company, San Francisco, 1971.
Webster's Seventh New Collegiate Dictionary, G. & C. Merriam Company,
Springfield, MA, 1970. .
Mausel, P.W., Leivo, C.E., and Lewellen, M.T., IIRegional Land Use Class-
ification from Computer-Processed Satellite Datall, Journal of the Ameri-
can Institute of Planners, 42(2):153-164, April 1964.
Fitzpatrick, A., Cost, Accuracy and Consistency Comparisons of Land Use
Ma s Made from Hi h-Altitude Aircraft Photo ra hand ERTS Ima er ,
Final Re ort, Central Atlantic Ecolo ical Test Site CARETS Pro'ect,
U.S. Geological Survey and National Aeronautics and Space Administration,
Reston, VA, September,1975.
Branch, M.C., City Planning and Aerial Information, Harvard University
Press, Cambridge, MA, 1971.
Lindgren, D.T., Dwelling Unit Estimation from Color Infrared Photography,
Dartmouth College~ Department of Geography Paper No.1, Hanover, NH,
May 1970.
Collins, W.G.,. El-Beik, A.H~, liThe Acquisition of Urban Land Use Informa-
tion from Aerial Photographs of the City of Leeds (Great Britain)lI,
Photogrammetria, ~:71-92, 1971.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
6-2
-------
36.
Simpson, R.B., Line-Scan vs Optical Sensors for Discrimination of Bui1t-
Up Areas, Dartmouth College, Department of Geography Paper No.2, Hanover,
NH, May 1970.
Personal communication, Ivan Harden, Geography Applications Program,
USGS, U.S. Department of the Interior, Reston, VA, May 31,1977.
Personal communication, Katherine A. Fitzpatrick, Geography Applications
Program, USGS, U.S. Department of the Interior, Reston, VA, June 6,
1977 .
37.
38.
39.
Avery, T.E., Interpretation of Aerial Photographs (2nd Edition), Burgess
Publishing, Co., Minneapolis, MN, 1968.
U.S. Department of Agriculture, Aerial-Photo Interpretation in Classify-
ing and Mapping Soils, Soil Conservation Service, U.S. Department of
Agriculture, Agriculture Handbook 294, Washington, DC, October 1966.
Personal communication with Katherine A. Fitzpatrick, Geography Applica-
tions Program, USGS, U.S. Department of the Interior, Reston, VA,
June 7, 1977.
Office of Air Quality Planning and Standards, Guidelines for Air Quality
Maintenance Planning and Analysis; Volume 2: Plan Preparation, EPA
Publication No. EPA-450/4-74-002, Research Triangle Park, NC, 1974.
Office of Air Quality P1anni~g and Standards, Guidelines for Air Quality
~~~ n~~~~~~ ~~ ~n~~~~4~g~ 4~~:~b~~ ~; :eOs1~am:ch3 \riC~~gtreO 1paS:;:t~(,i e1s~h f.PA
Office of Air Quality Planning and Standards, Guidelines for Air Quality
Maintenance Planning and Analysis; Volume 4: Land Use and Transporta-
tion Considerations, EPA Publication No. EPA-450/4-74-004,
40.
41.
42.
43.
44.
45.
Office of Air Quality Planning and Standards, Guidelines for Air Quality
Maintenance Planning and Analysis; Volume 6: Overview of Air Quality
Maintenance Area Analysis, EPA Publication No. EPA-450/4-74-007, Research
Triangle Park, NC, 1974.
Office of Air Quality Planning and Standards, Guidelines for Air Quality
Maintenance Planning and Analysis; Volume 9: Evaluating Indirect Sources,
EPA Publication No. EPA-450/4-75-001, Research Triangle Park, NC, 1975.
Office of Air Quality Planning and Standards, Guidelines for Air Quality
Maintenance Planning and Analysis; Volume 12: Applying Atmospheric
Simulation Models to Air Quality Maintenance Areas, EPA Publication No.
EPA-450/4-74-013, Research Triangle Park, NC, 1974.
National Environmental Policy Act of 1969, 42 U.S.C., Section 4321 et seq.
Council on Environmental Quality, "Guidelines for Preparation of Environ-
mental Impact Statements," Federal Register, 38(147), Part II, August 1,
1973.
46.
47.
48.
49.
50.
Personal communication, Mr. Bruce Bane, Assistant City Engineer, Los Altos,
CA, December 3, 1976.
Personal communication, Mr. Arthur Vondrick, Water and Sewers Director,
Phoenix, AZ, December 8, 1976.
51.
6-3
-------
52.
Grants Administration Division, Project Register - Wastewater Treatment
Construction Grants, U.S. Environmental Protection Agency, Washington,
DC, June 30, 1973.
Promise, J. and Leiserson, M., Water Resources Management for Metropoli-
tan Washington: Analysis of the Joint Interactions of Water and Sewer
Service, Public Polic and Land Develo mentPatterns in an Ex andin
Metro olitan Area includin A endices , Metropolitan Washington Council
of Governments, Washington, DC, 1973.
Grava, S., Urban Planning Aspects of Water Pollution Control, Columbia
University Press, New York, NY, 1969.
Personal communication, Mr. Robert Richmond, Tri-State Regional Planning
Commission, New York, NY, December 8, 1976.
U.S. Census Bureau, IIFinances of Special Districts,1I 1972 Census of
Governments, i(2), Washington, DC, 1972.
Milgram, G., The City Expands - A Study of the Conversion of Land from
Rural to Urban Use in Philadelphia, 1945-62, University of Pennsylvania,
Philadelphia, PA, March, 1967.
Rabe, F.T. and Hudson, J.F., IIHighway and Sewer Impacts on Urban Develop-
ment,1I J. Urban Planning and Development - ASCE, 101 :217, November 1975.
Tabors, R., Shapi ro, M., and Rogers, P., Land Use and the Pi pe: P1 anni ng
for Sewerage, Heath, Lexington, MA, 1976.
Stansbury, J., IISuburban Growth - A Case Study,'~P()pulation Bulletin,
28:5, February 1972.
Bascom, S.E., et al., Secondary Impacts of Transporation and Wastewater
Investments: Review and Bibliography, EPA Publication No. EPA-600-5-
75-002, Washington, DC, 1975. .
Council on Environmental Quality, The Fifth Annual Report of the Council
on Environmental Quality, Washington, DC, 1974.
Urban Systems Research and Engineering, Inc., Interceptor Sewers and
Suburban Sprawl: The Impact and Construction Grants on Residential Land
Use, Volume I: Analysis, and Volume II: Case Studies, prepared for the
Council on Environmental Quality, Washington, DC, September 1974.
Urban Systems Research & Engineering, Inc., The Growth Shapers - The Land
Use Impacts of Infrastructure Investments, prepared for the Council on
Environmental Quality, Washington, DC, May 1976.
Hudson, J.F., Demand for Municipal Services: Measuring the Effect of
Service Quality, Ph.D. thesis, MIT Department of Civil Engineering,
Cambridge, MA, June 1975.
Bascom, S.E., Cooper, K.G., Howell, M.P., Makrides, A.C., and Rabe, F.r.,
Secondary Impacts of Transportation and Wastewater Investments: Research
Results, EPA Publication No. EPA-600j5-75-013, Washington, DC, 1975.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
6-4
-------
67.
Real Estate Research Corporation. The Costs of Sprawl -Environmental
and Economic Costs of Alternative Residential Development Patterns at
the Urban Fringe. prepared for the Council on Environmental Quality.
Washington. DC. April 1974. '
Phillips. M.B.. "Developments in Water Quality and Land Use Planning:
Problems in the Application of the Federal Water Pollution Control Act
Amendments of 1972". Urban Law Annual. 10:43. 1975.
Personal communication. Dr. Peter Beaulieu. Puget Sound Council of Govern-
ments. Seattle. Washington. December 1. 1976.
Environmental Reporter. December 27. 1974. pp. 1328-1330.
Tri-State Regional Planning Commission. The Extent of Public Water and
Sewer Systems in the Tri-State Region. New York. NY. May 1973.
Rivkin/Carson. Inc.. Population Growth in Communities in Relation to
Water Resources Po,licy. prepared for the National Water Commission.
Arlington. VA. October 1971.
Hill. D.. "A Growth Allocation Model for the Boston Region". Journal of
the American Institute of Planners. 31(2). 1965.
Train. R.E.. liThe EPA Programs and Land Use Planningll~ 'Columbia Journal
of Environmental Law. ~:255. Columbia University. New York, NY. 1976.
Breindenbach. A. and Strelow. R.. EPA memorandum to Regional Administra-
tors. Washington. DC, November 15. 1976.
Pail thorp. R.. "Joint Treatment Rate Structures, II CH2M Hill. Corvallis.
OR. February 1977.
San Diego Building Contractors Association. Builder. 29(11): 13. November
1976. --
Abt Associates. Manual for Evaluating Secondary Impacts of Wastewater
Treatment Facilities. prepared for EPA Office of Air. Land and Water
Use. Washington. DC. January 1976.
Booz-Allen & Hamilton. Inc.. Methodologies for the Analysis of Secondary
Air Quality Impacts of Wastewater Treatment Projects Located inAQMA.
prepared for EPA Region II. New York. NY. March 1976.
Bureau of Regional Planning. Secondary Impact of Regional Sewerage Sys-
tems. State of New Jersey Department of Community Affairs, Trenton, NY,
1975.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
Greenwood. E.. liThe Relationship of Science to the Practice Professions".
Journal of the American Institute of Planners, 28:223-232, 1958.
Brodbeck. M., Readings in the Philosophy of the Social Sciences, The
Macmi~lan Company. New York. NY, 1968.
Popper. K., The Logic of Scientific Discovery. Harper and Row Publishers,
New York. NY. 1959.
Lowry. I.. A Model of Metropolis. The RAND Corporation. Report No. RM-
4035. RG. Santa Monica. CA. 1964.
82.
83.
84.
6-5
-------
100.
101.
102.
85.
86.
Forrester,J., Urban Dynamics, The MIT Press, Cambridge, MA, 1969.
Seidman, D.R., The Construction of an Urban Growth Model, Delaware Valley
Regional Planning Commission, Philadelphia, PA, 1969.
Center for Real Estate and Urban Economics~ Jobs, People, and Land, Bay
Area Simulation Study, University of California, Berkeley, CA, 1968.
Ohls, J.C. and Hutchinson, P., IIMode1s in Urban Deve10pmentll pp. 165-200
in: Saul I. Gass and Roger L. Sisson (eds), A Guide to Models in
Governmental Planning and Operations, Sauger Books, Potomac, MD, 1975.
Blalock, H.M., Jr., Theory Construction, Prentice-Hall, Englewood Cliffs,
NJ, 1969.
Meisel, W.S. and Collins, D.C., IIRepro-Modeling: An Approach to Efficient
Model Utilization and Interpretationll, IEEE Transactions on Systems, Man
and Cybernetics (1973) l:349-358.
Personal communication, Mr. Robert McMahon, Old Colony Planning Council,
Brockton, MA, November 16, 1976.
Nie, N.H., Hull, C.H., Jenkins, J.G., Steinbrenner, K., and Bent, D.H.
Statistical Package for the Social Sciences (2nd Edition), McGraw-Hill,
New York, NY, 1975.
Andrews, D.F., IIA Robust Method for Multiple Linear Regression,11 Tech-
nometrics, ~(4):523-531, November 1974.
McDonald, G.D. and Schwing, R.C., IIInstabilities of Regression Estimates
Relating Air Pollution to Morta1ityll, Technometrics, 15(3):463-481,
August 1973. --
Hoer1, A.G. and Kernard, R.W., IIRidge Regression: Biased Estimation for
Non-orthogonal Problems,1I Technometrics, ]1.:69-82, 1970.
Marquardt, D.W., IIGeneralized Inverses, Ridge Regression Biased Linear
Estimation, and Nonlinear Estimationll, Technometrics,]1.:591-612, 1970.
Theil, H., IIA Simple Modification of the Two-Stage Least-Squares Pro-
cedure for Undersized Samp1esll, pp. 113-129, Arthur S. Goldberger and
Otis D. Duncan (eds), Structural Equation Models in the Social Sciences,
Seminar Press, New York, NY, 1973.
Office of Land Use Coordination, Mitigating Secondary Impacts from the
Wastewater Facilities Program, U.S. Environmental Protection Agency,
Washington, DC, 1977..
Abbett, R.W., American Civil Engineering Practice, John Wiley, New York;
NY, 1956.
Metcalf & Eddy, Inc., Wastewater Engineering,. McGraw-Hill, New York, NY,
1972.
87.
88.
89.
90.
91.
92.
93.
94.
95.
96.
97.
98.
99.
Personal communication, Mr. Robert L. Michel, Municipal Construction
Division, EPA, Washington, DC, November 1976.
Hudson, J.F., and Alford, M.~ Growth Effects of Major Land Use:P~oje~ts:
Wastewater Investments, Area of Analysis and Causal Mode1Spec1f1catlons,.
Urban Systems Research & Engineering, Inc., Cambridge,MA, July 1977. ThlS
report was based in part on data from reference [63].
6-6
-------
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
117.
118.
119.
Great Lakes--Upper Mississippi River Board of State Sanitary Engineers,
Recommended Standards for Sewage Works, Albany, NY, 1970.
Lexington-Fayette County Planning Commission', A Growing Community (1973
Updatet, Lexington, KY, June 1973.
Lerman, S., A Disaggregate Behavioral Model of Urban Mobility Decisions,
Center for Transportation Studies Report No. 75-5, MIT, Cambridge, MA,
1975.
Peterson, G.E., Tax Policy and Land Conversion at the Urban Fringe, Land
Use ,Center Working Paper No. 0875-04, Urban Institute, Washington, DC,
1974. '
Environmental Impact Center, Secondary Effects of Public Investments in
Highways and Sewers, Council on Environmental Quality, Washington, DC,
1974.
Personal communication, Mr. Michael Cook, Chief, Facility Requirements
Branch, EPA, Washington, DC, August 26, 1977.
U.S. Urban Renewal Administration and Bureau of Public Roads, Standard
Land Use Coding Manual, Washington, DC, January 1965.
Scott, G.R., "New Sewage Treatment Works for Boulder, CO, The American
City, 74(1): 129-131, January, 1959.
Area Redevelopment Administration, Accelerated Public Works Program
Directory of Approved Projects as of July 1, 1964, U.S. Dept. of Com-
merce, Washington, DC, August 1964.
Office of Water Programs, 1968 Inventory, Municipal Waste Facilities,
Volumes I-X for each Federal region, EPA Publication No. OWP-1,
Washington, DC, 1971.
Personal communicaction, Ms. Carol Wegrzynowicz, Municipal Construction
Division, EPA, Washington, DC, June 21,1977.
Facility Requirements Branch, Guidelines for 1976 Update of Needs for
Municipal Wastewater Facilities, Environmental Protection Agency,
Washi ngton, DC, March 1976.
Evans, R., Cothran, C.L., and Cunningham, R.O., Multi-Model Transpor-
tation Site Selection Study, prepared for the Orange County Trans-
portation District, CA and the City of Anaheim, CA, August 1976.
U.S. Departmentof Transportation, Guidelines for Trip Generation Analy-
sis, Washington, DC, April 1973, p. 40.
Reference [44], pp. 30-37.
National Association of Regional Councils, Directory 177, Washington,
DC, September 1976.
Chatterjee, and Cribbins, P.L, "Forecasting Travel on Regional High-
way Network," Trans ortation En ineerin Journal, Proceedings of the
American Society of Civil Engineers, 98 TE2 :210.
6-7
-------
120.
121.
122.
123.
124.
125.
126.
127.
128.
129.
130.
131.
132.
133.
Maricopa Association of Governments, Trip Generation by Land Use, Pre~
pared for the U.S. Department of Transportation, Federal Highway Ad-
ministration, and Urban Mass Transit Administration, Phoenix, AZ,
Apri 1 1975.
Tardiff, J., "Trip Generation as a Choice Process, II Transportation En-
gineering Journal, 103(TE2):338-341, March 1977.
Gustafson, L., "An Empirical Study of Factors Influencing Trip Attrac-
tion and Trip Generation," High Speed Ground Transportation Journal,
2(3):314, 1973.
Meyerowitz, Trip Generation: Regression Analysis, Procedural Manual,
Michigan Department of State Highways, Lansing, Michigan, May 1970.
Anderson, J.R., Hardy, E.E., Roach, J.T., and Witmer, R.E. ,ALand Use
and Land Cover Classification System for Use with Remote Sensor Data,
U.S. Geological Survey Paper #964, Washington, DC, 1976.
Harvard Institute of Economic Research, Time Series Processor, Harvard
University, Cambridge, MA, December 1974.
Dixon, W.J., "Ana1ysis of Extreme Values, II Annals of Mathematics
Statistics, ~:488-506, 1950.
Comsis Corporation, Travel Estimation Procedures for Quick Response to
Urban Policy Issues, Volume II, prepared for Transportation Research
Board, Washington, DC, February 1977.
Bates, J.W., Development and Testing of Synthetic Generation and Dis-
tribution Models for Urban Transportation Studies, State Highway De-
partment of-Georgia, Atlanta, GA,Ju1y 1971.
Jeffries, W. and Carter, E., Simplified Techniques for Developing Trans-
portation Plans - Trip Generation in Small Urban Areas, Technical
Bulletin #84, Engineering Experiment Station, West Virginia University,.
Morgantown, WV, December 1966.
Ashford, N. and Holloway, F., The Performance of Trip Generation Equa-
tions, UMTA Report No. URT-52(70)-71-2, Washington, DC, August 1971.
Asin, R.H., "Purposes for Automobile Trips and Travel II , Nationwide Per-
sonal Transportation Survey Report #10, Federal Highway Administration,
Washington, DC, May 1974.
Gibbs, L.L., Zimmer, C.E., Zoller, J.M., Source Inventory and Emission
Factor Analysis: Volume I and II, EPA Publication No. EPA-450-3-75-082,
Research Triangle Park, NC, 1974.
Turner, D.B., Workbook of Atmospheric Dispersion Estimates, EPA Publica-
tion No. AP-26, Research Triangle Park, NC, 1970.
6-8
-------
v.
SUMMARY OF RESULTS AND CONCLUSIONS
A theoretical model of the Growth Effects of Major Land Use Projects
has been developed. This model represents the total land use in the drainage
basin of a wastewater collection and treatment system (the major project),
ten years after its construction. The model represents the process of induced
land use growth in the following 9 land use categories.
Residential
Commercial
Offi ce - Profess i ona 1
Manufacturing
Highways (Non-expressway)
Educa ti on
Recreation
Wholesale/Warehouse
Other
The assumption of a single basic causal structure for induced development,
and the use of cross-sectional data from 40 diverse case study major projects
throughout the United States, allowed a static approach to the testing of the
theoretical model, using path analysis.
Path analysis is a set of statistical techniques useful in testing
theories and studying the logical consequences of various hypotheses involv-
ing causal relations. It is not capable of deducing or generating causal
relations, only testing them. The causal analysis of induced land use
development in the current study involved the use of two basic statistical
techniques: two-stage least squares and stepwise ordinary least squares
(multiple regression). The first technique was required to produce consis-
tent estimates of the path coefficients in a system of simultaneous equations
involving feedback loops (or reciprocal causation) in the models. The second
technique was used to solve the remaining recursive portions of the models.
The dependent variables in these regression analyses represented the total
land use in the previously noted 9 categories. Both linear and non-linear
forms were tested and the linear form was found to produce the best fit.
Specific statistical criteria were developed to identify model paths that
were insignificant or redundant, and these 'criteria were used in an iterative
process to trim unneeded and undesirable paths from the models. The trimming
process eliminated almost half of the paths in the models as originally
5-1
-------
specified. The statistical problems of multicollinearity, suppressor variables and
identification were eliminated through the approach used to trim the initial model.
---
The final models of land use development show that strong statistical
relationships exist between the variables representing the 9 categories
of total land use and the other model variables representing induced and
non-induced land use growth processes. The results indicate that the final
model explains the majority of the variance in the case study data with R2
values ranging from 0.27 to 0.82 and averaging 0.54. The residuals of the
final regressions do not exhibit any trends or patterns, indicating the
remaining unexplained variance (1-R2) is not due to poor specification of
the model, but rather due to wide variance in the case study data (i.e.,
the problem of trying to develop one generalized model for a broad range
of situations). An analysis of the stability of the model coefficients
determined that, in general, the coefficient values have low variance (f15%)
and exhibit no extreme instabilities. An analysis of the net causal effects
. .~-~.. ._---. ---.. .,
in the land use model indicates that reserve capacity in the collection system
of a wastewater major project is a significant causal factor for induced land
use growth, principally in the residential, manufacturing, education and high~
ways categories. It is interesting to note that treatment plant capacity was
not found to be an important causal factor.
The CEMLUP-I VMT model [19J was validated using transportation data from
11 of the 40 case study major projects. Based on the validation results,
it was concluded that revisions to both the default predictive equations for
trip length and the default values for trip generation rates were necessary.
The revised VMT model was validated and found to have an average error
(imprecision) of 23%, with no statistically significant bias.
5-2
-------
-.---. -.-- ---
, '
" ~. .
;, ...
FIGU~E 4-3
, SCATTERGRAM OF, ACTUAL VERSUS PREDICTED OTHE,R TRIphLENGTHS
~ :". '.
-. ,
FILE NUMBERS (CREATION DATE = 02/01178) ,
SCATTERGRAMOF (DOWN,)" ,ACTUAL", '
0:50 ' 1".59' 2.50"'" 3.50
4' .'5.0 '
. j
, (ACROSS) t'ODEL
5~ISO'n.5b '
7~50
8.50
9.50
~+--~~~---+----+~---+---~+----+----+-~--+----+~---+----+----+----+----+----+----+----+----+----+----+.
, + " . '. . , +
l' 1. I
I 1
I 1
l' , '1
+
1
1
1
1
, '''18'.00' .-
1
I
,- I
. .1, ..',
.j~.Oo. .'"
", ;';' I , ..
" "" I ,1:>. .';.
. . 'I
- ,
, '. '. '1. ' . :.' , . .
14..00, +''-
I
,. , ',I
1', " --.
. I' -
"12.'00, +
. I
,1
, , 1.
1
10.00 +'
.1
. : I'
1
. ,I
8.06 . +
I
'I
I
. I
2<5'.00
~
, 6..00
4.00
2.00
.20 .pO,
18.00
+ . , 16.00 , ; ~
1 '
I " .: "
I ,
l' .'
+
1
t
I
I
.
+'
r
I
1
I
)
+
I
I
I
I
.
1.46 MODEL
+,
1
, I'
I
I
1
1
1
I
+
1
1
1 ,
1
.
+
.
1
1
I'
. I '
+.
1
1
1
1
.
.
.
+
+
1
1
1
I
+
I
I
1
1
+ " . , , +
.+----+----+---~+~---+-~--+----+----+----+----+----+~---+----+----+----+----+----+----+----+----+----+. .
0.0 '1.00 2.00 ; 3'.00 . 4.00' '5.'00 6.00" 7.00 8.00 9.00 10.00
OTHEH TRIP LENGTHS .., -,,' , 02/01/78 PAGE 3 '
L
0.0
14.00 .
12.00
10.00
8.00
6.00
I~ . (jQ
2.00
0.0
-------
. .
I
-------;.----.(
A
1,
. .1
" .'
.'1'
. . ,,- ."'. ,r.',"" :.<.,-. .. ,,~'-7.',.":'.FIGURE4~4' '.:.~ .
" : ' ' . ;' . ," ::,~. SCATIERGRAM ,OF ACTUAL~WORIS; TRIP lEr~GT.HS. VERSUS MODEll RESI DUALS
(CREATION DATE. = ,02/01/78)':..<, "'."". ...: ,,' ,.' .,.' .
(DOWN) RESD'UL' . . ','. ' " ;: ,. '.' . '(ACROSS) ACTUAL'. ,
1..25.. , ,',3.75: . .6..?5 . :,' 8.7? 11.25. 13.75.:16'~25:. 18.75,.,21.25 23.75" .
6; 00' ':. : +_..:_...+--;.;.+---..,.'+-:---,-+:-~--:+;;.---:+..._--+----+----,+----+----+..:_--+-;..-~+-~-:-:+----:+..._..:_+----+..:_--+----+----+:.
1 ' ' 1
, I' 1
1 . 1
I ' , I '
. ,
FILE:, ,NUNSERS
SCATTERGRA~1 OF
4.~0 +
I.
I ,
,I .'
., , '1
+
.
I
I
1
T
. 3.6Q
+
"",.
+
I
1
1
I
2~40+
'I
'. I
I
I
+
I
I
I
I
+
I
1
I
I
.
I
I
. I
I'
,1.20
,+
I
I
, I
I
+
I
1
I
I
'-
I
I
.1 '
1
+,
T'
.
.
,J::-
,R3
, 0.00
+
-,1.20
.
+
I
I
, I
'I
':'2.40,
.
I
I
'r
+
-3.60 '
'+
I
1
1
I .
+ .
I
1
1
I
+
I
1
I
I
1
I
J
I
I'
I
I
I
+
-4.80
+
I
I
I
I
-6.00 +.,' , . .' ' . " +
,+----+-:-~-+----+..._.,.-+----+----+----+"'---+----+----+-'"--+----+-.,.--+----+.,.---+----+----+----+-.,.-.,.+----+.
, 0.0 2:50 5.00 7~50 .10'.00 12.50. 15.00' rr.50 . 20.00,' 2;:>.50. ' 25.00
~ORK TRIP LENGTHS ' 02/01/78 PAGE ~
I
L.
:. .~,
, .
..
, 6.00'
4.80.
3.60
i:40 '.
1.20
'0..00
. -1.20
, -2.40
-3.60 '
-4.80
. -0.00
-------
"'~:;.-:z.":"':~""i!.~ ":. :v..; ..'"."
':"-,~.~~.>.r ',:,,~ 'FIGURE 4-5' '::-, " "
.: " ., ::' ,.~ ~~~R48,OF-,ACTuAl.":B1;HER TR!! L.E~.? VER$lJS !KlpE~ RESIDUALS
FILE NUMBERS (CR'~ATION DATE :: 02jo~iI:nbf!:.,-" .''- '" "..., . ,.,. ,.., ,I' '. .
SCA TTERGRAM OF (DOWN 1 RE3'DUL. "';,: -~-: I .:: .~, . ':;;" (ACROSS) 'ACTU'A.L :
1.00 3'.00 '5".'00 '::. 7.00. . ~:9'.00 11,p"O 1'3.00. 15.00 17.00 ,19.00 '.
.+----+----+----+----+----+----~~---+----+--~-+----+----+~~--+--~-+---~+----+----+----+----+----+----+.
6.00 '+ ,,' '" '. " +
".'
.~: i
I
I
I
I
4.80 +
I
I
I
',. 1.
, "
I
I
I
!
+
I
I
I
I
+
I
I,
I'
I
+
I
I
I
I
+
I
I
I
I
+
I
I
I
I
+
I
I
I
'- I
+
,I
I
I
I
+
I
I
!
I
-+
I
I
I
I
..
, ,~- ,
,,;
'.
'. .'.
'. ,
. '."
, .
,: .
:, ~. . I.
.; ~:.. '
- ;". ~! . , .
. \, .'
, .
.
3.60 +
I
I
, "'1
I
,1
. --.'
" ,'.
..;.
-. :,. .:- . ':
, ,
; i
. . ~ .
2:40
, ..
'.
,+
I
I
I
I
.
1.20
+
I
I
I
I
.s::
81
0.00
+
I
I,
.1
I
+
I
I
I
I
.
.
-1.20
'-2.40
.
,+
I
I
I
I
-3.60
.
+
I
I
I
I
-4.80
+
I
I
I
I
-6.00, + +
.+~---+----+--~-+----+----+-~--+----+----+----+----+----+----+-~--+-~--+----+----+----+----+----+----+..
0.0 2.00 4.00 6.00' 8.00 10.00 12.00 .14.00 10.00 18.0020.00
OTHER TRIP LENGTHS 02/01/78 PAGE 6
,I
I.
6.00
4.80 .
"
3.60.
2.40
1.20
. .
0.00,
-1 ;20
-,2.40
-'3.60
. , '"
-4.80
-5.00'
-------
b.
Trip Generation Rates
Actual data on trip generation rates were available from only
2 case studies. These data are summarized in Table 17, along with the
default values used in the VMT model. Since few data were available,
averages computed from several hundred trip generation studies done nation~
wide and reported by the Maricopa Association of Governments [120] have been
included in Table 4-24, as well. Due to the lack of sufficient data for sta-
tistical testing, the maximum error criteria were used only as a guide in
determining whether VMT model trip rates are reasonable. On this basis,.
the rates for the first two commercial categories are overestimated, while
the rate for recreation is underestimated. All other default values are
valid. Revisions made to the trip rate values in the VMT model were per-
formed so as to bring the default values into line with nationwide averages,
name ly:
,~ ..
" ~
. Commercial,
-------
TABLE 4-24
COMPARISON OF ACTUAL AND PREDICTED
TRIP GENERATION RATES BY LAND USE
~ " Land Use Type/ Trips Per Auburn, St. Louis, Nationwide VMT
Tri p Purpose Measure WA MO Average[120] Mode 1
.'
Residential/Work OWe 11 i n 9
SF Detached Unit 1.7** 1.8
SF Attached 1.5
MF Low Ri se 1.2
MF Hi gh Ri se 0.8
Mobile Homes 1.8
Residential/Other Owe 11 i ng
SF Detached Unit 4.4** 8.0* 9.5* 9.0
SF Attached 7.0* 7.0* 7.0
MF Low Ri se 7.0* 7.0* 6.0
MF Hi gh Ri se 7.0* 4.0
Mobile Homes 5.5* 6.1 * 5.0
Commercial, 103ft2 25-80 67 130
<50K GLA
Comme rci a 1 , 103ft2 25-80 64 80
50-lOOK GLA
Commercial, 103ft2 -30-65 34-46 40
>lOOK GLA
Offi ce 103ft 2 12-20 10- 25 16
Manufacturi ng 103ft2 5.0 4.2-9.3 5.0
Wholesale/Warehousing 1 03 ft 2 9.0 5.5 4.0
Culture 103ft2 2.0
Ch urches 103ft2 2.0
Hotel/Mote 1 103ft2 10
Hospitals 103ft2 16
Educa ti on 103ft2 4.0
Recrea ti on Acre 42 10
* Total for work and other trip purposes
**Average for all residential types
. .
4-65
-------
that the lIactua'" data are in reality model estimates themselves, it is
expected that a large amount of imprecision exists in these data. Thus,
applying maximum error criteria or minimum correlations were not appropriate
However, the comparison still provided a method for estimating model bias.
Due to the mis-match between the land use and trip type categories of avail-
able data from transportation agencies and those used in the VMT model, only
3 combined categories provided sufficient data for comparison. Table 4-25 .
summarizes the test data by case study. The average bias of the model pre-
dictions are:
-6% for Residential/Work trips
+10% for Residential/Other trips
+22% for Other Trips
These values are well within any reasonable limits for this type of model and
indicate, if anything, that the model is conservative in general, i.e., it
slightly overestimates total vehicle trips.
c.
VMT
Direct data on total VMT were available for only 2 case
studies. However, indirect estimates of VMT, based on zonal origin/desti-
nation data, were available for an additional 6 case studies, providing a
total data base of 8 samples for the validation exercise. Table 4-26 summarizes
these actual VMT data, along with the estimates produced by the revised VMT
model. All VMT estimates used a value for Lr based on a circle with an
area equivalent to that of the actual area of analysis. Thus,
L =((AREA/640)/TI)1/2
r
A scattergram of actual versus predicted VMT is given in Fig-
ure 4-6. The correlation coefficient is 0.93, indicating excellent predictive
ability for the revised VMT model. The regression line of best fit has
slope of 1.05 z 0.19 and intercept of 35,792 z 65,043. Since the standard
errors Qf estimate for the slope and intercept exceed the differences between
4-66
-------
TABLE 4-25
COMPARISON OF ACTUAL AND PREDICTED* DAILY TRIPS
IN THE AREA OF ANALYSIS BY LAND USE AND PURPOSE
Transporta ti on Res i denti a 1 Other Land Use
Case Study Work Tri p Other Trip All Trips
Lexington, MA 7 , 1 32 17,317 36,202
(7,490) (22,380 (33,984)
Benni ngton, VT NO NO NO
(4,590) (13,320) (89,803)
Richmond, VA 40,698 87,008 88,946
(29,560) (88,350) (98,489)
Charlotte, NC 3,250 NO NO
(7,500) (22,440) (41,033)
Clearwater, FL 24,455 73,500 . 115,849
(49,940) (148,980) (202,675)
Roseville, MN 36,466 100,731 ~ 36,971
(21,890) (65,580) (213,784)
St. Louis, MD NO NO NO
(95,950) (287,340) ( 438, 149)
Bo u 1 de r , CO 28,729 144,296 184,081
(38,430) (114,720) (139,428)
S. Phoenix, AZ 24,057 48,965 80,728
(29,700) (88,980) (23,568)
Va llejo, CA . 35,828 88,420 63,864
(8,400) (97,680) . (145,802)
Auburn, WA 19,467 51,906 125,748
(14,650) (43,860) (36,342)
NO = No Data
*Predi cted values are in parantheses.
I
, .
4-67
-------
TABLE 4-26
COMPARISON OF ACTUAL AND PREDICTED TOTAL
VMT FOR THE AREA OF ANALYSIS
Transporta ti on
Case Study Actual
Lexington, MA 97,810*
Benni ngton, VT NO
Richmond, VA 608,676*
Charlotte, NC ND**
Clearwater, FL 1 , 1 36 ,800
Rosevi lle, MN 519,746
St. Louis, MO N 0***
Bo u 1 de r , CO 1,126,616*
S. Phoenix, AZ 554,035*
Vallejo, CA 484,758*
Auburn, WA 273,559
Total VMT
Predi cted
102,975
607,870
1,123,422
898,990
1,267,922
512,588
649,090
175,076
NO = No Da ta
* Vtn estimated from zonal origin/destination statistics and area of
ana 1 ys i s size.
** VMT data for entire city, not just area of analysis.
***Land use data incomplete to translate trip rates to trip totals.
4-68
-------
the parameter values and the set (1.0, 0.0), the values are not significantly
from (1.0, 0.0). Hence, the revised VMT model contains no statistically
significant bias. The average error (imprecision) of the revised VMT model
is 18%. As this value does not exceed the acceptable criteria limit of 23%,
the revised VMT model was judged suitable for use in the GEMLUP-II model.
The residual errors from the VMT val idation are graphed versus
actual VMT data in Figure 4-7. Examination shows no evidence of hetero-
scedas ti ci ty.
4-69
-------
FIGURE 4:-6
SCATTERGRAM.OF~ ~t'ruAL VERSUS -PREDICTEDVMT
SCi\TTERGRlI!.j' OF (l.;OW~;) ACTUAL - - (Acr;CS:,)PR::C
__no. .-. -.- -n --., 5000'. QO -2 25UOU-:OO -TI5-COO. 00 525000: COO 67500C. 00- S 25000.00 '17',000:; do 11 25r'Or . ('(11 :'7". r'no . (lC 1 !! ?'>000 . 00
.+----+----+----+---~+----+----+----+~--~+----+----+----+----+----+----+----+----+----+----+----+----+
1500000.00 + +
-.---'--"---'---"----."'----r - ,.. -- __.__0_._._----.----"-" .0. '.-- -.. -- -- _0._' ._.0 ".
I
I
+
I - T
l~- -- . ------ ~- _0 --- - - - -. - - . - - - - 1
+ - , ' " - - +
- ----+----+----+----+----+--7~+----+----+----+----+----+----+----+-~--+----+----+-~--+----+----+---~+.
- ~.(' lSGUOo.ou-~r~noo~oo "SOGOO.DO 5nC000.C375ra~C.Co 9~~OCG.~11D~Jnoc.~n12n0n~0.nn1~50C00.Gn1~n0~~C.0C
.::CA'1"lT 1:(; riM: LI STI [ :.; - ::-,21P(~ /'( f1 Dr': f 3
---.----- --'U -- ..,. 0 - r
. <~<_..- -. -
--. - _-0 - .,
1350Cd:;O.CiO +
I
--- 0.'-- _...,- -.. ..0_-"10.' - ----
r
- r ,
.-., 2f;COOC; 00' +' -
I
. - ...~--_._.. .-- "'-~'---'."~"-"----'''-' .~." ._.~-~-,
'--... .... -, ..
---:....-.. .~~_. ',' .-..-.-, . ''''':'';'''
- ,
I -
---r-".- ,,_..
r
...,..... "-"-._..u_, .-"--' -_...._-~. ... .--..
- . . ,
. .'..,.... ..._~,_.
.i 0500tJO. 00 +-
_0_.._-... -- 0'- -..--... - -., I' -
.l
r
------_.._~.- ---.- --..----4--.
- ~
. -- . -- ._- .--.-. - ,.. " --._-- .~. .-- ...~
.::-
--~..
_0 -
9000tJO.OO +
r
I' -
r
I
T,roooo; co .- "" .-0
. I
r
r
I
t ... ,.-..-
6UUOO,LOtJ +
i .._--. - - .... .. ,. "." -...- I
r~ I
I
I
1::;(j()(jO.OO
+
I
-1
I
r
+
T .
.l
I
1
1
+
j'OC{)OO'OOO
l:)(;OUO. GO
O c-
. J
. .-- .', ... - .~ -
. ....~. .'
.
.-
.- .
T
I
,
+
T
I
I
I
+
. I
T
T
+
I
I
I
1
+
T
T
I
+
I
I
T
T
+
:r
T-
I
T
+
T
J
T
T
1
+
1 S!)('~~)(:r.:'O 00
1 ~~lDr~()o.. ~JO
l?oncnc'Oo()'-
1050000.00
9pt)0IJC.00
7",C'O~JO; 0(1 -.
rsnCD(H''OOO
)1 '}-"onG . 00
'3(~o(~cn'O r.(')
T
, 5C:""10C:'O 00
0..0
-------
. 0....--- 0-
,.-' - ,
. '--"1 ...-.,
---- . -- ----.
-.--. _._. .-. - .'
. ~
.;
----. -.- '-'--'~'''---'-''-'--
-------.-- -- -.- ---"_._-_.~---_.,
FIGURE 4':'7
SCATTERGRAM OF ACTUArVMT~VERStJS MODEL RESIDUALS
SCI"i"i'f::H(j~;M; OF (DC~it:) RESDUL (ACROSS) ,\CTlJAL
--- -----.-- ___.h" ....15.0o.0~.OJL2250ilO....Qo...:n5000.. 00. 'i25.oo.o..0.D. 6., 5.Q1).D:. 0(L 825.000...00..!D'.5coo._L:O.1.12S00G . Q(: 127 50 or. . co 1 iI25JC-O. OC
.+----+----+----+----+----+----+----+----+----+-~--+----+~---+----+----+----+----+----+----+----+----+.
500000.00 + +
1 I
I ,
I l
1.. . - ... -"~--'- "'h .. --..... --.. T
I
I
I
I
_.--._--}O..o1.QC_illL..+.~. ---.- ..
I .
I
I
I
2l!Onoc.uu +
_..~--_. -.-- ..~.~.- 1...__- ._h .
. I
I
I
.-. ---.. - - -
- -. ~ -- .__.
400000.00
-..--- -..
-'-.---. - .- _. ...
10DOOO.CO +
I
-..- ._~... "'h --. '" I
I
I
~
I -...
~
-----_.~.. ~_. ..-- .-. . -...,
-100000.00
1'- -
! .--. ..----... ....-.. - ~- .
-2000r.O.OO
+
-,..- .. '...---'----" - ...
- -. ~--- , .
. ,- .
.:.
.. .~._-.__..._.~ . ._-....- .,
.
--..- ".. .~....
.
.
a.DO
.
.
+
I
I
." I
I
+
I
T
J.
.
I
.1 .
+
I
.1
I
.
I
- j~J_('_l.UG . ~:D u- + u
I
I
I
I
+
I
-4(;uJ0u. c-j
-~.')~J(jGC. Cl:
.
I
1 .
j
~ - -.- -----.-.---
+
.+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----~----+----+----+----+----+.
(1.0 . 1 '.) i ~(', OC . ;~'C 3 CQD DC. Of) 45 r. 0 GO. oe. ') (: G DOC, ..c.( 7 S (1(;_00. t:C 9 G;JG.o~; . 0.0-1 r:-~. {)O:JC.. Q~': 1 z-..) ,) (: (~C . ["1(' 1 ~ S C'CH:~~ . :) C 1 S OG C' 'JC . c.;~
SO'iTEhCfi;'\:': LlSTlNG o ?.I(J-/'n .),'~r ,.
')()(lU(JLJ. ()Q
+
1100000.00
I
T
+
I
T
3Q~OD:j. ':.G
. I
T
+
?on,:'iJ(; . on
1
T
T
I
+
T
I
J (:,GCr.~o. nn
T
1
+
I
0.80
I
I
T
+
I
-100nOC.OO
J
.,.
.1
+
-:?(1C1()(I.OO
I
J.
I
+
-:iOCOvO.DO
T
.1
T
:;
:;
+
-;~,~~;f~OrJ. cn
J
T
!
I
_1)~'"1(',nf)iJ. (;0
------- |