vvEPA
United States
Environmental Protection
Agency
Office of Energy, Minerals, and
Industry
Washington DC 20460
EPA-600/7-79-119
May 1979
Research and Development
The SEAS
Region Model
An Assessment of
Current Status and
Prospects
Interagency
Energy/Environment
R&D Program
Report
-------
ABSTRACT
This document describes the research performed in response to a recog-
nized need .for improvement of the regional component (REGION) of the U.S.
Environmental Protection .Agency's Strategic Environmental Assessment System
(SEAS), a..computer model of the nation's economic-environmental-energy inter-
actions designed to assess alternative environmental and energy policy
scenarios.
The research reported here consists of two parts. Part I is an assess-
ment of the quality of the regional economic data base developed by the
REGION module of SEAS. A comparison of this data base with regional output
data computed by U.S. Census Bureau leads to a recommendation that SEAS base
year data base be recalibrated according to the Census data, taking into
account spatial variations in sector output worker.
Part II examines alternative regional projection methods for possible
inclusions in SEAS. The method of location quotients to derive regional
input-output tables from National 1-0 is re-evaluated with encouraging
results. Reviews of alternative existing models were then performed, with
their handicaps against inclusion in SEAS discussed. A new Multiregional
Model of Regional Economy and £nergy Demand (MREED) is developed as an
alternative. FinaTly, a model to assess regional impacts of changes in a
region's energy prices or final demand is presented.
This report was submitted in fulfillment of Grant No. R8048690-01 by
The Johns Hopkins University under the sponsorship of the U.S. Environmental
Protection Agency. This report covers the period from October 1976 to
February 1978 and work was completed as of February 1978.
ii*
_
-------
CONTENTS
Page
Abstract
Figures vi
Tables vii
Acknowledgment viii
1. Overview and Summary 1
2. An Assessment of the Data Base of REGION 19
3. An Assessment of Location Quotient Methods of
Regionalizing the National Input-Output
Model 50
4. A Review of Alternative Regional and
Interregional Models 95
5. A Multi-Regional Model of Regional Economy
and Energy Demand (MREED) 116
6. A Model to Assess Regional Economy Effects
of Changes in Final Demand on Energy
Prices 130
7. Appendix -- Data Sources for Chapter 3 140
References 146
-------
FIGURES
Number
Page
1-1
4-1
4-2
5-1
5-2
The SEAS System 3
Flowchart of READ Model 110
Energy Demand i n READ Ill
Structure of MREED Model 120
The MREED Model 121
v1
-------
TABLES
Number
Page
1-1 SEAS/CM Functions for Selected States 7
2-1 Purification of Census of Manufactures
Values of Shipments 25
2-2 1972 Price Indices for INFORUM Sectors 26
2-3 Selected National Sector Outputs for 1972 27
2-4 SEAS/CM Fractions for Selected States 28
2-5 Effect of 0/W Modifier on SEAS/CM Fractions
for Selected States and Selected Sectors 31
2-6 - 2-39
Detailed Tables for each of the 34 Sectors 33
3-1 1967 Washington 50-Sector Input-Output Table 60
3-2 Employment, Output, and Location Quotients
for Washington and the U.S 66
3-3 Comparison Between Direct Coefficients and
Tables Obtained from Survey and Short-Cut
Methods 1 78
3-4 Comparison Between (I-A)"1 Matrices Obtained
From Survey and Location Quotient Methods 80
3-5 Comparison Between Actual and Estimated Output
Computed with Final Demand as Given 81
3-6 Comparison Between Intermediate Outputs
Obtained From Survey and Snort-Cut Methods 83
3-7 Comparison Between Exports and Imports
Obtained From Survey and Short-Cut Methods 85
3-8 Comparison Between Tables Obtained From
Survey and Balancing Procedure 87
5-1 Sectors Proposed for MREED Phase 1 119
7-1 Sources Used for Employment Data for
Industry Sectors 140
7-2 Sources Used for Employment Data, by
Industry Sector 141
vii
-------
ACKNOWLEDGEMENTS
The authors would like to take this opportunity to acknowledge the
invaluable help received during the conduct of the research reported here
from many individuals too numerous to mention. Professor Anne Carter of
Brandeis University provided invaluable help throughout the project
reacting to and refining many ideas presented to her in her own inimitable
style. This study owes a deep debt to her. We are also thankful to
Professor William Miernyk of West Virginia University, who reviewed our
efforts at assessing the location quotient method of regionalizing the
national 1-0 table.
This study would not have been possible without the encouragement,
discussions and reviews provided by some key members of the SEAS model
development and policy analysis team. Particular thanks are due in this
regard to Dr. Richard Ball, the project monitor at the U.S. Environmental
Protection Agency (EPA), Mr. Edward Williams of U.S. Department of Energy
and Mr. Sam Ratick, of U.S. EPA.
We are also grateful to anonymous reviewers of our draft report for
valuable comments. The responsibility for any remaining errors of
omission and commission is entirely ours.
T. R. Lakshmanan
viii
-------
CHAPTER 1
OVERVIEW AND SUMMARY
The last decade has witnessed a widespread concern with the quality of
the natural environment -- the quality of air, water and land and the
adequacy of the natural resources (particularly energy) to sustain economic
growth. This concern stems from such factors as the adverse environmental
factors of economic growth, the unrelenting growth of energy use, and the
increasing participation of the government in decisions pertaining to
environmental protection and energy supply.
This concern has expressed itself in the passage and implementation of
a variety of environmental protection laws and regulations and energy
conservation and supply augmentation measures. Such laws and regulations
have far-reaching economic and environmental consequences. For instance,
the regulations requiring industries to abate environmental pollution lead
to increased costs of production, and in turn to higher prices, falling
output in those industries and reduced employment and income in the region
where such industries are located. There may be, at the same time, growth
in industries that supply pollution abatement equipment and services in
those or other regions. Over time, the health and economic benefits of
higher environmental quality express themselves in changing patterns of
consumption. Thus as environmental standards pertaining to industries,
automobiles and municipalities are implemented, the mix of goods and
services in the nation and the region changes. Again, it is well known that
the magnitude of energy demand and the mix of sources of fuel or energy in
the future will greatly affect the quality of land, air and water bodies.
In the context of this intimate association between the economy,
environment and energy, a number of public policy questions are posed by
recent and proposed environmental and energy development measures. What
are the full range of socioeconomic effects of existing and proposed air,
water and land pollution abatement provisions? What are the likely effects
on the economy of (and the policies to facilitate) the transition from
cheap, or abundant energy and from a reliance on oil and gas to other more
expensive sources of energy? What is the nature of the trade-off between
additional, or lower cost energy and environmental quality?
Considerable analytical effort has been devoted, in the last half
decade, to these policy issues by the development and use of large models
at the interface between the economy, environment and energy (Hogan and
Parikh, 1977; Hitch, 1977; House, 1977; Lakshmanan, 1978). These models,
-------
reflecting the vast scope of economic-environmental-energy interactions, are
concerned not- only with economic issues but with a range of issues of public
policy, physical environmental quality processes (e.g. emissions and spatial
diffusion of pollutants), flows of energy and alternative technological
processes of energy supply. This overwhelming diversity and the inter-
mingling of socioeconomic and physical relationships in economic-environ-
mental-energy modelling has encouraged cooperative endeavors among social
scientists, engineers, and physical scientists. The resulting models often
represent an integration of the different analytical approaches the
different analytical approaches the different disciplines bring.
One of the earliest and more significant examples of such integrated
economic-environmental-energy models is the Strategic Environmental
Assessment System (SEAS) model. The SEAS model has evolved from a test
version implemented in 1973 in the U.S. Environmental Protection Agency
(EPA), to a large system of interlinked models of the economy, environment
and energy, operationally used for policy analysis at the U.S. EPA and U.S.
Department of Energy (DOE) (Lakshmanan and Krishnamoorthi, 1973; House,
1977; Lakshmanan and Ratick, 1977).
The SEAS model has been developed by a large number of individuals
drawn from different disciplines -- economics, geography, engineering,
systems analysis, policy sciences, and energy technology and different
institutions universities, private research organizations and the U.S.
Government.1 This multidisciplinary, multi-institutional and development
environment has made the SEAS model comprehensive in scope and closely
linked to policy analysis. SEAS comprises five component subsystems
(Figure 1-1):
1. The National Economic Subsystem (consisting of a linked
National Input-Output Model (INFORUM), a sector-process
disaggregation model and a pollution abatement cost
model).
2. The Environmental Subsystem (generating gross and net
residuals from production, energy conversion and trans-
portation).
3. Regional Economic and Environmental Subsystems.
4. The Energy Subsystem (generating residential and
commercial, industrial and transportation energy demand,
total energy resource requirements and energy
investment).
1
A selected list of these individuals includes Peter House, T. R. Lakshmanan,
Ted Williams, Richard Ball, Roger Shull, Phil Patterson, Sam Ratick, Ron
Ridker, William Watson, A. Shapanka, N. Dossani, S. Krishnamoorthi, R. Meyer,
R. Doggett, M. Dramer, M. Stern, R. Anderson, B. Wing, E. Lake, R. Ubico,
and Peter Krol1.
-------
-------
5. National Resource Subsystem {accounting for the effects
of depletion of critical raw materials).
From the early steps of its development, this large model has been used for
analysis of a wide range of public policies in the U.S. EPA, U.S. DOE, and
other agencies. Illustrations of such policies include the economic and
environmental effects of resource recovery policies, the annual economic
effects of the implementation of the U.S. Clean Air Act and the environ-
mental consequences of the National Energy Plan.
In view of this important role in policy analysis, the SEAS model has
been the subject of evaluation. A SEAS Review Panel, chaired by Professors
Wassily Leontief and Thomas Crocker, expressed confidence in the overall
structure and data bases of SEAS as used at the National level.2 The
panel, however, was less sanguine about the Regional Module (REGION) of
SEAS and recommended an improvement of the SEAS Region Model.
In response to this suggestion, a systematic analysis of SEAS Region
model structure and data bases was conducted in Johns Hopkins University
during 1976-78, under a grant from the U.S. Environmental Protection Agency.
The result of this inquiry is detailed in their report.
Two objectives have guided the SEAS Region assessment project. The
first objective was to assess'the quality of the data bases used in the
SEAS Region -- an effort reported in Part I of the document.
The second objective was to recommend an improved structure of the
SEAS Region Model. The research work directed at the second objective was
organized into tasks. The first task focused on assessment of available
alternative models to regionalize national economic data. The second task
was the formulation of a regional model designed especially to facilitate
environmental and energy policy analysis in the structure and policy
analysis context of SEAS. The outcome of the research in both these tasks
is the theme of Part II of this report.
We highlight in the remaining portion of this chapter, the salient
features of research directed at both of the above objectives. We present
the same results in greater detail in the following chapters.
PART I. ASSESSMENT OF THE DATA BASE OF REGION
Introduction
The first task of the research work was to assess the current data
base of REGION, the regionalization module of the Strategic Environmental
assessment System (SEAS) Model. A brief description of REGION will be
2
The Ad Hoc SEAS Review Panel {Co-chairmen: Wassily Leontief and Thomas
Crocker), Quality Review of SEAS, (Mimeo), December 3, 1975.
-------
presented here. It is followed by a comparison of economic output data for
1972 for selected INFORUM sectors in 15 major states generated by SEAS Region
Module against data collected by the U.S. Bureau of the Census for the same
year. Discrepancies between the actual state values and the SEAS regional
estimates are then interpreted in terms of variations in state labor pro-
ductivity from the national norm. From the results of this analysis we
present some suggestions for future approaches to strengthening REGION'S
current data base.
The Regionalization Module of SEAS
REGION allocates national projections generated by other SEAS modules --
economic output (from INFORUM), and pollutant emissions (from RESGEN) -- to
counties and reaggregates the estimates to any of several levels of regional
classification. No analytical representation of intraregional or inter-
regional relationships of production or consumption is contained in REGION'S
structure. Rather, national sector output is allocated to regions according
to each region's proportional share of national employment (in most cases)
in that sector. REGION contains a set of these fractions for each regional
classification scheme, for economic output and abatement costs (E-SHARE files)
and for residuals generation (P-SHARE). The P-SHARE file, applied to all
residual types generated by each industry therefore allocates to a region the
same fraction of each pollutant generated by a given national industry.)
These base-year (1971) SHARE files are based on county-level employment data
and aggregated to the appropriate regional level.
SHARE values for future years are adjusted by indices developed from
1972 OBERS Projections: Regional Economic Activity in the U.S. (Series E
Population, BEA/ERS, U.S.6.P.O., 1974).This "OBERS Growth Index" is a
ratio of projected earnings of the 2-digit SIC code containing the INFORUM
sector to those of the base year (1971).
then normalized to sum to 1.
The adjusted SHARE fractions are
Several disadvantages are present in sich use of employment as an index
in constructing shares for allocation. First, by using employment as a
surrogate for output, SEAS incorporates the implicit assumption of constant
labor projections themselves employ 'shift-share' methods, extrapolating
present trends of industrial location. Third, the pollutants SHARE file
assumes that (in most sectors and most regions) the national technology
within an industry holds for all regions (although provision exists for the
use of other values if data were available). All of these weaknesses raise
doubt about the accuracy of the current SEAS regional allocation procedure.
Comparison of Regional Data Bases
SEAS-generated estimates of industrial output at the state level for
1972 were compared with Census Bureau data for that year, presented in the
Census of Manufactures. Comparisons of these two data sets were made for
34 selected industries in a sample of 15 large states. However, several
preliminary adjustments were first required. The SEAS output (reported in
-------
1971 dollars, hence adjusted for 1972 prices) is categorized by sector
according to "purified" sector used in the SEAS INFORUM model. In INFORUM,
all output by that industry consists of products of that industry. The
Census of Manufactures (CM) data, however, is reported on a state basis
only with all products produced by an establishment assigned to the
industrial sector manufacturing the product whose output is greatest at that
establishment. As a result, the state CM data required "purification" by
multiplication by the national ratio of purified to unpurified output by
sector, both of whose values are available at a national level in the Census
of Manufactures. Finally as the 1972 SEAS outputs are projections of a base
year (1971), the national SEAS and CM output figures cannot be presumed to
coincide. The state SEAS figures are simply allocations of national pro-
jections, so in order for these allocated shares to be comparable to state
CM data, the state SEAS estimates were scaled by the ratio of national CM
sector output to national SEAS sector output. In this way, REGION'S allo-
cation of the correct national totals could be tested.
Once these adjustments were made, the state CM and estimated SEAS
values were compared, w-ith substantial variation observed. Measures of
these variations are provided in Table 1-1 and detailed tables in Chapter
2. From these results, it is clear that SEAS' allocation of national output
to regions involves substantial inaccuracy.
Accounting for Output Per Worker Variation
We had suggested earlier that the SEAS REGION Model, by allocating
national output to region using employment shares, ignores regional
variations of output per worker in an industry. The consequences of this
procedure were explored next.
The ratio of state to national output per worker by industry (obtained
from the Cejisus of Manufactures) was used as an adjustment factor applied
to SEAS state~ eslfimateis. This adjustment step generally caused the SEAS
REGION estimates to move closer to Census data, but frequently the full
difference was not accounted for. If it could be incorporated as a second
factor in REGION'S shares, however, substantial improvement may be obtained,
at least for those sectors for which regional data are available. This
suggestion is made under the presumption that regional output per worker
differentials do not vary greatly in the four or five year intercensal
period.
Implications
Considerable differences between SEAS estimates and Census of
Manufactures result from the current operations of SEAS' Regionalization
module.Output per worker variations among regions may account for some
discrepancy, and could be partially accounted for in the REGION module by
adjusting the SHARE factors where data are available,1assuming regional
stability of output per worker relationships. Additional error may be
attributable to noise derived from basing the shares on county-level
employment, aggregated to the appropriate regional configuration. The
-------
TABLE 1-1. SEAS/CM FRACTIONS FOR SELECTED STATES
SEC-
TOR NY OHIO ILL CAL TEX
23
24
25
26
27
31
39
40
45
SI
52
53
54
59
66
67
69
2.03 1.77 1.34 .98 1.26
.57 1.12 .60 .74 2.49
.74 .68 .98 .86 1.23
.97 .90 1.23 .78 1.38
.94 .93 .87 .92 1.45
.74 1.18 1.13 .66 D
.43 0 1.31 D D
1.13 .61 1.03 1.20 1.51
1.42 .79 1.33 .80 1.02
1.17 .98 1.15 .75 .85
.98 1.07 1.11 .90 1.08
.62 1.72 1.09 .61 D
b
.79 1.06 .99 1.16 2.11
1.21 D D D .61
.62 1.06 1.32 1.28 1.5CT
1.71 1.29 .7? 1.39 1.18
D .61 1.41 .93 1.12
SEC-
TOR NY OHIO ILL CAL TEX
76
81
84
89
94
95
97
98
101
103
107
113
118
119
123
127
128
1.00 D .77 .73 .66
D 1.16 1.23 .85 .98
.83 1.67 1.10 D D
.45 1.27 ODD
.96 .99 1.20 D D
1.20 .89 1.19 .95 1.14
D .88 1.08 1.51 3.02b
1.04 .84 1. 00 .90 2.41b
D D .83 D D
b b
1.09 1.02 1.16 D 1.77
1.08 .70 1.03 D .7$
.90 D 1.35 D 1.126
1.03 1.17 .96 D .69
1.56 .95 1.31 1.00 .90
D 1.38 1.03 .75 .55
1.06 1.13 .98 .99 D
.50 .93 1.29 1.16 D
aSEAS is modified by the price index and.scaling, factor; CM is modified by the purification
correction factor. No modification has been made for 0/W differences.
"State accounts for less than 17. of total sector output.
D « Disclosure limitation
-------
flexibility of the current SEAS regionalization procedures allowing various
regional classification schemes can be viewed as a trade off against the
more tractable alternatives of data collection and modeling on the scale of
a larger regional level such as a state or BEA economic area.
These observations lead to the proposition that a regional data base
for economic output be calibrated for a base year, say a census year such as
1972. By calibrating to a state level, the best available regional output
data can be utilized supplemented by any additional sources where census
data are insufficient. By conducting a careful inventory of regional
economic data for a base year, the credibility of SEAS' regional output
estimates can be enhanced, and its measurement of regional output can be
used with greater confidence in regional analysis and by regional users.
PART II. APPROACHES TO IMPROVE SEAS REGION MODEL STRUCTURE
Our efforts at exploring alternatives to the current practice of using
OBERS regional projectors in SEAS were structured into three tasks.
First, we explore the potential of extending the input-output (1-0)
framework (used at the national level in SEAS) to the regional level. The
depiction of a regional economy by an 1-0 table permits detailed analysis
of the structural interrelationships and is useful in a descriptive and
predictive sense. However, data collection for such detailed tasks requires
expensive surveys and the data is often dated by the time it is collected
and used. Hence, various short-cut methods have been proposed to develop
regional tables from a given national table, using only a limited amount of
regional data. Such methods would be attractive, as they are relatively
simple, quick and inexpensive. However, these methods have yet to be proven
effective. In this study we test empirically one nonsurvey -- the location
quotient method of regionalizing the national 1-0 table to the state
level for the State of Washington with encouraging results.
Second, we review various existing regional and interregional models --
econometric and inputroutput -- for potential adoption with or without
modification as the SEAS REGION model. The primary concern of this review
was the suitability of these models as tools for analyzing environmental
and energy policies at the regional level.
Next, we present a regional model that is specifically designed for
environmental and energy policy analysis in the analytical context of SEAS.
Finally, we discuss a model to assess regional economic effects of
final demand or energy price changes. We present the results of this
study in brief in the rest of the chapter and in detail in Chapters 3 through
6.
-------
A. An Analysis of a Shprt-Cut Method of Developing RegionalInput-Output
Models
Introduction
Various methods have been suggested to derive regional input-output
tables without undertaking expensive and time-consuming surveys. The
common procedure is to modify a national table using one of a variety of
methods. In this study, location quotients were used to attempt to adjust
the 1967 U.S. national table3 to derive the 1-0 table for the state of
Washington for that year.4
The Use of Location Quotients- to Regionalize National Input-Output Tables
In regional studies, location quotients have been used as a measure of
an industry's representation or importance in a region, relative to its
importance in the nation: LQ. = (N.R/NR)/(N./N) , where for industry i, N^
represents some measure of the industry's
importance (e.g., output, value added, or employment) in region R; NR is
total employment in R;
employment.
is national employment in i; and N is total national
If LQ. = 1, then the industry has the same importance within the region
as it does^ithin the nation a? a whole. If LQ^l, its relative importance
is less than it is nationally, and if LQ.>1, it is greater.
The application of this simplest location quotient to a national
input-output table requires the assumption that regional technology (i.e.,
input requirements of industries) is similar to that of the nation. Hence,
if the region could supply all of regional intermediate demand, the matrix
of direct coefficients A would be identical to that of the nation. In the
application of location of quotients to national tables, LQ.>1 is inter-
preted to mean that this situation is indeed the case; industry i
is more important within the region than in the nation, and so can provide
sufficient supply to all local intermediate demand. Hence, the inputs
required by all regional industries from industry i will be met, and
national technology (i.e., all A-matrix coefficients A., in row i) is
valid in depicting regional interindustry requirements^
On the other hand, an industry's location quotient LQ.<1 is taken to
mean that regional production of i is insufficient to 1 supply
regional demand for industry i's goods, and so only the fraction of the
national production requirements, a-. = LQ.A. . for all purchasing sectors
j, can be met within the region ^ 1 1J by industry i. The
Social and Economic Statistics Administration, Bureau of Economic Analysis,
U.S. Dept. of Commerce, Input-Output'Structure of the U.S. Economy: 1967,
v. 1-3, Washington, D.C.: U.S. GPO, 1974.(Magnetic Tape obtainea from
B.E.A.).
;
W. B. Beyers et al., Input-Output Tables for the Washington Economy, 1967,
Seattle, Wash.: Graduate School of Business Administration, University of
Washington, 1970.
9 U.S EPA Headciuarters Library
Mail coot ^MT
1200 PennstfWsmc Avenue NW
Washinoion. DC 2o4bU
202-566-0556
-------
remainder of the region's requirements must be made up by imports from the
nation. To summarize*
letting LQi = (N^/N^/dyN)
A.., if LQ.>_ 1
LQ:
then a.. =
I ij
'i
where
= national direct coefficient from industry i to industry j, a.. =
regional direct coefficient, LQ. = location quotient 1J
for producing industry i, N indicates a l measure of economic
importance, and superscript R indicates the region.
By adjusting the rows of the national direct coefficients matrix in
this manner, an estimate of the regional counterpart of the A-matrix table
can be derived with a minimum of information, namely only N^R and N.J for all
industry sectors.
Approach to Assessment of Location Quotient Method --
While methods of regionalizing national input-output tables in general,
and location quotient methods in particular, have been previously studied
with less than satisfactory results obtained, the potential was seen for
areas of further examination and possible improvement.
First, employment has traditionally been used as the location
quotient's measure of economic importance, due to the difficulty of
obtaining other economic data. In the present study, the use of total
gross output figures, an intuitively better measure of economic importance,
was compared with the use of industrial employment figures.
Second, the necessity of having both a national table and survey-based
regional table (with which to compare the eventual estimate) in compatible
forms (in terms of sector definitions, conventions, and aggregation) often
required a high degree of ^aggregation in order to reconcile the sectors of
the two tables. We began "with a national table in 367-sector detail, and
were therefore able to combine sectors to match those of the Washington
table as closely as possible while keeping the estimated table at a level
of 50 industrial sectors (compared to the original 52 of Washington). While
some important discrepancies between the national and Washington conventions
could not be removed, it was felt that the definitional closeness, homo-
geneity of sectors, and sectoring detail of the present study of location
quotients provides a substantial improvement of weaknesses of some previous
studies. Also, as is evident, a more detailed table is of greater use in
regional analysis.
Third, in testing the accuracy of the estimation of the true regional
table, no consistent measure has been used by all analysts. As a result,
their results are difficult to compare. We- decided to use as many of these
10
-------
tests as possible in order to note any improvement attributable to our
procedure.
Fourth, in order to gain some idea of the cost-effectiveness of using
this short-cut method to obtain a regional input-output matrix, it was
decided to test the estimated table against the "naive" alternative of
simply using the national table as a model of regional interindustry
relationships. Thus, a comparison can be made of three tiers of estimates
of the regional table: the naive model using no regional information (NAIVE);
one based on employment location quotients (EMPLQ) requiring only data on
employment by industry; and an estimate somewhat more demanding of data, one
based on output location quotients (OUTPLQ).
Estimation of the Washington State Input-Output Matrix --
As described previously, the 367-sector 1967 national table (Bureau of
Economic Analysis) and the survey-based 1967 Washington State table were
both aggregated to a compatible scheme of 50 sectors. The table of direct
coefficients was then developed from this 50-sector array by dividing each
interindustry flow by the purchasing sector's total gross output. At this
point, there were then the "true" survey-based Washington table (WASH), and
the national table which we also term the "naive" estimate of the Washington
economy.
Then, using employment data and output data, respectively, for both
the nation and Washington state by industry, location quotients were
computed and applied to the national direct coefficients matrix. The two
new tables derived, EMPLQ and OUTPLQ, were the basic location quotient-based
estimates of the Washington table. For each model, the A-matrix was also
used to calculate (I-A)-I. (Because the application of location quotients
to the national table can never cause the national coefficients to be
exceeded, strict multiplication across a row even if LQ-> 1 was examined
as an auxiliary test.)
Once A and (I-A) were developed for the three models, plus WASH,
their ability to predict total gross output with final demand F given was
tested, by multiplying (I-A)-1F = X, the familiar input-output relation-
ship.
While this derivation and application (and analysis) of location
quotient estimates of the regional direct coefficients matrix was the main
aim of our study, some additional implications of the estimates were
pursued. With each sector's total output known, along with final demand
and the intermediate transactions AX, exports can be computed as residuals
for each row. These were calculated. In addition, a "balancing" procedure
was performed (under two separate conventions) to preserve the consistency
of the table, adjusting for exports calculated as negative residuals (and in
one case not permitting any exports where LQ.<1, by strict interpretation
of the theory behind location quotients). This balancing caused
some adjustments in the A-matrices which were recalibrated.
11
-------
The results of all these models of the Washington State table were
compared with the values of the true (survey-based) Washington table. The
results and conclusions are described in the next section.
Analysis of Results --
As previously stated, our main concern in the test of the use of loca-
tion quotients to regionalize national input-output tables was to accurately
derive the matrix of direct coefficients which, given a vector of regional
final demands, can be used to calculate total gross output by industry:
(I-A)~'F = X. By such a procedure, regional outputs can be predicted by a
means other than allocating national production. The values of the estimated
and survey-based tables were compared using a variety of tests, and some
encouraging results were obtained.
First, the survey-based table was taken as the true depiction of the
1967 Washington economy. Each of the NAIVE, EMPLQ, and OUTPLQ direct
coefficients matrix estimates was compared with the Washington matrix (WASH)
and each other. As hypothesized, the NAIVE model proved the least accurate
estimate in all but one of the tests. EMPLQ showed substantial improvement
in estimation accuracy, with OUTPLQ exhibiting the smallest error of all.
Hence, our results indicated that the coefficients displayed a definfte
increase in accuracy when location quotients were applied to the national
table, particularly when output rather than employment was used. For
example, from NAIVE to EMPLQ to OUTPLQ, the mean percentage error5 decreased
from 314% to 162% to 135%. However, these raw figures taken by. themselves
still do not indicate very close estimation. In fact the correlation
coefficient ranged, respectively, from .649 to .717 to .733.
The real-life test of the estimated A-matrix is its predictive value,
obtaining gross outputs using A and the final demands vector: X = (I-A)-'F,
where (I-A)~1 is the matrix of direct and indirect coefficients. Again
significant improvement was evidenced in progressing from NAIVE to EMPLQ to
OUTPLQ.
First, in testing (I-A)~1, the statistics again showed some variation
between estimated and true coefficients. For example, mean absolute
difference, reduced by a factor of about two-thirds in moving from NAIVE to
OUTPLQ, measured .0056, on the same order as its value for the A-matrix
(.0043). The very high correlation coefficients .976, .991, and .993
are likely due to the extremes of coefficient values, concentrating either
at small fractions or greater than one on the matrix's diagonal, allowing a
least squares regression to explain much of the variation about the mean.
The ability of the estimated tables to predict actual total output
with all final demand given was quite impressive. Mean percentage error
(and standard deviation) decreased from 161.95% (239.28%) for NAIVE to only
32.79% (53.59%) for EMPLQ and 16.63% (17.17%) for OUTPLQ. All three
Defined in Chapter 3.
12
-------
estimates overstated total output and a least squares regression line fit
very closely, with correlations of .965, .988, and .994 with slopes
fractionally greater than one, possibly suggesting some type of bias.
Tests of estimated intermediate output AX were also made, resulting as
with the coefficient matrices in significant improvement from NAIVE to EMPLQ
to OUTPLQ, but with substantial error present. Exports calculated as
residuals were also highly correlated with the true values, although highly
understated, again a possible indication of some sort of bias. The
previously described balancing procedure to insure internal consistency,
while not providing drastic changes in the results, does in effect virtually
nullify the influence of location quotients in the regionalizing of the
national table and so will not be discussed here.
Conclusions and Summary --
In this description of our effort to examine the ability of two types
of location quotients to adjust national input-output tables to reflect
intraregional interindustry transactions, we have attempted to highlight
our efforts at improving the methods used in previous studies.
First, the input-output matrices estimated using location quotients
were shown to be a significant improvement over the naive assumption of
national interindustry relationships within the region.
Second, location quotients based on output were shown to provide better
adjustment factors for the national table than did those in which employment
was the measure of importance of an industry.
Third, the application of a large number of previously-used tests of
the accuracy of the estimated table allowed some degree of comparison to be
made with previous tests. In this way, for example, the effects of a large
number of sectors (and hence more homogeneous industry) and the use of out-
put as opposed to employment in computation of location quotients could be
examined. It was found in virtually all tests that output data provided
better results than employment. And, while in general the results of the
present study indicated improved estimation ability over previous studies,
contrary conclusions based on some comparisons or other tests may be
explained by the increased cumulative error in the summary comparative
statistics due simply to the increase in data points (number of sectors,
squared).
Finally, in the desire to estimate accurately regional outputs using a
short-cut method to regionalize national input-output tables, the most
important goal is to be able to accurately estimate regional outputs. Our
application of location quotients achieved quite good results using the
given Washington final demand vector.
By removing some handicaps suffered by other studies, and achieving
improved results, it is tempting to suggest that further improvements may be
attained if the problems we could not remove were eventually eliminated.
Specifically, definitional differences in sectors and conventions between
13
-------
Washington and national tables may account for additional error. The
transfers of secondary products and imports within the national table in .
particular could be expected to influence the effect of location quotient
adjustments. Further, the consistent overstatement of all three estimates
suggests some possible bias. It may be feasible to apply some sort of
adjustment factors to account for this bias. While these problems have not
yet been solved, the success achieved in the present study suggests that
location quotients can indeed be viable means of regionalizing national,
input-output tables.
When an individual state or region wants to explore the effects of
national economic trends, environmental standards or energy development on
the economy and environment of its state or region, the preparation of a
state or regional 1-0 table by the location quotient method may be warranted.
While the resulting 1-0 table may not accurately reflect a survey-based
regional table, it may be sufficiently close to permit approximate estimates
of the direct and indirect effects of exogeneous developments on the regional
economy. Thus the location quotient method may provide an additional tool
that allows a first order estimate of effects that an elaborate interindus-
try, interregional model (along with current and projected data bases)
could provide if it were available. In other words, location quotients
would be, in our view, an acceptable method for regional analysis from the
perspective of a regional or state administration, until such time as a
satisfactory regional model becomes available.
B. A Review of Alternative Regional and Interregional Models
We have reviewed a variety of existing regional models with the objec-
tive of determining which of these models can be incorporated, with or with-
out modification as the Regional Module of SEAS. Due to the complexity of
the problem, and the not too infrequent lack of reliable data, most regional
models are 'linked' in some fashion to a national model. Examples of these
models are the (1) Shift-Share, and (2) Location-Agglomeration Models,
(3) Regional and Interregional Input-Output Models, and (4) Regional Macro-
Econometric Models. Each of these models is designed to explain and project
regional economic activity. At the original time of their formulation, how-
ever, these models were not designed for the purpose of analyzing the
economic and environmental impact of alternative government energy policies.
Existing Regional Forecasting Models --
The existing regional forecasting models can be classified into four
relatively distinct categories and these are: 1) the OBERS "shift-share"
type model; 2) the Harris "Location-Agglomeration" type model; 3} the Klein-
Glickman type macro-regional econometric models; and 4) the Polenske MRIO
model. These categories are not, however, entirely distinct since some of
the models utilize a number of different techniques to conduct forecasting
analysis. For example, the Harris model and the OBERS projections combine
regression analysis with the regional shares, and 1-0 approaches to determine
the effect of alternative transportation systems and water resource develop-
ment on a regional economy, respectively. MRIO is strictly an Input-Output
technique developed to project regional and interregional economic activity,
14
-------
and the Glickman model is strictly an econometric type model using regres-
sion analysis to project economic activity in the Philadelphia region.
Energy and the Environment
None of the models, however, contains explicit formulations to project
the environmental impact and the effect on energy demand of alternative
energy scenarios. The Glickman model and MRIO are 'pure1 regional economic
impact models. The OBERS model tries to project the environmental impact of
water resource development on land use, and the Harris model introduces
environmental quality variables to explain regional variation in output.
Structure of the Models --
All of the models, except MRIO, consider the national economy as the
'driving force behind the regional economies.' The OBERS projections and
the Harris model project regional shares of national values - output, em-
ployment, income - into the future. Changing shares over time reflect the
dynamic character of the model and the shifting location of regional
industries. The Glickman regional model of the Philadelphia SMSA was
designed to be part of a satellite system with the national economy in the
center of the system. HRIO, on the other hand, was designed to estimate
each region's level of output, employment, income and expenditures, and
then obtain the national totals as the sum of all the regional values.
Geographical Unit of Analysis --
To project economic activity levels OBERS bases its analysis on the
BEA Economic Area. The Harris model has also been applied to the 173
Economic Areas, but was initially intended for the county subdivision. The
MRIO region is the state, a political unit. The Glickman model is based on
the Philadelphia SMSA. Of these modular regions, the MRIO state is probably
the most open since SMSA's extend beyond political boundaries. The BEA
economic areas are the most self-sufficient in the employment sense since
they usually contain labor's place of residence and place of employment in
the same area.
c- A Multi-Regional Model of theRegional Economy and Energy Demand
(MREED)'
In view of the significant drawbacks of the OBERS model currently in-
corporated in regional forecasting in SEAS, the need for an improved regional
forecasting tool in REGION became clear. A survey of existing regional
models -- regional and interregional input-output models, regional econo-
metric models, location-agglomeration models -- suggests their limited use-
fulness for the specific regional information for environmental and energy
policy analysis and the computational needs of SEAS.
In the SEAS policy context, it is important to understand the regional
economic and environmental effects of alternate scenarios. How do the
various regions fare in attracting different primary and manufacturing
industries? How will they be affected when different pollution abatement
policies are implemented? How do the increased costs of abatement, or
15
-------
changing energy prices affect regional comparative advantage? Which indus-
tries and regions are most likely to be sensitive to these changes? What
are the effects of all of this on regional income, consumption, local govern-
ment revenues and taxes? What are the likely consequences on unemployed in
the region? What are the likely consequences of energy price increases on
consumption of residential, industrial, and transportation energy in various
regions? What are the feedback effects of energy price increases on regional
growth? What are the likely consequences of large scale energy development
in one region (such as ORBES) on the economy and environment of that region
and on those of other regions? These are the kinds of questions that a
Region Module of SEAS needs to address and the models reviewed above cannot
satisfactorily address. Regional models designed for different objectives
do not serve SEAS' purposes.
As a direct consequence of the above finding, we have outlined a Model
of the Regional Economy and E_nergy Demand {MREED} whose structure and data
output are tailored to the regional information needs of environmental and
energy policy analysis. MREED incorporates the key elements of regional
economic theory and the more attractive features of existing regional models
in a framework of economic-environmental-energy analysis. Large portions of
HREED's structure have been previously estimated by the senior author and
the relevant data base is available. The proposed incorporation of HREED
in the REGION module will follow the generalSEASstrategy of improving"
existing models for incorporation into various modules of SEAS7
Chapter 5 outlines a proposal of continuation of the Johns Hopkins
University work on the REGION module directed at developing the core
components of MREED. It provides a description of the proposed full struc-
ture of HREED. T. R. Lakshmanan has developed and used at the U.S. EPA an
earlier regional model for the 100 largest SMSA's.6
While the model structure, data base, and research strategy are
detailed in Chapter 5 a few salient features of MREED may be highlighted.
1. MREED is a multi-industry, multi-regional model of the U.S.
that provides a description of the evolution of various
regional economies (BEA economic areas) in the national
economic setting. It is formulated as a regional module
of SEAS, so that the regional estimates obtained from the
model are normalized to the national totals obtainecT from
SEAT! MREED can be used Tn two ways: Firslf, it wi 11 help
allocate national (INFORUM) estimates of output by sector
to regions. Second, using observations over 10-15 years,
the model can be estimated also for each BEA area so that
each region's output by sector, income, consumption,
government expenditures, labor force, employment population,
6
T. R. Lakshmanan and Fu-Chen Lo. "A Regional Economic Model for the
Assessment of Effects of Air Pollution Abatement," Environment and
Planning. Vol. 4, 1972, pp. 73-97.
16
-------
and energy demand by major end use can be estimated. Thus,
the model could be used for assessing economic and environ-
mental and energy effects in the SEAS framework either for
one region such as ORBES Region or for allocating national
totals to regions. Since MREED will produce economic in-
formation at the BEA area level, allocation of some sort to
the county level is necessary to permit the flexibility in
regional area data in SEAS studies.
2. MREED is a 26-equation econometric model organized at the
level of 173 BEA areas -- each of which is a labor market
area. It is organized into four blocks output, income
and expenditures, labor market and energy. The equations
in those blocks incorporate principles of regional
location theory and interregional interaction theory in a
framework where environmental quality and energy prices
affect comparative regional advantage. Thus, output in
manufacturing and primary sectors is determined in national
markets (viewed multi-regionally) under the influence of
regional environmental quality and different regional
energy prices. Energy demand is in turn related to income,
industrial structure, and price. While the resulting
structure of the model iselaborate, it must be noted that
the greater portion of the model has been already estlmaled
(inthe form presented here or q'uite clp'se""to it) by the
principal inves'tigator and others (L^lcshmanan^ and Lo 1972,
Harris 1973V G1ickman 1974). The exterisio₯ p^this previous
work embodiedin MREED" consists of adjduig and incorporating
en v 1 ronrnen ta 1 ancT jene rgy var i abl es as factors af fecti jig
region 1ocation of industries and households and in identi;-
fying energy consequences of regional economic growth.
Thus the developmental work in" MREED is far morelimited
than the model structure described in the next section would
suggest.
3. The strategy of development of MREED should follow the
evolutionary approach and be organized in phases. The first
phase can be visualized as the development of only two
blocks manufacturing output, and energy -- at a sector
detail for which BEA area data are readily available. This
work will permit the assessment of the role of regional
environmental quality and energy input costs in industrial
location and the level of regional energy demand industrial,
residential, and transportation. The information developed
here will greatly augment the utility of the regional module
of SEAS. In the next phase, the remaining blocks can also
be estimated at the BEA region level.
17
-------
D. A Model to Assess Regional Economic Effects of Final Demand or Energy
Price Changes
Despite the limitations of input-output analysis and the resource
constraints involved in collecting national and regional data, the model
developed by Mlyazawa and generalized by Yamada and Ihara and discussed in
Chapter 6 will prove to be an invaluable instrument to analyze national,
regional, and interregional economic repercussions of alternative energy and
public expenditure policies.
This model can be used to analyze the effects on regional output,
income, private expenditures, and interregional trade flows of an exogenous
change in public expenditures in either of the two regions.
Furthermore, we can analyze the effect of an exogenous change in a
region's vector of prices. The impact will first be felt in that region's
level of income, and will then influence output, private expenditures and
interregional trade flows through the income multiplier process.
This model can also be used in an entirely different way. The two
regions can be viewed as two distinct industrial sectors industrial goods
and agricultural goods producing sectors, energy-intensive and non-energy
intensive sectors, etc. As with the analysis in the two region case, the
effect on output, income, prices in a given sector, and intersectoral flows
can be analyzed. By categorizing the industries in a national 1-0 table
into energy producing and non-energy producing industries, for example, we
can examine the degree of dependence (independence) of non-energy producing
industries on the energy sector. This can be done by examining how non-
energy output and intersectoral trade flows respond to an expansion in the
energy sector. Assuming, however, fixed production and trade coefficients,
one must realize the dependence of one sector on the other will be over-
stated.
Similarly, 'cost-push' inflation caused by increased energy prices can
be analyzed. Again by dividing the national 1-0 table of prices and wages
into those pertaining to energy and non-energy producing sectors the effect
of energy price increases on other prices can be determined. However, we
must again realize that by ignoring the price sensitivity of inputs, the
extent of 'cost-push' inflation caused by energy price increases will be
overstated.
18
-------
CHAPTER 2
AN ASSESSMENT OF THE DATA BASE OF REGION
This chapter presents the results of an assessment of the data base used
in the Regionalization (REGION) Module of the Strategic Environmental
Assessment System (SEAS) Model. It opens with a brief summary of the REGION
Model and its assumptions. We proceed then to a comparison of the state
output data developed by SEAS for 1972 with the state output data for the
same year developed by the U.S. Bureau of Census for 34 INFORUM sectors in
15 major states. We interpret the differences in regional output between
these two sources in terms of sector productivity variations at the state
level. From this empirical exercise, we draw some inferences for future
directions in strengthening the regional data base in REGION.
I. THE REGION. MODEL
The purpose of REGION is to allocate national economic output, abate-
ment costs and pollutant emissions developed by other SEAS modules (INFORUM,
ABATE, RESGEN) to various regional units. REGION does not incorporate an
analytical description of the structure of production and consumption in the
regions or the economic interactions among regions in the country. Instead,
it provides for a straightforward proportional allocation of a base year
national sector output to regions based on the letter's proportional share
of national employment in that sector.
(1)
where X and L represent output and employment in industrial sector i, j is
the regional unit and t is the time period.
The use of the more readily available regional data on employment in
Equation (.1) implies an assumption of equal output per worker in a sector
in all regions. (In some instances, plant capacity data are substituted for
employment in Equation (1).)
For a base year, the REGION model develops 18 sets of regional fractions
termed SHARE ( = L.../L in Equation (1)) to be applied to the national
estimates to J ^ determine regional output and pollutant
emissions. Nine of these SHAREs are intended to regionalize the output and
19
-------
abatement costs to any of the following regional levels:
1. EPA Federal Regions
2. States
3. Standard Metropolitan Statistical Areas (SMSA)
4. Air Quality Control Regions (AQCR)
5. Major River Basins
6. Minor River Basins
7. Aggregated Subareas {ASA]
8. Bureau of Economic Analysis Regions (BEA)
9. Non SMSA BEA Regions
(10)
(51)
(254)
(247)
(19)
(221)
(101)
(173)
For each of the above, E-SHARE is computed as follows:
E-SHARE (t) = the proportion of output and abatement
ij activity associated with sector i in
regional unit j in year t.
The other 9 SHAREs are for regionalizing pollutant emissions to any
of the above 9 regional units and are computed as follows:
P-SHARE (t) =
ij
the proportion of pollutant emissions
derived from sector i in region j in
year 6.
The definition of P-SHARE in this manner implies that the national
shares of all pollutant types associated with a sector in any one region
are the same. (This is modified in selected sectors, however.)
The E-SHARE and P-SHARE files are developed for a base year (1971)
using employment data (.as indicated earlier, for a few sectors plant
capacity data are used) at the county level and aggregated to the desired
regional level. Thus
"n
m
SHARE (1971) =
ij
c=l
L1C(1971),
L1C(1971)
(2)
where c refers to a county and there are m counties that make up a regional
unit j and n counties make up the nation.
Economic Information System (EIS) is the source of REGION'S base year
data on L. , ^. EIS is a private company deriving its data for 115,000
lc(k ' industrial establishments by name and address, state and
county, and by 4 digit SIC. Since these data are drawn from public sources,
there are no disclosure problems. The data base, also known as the "Gould
Tape" (after the President of EIS) is considered a reasonably complete
survey of all plants in the continental U.S. with more than 20 employees.
EIS claims completeness of data for plants with more than 250 employees,
98% accuracy in the range of 50-250 employees and 85% to 98% for plants
with 20-50 employees. It does not cover plants with less than 20 employees.
20
-------
The SHARE values are obtained for future years by utilizing the 1972
Office of Business Economics (OBERS), Projections of Economic Activity in
the U.S. These projections of earnings put out by the U.S. Department of
Commerce at the 2 digit SIC are used to compute an OBERS growth factor
(OG.,.). The latter is defined as follows:
I *J
OG (t) = OBERS (t) /OBERS (1971)
(3)
where OG. . is the OBERS growth factor associated with the 2 digit SIC
sector 1 »J i1 in regional unit j. OBERS.. _. is defined as the industry
earnings associated with i1 in region j.
1.J
For each INFORUM sector included in the OBERS sector i', the projection
year (t) SHARE values are computed:
SHARE (t) = SHARE (1971) x OG
ij ij 1.j(t)
(4)
The SHARE values in Equation (4) are normalized over all regions in sector
i to sum to 1. Thus the growth patterns in each of the 3 or 4 digit INFORUM
sectors are constrained to that of the 2 digit OBERS sector they are
contained in. OBERS makes projections for 37 economic sectors compared to
INFORUH's 185.
The allocation procedures used in REGION have a number of drawbacks.
The use of employment as a surrogate for output in REGION is acknowledged as
introducing errors whenever output per worker levels vary significantly
among regions, but was judged in the design phase of SEAS as not producing
unreasonably great inaccuracies. Output per worker within a sector does
vary considerably in response to a number of factors, such as process or
product mix. In an industry with high capital investment and relative
"stability" in the markets for the product, older plants are likely to
continue producing in the face of competition from newer plants using
different processes and capital/labor ratios. While each plant produces
essentially the same type of product, their production processes differ. If
that sector has experienced regional location shifts (such as textile plants
moving from the Northeast to the South) differences in plant age, and
processes may be highly correlated with specific regions. Another problem
is that the type of products classified under a single INFORUM sector (and
indeed under any classification scheme with a reasonably limited number of
sectors) may differ widely. "Apparel" produced in New York may involve
expensive fashionable clothing with relatively high value added compared to
the work clothing produced in Wisconsin. Apparel in Texas may be relatively
labor intensive, taking advantage of the availability of less expensive
labor. The point made here is that output per worker is sufficiently varied
for a number of sectors that the number of employees is a poor indicator
of levels of output.
21
-------
Second, the OBERS method of projecting regional growth to be built
around a "shift share11 model which extrapolates past locational patterns.
(5)
where LS = locational shift of industry T in region j between time periods
tj o and t.
E and E are levels of economic activity in sector i, region j at
ijt ijo time periods t and o.
E and E are national levels of economic activity in sector i at time
it io periods t and o.
OBERS trends LS into the future with the result that future locational
shifts mirror ij only past ones and do not take into account the
effects of dynamic factors on changes in regional comparative advantage.
Regional pollution emissions are obtained in SEAS on the assumption that
a region has the same share of national emissions as it has of the national
output. This assumption ignores differences in process mix, industry mix
and environmental standards, plant age, etc., among regions, but still
serves as a useful first approximation.
The objective of the study reported in this chapter is a limited one.
It is not to improve the workings of REGION model structure, but to suggest
improvements in the economic data base, so that regional shares (and in turn
pollutant estimates) can be made more reliable. For this purpose, the
comparison of a base year SEAS regional data base with an independent
regional data base is made in the next section.
II. COMPARISON OF REGIONAL DATA BASES
Regional estimates of economic output for 34 INFORUM sectors were
compared for a sample of 15 large states with data from the 1972 Census of
Manufactures published by the Bureau of the Census. The Census of
Manufactures involves a complete and detailed survey of all manufacturing
establishments with 20 or more employees, and less detailed data on plants
with less than 20 employees. It provides data down to the 5-digit Standard
Industrial Classification (SIC) code for individual states, SMSA's and
other regions, including counties.
However, the CM data is considered confidential and so disclosure
limitations preclude the publication of data which would reveal specific
information about an identifiable plant. Obviously, at the level of the
county, disclosure problems become more significant. Because the most
complete sector detail is published at the state level, the present study
chose this level as the basis of comparison with SEAS. The CM data
22
-------
gathering effort is extensive and the publications include data on a wide
variety of industry related subjects. The comprehensiveness of the effort
also means that the data is made available several years after the period of
interest (1972 data was published in book form in 1976). Of the 121 INFORUM
sectors (Numbers 20-150) for which CM data are available, 34 sectors were
chosen for this comparison.
A directory of 1972 SIC sectors comparable to given INFORUM sectors
was obtained from the U.S. Environmental Protection Agency. The Value of
Shipments for each SIC sector in a given state included in an INFORUM
sector was compared with the value of output provided in a SEAS Base Case
Run.
It should be noted that the use of the Census of Manufactures data as
"ground truth" may be open to some question. As in any large scale data
gathering effort, sampling and aggregation errors are probably inevitable.
The Census of Manufactures does, however, have the advantage of a long
period of data gathering and preparation and of being the most recent effort
of a data series which has been produced for a number of years. In
contrast, tfie EIS Gould Tape was used much sooner after the period under
study, and the 1971 data was one of the early ones undertaken by EIS.
Further, the Gould Tape aggregates from the county level, a relatively small
geographical area, which increases problems of randomness in the data and
the possibilities of aggregation errors particularly due to incomplete
coverage of smaller firms. Such factors certainly do not ensure the
greater accuracy of the Census of Manufactures, but are in its favor as a
benchmark. Thus while the text will refer to the "accuracy" of the SEAS
data, such references are premised on the validity of the CM figures.
What follows is a description of the procedures used to determine how
closely SEAS estimates correspond with data from the Census of Manufactures;
a description of the degree of correspondence by sector and state; an
analysis of output per worker differences as an explanatory factor for
differences and the development of a modifier based on the output per worker
variations which could be determined; and an appendix listing the specific
information used in each sector.
Several discrepancies between SEAS and CM conventions required adjust-
ment before a comparative analysis could be made. SEAS INFORUM projections
are classified according to a "purified" sectoring scheme -- a sector's
outputs consist only of that product. However, CM data on a state level are
only available classified at an establishment level -- all outputs of an
establishment are assigned to a sector if its production of that sector's
products is greater than that of any other sector's products. National
CM data are available on a "purified" basis,' however, so a "purification
correction factor" the ratio of Purified National CM Value of Shioments
Census of Manufactures. Industry Series, Tables 5-B. Purified Values of
Shipments are obtained from addition of an industry's primary products and
miscellaneous receipts.
23
-------
to Unpurified National CM Value of Shipments was determined for each INFORUM
sector (Table 2-1). This ratio was then used to modify the available state
CM data to correct for the lack of purified figures. (This adjustment
implicitly assumes the national ratio of "purified" to "unpurified" pro-
duction in an industry holds in each state. However, it was seen as the
only means of making the adjustment with the data available.)
It would have been preferable if the SEAS base year (1971) coincided
with the data published for every fifth year (1967, 1972, 1977) by the CM.
Thus it was necessary to compare a projection year estimate (1972) by SEAS
with the CM data. First, as SEAS output is reported in 1971 dollars, adjust-
ments for inflation were made by multiplication of SEAS projections by
appropriate price indices (Table 2-2). Secondly, since the purpose of this
research effort Is to validate base year data rather than to check the
accuracy of projections, an attempt was made to reduce errors due to
"inaccurate" projection. Within SEAS, the 1971 base year totals agree with
known levels and are used as controls for the regionalization (i.e., the
sum of all the regional outputs in a given sector must equal the known
national total). The regional output data of SEAS were summed for each
sector in 1972 and were compared with the totals published in the CM. The
SEAS state-level estimates were then modified by national sector scaling
factors which reflect the differences between the SEAS-estimated sector
growth and the CM sector growth in the nation for each sector (Table 2-3).
Of course, REGION uses regional differential growth rates from OBERS and so
a national correction factor would not completely correct for the projection
inaccuracies in each region. However, since only a projection of one year
is involved, such considerations are believed to be very minor.
The 34 sheets presented at the end of this chapter display the
comparisons between CM state output data and SEAS 1972 data modified for
national sector total differences in 1972 between CM and SEAS. In addition,
the tables indicate the percentage of national output accounted for by each
state in the sample, and the national and state output per worker for each
of the 34 sectors.
Table 2-4 summarizes, for 5 major states, the SEAS-CM comparisons
for 34 sectors displayed in these Appendix tables. Substantial variation
between the estimated (SEAS) and true (CM) values can be seen. Mean
statewide SEAS/CM ratios range from .95 for California to 1.31 for.: Texas.
Variation about the means is large as well; standard deviations for the five
states are .20 (Illinois), .23 (California), .30 (Ohio),,,37 (New York, and
.63 (Texas) over the 34 ratios.
III. OUTPUT PER WORKER MODIFIER
Output per worker (0/W) levels were examined to determine if differences
in such levels could explain all or part of the observed SEAS/CM differences.
The 0/W was calculated from the CM data for each sample state for each
of the 34 INFORUM sectors. These values were then compared with national
24
-------
TABLE 2-1. PURIFICATION OF CENSUS OF MANUFACTURES VALUES OF SHIPMENTS
(1)
Census of Manufactures
(2)
Purified Value
Purification
INFORUM
Sector
23
24
25
26
27
31
39
40
45
51
52
53
54
59
66
67
69
76
81
84
89
94
95
97
98
101
103
107
113
118
119
123
127
128
Value of Shipments
($ millions)
31,447.9
16,311.5
11,478.6
12,162.2
7,895.5
6,925.8
22,709.7
5,099.5
7,409.6
8,104.7
8,262.8
3,510.6
3,794.5
2,778.3
8,018.5
9,777.6
26,761.8
3,478.7
9,537.6
8,432.4
1,244.6
2,136.8
14,179.6
8,318.3
5,094.3
4,288.3
5,590.4
696.3
1,813.1
1,540.8
3,585.4
6,938.6
13,665.0
8,836.6
of Shipments
($ millions)
38,454.7
15,998.2
11,295.2
13,011.4
7,943.8
7,046.1
24,500.4
5,887.8
7,385.0
8,148.1
8,200.0
3,584.7
4,140.4
2,909.8
7,830.4
9,998.4
28,163.4
3,486.1
9,398.0
12,734.6a
1,197.7
2,043.3
14,525.4
8,351.1
5,111.2
b
N.A.
5,501.7
709.7
1,842.4
1,408.6
3,605.2
6,973.6C
13,353.6
9,038.0
Correction
Factor
1.22
.98
.98
1.07
1.01
1.02
1.08
1.15
1.00
1.01
.99
1.02
1.09
1.05
.98
1.02
1.05
1.00
.99
1.51
.96
.96
1.02
1.00
1.00
(1.00)
.98
1.02
1.02
.91
1.01
1.01
.98
1.02
^Extensive duplication in shipments.
bNA=Not available.
cSmall undisclosed miscellaneous receipts entry not included.
25
-------
TABLE 2-2. 1972 PRICE INDICES FOR INFORUM SECTORS (1971 = 1.00)
INFORUM
Sector
23
24
25
26
27
31
39
40
45
51
52
53
54
59
66
67
69
76
81
84
89
94
95
97
98
101
103
107
113
118
119
123
127
128
Wholesale Prices and Price
Price
Index
1.12
1.03
1.05
1.04
1.03
1.02
1.02
1.05
1.02
1.03
1.02
1.05
1.02
1.01
1.01
1.00
1.02
1.07
1.05
1.04
1.02
1.03
1.04
1.05
1.03
1.05
1.04
1.02
1.02
1.01
.99
1.00
1.00
1.00
Indexes (_1971 , 1972), Bureau (
Labor Statistics.
Handbook of Labor Statistics 0974), Bureau of Labor Statistics
26
-------
TABLE 2-3. SELECTED NATIONAL SECTOR OUTPUTS FOR 1972
SEAS
SECTOR #
23
24
25
26
27
31
39
40
45
51
52
53
54
59
66
67
69
76
81
84
89
94
95
97
98
101
103
107
113
118
119
123
127
128
PURIFIED
C.M.
($1972)
38,454.7
15,998.2
11,295.2
13,011.4
7,943.8
7,046.1
24,500.4
5,887.8
7,385.0
8,148.1
8,200.0
3,584.7
4,140.4
2,909.8
7,830.4
9,998.4
28,163.4
3,486.1
9,398.0
12,734.6a
1,197.7
2,043.3
14,525.4
8,351.1
5,111.2
4,288.3b
5,501.7
709.7
1,842.4
1,408.6
3,605.2
6,973.6
13,353.6
9,038.0
SEAS
($1972)
25,557.9
16,095.6
12,610.5
9,974.4
7,861.0
6,362.9
23,558.1
4,712.2
7,565.4
7,820.3
7,982.4
4,563.6
3,641.0
1,900.4
7,642.3
9,283.6
27,975.3
4,015.8
9,036.6
6,342.5
978.1
1,959.1
14,105.0
7,741.3
5,206.2
5,052.0
4,460.8
772.3
2,001.3
1,308.3
3,613.6
7,611.9
15,372.5
9,050.6
CM _ Scaling
SEAS Factor
1.50
.99
.90
1.30
1.01
1.11
1.04
1.25
.98
1.04
1.03
.79
1.14
1.53
1.02
1.08
1.01
.87
1.04
2.01
1.22
1.04
1.03
1.08
.98
.85
1.23
".92
.92
1.08
1.00
.92
.87
1.00
j*See notes in Table 2-1.
Value of shipments not purified due to unavailability of data
27
-------
TABLE 2-4. SEAS/CM FRACTIONS FOR SELECTED STATES
SEC-
TOR NY OHIO ILL CAL TEX
23
24
25
26
27
31
39
40
45
51
52
S3
54
59
66
67
69
2.03 1.77 1.34 .98 1.26
.57 1.12 .60 .74 2.49
.74 .68 .98 .86 1.23
.97 .90 1.23 .78 1.38
.94 .93 .87 .92 1.45
.74 1.18 1.13 .66 D
.43 D 1.31 D D
1.13 .61 1.03 1.20 1.S1
1.42 .79 1.33 .80 1.02
1.17 .98 1.15 .75 .85
.98 1.07 1.11 .90 1.08
.62 1.72 1.09 .61 D
b
.79 1.06 .99 1.16 2.11
1.21 D D D .61
.62 1.06 1.32 1.28 1.5&
1.71 1.29 .77 1.39 1.18
D .61 1.41 .93 1.12
SEC-
TOR NY OHIO ILL CAL TEX
76
81
84
89
94
95
97
98
101
103
107
113
118
119
123
127
128
1.00 D .77 .73 ,66b
D 1.16 1.23 .85 .98
.83 1.67 1.10 0 D
.45 1.27 ODD
.96 .99 1.20 D D
1.20 .89 1.19 .95 1.14
0 .88 1.08 1.51 3.02
1.04 .84 1.00 .90 2.41b
D D .83 D D
b b
1.09 1.02 1.16 D 1.77
1.08 .70 1.03 D .72*
.90 D 1.3S D l.lP
1.03 1.17 .96 D .69
1.56 .95 1.31 1.00 .90
D 1.38 1.03 .75 .55
1.06 1.13 .98 .99 D
.50 .93 1.29 1.16 D
aSEAS is modified by the price index and scaling factor; CM is modified by the purification
correction factor. No modification has been made for 0/W differences.
6State accounts for less than 17. of total sector output.
0 = Disclosure limitation
28
-------
0/W values for each sector. < If the differences between SEAS and CM are to
be explained by differences in the 0/W among the states, an 0/W for a state
which is higher than the national average for that sector should lead SEAS
to proportionally underestimate production in that state, and a lower than
national average 0/W for a state should lead SEAS to proportionally over-
estimate production. The ratio of state to national 0/W was used to adjust
the SEAS/CM fraction to test this hypothesis; the adjusted SEAS estimates
would then approach those of CM (and SEAS/CM would approach 1.00). For the
most part, the trends in the state 0/W and the corresponding differences in
the SEAS/CM fraction tended to follow the expected path. However, there
were frequent cases where 0/W failed to explain the full difference in the
SEAS/CM.
Although the differences in output per worker among the sample states
does not always correct for the differences in the SEAS/CM figures, it still
provides a very useful explanation for much of the observed differences.
Besides OBERS growth, a second modifier within SEAS based on the variation
in output per worker would serve to bring most of the SEAS/CM figures more
closely in line. Table 2-5 demonstrates the effect of this second modifier
for sample states in 3 sectors. The sectors were chosen because of the
relatively wide variation on 0/W levels within each sector. States with
disclosure problems or which accounted for ]% or less of sector output are
not shown. The ratio of state 0/W and national 0/W is multiplied by the
SEAS/CM fraction. If the state has less than a national average output per
worker level, this modifier will be less than one and so the SEAS/CM
fraction will be deflated. Although some SEAS/CM fractions actually move
further from 1 (the standard of perfect agreement), most of the fractions
moved closer to 1 when multiplied by the 0/W modifier, and a number showed
significant improvement. For example, the difference between upper and lower
values for sector 23 narrowed from 1.4 to 0.9, with standard deviation of
.44 around a mean of 1.24 dropping to .32 about a mean of 1.22.. In sector
26, the mean (and standard deviation) improved from .955 (.27) to 9.24 (.18).
For sector 65, the change was from 1.04 (.31) to 1.01 (.21).
If state level output data were available without disclosure problems
it would be relatively simple to incorporate the differences in output per
worker into the regional allocation factors. Given the disclosure limita-
tions it is not possible to correct for 0/W differences in all states in
all sectors. One issue then is whether to use available data on 0/W
differences or not to use it because it is not complete. One possibility
is to perform a two-part allocation. In the first part states would
receive their share of a sector's output based on employment modified by
state level output per worker. The sum of output of all states for which
the 0/W was available could then be subtracted from the national control
"0/W fractions based on unpurified values of shipments were used, but both
state and national output are purified by multiplication by the same
purification correction factor for a given sector, so for comparative
purposes, no conversion was required.
29
-------
total, and the remaining states would receive allocations from the remaining
national output. On the basis of the work performed for this project, it is
estimated that 0/W levels could be developed for between one half and two
thirds of all states in each of the 121 SEAS industrial sectors.
Of course, the calculation of state level output per worker levels would
only be useful if the differences among the states were relatively stable
over the projection period. It would seem that for a period of up to a
decade the regional differentials in 0/W may not vary to a significant
extent. If the output per worker differences are due to differences in
product and process mix, the forces which affect this mix would tend to act
over relatively long time periods. This assumption needs to be verified by
an analysis of past Census of Manufactures data before this approach can
be seriously advanced.
IV. IMPLICATIONS
It appears that the SEAS output data at the state level obtained
using the Gould tape display considerable differences with CM data for
quite a few sectors and states. Apparel (#39) and electronic components
(#128) are examples of INFORUM sectors in which the output estimates for
a few major producing states are widely divergent from the CM data.
Some of these differences are attributable to the significant
departures of output per worker in those states from the national average.
Corrections for such differences bring the SEAS and CM estimates closer.
The remaining differences, assuming the veracity of CM data, may be attri-
butable to the fact that SEAS aggregates state data from county level
data. The "noise" at the fine geographical detail of a county is another
source of potential differences.
There is an obvious trade-off between flexibility and accuracy in the
choice of the county as the basic building block. The use of the county
as the basic unit permits flexible assembly of regional data in any one
of 9 different regional units. However, data bases and modelling of
locational and interregional relationships are more tractable at a larger
regional level such as a state or BEA economic area.
Consequently, it may be necessary to explore an approach complementary
to the current regionalization work in SEAS. What we would like to consider
is an assembly of a base year economic output data base at the state level
for the INFORUM sector. Such a calibration of base year regional data base
will utilize the best available output data. The base year could be a
census year such as 1972 and CM data can be supplemented with data for
other sectors. Where output data are not available because of disclosure
problems, other sources such as EIS can be used in a supplementary fashion.
Such a careful inventory of economic data base will increase the credibility
and usefulness of SEAS regional information.
30
-------
TABLE 2-5. EFFECT OF 0/W MODIFIER ON SEAS/CM FRACTIONS FOR
SELECTED STATES AND SELECTED SECTORS9
SECTOR
STATE
SEAS
CM
STATE 0/W
NATL. 0/W
#23
NY
NJ
PA
MICH
OH
ILL
WIS
NC
TEX
COL
CAL
.03
.89
.47
.59
.77
.34
.41
.24
.26
.66
.98
1.57
.85
1.33
.79
1.58
1.44
1.47
.67
1.17
1.34
1.18
#26
#66
NV
PA
MICH
OH
ILL
WIS
NC
ALA
TEX
CAL
NY
NJ
PA
MICH
ILL
CAL
.97
.86
1.31
.90
1.23
.50
.78
.84
1.38
.78
,62
.96
.75
.29
.32
.05
84
.02b
02b
62
OOb
12
1.28
1.31
.91
.73
.04
,05b
.36
These 3 sectors were chosen as illustrations because 0/W among
states varies considerably in each. The number of states shown
vary from sector to sector because of disclosure problems. Also
states with 1% or less of sector output are not shown.
Significant Improvement in SEAS/CM.
31
U.S EPA Headquarters Library
Mail code 3404T
1200 Pennsy'vani? Avenue NW
Washington. 1C 20460
202-566-0556
-------
V. APPENDIX DETAILED TABLES FOR EACH OF THE 34 SECTORS
The following tables list for each of the 34 INFORUM sectors examined,
individual state figures for 1972 output according to the Census gf
Manufactures, those projected by SEAS, and their ratios. Output per worker
and percentage of national output are also listed, as are national summary
figures and adjustment factors (described in the text) for each sector.
The following terms are used in the tables. The national purification
correction factor was used to adjust the Census of Manufactures data to
conform to INFORUM's sector definitions. The scaling factor corrects for
SEAS' projection from 1971 to 1972. The 1972 price index is based on
1971 = 1.00. The output per worker calculations were based on unpurified
sectors. The purified CM column was calculated by multiplying state
Census of Manufactures data by the national purification correction factor.
The modified SEAS column was obtained by multiplying the state SEAS figure
by the price index (to adjust for SEAS' 1971 dollars) and the scaling factor
which accounts for projection error. The appearance of a D in the tables
indicates the data was not listed in the Census due to disclosure limitations
while NL indicates no listing.
32
-------
TABLE 2-6. SECTOR #23 MEAT PRODUCTS
Equivalent S.I.C. Sector(s): 201
Census, of Manuf. (CM) Nat'l Purified Value of Shipments ($xlQCQ; 38,454.7
National Sector Output per Worker ($x!03): 102.2
Nat'l Purif. Corr. Factor:1.22 Scaling Factor: 1.50 1972 Price Index:1.12
STATE SEAS(modified)
CM(purified)
Purified
CM Amt.
($x!06)
Modified
SEAS Amt.
($x!06)
% of Nat'l
CM Output
State 0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.55
2.03
.89
1.47
.59
1.77
1.34
1.41
1.24
1.44
1.66
1.26
.61
.66
.98
405
947
622
1,514
1,170
1,299
2,206
1,248
736
479
575
1,974
329
1,549
2,339
626.3
1,918.6
555.4
2,219.3
696.0
2,305.0
2,950.1
1,765.7
914.4
691,0
952.0
2,489.8
200.4
1,023.1
2,299.9
1
2
2
4
3
4
6
3
2
1
1
5
1
4
6
87.4
79.2
98.1
92.6
137.0
91.0
109.6
106.6
54.9
78.6
50.1
94.6
207.7
208.2
122.9
TABLE 2-7. SECTOR #24 DAIRY PRODUCTS
Equivalent S.I.C. Sector (s) : 202
Census of Manuf. (CM) Nat'l Purif'ied Value of Shipments
National Sector Output per Worker ($x!03) : 86.5
Nat'l Purif. Corr. Factor: .93 Scaling Factor: .99
STATE SEAS (modified) Purified Modified %
($x!06): 15.998.2
1972 Price
of Nat'l
Index:
State
1.03
0/W
CM (purified) CM Amt. SEAS Amt. CM Output
($x!06) ($x!06)
Mass .
N.Y.
N.J.
Pa.
Mich.
.Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.53
.57
.83
1.63
.76
1.12
.60
.42
1.21
1.14
2.05
2.49
.84
1.23
.74
325
1,157
267
943
570
857
907
2,059
224
347
168
455
90
114
1,290
497.3
655.1
222.5
1,533.7
434.6
962.1
543.0
865.3
270.4
397.1
344.6
1,130.9
75.8
140.0
959.4
i
7
2
6
4
5
6
13
1
2
1
3
1
1
8
87.4
93.0
90.7
66.3
93.9
59.5
95.5
137.3
71.6
73.7
48.9
62.7
92.0
72.5
105.3
33
-------
TABLE 2-8. SECTOR #25 CANNED AND FROZEN- FOODS
Equivalent S.I.C. Sector(s): 203
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlQQ): 11.295.2
National Sector Output per Worker ($x!03); 49.2
Nat'l Purif. Corr. Factor: .98 Scaling Factor: .90 1972 Price Index: 1.05
STATE SEAS(modified)
CM(purified)
Purified
CM Arat.
($x!06)
Modified
SEAS Amt.
($x!06)
% of Nat' 1
CM Output
State 0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
2.41
.74
1.01
.94
.96
.68
.98
.88
1.28
.85
1.07
1.23
--
2.24
.86
71
779
527
700
413
517
539
427
85
917
60
340
D
26
2,504
171.1
579.3
530.4
659.8
397.3
353.7
530.7
374.7
109.2
777.7
64.2
418.9
58.3
2,156.5
1
7
5
6
4
5
5
4
1
8
1
3
<1
22
~51.4
61.6
53.8
52.1
50.7
71.3
52.9
41.5
51.9
76.7
43.6
50.3
27.0
51. 1
TABLE 2-9. SECTOR #26 GRAIN MILL PRODUCTS
Equivalent S.I.C. Sector (s): 204
Census of Manuf. (CM) Nat'l Purified Value of Shipments
National Sector Output per Worker ($x!03) : 109.3
Nat'l Purif. Corr. Factor: 1.07 Scaling Factor: 1.30
($xlO&): 13,011.4
1972 Price Index:!. 04
STATE SEAS (modified) Purified Modified % of Nat'l State 0/W
CM (purified) CM Amt. SEAS Amt. CM Output
($x!06) ($x!06)
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
2.20
.97
1.48
.86
1.31
.90
1.23
.50
.78
1.29
.84
1.38
1.36
5.68
.78
48
584
83
471
616
547
1,464
289
330
107
281
844
35
111
1,056
104.1
565.2
122.8
407.0
804.8
490.6
1,796.8
143.9
256.3
138.4
235.7
1,163.8
47.6
630.7
824.3
<1
5
1
4
5
4
11
2
3
1
2
6
<1
1 j
8
75.0
118.7
78.0
107.3
84.7
121.7
90.6
135.0
140.0
83.3
146.1
103.8
110.0
80.0
135.2
34
-------
TABLE 2-10. SECTOR #27 BAKERY PRODUCTS
Equivalent S.I.C. Sector(s): 205
Census of Manuf. (CM) Nat'l Purified Value of Shipments (flxlQfr): 27794378
National Sector Output per Worker ($x!03) : 3J5.J
Nat'l Purif. Corr. Factor: 1.01 Scaling Factor: 1.01 1972 Price Index:1.05
STATE SEAS(modified)
CM(purified)
Purified
CM Amt.
($x!06)
Modified
SEAS Amt.
($x!06)'
% of Nat'l
CM Output
State 0/W
Mass .
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
.88
.94
.65
1.14
.98
.93
.87
, 1.48
1.11
.91
.78
1.45
.84
1.07
.92
232
559
485
652
307
521
715
129
270
186
97
322 H
39
117
750
204.8
525.4
313.1
740.2
299.7
486.4
621.7
191.5
300.9
169.2
75.6
467.0
32.8
125.6
688.0
3
7
6
8
4
6
9
2
3
2
1
4
<1
1
9
33.8
30.7
40.7
33.3
34.9
34.9
38.9
27.8
27.2
29.7
34.3
30.7
32.5
36.2
35.7
TABLE 2-11. SECTOR #31 SOFT DRINKS AND FLAVORINGS
Equivalent S.I.C. Sector(s) : 2086,2087
Census of Manuf. (CM) Nat'l Purified Value of Shipments
National Sector Output per Worker ($x!03) : 52.75
Nat'l Purif. Corr. Factor: 1,02 Scaling Factor: 1.11
($xlOO): 7.046.1
1972 Price Index:
STATE SEAS (modified) Purified Modified % of Nat'l State
CM(purified) CM Amt. SEAS Amt. CM Output
($x!06) ($xlQ6)
1.02
0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
.74
.50
1.21
.72
1.18
1.13
l_ 1.38
1.00
1.05
.66
D
626
346
L 300
256
363
277
100
D
240
D
D
50
D
609
463.1
173.1
362.4
184.9
427.4
314.3
137.6
240.5
52.5
403.3
9
5
4
4
5
4
1 ,
3
1
9
73.1
84.7
49.0
59.8
48.8
34.0
40.8
53.4
40.8
74.6
35
-------
TABLE 2-12. SECTOR #39 APPAREL
Equivalent S.I.C. Sector(s); 251 through 238
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO°):24,S00.4_~
National Sector Output per Worker ($x!03): 19.2
Nat'l Purif. Corr. Factor: 1.08 Scaling Factor: 1.04 1972 Price Index:1.02
STATE SEAS(modified) Purified Modified % of Nat'l State 0/W
CM(purified) CM Amt. SEAS Amt. CM Output
($x!06)
($x!06)
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
.43
1.04
1.31
1.31
D
7,454
1,152
2,169
D
D
450
D
D
D
D
D
D
D
D
3,236.5
1,194.4
2,856.3
589
30
5
9
2
31.8
18.3
13.7
19.0
TABLE 2-13. SECTOR #40 HOUSEHOLD TEXTILES
Equivalent S.I.C. Sector(s): 239
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO&); 5,887.8
National Sector Output per Worker ($x!03): 27.4
Nat'l Purif. Corr. Factor:1.15 Scaling Factor: 1.25 1972 Price Index: 1.05
STATE SEAS(modified)
CM(purified)
Purified
CM Amt.
($x!06)
Modified
SEAS Amt.
($x!06)
% of Nat'l
CM Output
State 0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.L.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.31
1.13
1.33
2.11
.47
.61
1.03
.96
.55
1.11
.46
1.51
.08
3.42
1.20
219
926
311
198
894
214
281
48
595
54
135
138
13
13
368
286.4
1,051.0
412.6
418.1
418.3
131.0
290.5
46.0
325.6
60.0
h 62.5
i 208.8
1.0
44.5
441.1
4
16
5
3
.. - 15
4
5
<1
10
<1
2
2
<1
<1
6
22.3
26.1
23.9
17.9
48.6
31.0
28.4
22.1
32.1
15.2
37.7
22.2
15.7
18.3
21.2
36
-------
TABLE 2-14. SECTOR #45 HOUSEHOLD FURNITURE
Equivalent S'.I.C. Sector(s): 251
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO°): 7,58570
National Sector Output per Worker ($x!0^): 25.55
Nat'l Purif. Corr. Factor: 1.00 Scaling Factor: .98 1972 Price Index;1.02
STATE SEAS(modified)
CM(purified)
Purified
CM Amt.
($x!06)
Modified
SEAS Amt.
($x!06)
% of Nat'l
CM Output
State 0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.39
1.42
1.46
1.53
1.57
.79
1.33
.97
.84
1.12
.79
1.02
.61
1.40
.80
172
320
144
300
143
232
336
97
1,374
140
109
247
23
28
752
238.5
455.2
210.3
398.9
224.6
183.0
446.6
94.6
1,157.5
.157.5
86.7
251.9
13.9
39.4
600.7
2
4
2
4
2
3
4
1
18
2
1
3
<1
<1
10
23.9
22.9
29.4
23.6
22.3
28.6
28.5
22.0
21.7
20.9
21.0
23.7
20.9
25.4
25.7
TABLE 2-15. SECTOR #51 PAPERBOARD PRODUCTS
Equivalent S.I.C. Sector (s ): 265
Census of Manuf. (CM) Nat'l Purified Value of Shipments
National Sector Output per Worker ($x!03) : 36.2
Nat'l Purif. Corr. Factor: 1.01 Scaling Factor: 1.04
($xlO&): 8,148.1
1972 Price Index:
STATE SEAS (modified) Purified Modified % of Nat'l State
CM(purified) CM Amt. SEAS Amt. CM Output
($x!06) ($x!06)
1.05
0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz .
Col.
Calif.
1.18
1.17
1.37
1.00
1.07
.98
1.15
.89
.87
.91
1.10
.85
.75
299
579
498
566
385
590
741
286
207
179
42
267
D
D
759
352.2
675.1
681.4
567.4
410.8
579.5
853.8
254.1
80.7
162.2
46.3
226.4
570.4
4
7
6
7
5
7
9
3
2
2
<1
3
9
29.8
28.5
32.6
32.9
41.4
36.3
36.0
38.2
33.6
40.2
26.2
38.8
45.5
37
-------
TABLE 2-16. SECTOR #52 NEWSPAPERS
Equivalent S.I.C. Sector(s); 271
Census of Manuf. (CM) Nat'1 Purified Value of-Shipments ($x!0°); " 8., 2'60JP~ _._.
National Sector Output per Worker ($x!03): 23.68
Nat'l Purif. Corr. Factor: .99 Scaling Factor; 1.05 1972 Price Index: 1.02
STATE SEAS(modified)
CM(purified)
Purified
CM Arat.
($x!06)
Modified
SEAS Amt.
($xlb6)
% of Nat'l
CM Output
State 0/W
Mass.
N.Y.
N.J.
Pa.
Mich .
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.04
.98
1.04
1.02
.67
1.07
1.11
.73
1.28
.90
.91
1.08
.80
1.06
.90
315
921
273
496
355
452
545
170
136
355
82
403
93
107
883
327.9
903.5
283.0
-504.0
239.5
485.4
605.4
123.4
173.9
318.2
74.9-
434.7
74.6
113.1
792.9
4
11
3
6
4
5
7
2
2
4
1
5
1
1
11
24.1
28.1
24.6
21.8
27.0
23.8
24.7
20.2
20.4
25.6
24.4
23.1
26.9
22.0
26.8
TABLE 2-17. SECTOR #53 PERIODICALS
Equivalent S.I.C. Sector(s): 272
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlQQ): 3,5847 I-
National Sector Output per Worker ($x!03): 52.71 .
Nat'l Purif. Corr. Factor:1.02 Scaling Factor: .79 1972 Price Index: 1.05
STATE SEAS(modified)
CM(purified)
Purified
CM Amt.
($x!06)
Modified
SEAS Amt.
($x!06)
% of Nat'l
CM Output
State 0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N-.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.0.5
.62
1.48
1.72
1.09
1.05
.66
2.63
.61
28
1,776
D
D
20
112
325
23
D
30
NL
D
NL
19
179
29.5
1,105.7
-
29.6
193.0
354.7
24.2
19.8
50.0
108.4
1
50
<1
3
9
1
1
<1
5
33.7
67.5
33.0
36.7
47.6
32.9
72.5
38.0
46.0
38
-------
TABLE 2-18. SECTOR #54 BOOKS
Equivalent S.I.C. Sector(s): 273
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlQQ): 4.140.4
National Sector Output per Worker ($x!03) : 58.72
Nat'l Purif. Corr. Factor: 1.09 Scaling Factor: 1.14 1972 Price Index: 1.02
STATE SEAS(modified)
CM(purified)
Purified
CM Amt.
($xlO&)
Modified
SEAS Amt.
($xlQ6)
% of Nat'l
CM Output
State 0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz .
Col.
Calif.
.77
.79
2.53
1.10
.93
1.06
.99
1.01
1.10
.40
2.11
.90
1.16
278
1,675
133
164
69
207
571
83
16
22
D
56
NL
19
161
214.3
1,321.0
334.8
180.8
63.9
220.2
563.3
83.8
17.7
8.8
118.3
17.0
186.2
6
40
3
4
2
5
14
2
<1
<1
1
<1
4
35.1
54.5
38.1
24.6
28.6
41.3
58.2
40.0
18.7
22.2
23.2
21.2
29.6
TABLE 2-19. SECTOR #59 FERTILIZERS
Equivalent S.I.C. SectorCs): 2873. 2874. 2875
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlQQ): 2^909.8
National Sector Output per Worker ($x!03): 77,85
Nat'l Purif. Corr. Factor: 1.05 Scaling Factor: 1.53 1972 Price Index:
STATE SEAS(modified) Purified Modified % of Nat'l State 0/W
CM(purified)CM Amt. SEAS Amt. CM Output
($x!06) ($x!06)
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.21
4.39
1.17
.72
1.03
.61
NL
21
9
D
D
D
D
17
133
476
D
200
NL
NL
D
24.3
39.5 ,
19.9
96.2
492.0
122.1
1
<1
1
4
16
7
65.3
43.5
80.5
57.7
74.6
86.4
39
-------
TABLE 2-20. SECTOR #66 DRUGS
Equivalent S.I.C. Sector(s): 285
Census of Manuf. (CM) Nat'l Purif fed" Value of Shipments ($xlOp): 7,830747
National Sector Output per Worker ($x!03): 61.73
Nat'l Purif. Corr. Factor: .98 Scaling Factor: 1.02 1972 Price Index; 1.01
STATE SEAS(modified)
CM(purified)
Purified
CM Amt.
Modified
SEAS Amt.
($x!06)
% of Nat'l
CM Output
State 0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.99
.62
.96
.75
1.29
1.06
1.32
1.08
.31
1.50
1.28
28
1,169
1,865
1,108
466
119
539
36
D
65
NL
50
D
D
376
55.7
719.7
1,788.5
835.2
599.2
126.4
712.2
38.7
20.4
75.1
480.9
1
15
24
14
7
1
7
<1
1
1
5
36.3
72.7
67.0
86.3
65.2
40.3
45.8
37.0
60.0
36.4
44.1
TABLE 2-21. SECTOR #67 CLEANING AND TOILET PRODUCTS
Equivalent S.I.C. Sector (s): 284 __^
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO&) : 9,998.^4
National Sector Output per Worker ($x!03): 87.5
Nat'l Purif. Corr. Factor: 1.02 Scaling Factor: 1.08 1972 Price Index: 1.00
STATE SEAS(modified) Purified Modified % of Nat'l State 0/W
CM(purified) CM Amt. SEAS Amt. CM Output
($x!06)
($x!06)
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
.54
1.71
1.07
1.51
1.29
.77
.69
.45
1.18
.66
1.39
438
741
1,542
187
D
1,044
1,414
D
115
58
NL
208
D
20
956
236.7
1,266.8
1,653.4
282.3
1,345.7
1,089.7
79.5
26.3
245.0
13.2
1,326.9
4
7
15
2
10
14
1
<1
2
<1
10
104.7
64.3
73.1
50.7
146.3
112.7
S3. 4
52.0
88.6
100.5
86.8
40
-------
TABLE 2-22. SECTOR #69 PETROLEUM REFINING
Equivalent S.I.C. Sector(s) : 291,299
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO°): 28,165.4
National Sector Output per Worker ($x!03): 243.29
Nat'l Purif. Corr. Factor: 1.Q5 Scaling Factor: 1.01 1972 Price Index: 1.02
STATE SEAS(modified) Purified Modified % of Nat'l State 0/W
CM(purified) CM Amt. SEAS Amt. CM Output
($x!06) ($x!06)
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
.73
.61
1.41
3.90
1.12
.93
D
D
D
1,850
D
1,221
1,869
D
, NL
NL
61
7,659
D
D
3,390
1,345.6
745.3
2,632.2
237.7
8,603.2
3,143.2
7
4
7
<1
27
12
169.4
314.3
240.5
193.3
238.4
230.6
TABLE 2-23. SECTOR #76 FOOTWEAR
Equivalent S.I.C. Sector(s): 313,314
Census of Manuf. (CM) Nat'l Purified Value ofShipments ($xlQ6); 3,486.1
National Sector Output per Worker ($xlO^): 18.8
Nat'l Purif. Corr. Factor:
STATE SEAS(modified)
CM(purified)
1.00 Scaling Factor: .87 1972 Price Index: 1J37
Modified % of Nat'l State 0/W
Purified
CM Amt.
($x!06)
SEAS Amt.
($x!06)
CM Output
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.50
1.00
1.44
1.18
.77
.49
1.49
.66
.73
413
230
24
326
D
D
139
D
65
24
D
50
NL
NL
53
617.6
229.7
34.5
384.1
106.6
32.0
35.7
33.2
38.5
12
7
1
9
4
2
1
1
1
19.0
30.1
20
16.7
20.7
21.7
12.6
18.5
16.6
-------
TABLE 2-24: SECTOR #81 CEMENT, CONCRETE, GYPSUM
Equivalent S.I.C. Sector(s): 524,527
Census of Manuf. (CM) Nat'l Purified Value of Shipmen
National Sector Output per Worker ($x!03) : 42.69
Nat'l Purif. Corr. Factor: .99 Scaling
STATE
SEAS (modified)
CM (purified)
Purified
CM Amt.
($x!06)
Factor: 1.04
Modified
SEAS Amt.
($x!06)
ts ($xlO&): 9,598.0
1972 Price
% of Nat'l
CM Output
Index:
State
1.05
0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.15
___
.94
1.12
.70
1.16
1.25
1.21
» _>
-,-
1.47
.98
.85
149
D
256
479
462
469
576
175
D
D
168
714
D
D
941
171.7
259.6
555.4
524.5
545.2
462.7
211.5
247.1
697.9
797.9
2
5
5
5
5
4
2
2
8
10
59.5
45.9
41.7
51.5
41.6
45.7
58.5
57.0
45.2
49.7
TABLE 2-25. SECTOR #84 COPPER
Equivalent S.I.C. Sector(s): 5551, 5541, 5551^ 5562
(CM) Nat'l Purified Value ofShipments ($xlO&):
12,754.6
Census of Manuf.
National Sector Output per Worker ($x!03): 144.58
Nat'l Purif. Corr. Factor: 1.51 Scaling Factor; 2.01 ~197~2 Price Index:_K04_
STATE SEAS(modified) Purified Modified % of Nat'l State 0/W
CM(purified)CM Amt. SEAS Amt. CM Output
($x!06)
($x!06)
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
L 1.25
1.67
__-
_~_
D
D
D
489
D
294
D
D
D
NL
D
D
NL
NL
D
602.7
492.5
4
5
,
55.5
70.2
42
-------
TABLE 2-26. SECTOR #89 OTHER NONFERROUS ROLLING AND DRAWING
Equivalent S.I.C. Sector(s): 3356 .
Census of Manuf. (CM) Nat'l PurifiedValue of Shipments ($x!Q°): 1.197.7
68.76
rational Sector Output per Worker ($xl03):
Nat'l Purif. Corr. Factor:_.96_ Scaling Factor:
SEAS(modified)
CM(purified)
STATE
Purified
CM Amt.
($x!06)
Modified
SEAS Amt.
($x!06)
1972 Price Index: 1.02
% of Nat'l State 0/W
CM Output
Mass .
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
.87
.45
.54
.58
1.12
1.27
M _ *>
108
147
105
86
53
111
D
NL
D
NL
D
D
NL
NL
D
93.6
66.3
57.1
50.2
59.3
141.1
9
12
9
7
4
9
113.0
127.5
68.1
56.2
68.7
44.6
TABLE 2-27. SECTOR #94 PLUMBING AND HEATING EQUIPMENT
Equivalent S.I.C. Sector(s): 543
Census of Manuf. (CM) Nat'l Purified Value of Shipments (.$xlOG) :
National Sector Output per Worker ($x!03):35.2
Nat'l Purif. Corr. Factor:
STATE SEAS(modifled)
CM (purified)
2.045.3
.96
Scaling Factor: 1.04 1972 Price Index:i.pj
Purified Modified % of Nat'l State 0/W
CM Amt. SEAS Amt. CM Output
($x!06) ($x!06)
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
.64
.96
1.33
.90
1.02
.99
1.20
1.07
.12
.52
30.
131
56
180
95
203
196
132
29
9
D
D
NL
NL
D
19.3
126.4
74.3
161.1
97.2
200.1
. 234.8
141.3
3.4
4.7
1
6
3
9
5
10
9
6
1
<1
34.8
33.2
33.9
33.3
59.5
39.0
34.6
32.7
33.2
43.5
43
-------
TABLE 2-28. SECTOR #95 STRUCTURAL METAL PRODUCTS
Equivalent S.I.C. Sector(s)= 544
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xl06)t_
National Sector Output per Worker ($x!03): 55.94 '
Nat'l Purif. Corr. Factor: 1.02
14.525.4
Scaling Factor; 1.03 1972 Price Index:1.04
STATE SEAS(modified)
CM(purified)
Purified
CM Amt.
C$xl06)
Modified
SEAS Amt.
($xl06)
% of Nat'l
CM Output
State 0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.12
1.20
1.00
1.18
.95
.89
1.19
1.31
.79
.67
1.08
1.14
1.13
1.65
.95
264
829
476
1^237
582
1,278
870
356
322
556
312
1,009
93
97
1,195
296.9
991.7
474.5
1,457.9
551.9
1,136.5
1,032.3
467.9
254.4
374.5
336.5
1,150.5
105.4
159.9
1,131.1
2
6
3
8
4
9
6
2
2
4
2
7
1
1
8
32.4
31.3
35.9
36.0
36.8
38.9
40.0
34.2
41.6
34.5
37.8
33.9
33.7
32.8
37.6
TABLE 2-29. SECTOR #97
Equivalent S.I.C. Sector(s)
METAL STAMPING
: 3465, 3466,
3469
Census of Manuf. (CM) Nat'l Purified Value of Shipments
National Sector Output per Worker ($x!03) : 37.2
Nat'l Purif. Corr. Factor: 1.00 Scaling
STATE SEAS (modi f ied)
CM (purified)
Purified
CM Amt.
($x!06)
($xlO<>) :
Factor: 1.08 1972 Price
Modified %
SEAS Amt. CM
($X106)
of Nat ' 1
Output
8,351.1
Index :
State
1.05
0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz,
Col.
Calif.
_»_
1.79
.90
.88
1.08
1.65
3.02
1.51
D
D
187
D
2,424
1,676
998
303
D
D
D
17
NL
D
227
333.8
2,189.7
1,474.2
1,081.7
498.7
51.3
343.7
2
29
20
12
4
<1
3
30.2
43.1
40.5
40.1
28.9
23.7
30.3
44
-------
TABLE 2-30. SECTOR #98 CUTLERY, HANDTOOLS, HARDWARE
Equivalent S.I.C. Sector(s): 542
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xl06):
National Sector Output per Worker ($xlQ3): 31.9
Nat'l Purif. Corr. Factor:
SEAS(modified)
CM(purified) CM Amt. SEAS Amt. CM Output
5,111.2
STATE
1.00 Scaling Factor: .98 1972 Price Index:1.03
Purified Modified % of Nat'l State 0/W
($x!06)
C$xl06)
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.27
1.04
.93
1.33
.70
.84
1.00
.42
.80
.65
1.20
2.41
.86
.90
335
390
268
193
731
672
537
199
55
32
11
17
NL
29
415
426.1
405.5
248.9
257.2
509-. 1
563.0
536.6
84.4
43.8
20.7
13.2
40.9
24.8
371.9
7
8
5
4
14
13
10
4
1
<1
<1
<1
<1
8
37.6
27.9
31.2
25.4
43.0
41.7
29.2
38.3
25.0
24.6
15.7
28.3
26.4
29.9
TABLE 2-31. SECTOR #101 OTHER FABRICATED METAL PRODUCTS
Equivalent S.I.C. Sector(s) : 347, 3493. 3497. 3499
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO&) :
4.288.
3
National Sector Output per Worker ($x!03) : 27_JJ
Nat'l Purif. Corr. Factor: Scaling Factor: .85 1972 Price
STATE SEAS (modified) Purified Modified % of Nat'l
CM (purified) CM Amt. SEAS Amt. CM Output
($x!06) ($x!06)
Index:
State
1.05
0/W
Mass.
N.Y.
N.J.
Fa.
Mich.
Ohio
111.
WTs.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.09
1.22
1.14
.83
2.67
.94
1.07
1.86
1.12
91
D
202
D
312
D
670 .
64
28
49
D
D
5
22
D
99.4
245.5
354.3
553.7
170.9
26.4
52.6
9.3
24.7
2
5
7
16
1
<1
<1
,
<1
<1
20.1
28.5
24.8
34.0
20.5
19.9
22.2
16.0
27.9
45
-------
TABLE 2-32. SECTOR #103 FARM MACHINERY
Equivalent S.I.C. Sector(s): 352
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO&):
National Sector Output per Worker ($x!03) : 44.62
Nat'l Purif. Corr. Factor: .98 Scaling Factor: 1.
SEAS(modified] Purified Modified
CM(purified) CM Amt. SEAS Amt. CM Output
5,501.7
STATE
1972 Price Index: 1.04
% of Nat'l State 0/W
($xlQ6)
($x!06)
Mass .
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.09
.74
1.02
1.16
.77
1.29
.59
1.77
1.57
.._
NL
76
D
D
339
254
970
643
60
25
D
59
D
25
D
83.2
251.1
260.2
1,120.6
493.3
77.6
14.7
104.5
39.2
1
6
5
18
12
1
<1
1
<1
51.7
93.6
45.4
45.9 .'
51.3
27.7
41.7
31.4
28.8
TABLE 2-33. SECTOR #107 MACHINE TOOL, METAL FORMING
Equivalent S.I.C. Sector(s): 3542
Census, of Manuf. (CM) Nat'l Purified Value of Shipmen
National Sector Output per Worker ($xlQ3) : 28.78
Nat'l Purif. Corr. Factor: 1,02 Scaling
STATE SEAS (modified)
CM (purified)
Purified
CM Amt.
($x!0&)
Factor: .92
Modified
SEAS Amt.
($x!06)
ts ($xlO&) :
1972 Price
% of Nat'l
CM Output
709.7
Index:
State
1.02
0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
Cf.
Fla.
Ala.
Tex.
Ariz.
Col.
CalTfT
1.08-
1.23
1.15
.74
.70
1.03
1.18
.72
D
63
10
34
88
161
204
NL
D
4
NL
5
NL
D
D
68.1
12.3
39.1
65.3
112.0
210.6
*
4.7
3.6
9
1
..5
12
23
29
1
1
26.8
25.8
27.4
33.2
30.4
30.3
19.0
24.0
46
-------
TABLE 2-34. SECTOR #113 INDUSTRIAL PATTERNS
Equivalent S.l.C. Sector(s) : 3565. 3567. 3569 ^___
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO&):
National Sector Output per Worker ($x!03): 50.7
Nat'l Purif. Corr. Factor: l^QJ
SEAS(modified) Purified
CM(purified)CM Amt. SEAS Amt. CM Output
1842.4
STATE
Scaling Factor:^%9jZ 1972 Price Index: 1.Q2
Modified % of Nat'l State 0/W
C$xl06)
Mass .
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
1.02
.90
.53
.69
.99
1.35
.49
1.12
**
60
126
124
224
192
D
136
D
: 22
D
D
22
NL
NL
D
61.0
113.7
65.4
153.9
190.1
183.8
10.8
24.6
3"
7
7
12
10
7
1
1
32.8
31.9
32.9
31.0
31.3
28.4
27.1
24.9
TABLE 2-35. SECTOR #118 ELECTRONIC MEASURING INSTRUMENTS
Equivalent S.l.C. Sector(s) : 3825
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO&) :
National Sector Output per Worker ($x!03) : 28.2
Nat'l Purif. Corr. Factor: .91 Scaling Factor: 1.08 1972 Price
STATE SEAS (modified] Purified Modified % of Nat'l
CM (purified) CM Amt. SEAS Amt. CM Output
($x!06) C$xl06)
1,408.6
Index:
State
1.01
0/W
Mass.
N.Y.
N.J.
Pa.
Mich.
Ohio
ill.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz .
Col.
Calit.
.69
1.03
1.81
1.16
.62
1.17
.96
3.44
.69
_
86
110
53
44
25
63
132 »
D
D
5
NL
34
NL
D
D
59.2
113.4
95.8
51.1
15.6
73.5
126.4
17.2
23.3
6
8
4
3 _,
2 1
5
10
<1
2
27.6
35.6
24.2
24.0
24.5
24.6
30.2
12.5
28.5
47
-------
TABLE 2-36. SECTOR #119 TRANSFORMERS AND SWITCHES
Equivalent S.I.C. Sector (s) :_361
Census of Manuf. (CM) Nat'l Purified Value of Shipments
- 3,605.2
National Sector Output per Worker ($x!03): 51.23
Nat'l Purif. Corr. Factor: 1,01 Scaling Factor: 1_.QQ 1972 Price Index:j).99
STATE SEAS(modified) Purified Modified % of Nat'l State 0/W
CMCpurified)CM Amt. SEAS Amt. CM Output
($x!06) ($x!06)
Mass.
N.T.
N.J .
Fa.
Mich.
unio
in.
WlS.
N.L,
Ma.
Ala.
lex.
Ariz.
(_OI .
Lam.
.33
1.56
1.23
1.41
1.38
.95
1.31
.51
1.00
.90
.62
1.00
187
79
142
799
37
154
315
192
103
D
D
60
D
6
194
62.5
123.4
175.3
1,122.6
50.9
146.8
412.1
97.6
103.5
54.1
3.7
193.8
5
2
4
22
1
4
9
5
2
2
<1
5
25.3
30 . 0
25.2
30.0
33.6
32.3
26.2
40.4
30.0
39.3
30.0
32.5
TABLE 2-37. SECTOR #123 HOUSEHOLD APPLIANCES
Equivalent S.I.C. Sector(s): 365
Census of Manuf. (CM) Nat'l Purified Valueof Shipments ($xlO&):
National Sector Output per Worker C$*10^):__42.72
Nat'l Purif. Corr.
6973:6
Factor: 1.01 Scaling Factor: .92 1972 Price Index: 1.00
STATE SEAS(modified)
CM (purified)
Purified
CM Amt.
Modified
SEAS Amt.
($x!06)
% of Nat'l
CM Output
State 0/W
Mass .
N.Y.
N.J.
Pa.
Mich.
Ohio
111.
Wis.
N.C.
Fla.
Ala.
Tex.
Ariz.
Col.
Calif.
.55
.85
1.38
1.03
.98
.78
.27
.55
.75
75
D'
103
D
D
1,165
863
282
171
D
19
43
NL
D
256
41.5
87.5
1,602.6
891.5
276.3
132.8
5.1
23.6
191.9
1
1
17
12
4
2
<1
1
4
38.9
24.3
49.5
37.0
35.3
36.7
38.0
30.7
41.5
48
-------
TABLE 2-38. SECTOR #127 COMMUNICATION EQUIPMENT
Equivalent S.I.C. Sector(s): 366
Census of Manuf. (CM) Nat'l Purified Value of Shipments ($xlO°):
National Sector Output per Worker ($xlQ3) : 30.15
Purif.
13.355.6
Nat'l Purif. Corr. Factor:
STATE SEAS(modified]
CM(purified)
.98 Scaling Factor: .87 1972 Price Index: I.QQ
Purified Modified % of Nat'l State 0/W
CM Amt.
C$xl06)
SEAS Amt.
C$x!06)
CM Output
Mass.
ST. V.
N. J .
Fa.
Mien..
Unio
III.
ffis.
N.C.
Fla.
Ala.
lex.
Ariz.
Col.
Cant.
1.60
1.06
1.08
.80
.54
1.13
.98
1.39
1.26
.47
_-_
.99
820
1,231
1,178
466
101
387
1,442
83
D
425
177
D
D
D
2,400
1,313.7
1,303.7
1,270.2
373.5
54.2
437.8
1,411.2
115.6
537.2
83.1
2,383.8
6
9
9
4
1
3
11
1
3
1
18
29.2
27,0
30.3
31.7
42.9
30.6
31.1
25.0
26.1
27.5
32.4
TABLE 2-39. SECTOR #128 ELECTRONIC COMPONENTS
Equivalent S.I.C. Sector(s); 367
Census of Manuf. (CM) Nat']
National Sector Output per
Nat'l Purif. Corr. Factor:
STATE SEAS (modified)
CM (purified)
L Purified Value of Shipment
Worker ($x!03) : 26.35
1.02 Scaling
Purified
CM Amt.
C$xl06)
Factor: 1.00
Modified
SEAS Amt.
C$xl06)
;s ($xlO&) :
1972 Price
% of Nat'l
CM Output
9038.0
Index:
State
1.00
0/W
Mass.
N.V.
N.J.
Pa.
Mich.
Uhio
Til.
Wis.
N.L.
Fla.
Ala.
Tex .
Ariz .
Col .
Calif.
1.86
.50
2.24
1.04
1.32
.93
1.29
1.66
.74
.89
3.52
1.16
396
2,046
361
882
53
263
580
69
D
219
D
D
327
14
1,365
735.6
1,020.0
809.4
914.1
69.8
245.8
748.2
114.3
162.1
292.2
49.3
1,585.3
4
23
4
10
<1
3
6
<1
2
4
<1
15
21.8
41.7
19.7
27.5
20.8
32.2
22.6
21.9
23.4
19.8
14.0
25.2
49
-------
CHAPTER 3
AN ASSESSMENT -OF LOCATION QUOTIENT METHODS
OF REGIONALIZING THE NATIONAL INPUT-OUTPUT MODEL
I. INTRODUCTION
The depiction of a regional economy by an input-output table is useful
in both a descriptive and predictive sense. The interrelationships of all
parts of a regional economy are comprehensively detailed in such a model,
and its structure permits detailed analysis of the workings of that economy.
However, data collection for such a detailed table requires expensive
surveys, and construction of the table often requires such a lengthy period
that it is often obsolete before it is published.
Various short-cut methods have been proposed and tested for deriving a
regional table from a given national table, using only a limited amount of
regional data. Such an approach is attractive, as it is relatively simple,
inexpensive, and quick, and requires no extensive data gathering. However,
such techniques have yet to be proven effective.
Past attempts to regionalize national input-output tables have produced
less than satisfactory results. The requirement of having available a pre-
existing ("true") regional table for verification of the estimate has imposed
constraints on the degree of disaggregation of industry sectors, as the
sectors of the estimating table must correspond to those of the "true" table
in order for meaningful comparisons of results to be made. As a result,
typical sectoring schemes in previous studies have been forced to use less
than 30 sectors in order to be compatible with the "true" regional table
while national tables are available at a 367-sector level. The aggregation
necessary to reconcile the sectors of the national table with those of the
existing regional table introduces a wide industry (or product) mix within
most sectors. This, together with the inherent limitations in the value of
highly aggregated input-output tables, disturbs the premises necessary for
application and examination of short-cut regionalization techniques for
national tables.
The present study is intended to evaluate the method of location quo-
tients (measures of the relative importance of a regional industry) as a
means of modifying a national input-output table to reflect the structure of
a regional economy. If the application of such a method is indeed verified
as a valid means of estimating the interindustry structure of a regional
economy, requiring a minimum of data and avoiding expensive survey proce-
dures, then an alternative may be available for the existing method of
50
-------
regional projection of the SEAS system. As opposed to the current top-down
structure involving national projections allocated to regions by shares based
on employment, the easy availability of regional input-output tables would
allow accurate regional projections to then be aggregated to a national
level. Alternatively, the current method of national projections could be
retained, while specific projections for those regions under consideration
could be derived using those regional tables.
We will first survey quotient methods previously used to regionalize
national input-output tablesJ Following this, we describe our ideas for
improving the means of evaluating the effectiveness of one method in partic-
ular, that of simple location quotients. The next section will describe our
procedure for examining this method. The means of measuring the technique's
simulation ability will be detailed, and finally our results and conclusions
will be presented,
II. QUOTIENT METHODS: A BRIEF SURVEY
Morrison and Smith (1974) describe several quotient methods to
regionalize national input-output tables which use the principle that
a.,-..- = q.j.j A.-J, where a^ = the regional trade coefficient from industry i to
industry j (i.e. the ratio of i's input to j
divided by j's total output), A.. = the national technical coefficient (i.e.
the national ratio), and the J reduction quotient q,. is less than or
equal to 1. Exports are obtained as residuals: 1J
- y
R
if
where y. is final demand for industry i, x is output, the subscripts i and j
are ]' sectors, and the superscript R denotes the region.2 Imports of
product i are computed as the amounts necessary to satisfy production re-
quirements not met by regional industry i, assuming national production
technology holds in region R:
mR. = A..xR - a. .xR.
U U j iJ j
Simple Location Quotient
The simple location quotient (SLQ) measures the relative importance of
the industry in the region compared to its importance in the nation:
LQ. = (NR/NR) / (i
1
A summary of other previous studies of regionalization techniques is con-
tained in Section VIII of this chapter.
2
Equations and notation in the following pages will always be assumed to
refer to region R, and the superscript will be omitted except where confusion
may otherwise result.
51
-------
Here, N represents output, employment, or some other measure of economic
importance, If LQ^l, then the region is assumed to be self-sufficient in
industry i, and r^ = an--j, the national coefficient. If LQ,.<1, then
local requirements are
P,j - LQ, .,.,.
Purchases -Only .Location Quotient
not satisfied locally, and
The use of a "Purchases-Only Location Quotient" (POLQ) was suggested by
Tiebout (Consad, 1967, pp. 3.10-3.13) and differs from the simple location
quotient in that the summation of total output (or employment) includes only
those sectors purchasing from sector i. Consistent results were obtained
when used in the Consad study, and the POLQ ranked directly below the SLQ in
all five of the tests of Morrison and Smith, who note only a marginal
difference in results and point out that Schaffer and Chu (1969) do not
distinguish between the two in their discussion of results.
Consad Alternatives
In an appendix to the Consad study, 'two alternate methods were examined
and rejected in favor of POLQ (Consad, 1967, pp. F.1-F.6). In the first
method,
LQ1? = (N*/ENRA ) / (N./EN.A. )
1 ! J J ij J j 3 TJ
where j refers to consuming industries, and the national technical coeffi-
cients are incorporated into the definition to weight what would otherwise
be the Purchases-Only Location Quotient, In general, results obtained were
not significantly different from the POLQ method, with both positive and
negative variations in individual elements. In all cases, this alternate
method resulted in slightly higher values {1% to 5%) than POLQ. Consad con-
cluded that, as results were so similar and intra-state self-sufficiency was
already considered overstated, the POLQ results were to be used.
The second alternative considered by Consad, based on interstate imports,
used a location quotient of the form
LQR = NR/(NR
where I = imports of industry i into region R. In recognition of the
1 extensive data requirements, the direct estimate of the
Washington model's imports (considered of high accuracy) and imports imputed
from Census of Transportation data were compared. Large discrepancies
existed between the two sets of data in all cases, and as a result no
further consideration of this alternative was made.
52
-------
Cross-Industry Quotient (Morrison and Smith, 1974)
In order to include the relative size of the purchasing industry as
well as that of the producing sector, the Cross-Industry Quotient (CIQ)
compares the proportion of national output of producing sector 1 to that of
purchasing sector j:
CIQ.
(NR/N
i J j
(Hence CIQ.. = LQ^/LQ..) This, in a manner similar to those above, is used
t.n J J arliijst thp national rnpffiripnt
to
adjust the national coefficient.
Logarithmic Cross-Quotient
Round, using a Logarithmic Cross-Quotient (RND), takes into account the
relative sizes of region and nation while retaining the properties of the
CIQ:
RND = LQ./log (1 + LQ )
' J ^ J
applied in the manner of the other quotients (Morrison and Smith, 1974).
Round's Welsh Model
Round (1972), developing his Welsh model with two regions and stressing
the presence of a set of clearly defined assumptions (inputs and outputs of
regional industries to and from national industries are in the same propor-
tion as those of the national industries, and as far as possible, intra-
regional trade takes precedence over exporting), derives quotients for
regions R-, and R? similar to the cross-industry quotient, but where x = total
intermediate output and Y = total intermediate input,
and
Round stresses what he calls three essential points: that the Welsh model
is as much a quotient technique as the others, that it has appeal as being
based on well-defined assumptions lacking in other methods, and that a
trading coefficient in one region defines the corresponding one in the other
region and the model_is internally consistent. However, Round closes by
stating that " /_ i _/ t is difficult to believe that the highly complex
pattern of interregional trade can be adequately described by a set of such
extremely simple constructs. It is clear from the earlier tests of the
Welsh model and more recent tests, as yet unpublished, that a great deal
more research is yet to be carried out before one is in a position to assess
the prospects of utilizing non-survey techniques in regional input-output
53
-------
analysis" (Round, 1972, p. 9).
Cross-Industry Quotient Modifications
Morrison and Smith (1974), noting that CIQ.. = LQ./LQ. implies that
intrasectoral shipments of commodity i ij ' J (i.e., where
i = j) have CIQ.. = 1, and therefore that sector i is fully supplied
locally, 1J decided to substitute the simple location quotient
for the elements of the CIQ^ . along the main diagonals to remove this
problem in both ^ cross-industry quotient techniques. These
fifth and sixth quotient methods examined by Morrison and Smith were called
the modified cross-industry quotient (CMOD) and modified logarithmic cross-
quotient (RMOD).
Summary of Location Quotients
Isard (I960, pp. 123-126), succintly summarizes the characteristics of
location quotients. Their advantages include simplicity and the fact that
they can be based on readily available data. They possess four major
disadvantages, however. Tastes and expenditure patterns of households of
the same type and income differ over regions. For example, little fuel is
used in the South, and much in the North, so that a location quotient of 1
for fuel in a Southern state would imply it exports, while in the North the
state would be an importer. Income levels also differ over regions; so do
production practices and industrial mixes. These factors all affect the
location quotient, which is based on the assumption that all factors of an
industry are in the same relative proportion to its output over all regions.
We have used the simple Location Quotient Method in our present study.
A review of experience of other analysts with these and other methods of
regionalizing national input-output models appears in the appendix to this
chapter.
III. APPROACH TO ASSESSMENT OF THE LOCATION QUOTIENT METHOD
Previous evaluations of short-cut methods of regionalizing national
input-output tables have concluded that the results may be of some help in
estimating certain industry sectors, but that at this time the technique
cannot be used reliably as a basis for regional studies. Iterative approxi-
mating, techniques may be more intuitively satisfying than other methods, but
the cost of their implementation would probably be prohibitive for repeated
use in a system such as SEAS as a means of generating tables for all regions
in the nation. Further, some iterative techniques require more data which
must be obtained by survey methods, adding to the cost. Also, iterative
techniques (e.g., Schaffer and Chu's R-I-O-T) have not necessarily produced
better results than simpler techniques. In fact, location quotient methods
have been shown generally to be among the better methods of estimation.
Schaffer and Chu found them to be superior to their iterative procedure and
a pool technique. Morrison and Smith's comparative tests for eight methods
found only the semi-survey-based RAS iterative technique consistently
54
-------
better than the simple location quotient method. Because of the relatively
good showing of location quotient methods in comparison with others, coupled
with the simplicity and minimal data requirements of quotient models, it
was decided that for our needs these methods provided the most promise for
further examination.
In reviewing previous studies of location quotients, several handicaps
to testing of the method were observed. The difficulties and their potential
remedies are described below.
Computation of Location Quotients: Employment vs. Output
First, because of the problems of data availability, employment is
generally considered as the surrogate measure of "importance" of an industry,
rather than output. (Although Schaffer and Chu did use output in computing
location quotients, they were hampered by other data problems which will be
discussed shortly.) Intuitively, in the estimation of a regional economic
input-output model, gross output by sector would seem to give a better
measure of an industry's importance than would employment.
Compatibility of Nationaland Regional Input-Output Tables Used
Secondly, the verification of the validity of any estimating method
requires a valid data base (here, a national table) as a starting point,
and a valid true depiction of what is being modeled, to compare with the
estimate. This requirement has also proven to be a handicap to previous
studies of regionalization methods. In particular, it means that the
national table and the true regional table must be comparable in terms of
time frame and sector classification and aggregation. Unless the estimated
tables derived from the national table are comparable to the true regional
table in these terms, comparison of estimated and true results is impossible.
This problem has been a significant one in previous studies, in several
aspects. The "true" regional table itself is based on survey methods, and so
is itself actually only an estimate, containing a certain amount of error.
Similar criticism may be made of the national input-output table used to
begin the process. Also, the absence of a standard industry classification
scheme for regional tables has resulted in a variety of sectoring breakdowns
not comparable with national tables. Because of this, in order to compare
corresponding compatible sectors of the "true" and estimated tables, aggre-
gation of both the "true" and national tables are required before estimation
begins. This results in an often drastic reduction in the number of sectors
in the table, introducing a large product mix within a sector with all its
attendant problems in input-output applications. Further, a table with a
small number of highly aggregated sectors is of limited use to regional
analysts. Finally, a "true" regional table must be found such that the
economy it describes is of the same point in time as the national table on
which the estimate is based. (Schaffer and Chu used the 1958 national table
to estimate the 1963 survey-based table of Washington state.) Several of
these problems were encountered in a previous study (Walderhaug, 1971)
attempting to adjust the 1963 national table to conform to that year's
27-sector Washington table strictly by trying to achieve conformance
55
-------
between sector definitions and conventions. Walderhaug judged that the
coefficients of over half the sectors of the adjusted table (14 of the 27)
differed from the survey-based table due to technology and product mix,
definitions and conventions, and questionable state sample data.
These problems have hampered previous studies, but suggest that if they
could be overcome, the location quotient method could be tested more fully
on its merits, rather than be subject to the limitations of structural data
problems.
In fact, previous studies have recognized this, and suggested further
investigation and elimination of some of these handicaps:
In themselves the nonsurvey methods may prove useful
supplements to survey studies...
But it seems that, at the moment, there is still no
acceptable substitute for a good survey-based study.
This conclusion, however, is founded on a very
limited test. We deliberately avoided any but grossly
mechanical computations, and no systematic correction
procedures are available yet. Further, our tests
have been based on an aggregation procedure which
completely ignores the product mix in Washington.
Care in aggregation might provide quite different
results -- it is surprising to us that our techniques
yielded acceptable regional production coefficients
for over 25 per cent of the industry categories.
Further tests against other survey-based tables may
provide evidence of consistent differences between
our estimates and survey results. These differences
might be used as adjusting factors to render our
estimates more reliable' and to produce at least
pedagogically acceptable models of regional economies.
(Schaffer and Chu, 1969, p. 96)
Measurement of the Accuracy of the Estimated Table
One additional problem also existed in previous attempts to evaluate
regionalization methods. The measures of comparison between actual and
estimated tables used in such studies have been numerous, but no single
measure has been used by all. As a result, there is no common means to
compare the results of all previous studies, and many of the measures used
have little meaning on a cardinal scale. As no ideal comparative statistic
appears to have been found, the best approach appears to have been followed
by Morrison and Smith (1974) who used a variety of alternative.statistics
to rank methods (although publishing few numerical results).
56
-------
Use of a Control Hodel to Measure the Improvement Attributable to the
Technique
Another problem, pointed out and dealt with by McMenamin and Haring,
was the neglect of testing whether the estimated table performed better than
would a simple naive model. In a sense, this can allow a crude method of
cost-effectiveness analysis, and would further demonstrate in real terms any
advantage of estimating tables using location quotients. McMenamin and
Haring (1974), using the RAS technique to estimate the 1967 Washington table
from the 1963 Washington table, took the 1963 table as the naive model. In
our study, the national table, unmodified by location quotients, was used as
the naive model.
Modifications to the Simple Location Quotient
Modifications to the simple location quotient were presented in Section
II of this chapter. However, previous studies of their applications to
national tables have generally proven them to be inferior to the simple
location quotient. The "purchases-only location quotient" which sums only
over purchasing industries for that sector was reported to give "consistent
results" by Consad (1967), and ranked immediately below the simple quotient
in all of Morrison and Smith's tests (1974). As a result, no potential
benefit was seen in testing this modification. The "cross-industry location
quotient" (which takes into account the relative importance of both the
selling and purchasing industries) provided Schaffer and Chu with results
of comparable quality to those of the simple location quotient and best
estimated imports and exports. However, Morrison and Smith's comparisons
of regionalization methods ranked the cross-industry technique last or
next-to-last in all tests. Again, due to these pessimistic results, we felt
that further investigation of this modification was not a promising avenue
to pursue. RND, CMOD, and LMOD similarly ranked low in Morrison and Smith's
study.
Summary
So, based on our preliminary examination of the results and problems of
previous studies, we decided to evaluate the method of regionalizing national
input-output tables using simple location quotients in a comprehensive and
systematic manner, while working to remove as many of the obstacles encoun-
tered by previous researchers as possible. Summarized, our goals were to
test the simple location quotient technique with:
1) The least sector aggregation possible;
2) The most compatible sectoring scheme possible between the
national table (and therefore the estimated table) and
regional table;
3) The use of a large variety of statistical methods of
comparison;
57
-------
IV.
4) A comparative analysis of the advantages of the added infor-
mation of output by sector over that of employment by sector,
and of both of these over a naive model of unaltered national
coefficients;
5) An analysis of the ability of the location quotient method to
estimate a "true" regional input-output table, given the
efforts described above.
PROCEDURES USED TO ESTIMATE THE 1967 WASHINGTON STATE
INPUT-OUTPUT TABLE
In order to examine the questions raised in the previous section on the
usefulness of the location quotient method, a set of estimates of the 1967
Washington state input-output tables were developed. Examination was also
made of several alternatives to this simple application of location quotients
to the national direct coefficients matrix. Specifically, the tables were
adjusted for internal consistency (non-negative exports). A detailed
description of the procedure is presented in this section. The analysis of
the results is described in sections V, VI, and VII.
Reconciliation of National and Washington Tables
The survey-based input-output table for the state of Washington in 1967
was assumed to be the "true" table as the basis for comparison. The 1967
national table of the Bureau of Economic Analysis^ was used as the given data
3
W. B. Beyers, et^ aj_., Injput-Output Tables for the^^Washington Economy 1967,
Seattle: Graduate School of Business Administration, University of
Washington, 1970. The Washington intra-regional transactions table with
imports as a separate row was used, rather than the table incorporating
imports in the interindustry sales matrix. First, the rationale of applying
location quotients to national tables is to reflect the level of regional
self-sufficiency in supplying goods within a region where national technol-
ogy is assumed. As such, strictly intra-regional transactions are the
appropriate elements of the table. Second, it was felt that intra-regional
relationships are of more importance to regional users than strict technol-
ogical relationships are. Third, Czamanski and Malizia (1969) found in their
test of the RAS method that the intra-regional table could be estimated
more accurately than the one incorporating imports.
4
367-sector national tables are published in Social and Economic Statistics
Administration/Bureau of Economic Analysis, United States Department of
Commerce. Input-Output Structure of the US Economy: 1967. v. 1-3,
Washington: United States Government Printing Office, 1974; magnetic tape
is available from Bureau of Economic Analysis.
85-sector tables are also available on tape, and are published in Inter-
industry Economics Division, Bureau of Economic Analysis, United States
Department of Commerce, "The Input-Output Structure of the U.S. Economy:
1967," Survey of Current Business, LIV, no. 2, (Feb. 1974) pp. 24-56.
58
-------
base for application of the location quotient method. 'Location quotients
were to be applied to the national table,'and the results compared to those
directly obtainable from the "true" table.
The need to directly compare the 27- or 52-sector Washington table with
the 85- or 367-sector national table while preserving the most disaggregated
sectoring scheme possible led to the choice of a 50-sector classification
scheme. The detail of the 367-sector U.S. table permitted very close
reconciliation with Washington's 52 sectors, through aggregation (Table 3-1).
Only three Washington sectors, fishing, forestry, and miscellaneous agri-
cultural products, were forced to combine (ironically due to their aggrega-
tion within the U.S. table). While business and personal services sectors
also could not be exactly reconciled, they were left as separate industries
due to their individual importance within the economy. Both Washington and
U.S. interindustry transactions tables were aggregated to this 50-sector
configuration by pre- and post-multiplication by grouping matrices.5
4 (continued)"
Definitions, conventions and industry descriptions are contained in Social
and Economic Statistics Administration/Bureau of Economic Analysis, Inter-
industry Economics Division, United States Department of Commerce, Defini-
tions and Conventions of the 1967 Input-Output Study, October, 1974.
5
One convention of the national table is its treatment of secondary products
in primary producing sectors. Where this occurs, a transfer is created, de-
picting a flow of the secondary product from the producing industry into the
sector in which it is primary, and counting it as an interindustry transac-
tion. This transferred product is then distributed to its consuming indus-
tries from its new sector. In the aggregation process, any such transfers
between industries which are to be combined must be subtracted from inter-
industry flow, as such a transfer would become intra-industry. No such
transfers are performed in the Washington State tables.
Other conventions in the national tables should also be noted. Govern-
ment sectors are treated as separate sectors in the national table, while
incorporated in the corresponding industrial sector by the Washington table.
Only for the government sectors listed separately by function could proper
national aggregation into the corresponding industry sector be performed.
Competitive imports are transferred as fictitious purchases from the
imports row to the domestic industry, then distributed by that industry to
purchases.
See Beyers, e_t al., 1970, pp. 7-8 for Washington table's conventions,
and U.S. Department oT Commerce, Definitions for national table's conventions.
The national table's "special," "dummy," and unclassified government
industries which had no Washington state counterpart were not included in any
computations except total gross output by sector. Much of their sales were
direct to final .demand, so that their absence had little influence on the
derivation of the national and estimated direct coefficients matrices.
59
-------
LU
i
S3
1
r-
[X
i
i
r
2
*""
O
*
o
LU
I
g
5
1967 WASHINGTO
1
UJ
1
g
ec
C
H
£
U
ID
5
K
O
O
01
tn
«
o
z
u
I-l
n
(0
z
o
4J
U
41
cn
CA
°
W
4J IA
U CO
"**
to
o
a as
CD
z
^
o
1
Washington
.
*
ki >-^
O CM
o in
H
01 -o
JS 01
l> CQ
u a
r-l
r-l
o
C*J *^
r-l O
0
T-l ffl
O D.
a
o
u
*o
41
H
to
3
0
CO
«t
o
CM
ss
CM CM
to
O
0
CJ
0
«
f-4
C
H
U
a.
o
h
T)
r-l
01
rl
*-l
i-l
n o>
4J r-l
Q.i-1
CM
rj u
O O,
CM -3-
O O
st
3 ID
^
01 09
g ..
una,
« o
l<3 rH M
a U
CQ CB
"rl 0) O
3 to K
u a -H
h > E
O
CM
I
cS
CM
4J
§
fc4>
M
00
JJ
*r4
S
Ifcl
Vegetable,
crops
CM
CM
'
O>
~» -CM
"CM r-l C
CM r» O
-* O *J
-1 U O.
o - a. -
<»1 < Ct Cft
o o o o ^
/
u
o
00
U
^
^
u o
a s.
v»
f
^
u
o
V
rl
Livestock &
products
en
\
<*>
^
?
r-
O
C^ r^
r» O
0 j
CM" a
o
CM
O o
0>
§~
o «
C u
S-S
H O
ea M
. a
^j ^%
« u
O M
fc. 9
*-- C
O
a
4J
1
Vl
p.
u
0)
rl
U
01
a
^
*
CM
s
CO Oi
O C»
^ t0
r- aa
O O
~~ *
O O
v
fr
01
J=
CO
H
14
«e
is
« u
v< ^
0 O
fe 0.
Y
JL
a
H
in
-*
r-. co
o c*
o
CM in
M
r» CM
O r-
-» O
~». i-i
O (Xj CM
01
1 0)
« 0
0) -H
o £
, M
co M u
J-l 0} U
3 JS 3
*J a -rj
-« -rl 0
3 s iH
H C
Rt
O O fc
rl -f-l fl)
CO S S
^-^ W ftp
M& f*>
O r-J
en
CM
*
t
CM
w-t
m
e
rt
h
*
O
Beverage in
°
91
60
-------
o
a
fly
en
a.
a>
-O
i
4J
3
O
1
3
CL
C
o
u
I/)
1
o
in
Washington
-i»
r«.
r^
^~
JL
Of
-i
H
j
*~i
W
s
o
4J
u
kl
O CO
CO 1-4
>
5*
i
ca
a
o
u
01
w
§
JJ
UK
Washinj
§ 0
u Z
e n -~.
«p4 O CM
jj u in
U U 1
d 0) 1-4
y to v-'
L4 0>
s *
y jc o>
o u n
to to EJ
O»
o
CM
«,^
n
o
CM
^
a
1
o
kl
aj
JJ
o
(Various
s's'
t '
CO *^
,3 CM
p- rt
g>
1
O
^^
I*
0
^f
t-t
t »-<
^ n o »y
IN CM i-H CN
s
s-\ *t4 O
41 r4 O 4J
i-l 01 1*1 O
T4 ^» kl 0> -K
4J tQ 10 01 CQ kl
X U Q. O. O. 4J
a) o o. >, s e
u a at u q o
T> U U
wo a "
3 ki 3 3 ^ M M
oa. o o M e c
rl -4 a d e u MI
> -H > > -H O 0
s^ E - '^ B i-J 1-1
o
co en
1-4 r4
t *
i-l O O O O
tO CO 00 1 O
1 i-l -H lO CM
n
u
u
3
O
S
o.
1-4
1
,3 -3 ff
H ki a *
u « -H 00
K 0. C OB
« 3 3 5
CM ,
Sr-l
a
Sawmills
Mills, G
Veneer &
CM
>H
CM
. 1
I-l
IH kl
H OJ
E 41
» a
flj Q)
in >
r~ co
in o
f-4 1-1
f-4
CM
oC
CM 0*
-a- >*
CM n
kl
Vl 01
4) -- B
ja a -^
E u ca
3 U u
tH 3 e
a o
(Various
wood pro
wooden c
m
o
O 0
CM
m i*-"
O o
0 0^1
en
1-4
5
IO
CM
0)
*< Q)
3 4J
CM
CO
i-l
Papermil
es
O
*
CM
to
i-H
T-4
Papermi
04
C4
s
tD
CM
1
CO
to
CM
O
V
H Q
t ki
-^ «
*tf fl)
kl C
0) O
Q. U
a
o. «
j-t kl CD
(Various
products
Paperboa
and boxe
o
CM
i
o
-------
O)
03
t-
,f
**
10
01
1
-u
Q.
?
3
CL
r
lashinyton 50-Sector Ii
_
r^.
to
2
,_j
i
CD
S
10
*~
(O
o
£
j
«'
m
I
h
o
u
u
en
0 ^
CJ CO
V 1
CA i*-l
*£
i
Washington Sector R
§»
0
»*
CS r* / *
44 0 CM
.c tj in
B U 1
ti « i-i
» to «-
S
rl
o 1 .
W
0 O 0
W to 9
CM
o
S
Printing a
publishing
v£>
CM
fl
an
Printing, publishinj
allied Indus.
CM
CM
«M
(-1
CO
CM
4
10
« »-4
O -4
c e
I-t V
u
ii
o
R
Indus. Inorg. & Org
chemicals
m
CM
«n
CM
o\
CO
CM
1
CM
CO
CM
a
i-j
cd
u
1
«)
.e .-».
(Various c'
& products
el
a
o en
r» co
CM CM
B
rJ
O
o
u
01
X
o
vO
CM
<
en
CM
4
5
*w to
Petroleum
related In
en
Petrol. Refin. &
related Indus.
r*
CM
CM.
en Ov
1 1
i-l co
CM CM
en en
to
*J X
3 A
jj ifl
Ts
M O
n u
rH (0
00
« -rl
Js J
0 ~
CM
CM
en
!Q
en . CM
en en en
Honferrous metals,
a luminum
H
en
ex
CM
CM
m
en
en
-9 ^
en >o
en en
en en
§
e
1
p-t
(Various 8
products)
IK*
00
en
d- CO
o o
CO CO
en en
Aluminum
CM
en
e
en
^
en
«
U
S
1
0)
(Various h
products)
o
*
o
*
i
o
o
-*
r-l
O
i
"Heavy" fabricated
products
en
en
en
i
en
en
*
1
* *
en en
a
4J
§
U
a
-rl
(Various 1
products)
en
o
1
I-l
o
e CM
*«*
k *
O\ r*
en *
3
-------
o o /-.
t- *j-m
y 00
0> M rt
(tj ^^
Q. .
(U
CL.
O
I
I
L.
O
o
in
oi
C
* O
s> z
O
rl O JM
-X W in
i 10 O I
i a u IH
ro 3 w <
ai
-O
00
c
o X
01 u
to n ;
PI
»
O
en m
in in
1-1 en r-
Ok 00 O
i-i PI «
§
S 0.
oo a
5 3
M 10
§O. 01
H B
33
tr
O O
H 00
,5
3 B
o n
a)
1-4
V
a
3 .
O
«
O.
3 a
cr B
4) aj
a hi
u u
0 4J
* 0
eo
B
S
JO
a r-,
J S1
*J a>
is trx
u o; u
h U
s s
^ 0.
a
> §
oo
m
CO
*
n i£ H
3 <-v I *
o a
M *H a. a. a«
O h -H i u
4J « 3 x h
e
kO
o
v<
kO
(I
5
o
I
I
?
8
0) U
sg
s»
I "
I 0) <-k
' * CM O C4
V o o o
U -H
H 3
Wi cr
|
ess
chine
uppli
3
IT
a)
U 01
%3 I
o. u n,
-H o -H
3 v s
cr >H cr
01 MO)
^ S
s.
s
a
tf
3
a
-H
X
ufactaring
k
V
X
B O
O «
3»1
<»
63
-------
"3»
Q
55
*
u
M
*
tn
1
o
u
en
CO
a
M
T- O ys
U 00
cu m i
Dl CO tH
O- .
-^- CQ
o
CD A 55
XI
5- g
C Q
" S5
s- .,
i s
{ f
irt ^
ie
s _
a
r-*. o J2
to ** *
°> jf jj ^
TSSS
_ m o i
t 4 to t-i
f> 3 « »-
> hoc rt J: « x<
o H H uco oj u u oi
U OJC(JC>r9 Ccw
uMctfu z; S H 5 w x.a M a
n m
O 0
*
o o
1 1
S cf
*O [**» r^ CN Cft O O
* . * ,
9
5
^
w
*-i
H
«
d) U
§V
c vi
H O
U -rl
(0 AJ 0) V
O O "* U
H p a u c
c h » o o
oc oQ ^* to
'
p^
r*.
i
r»
^ m
co at o
* * >n
_
1 S.
S CM
00
tj to
t. «
01 (SI
Q. cn CNJ
w vo o
o c-
C to
o *>,§
o n. 3
5S * O
a>> «r c
x; 0,1-
Q.
(U QJ
«S "° *
c uS
C/) '"*J
CM CL
V) C§ «.
_^ -en
*u vo p*»
"CSJ
c en
O CXI 1C
+j en r-»
en co
c «
f~ - a>
x: to-o
S r-1j
x
Vt CO OJ
CM
V.
= iXui
«C ='=3
as xi
64
-------
Derivation of Location Quotient Estimates of Washington State
Direct Intra^Regional Coefficients
Once the Washington and U.S. transactions tables were aggregated to
similar 50-sector structures, the table of direct coefficients was computed
for each by dividing each interindustry flow by the purchasing industry's
gross output. Then, using the national direct coefficients table as a
basis, two additional tables were derived, one using location quotients
based on employment, and the other using location quotients based on output
(Table 3-2). As previously described, for each industry i, LQ. = (N^/N^) /
(N./N) is computed, where N is either employment or output,
and R indicates the region. Each_row of_the U.S. table of direct
coefficients is then multiplied by min /_LQ.,1_/. The employment by sector
was obtained primarily from the 1967 1 County Business Patterns..
Gross output data by sector was taken from the respective transaction
tables, whose sources are also standard government publications and readily
available. These data sources are listed in the Appendix to this report,
following the text.
At this point, then, we have four models of the Washington state
economy:
(1) The survey-based "true" table for Washington, 1967 (WASH);
(2) The table based on location quotients using gross employ-
ment figures (EMPLQ), derived from the 1967 national
table;
(3) The table based on location quotients using gross output
figures (OUTPLQ), derived from the 1967 national table;
(4) The 50-sector national table whose sectors correspond to
those of Washington, which can serve as what McMenamin
and Haring call the "naive model" (NAIVE). Namely, we
can use this model as a control, to see if the short-cut
method of regionalizing national input-output tables
provides any improvement over the naive use of simply
the national model as an approximation of regional
interindustry relationships.
A Slight Modification of the Application of Location
Quotients Also Examined
At this point, it should be noted that the theory behind application of
location quotients to national tables assumes that regional technology is
approximately the same as that of the nation hence, if every sector in
the region had enough capacity to supply any regional demand, no imports
would be needed, and national and regional direct coefficients would be
equal. If, however, sufficient production from within the region is not
available, that portion of the interindustry requirements not met by
65
U.S EPA Headquarters Library
Mail cow 3*041
1200 Periiiirylvornf Avenue NW
Wasnmrjiun, DC 20460
202-566-0556
-------
z
k-
D
1
&
4J
!
o
5
o
u»
VI
8
a
a.
t,
7.
Lu
X
V
a
o o «? O -* IT i«-, o>Oi^c^cnr>->w<5(^C'n'%«C'O-x<'"-O'>'o.'Or>r~i-irs;in it*. (\jec^-e?-oiimcof~o-r---H-"'*''ri'^co>ft'M
wir^>o.'Or>r~i-irs;in
is, ru .4 -
'"' '»* C c tr, _ «/iC
_ c »
ii- u.
C
^
C
66
-------
IVJ I
rv ,
-
I
m
c
o
*-»
cn
in
(M
rvi oj
-*--'?>A
3
CL
o
c
1C
Q.
s.
Ui
1
_J
D.
-< -C -^ -"
3> -C '.^
^i^G
in M
ft S. f
\j _ c. ~ ti. z O C c ^- = »- l:-
u-;.ai_i_i_uu.J-^<-<-c
X.
^
z
-O
<0
U-^«_2kH|H^^'Xl**U.^tf^*l_« ^^r^.^«^
_a-««J> ^.V'. ~<^ Z ". IK U.
u.1- a.a a: " t-<:a.ai-'aii-iii/:
_ v .if, a ottr = >-(-a;o£ «^ _,-)
a.rJi- = C;a.-j.!^ »i« <=_ >«£ i
Ei_ u.ii.K- i«i»5:^wi:i/'-<-w, ocH >-
C<2 U.OZKOt*-CC-J ZU.~UjH C
^ci>-u.
X
-
§.
67
-------
regional producers must be met by imports; hence the direct regional coeffi-
cient between any two such industries would be decreased from the level of
the national coefficient, and imports would become positive in the corres-
ponding column of that matrix to supply the necessary additional flow. On
the other hand, any production beyond regional (national technology) re-
quirements is assumed to be exported, so that in such a case the direct
regional interindustry coefficients are identical to the national coeffi-
cients, across that producing industry's row. From this interpretation it
can be seen that regional coefficient estimates derived using location
quotients can never exceed national coefficients. Clearly, this provides a
built-in error when regional technology significantly varies from national"
relations, where regional coefficients do in fact exceed those of the
national tables, a feature which cannot be captured by the standard location
quotient method. To test whether a slight modification of the standard
method could better estimate regional coefficients larger than national
coefficients, an alternative method was used (STRICTLQ), where location
quotients were multiplied across each row, regardless of whether they were
greater than one. Thus, for those industries with location quotients
greater than one, their regional coefficients were allowed to exceed the
national coefficients. Two additional matrices of direct coefficients (one
each for employment and output) were thus derived using this modification.
Estimation of TotalGross Output Using EstimatedTables
Next, the matrix of direct and indirect coefficients (I-A)~ was
computed for each of the WASH, EMPLQ, OUTPLQ, and NAIVE models. Each of
these matrices was then used to predict Washington gross outputs (already
known) by using the familiar relationship.
X = (I-A}~lF
where F = vector of final demands for regionally produced goods.
Simple Estimation with All Final Demand Given --
First, to test the predictive value of (I-A)"1 with all final demand,
including exports, given (known in fact from the survey table) the vector
of gross outputs X was estimated by multiplying (I-A)"' by the given 1967
Washington final demands vector.
Simple Estimation with Inferred Exports as a Component of Final Demand --
To this point, we have considered only the estimation of the matrix of
interindustry coefficients, and the use of a given final demands vector to
calculate estimated total gross outputs. These are areas of primary
concern. However, if we wish to construct other components of a regional
input-output table, further computations may be performed.
Exports and imports may not be known or confidently projected in an
actual practical application of location quotients, and they can be
derived using only the known gross outputs vector and local final demand
(and the estimated A-matrix). Schaffer and Chu (1969) describe a method
of calculating exports and imports as residuals:
68
-------
If LQ. > 1, the industry is perceived as more important within the
region 1 than it is in the nation, so the sector can supply all inter-
mediate demand and will export goods. For each such industry, exports
s t
e. = x. - £ x.-j- E y.f, where exports are computed as
1 ] -1 1J f-1 lf
the residual found by subtraction of all s intermediate industry demands
and all t final demands for good i from gross output x. of good i. No
imports are needed to supplement interindustry *'
across that row.
transactions
The case where LQ < 1 is interpreted to mean that industry i's pro-
duction is insufficient, and imports will be required to supplement inter
industry transactions across that row. For any such flow, the imports
required must be sufficient to compensate for the decrease in the direct
coefficient from the national technology (assumed to approximate the
regional technology), so for each element in such a row i,
where m.. = imports from industry i outside of the region to regional
1J industry j
A.. = national direct coefficient
a-- = regional direct coefficient
x-i = regional total gross output.
J i
By such a construction, row and column totals should hopefully equal the
proper gross output, but this may not necessary occur, as exports will not
necessarily be non-negative if LQ. >_ 1, and will not necessarily be zero
for LQ. < 1. However, to adhere strictly to the theoretical method
in this preliminary case, negative exports were allowed for
LQ,- >_ 1, and zero exports were assumed for LQ < 1.
Clearly, such strict application of the simple location quotient
technique will probably result in some cases of:
(2)
(LQ. <
1): rows summing correctly to total gross output
but with negative exports;
1): rows importing and therefore theoretically
having zero exports not summing to total
gross output.
If these exports are then incorporated in final demand, case (2) will
cause AX + F ? X, so (I-A)-lF f X. So, for the two estimates of A derived
using location quotients, we note that it is not likely that they will be
able to accurately predict the total gross outputs vector X where exports
as a component of final demand are imputed in such a strict interpretation.
However, the results obtained are of interest in determining the usefulness
of such strict direct application of the location quotient method. The
69
-------
balancing procedure described in the next section alters some of the strict
assumptions of the location quotient.method for the sake of internal con-
sistency in a complete table, causing substantial modifications to the
location quotient concept. (The NAIVE model, using national coefficients
implicitly has LQ. = 1 for all i, hence case (2) cannot occur and with
exports 1 (residuals) calculated for all rows, the predicted
X = (I-A) F should be-correct, although negative-exports may still arise.)
Balancing the Tables
The inconsistencies of (1) and (2) above can be remedied according to
either of two conventions. Each was applied to the original estimated
tables.
Non-Negative Exports (LQ >_ 1) / Zero Exports (LQ < 1) --
The first alternative used to balance the estimated tables was to
strictly follow the interpretation of LQ. < 1 to imply insufficient pro-
duction by industry i and hence none available for export. A
balancing procedure, based on that of Schaffer and Chu, was used to modify
the estimated direct coefficients matrix A and is described next.
Under the assumption that local final demand was to be taken as given,
the preliminary exports were computed for all 50 rows as total output less
local final demand less intermediate requirements based on the estimates of
A. If LQ. >_ 1 and exports were nonnegative, or if LQ^ < 1 and-exports were
zero, the row was left unaltered. .If this was
not the case, however, these residual-exports as a fraction of total inter-
mediate output were distributed back into the.A-matrix using a constant
multiplier across the row. In this manner, if LQ.->_ 1, negative exports
were redistributed as imports as a constant 1 fraction of row i,
and exports were raised to zero to compensate for the decrease in the row
of direct coefficients. Similarly, if LQ7- < 1 and preliminary residual
exports were calculated as either positive or negative (i.e.,
nonzero), they were distributed back in a similar fashion over row i to
yield zero exports and adjust row i's direct coefficients. If the regional
coefficients rose to the level of the national (technical) coefficients,
implying sufficient production to meet regional production requirements,
no increase of coefficients above this level was allowed, and positive
exports were then calculated.
It is to be noted that in fact such balancing is a form of the supply-
demand pool method (see section VIII of this chapter). To summarize the
preceding process, with local final demand taken as given and total gross
output known,
1) if LQ. < 1
exports must equal zero, hence all residual (total less
final) intermediate sales per unit output must be dis-
tributed by some constant fraction over the estimated
coefficients (which are already a constant fraction of
national coefficients) unless or until the national
coefficients are reached;
70
-------
2)
if LQ. >
the national coefficients are used, and hence the
residual exports (total output less known local final
demand less known intermediate demand by national
coefficients) must be the same as would be calculated
with the NAIVE table; as a basis. In the case where
such exports would be calculated as negative, their
redistribution over the intermediate row i coefficients
would be simply an adjustment of the national
coefficients.
From this it can be inferred that if the EMPLQ and OUTPLQ models both
have LQ. >_ 1 or both have LQ. < 1 for a row i, their corresponding revised
coefficients 1 will be identical. Since it is likely that
this situation actually holds for most rows, it is quite possible that the
balanced EMPLQ and balanced OUTPLQ tables may be identical.
The new A-matrix was then used in (I-A)" F = X to estimate total gross
outputs using both total final demand known, and known local final demand
plus imputed exports.
Non-Negative Exports (LQ >_ 1) / Positive Exports Allowed (LQ < 1)
The second balancing method used was identical to that of the method
above, except that (positive) exports were permitted even if LQ. < 1. In
actual fact, regional industries often do export products ^
while they do not fully supply local intermediate demand. This slight
modification permits such a characteristics to be captured.
A Comment on the Balancing Procedure --
It should be noted that this balancing procedure, necessary if one
desires to construct a full consistent regional table, redefines the
previous row multipliers (location quotients) of the national A-matrix so
that the influence of location quotients is lost. By in fact taking total
output and local final demand as given, and distributing the remainder of
a row's production as in-ermediate sales, labelling any surplus above
national technical requirements as exports, the initial application of
location quotients becomes moot and NAIVE, EMPLQ, and OUTPLQ should all
achieve the same results. (The allowance of positive exports with LQ. < 1
in the second alternative does, however, allow variation.) n
Summary of Procedures
Beginning with the application of the simple location quotient method
to a national table of direct coefficients, estimates of a table of direct
intraregional coefficients for the state of Washington were derived. Two
sets of tables, based on employment and output location quotients res-
pectively, were developed as was the NAIVE model of national coefficients.
Strict application of LQ^ > 1 was also tested. For each of the original
location quotient models and the NAIVE model, (I-A)'1 was calculated, as
were their estimates of total gross output using X = (I-A)~^F, where given
total final demand and then given local final demand plus inferred exports
were used. The models were then balanced for consistency under two
71
-------
\ 1
alternative conventions, and the revised A and (I-A) matrices were re-
derived.
For each of the steps, for all models, statistical comparisons were
made between the estimated and actual figures. A description of statistical
measures and the results of the tests are given in the following two
sections.
V. MEASUREMENT OF ACCURACY OF ESTIMATES
Two types of analyses of the-results are desired. First, in absolute
terms, we would like to know how closely our estimates compare to the
values of the "true" table. Secondly, we wish to note such comparisons
in relative terms is one method better than another? by how much? are
the additional data and computational requirements worth the effort?
Two types of estimates were produced by the previously discussed pro-
cedures: matrices of coefficients, and projections of flows. Each type
can be analyzed using different tests. In all cases, comparison is made
between the estimate and the "true" (survey-based) value.
Matrix Comparisons
The comparison of the elements of two matrices can be performed using
a variety of tests {see in particular Morrison and Smith, 1974). Several
measures were used in this analysis, but all have faults, which will be
noted. Comparisons are made element-by-element over the entire matrix, for
each row, and for each column. Computations over all elements in the
matrix provide an overall view of the accuracy of the estimate, while
statistics for each row and each column allow examination and evaluation
by sector. In the descriptions below,
a.. = the actual survey-based element
A
a^j = the estimate of a..
IJ . 1 j
n = number of elements considered:
2500 for whole-matrix calculations;
50 for row or column calculations
Summations £ are taken over i (rows), j (columns), or i and j (all elements
of the matrices) in all cases.
are:
The statistical measures of comparison of the matrices of coefficients
(1) Mean Percentage Error (HPE) and its Standard Deviation (SDMPE)
72
-------
MPE =
x 100%
SDMPE =
Several observations can be made:
(-
I
x 100%
(a) The statistic is weighted inversely by the size of the true coefficient,
so that an error in a small coefficient has a much larger effect on the
MPE than would an error of the same size in' a larger coefficient.
(b) If the actual a^ = 0 while the predicted
undefined. In J such cases, the
skipped in the computations.
0, then the fraction is
(aH,, a^) pair is
Clearly, a small
dicted matrices.
MPE indicates a closeness between the actual and pre-
The following five tests used were performed by Morrison and Smith
(1974) to compare several regional ization methods in relative terms.
(2) Mean Absolute Difference (MAD)
MAD = Elaij " aijl
This expression is "useful to gain some impression of the magnitude of the
deviations involved." Conclusions based on its results in absolute terms
may be difficult to achieve, but its magnitude for one estimate as compared
to that for another may be useful in determining relative similarity --
which is "closer" to the "true" table. A small MAD indicates closeness to
the "true" table.
(3) Mean Similarity Index (MSI)
la,, - a,J
MSI =
a . + a
n
Note that 0 <_ MSI <_ 1,
indicating no similarity.
and MSI can be graphed as
with 1 indicating perfect similarity and
For a given a.-, the relationship between
J
73
-------
1.0
MSI +
0
a.
Clearly an MSI as close to 1 as possible is desired.
(4) Correlation Coefficient (R) (also R2)
(
VVu
/ za.
('
R also ranges from 0 to 1, with 1 indicating a perfect correlation between
actual and estimated coefficients. However, R2 may equal one even if the
correlation between the two matrices is negative. Thus high R2 values do
not guarantee that the two matrices under comparison are necessarily
identical, only that a straight-line function can approximate the relation-
ship between them.
(5) Information Content (INFO)
This statistic is borrowed from information theory. The estimated
table is considered a forecast of the actual table, and the additional
information which is gained upon receiving the second (actual) table is
expressed as
I(A:A) = z
log.
I)
So, if the estimate is accurate and closely mirrors the actual table, then
the additional information received from the true table should be small.
Hence, a small value of INFO means a close approximation of the true table
by the estimate. Several observations are in order:
(a) The value of each term in INFO is weighted by the actual
coefficient a.., placing more importance on coefficients of
larger size. 1J
74
-------
(b) An absolute interpretation of its value is difficult, so
that it is primarily useful for making relative comparisons
between methods.
(c) In a situation converse to that of MPE, INFO is undefined
(d)
if the estimated value a
a.. is positive.
ij
is zero while the actual coefficient
Because of this, such elements
cannot be included in computation of INFO.
For the case where a.. = 0 and a.. > 0, the term in INFO equals
zero (a 1J A ""J characteristic which
can be found by applying I'Hopital's rule). That is, even if
an estimated coefficient is non-zero while the actual element
is zero, no additional information is contained in the actual
coefficient. Clearly, when such an error is not included in
the statistic, INFO becomes even weaker as a measure of
comparison between two matrices.
(6) "Chi-Square" - x2 (CHISQ)
CHISQ = I
Note that this statistic also is weighted inversely^by the size of the true
coefficient. Also, it is undefined (infinitely large) when a.. = 0 and
a.,-j > 0. Schaffer and Chu and McMenamin and Haring,
computing
over columns, treat' such cases by combining them into one cell to form a
new technical coefficient, and redefining the actual flows in the zero cells
as $49,000, the largest number which would be registered as zero in the
survey-based table. This admittedly arbitrary "adjustment" is credited
with avoiding zero division and extreme chi-square values. However, extreme
chi-square values can occur almost as easily with very small actually-
observed coefficients. Such a procedure of arbitrarily establishing a
minute flow in order to include a chi-square term appears to be more of a
testimony to the faults of the chi-square statistic than to be a consistent
method of overriding those flaws. For this reason, such coefficient pairs
were not included in the chi-square calculations; however, their number and
average magnitude were recorded, to provide some measures of those coeffi-
cients not included in the calculations. Morrison and Smith also chose not
to include them in their chi-square calculations.
Even if Schaeffer and .Chu's method for counting zero flows were used,
however, it should be noted that this statistic which the previous studies
call chi -square does not in fact have a chi-square distribution, unless
the a.., a., which were expressed in terms of percentages by McMenamin and
ij iJ Haring can in some way be interpreted in the customary
manner as frequencies rather than proportions. To arbitrarily use per-
centages rather than fractions {by multiplication by 100) multiplies CHISQ
75
-------
by 100, and the results may seem "reasonable" by evaluating levels of
significance. However, such multiplication by 100 implies that a sample
of 100 points (e.g., dollars) of inputs were purchased by industry j and
the number of dollars sold from each industry i to j was found to be that
particular amount 100 x a-. However, an analogous interpretation could
use any units at all to multiply a., and a,..-, (e.g., millions of
dollars input per millions of dollars J output) and implied
sample size would increase as would the CHISQ statistic, by that multiplier.
Hence, while CHISQ clearly measures a sum of weighted squared deviations
which is useful in comparative terms, the interpretation of a significance
level read off of a chi-square table appears to have little meaning, as,the
decision to force the coefficients (fractions) to appear as frequencies
introduces an arbitrary factor into the "chi-square statistic" with slim
theoretical justification. Calculation of these "pseudo-significance
levels" may only be useful in comparing Schaffer and Chu's and McMenamin
and Haring's number of "acceptable" sectors at "a" = .05, with those found
by using the same tests on our derived matrices.
Flow Vectors Comparisons
Comparisons of the vectors of flows predicted by the various estimated
tables can be made in much the same way as was done with the matrices.
Mean Absolute Difference, Mean Percentage Error, and Mean Similarity Index
can be used as previously defined, with no loss in their intuitive inter-
pretations. The Information Content can still be used for relative
comparisons between methods. While not distributed with chi-square distri-
bution, the formula *
"CHISQ" =
for any vector of flows 2 can also provide an intuitive impression of the
magnitude of deviations in values derived by different methods as a
"weighted sum of squared deviations" term. The linear least squares fit is
probably the set of statistics of most importance and calculations are made
for R, R2, standard error of estimate (STDERR), slope (BHAT), and y-inter-
cept (AHAT) of the least squares regression line. Ideally, BHAT =1, and
AHAT = 0.
VI. ANALYSIS OF RESULTS
As previously stated, our main concern in the test of the use of
location quotients to regionalize national input-output tables was to
accurately derive the matrix of direct coefficients which, given a vector
of regional final demands, can be used to calculate total gross output
by industry: ,
(I-A)"1 F - X.
76
-------
By such a procedure, regional outputs can be predicted by a means other
than allocating national production. The various models described in
Section IV.were analyzed with.a variety of statistical measures (detailed
in Section V). The coefficients of the A and (I-A)~^ matrices and the
predicted total gross outputs, exports, and imports vectors for each of the
models were compared with those of the "true" survey-based Washington
state table. The results are described below.
The Initial Estimates Using Location Quotients
Originally, it will be recalled, three estimates of the 1967 Washington
state input-output table were derived: 1) EMPLQ, estimated by multiplication
of the direct coefficients table of the national economy by location
quotients based on employment; 2) OUTPLQ, estimated by multiplication of
the national table by location quotients based on output; and 3) NAIVE, the
national table which assumes the state economy is identical to the national
economy.
Estimates of the Direct Coefficients Matrix --
The element-by-element comparison of direct coefficients from the
various models indicated less than impressive results for the matrix as a
whole. Each model's estimated matrix (NAIVE, EMPLQ, OUTPLQ) was compared
to that of the true table (WASH) (see Table 3-3). In absolute terms, no
measure yielded very impressive results (all R2 < .54, all MSI < .48, all
MPE > 135%). However, as hypothesized, improvement was noted of the
EMPLQ model's estimated coefficients over the NAIVE (U.S.) model, and of
the OUTPLQ's model over those of EMPLQ, in all tests except information
content. Mean Absolute Difference was reduced by 45% and CHISQ by about
75% in using OUTPLQ rather than NAIVE. The anomaly of the Information
Content's value is not very disturbing, as we have previously discussed its
shortcomings.
Examination of the individual sectors by column indicated significant
disparities between estimated and actual coefficients. Even for the closest
estimate (OUTPLQ), no MPE was less than 23%, only 7 sectors had less than
50% MPE, and 22 had values of MPE greater than 100%. Only one column's mean
similarity index was greater than 0.7 (on a scale of 0 to 1). Ordered by
output, 15 of the highest 21 sectors had R > .8 for OUTPLQ, compared to
12 for EMPLQ, and 13 for NAIVE, but no other statistics provided similar
encouragement.
Also, in order to compare these results with previous studies, Schaffer
and Chu's "chi-square" test was used, using the "$49,000" convention for
zero cells, and "pseudo-significance levels" determined from the CHISQ
statistics calculated (with the coefficients expressed as percentages).
Of the 50 sectors, only three were acceptable at "a" = .05 in the NAIVE'
model, 16 in the EMPLQ model, and 25 in the OUTPLQ model. It is recalled
that, using this test, Schaffer and Chu obtained only 6 of 23 sectors
significant at "
-------
to
o
S
ce
a
o
UJ
g
to
at
C
S
o
UJ
Of
o
o
9)
O
i
CK
I
U
UJ
I
CO
9
o
H
IA
,C
**
S
31
/ i
cy
2
o
H
,|j
tft
X
r**
V-*
j
i
£
rt
C
o
H
rt
S
UJ
tr
<
2
U
fl
in
H
rt
4-1
W
n
a
n
n
8
H
It
II
U
U
ft
n
n
u
u
n
n
5
n
n
n
u
n
n
n
0
n
n
n
n
n
n
n
n
a
ti
ti
n
ii
n
n
n
n
n
n
n
n
n
ii
n
n
n
n
11
11
n
g
II
u
ii
II
II
n
II
U
u
n
8
II
j[
II
II
II
1
u
tl
n
it
n
n
11
u
u
n
«o o ~4 ** %o \o -g- xr oo r* ^H in
|/> \& *? \& LT.K1 ^ SO \O O) *? 0»
ao to at M? co TJ* i«-'r-
T C4 o «* in oo o^io m ^
O O N O% O «-H O
M O m O 0
^"^ x""' ^^
^
oo O eo w o "tf* in rt c*j A ' -< to Tf
>o *»OLn>0o«-io
* i tn o o
w * ^
h» to \o i/) ND^T ro^to tn »~« i/j
"W Ch O \D " "OO^O v-'is.
^ FH iH -^ 00 r-t O *»-"«» IA ^
* -O oir-'voio
W O *0 O O
fp-l * *
\m^ **** Vwrf
Is* ^"^ ^ ^\ /\ cO f~* ^\
Ov o en O) *"< o r**1 ^ ^ o i*' ^n
\O "»t «H t^ ChOI p-4\£)O r^^-CJl
,i-t. *H. ift ^5 * »OOO * * ' h*
r-s. i/> o "v ~-> oo r^ ^-'tn M ««»
* «o »\o»-ih-o .*"* o
» 1-4 \D O O
«_/ * *
1/1 IS> (O O OOl TTlAtn
H
C 0
C) M
_C a>.
£ H~
1C
in
4^
O
§?
g" U
_
i-«C
« S
a
*> w
14 IN
^ o-
Irf ^
5 on
«H g
, 0)
11
rt DQ
V §
1-1
u. S
g |
* 4X
O
< C$ tf)
rt Cw
« O
o u
A^i
B 6
gs
-°
78
-------
using RAS and H-H, out of 27 sectors. Comparing this to the fully one-half
of our 50 sectors estimated as "acceptable," this suggests a substantial
improvement in our estimation of the Washington economy using the national
table.
Hence, our results indicated that the coefficients displayed a definite
increase in accuracy when location quotients were applied to the national
table, particularly when output rather than employment was used. For
example, from NAIVE to EMPLQ to OUTPLQ, the mean percentage error decreased
from 314% to 162% to 135%. However, these raw figures taken by themselves
still do not indicate very close estimation. In fact the correlation
coefficient ranged, respectively, from .649 to .717 to .733.
For the test of whether strict application of location quotients
(multiplying across rows by LQ-| regardless of LQ-| > 1), statistics indicated
poorer estimates of the coefficients resulted ("strict" column in Table
3-3).
Estimates of Direct and Indirect Coefficients --
The practical test of the estimated A-matrix is its predictive value,
obtaining gross outputs using A and the final demands vector: X = (I-A)-lF,
where (I-A)~1 is the matrix of direct and indirect coefficients.
The A-matrices found by WASH, NAIVE, EMPLQ, and OUTPLQ were used to
derive the tables of direct and indirect coefficients, (I-A)"' (see Table
3-4).
Again significant improvement was evidenced in progressing from NAIVE
to EMPLQ to OUTPLQ. For example, mean absolute difference, reduced by a
factor of about two-thirds in moving from NAIVE to OUTPLQ, measured .0056,
on the same order as its value for the A-matrix (.0043). However, the very
high correlation coefficients -- .976, .991, and .993 -- are likely due to
the extremes of coefficient values, concentrating either at small fractions
or greater than one.on the matrix's diagonal, allowing a least squares
regression to explain much of the variation about the mean. Also, even the
model yielding the best results, OUTPLQ, gives only MSI = .4924, a
statistic which considers each pair of elements separately and so should
not be exaggerated by the extreme elements along the diagonals. The results
of comparing the known values to those derived by the various methods of
estimation again indicate improvement with progression to the output-loca-
tion quotient model. However, absolute differences appear to be too high to
confidently accept the models' ability to depict accurately each of the
regional economy's interindustry relationships.
Estimates of Total Gross Outputs --
Next, the gross outputs were predicted by each estimate of the A matrix
using the assumption of all final demand (including exports) as given. In
Table 3-5, results are very impressive in the prediction of the total
gross outputs vector. Dramatic improvement was seen in movement from the
NAIVE to the EMPLQ to the OUTPLQ, but here, unlike the case of the coeffi-
cient matrix comparisons, the measures taken in absolute as well as
79
U S EPA Headauarters Library
I,',*; cock- WT
1200 Pith^vu^H; t««?iue NW
oiuri. UL 20460
202-566-0556
-------
(J
o
S"
a:
CD
O
O
oc
o:
«t
Q.
II
u
11
n
n
n
ii
n
n
u
n -»
Of It 00 P- CM
J ii CM in to
cu ri O) oo in
H 11 o* o^ o
S3 11 O
O II
II
II
n
it
it
n
tl
0
II
II
II
II
II
II
II
II
II
1)
II
II
II
u
O" II 00 00 l/>
-J II 0 -t «3
fi» II G\ CO I"*
E ii en 01 o
PJ n o
n
n
n
n
n
n
n
n
n
u
i^
IL
n
n
u
n
-H II
C II
O II
H II
OS II
Z II O "1 Oi
' ii >o
u) n o> o in
> II r-l
l-l 11 O
«£ H
Z |J
n
n
n
n
0 II
H II
4-1 II
«l II
H II
4-> II
rt fl Q
4J « t-J <
w o o: e: s:
/^^ ^^
r- »» 'J- in <-i
T \D O> O ^ CM
CM in i .«! h- f 1
ut r- o CM o o i
^ oocn r-»'*-fo ^^-*i
in M oo o -^
CM -H Ol ^
'"'
K5 "T O t O
03 ci ** to in tn
00 " M f~* K) O /"% 1
rO Ml"- *fMO «OI
H*X »-4 r- oo v°-/ o Is- *^ i
CM m oo o **
i* CM K>
in CM v_/
y^\ y^s
\o in in o in
Tf CO CM SO CO ^4
o o> /->en %£><->!
o -w en CM o o i
Kl OlOO CM^-^in ( ^^ 1
M O> M CM CM
V) Ol 00
*-l c* fM ^^-^
fO * ^
^ cy'oV, * 'o /-.
_-O co^oe o h oo
l-l UJ I-HI>> U. 0) >
CO CUIA S^NCd Sliced
*g; yr ^r t_> ^-^^^ I-H v-" >
J
c
V
i
1-4
a
4-1
^4
o
s
IA
-H M
O C
M «
f;
Bj JC
- 3
P *M
H O
O (U
*O DO
C n)
3 H
e
£ 5
* Ji
srs
t-i-i
8|
g
cd 4-^
e
s .5
o
*<$
4^ M
g O
00
J
-------
z
UJ
5"
3
VI
«E
O
z
£
Ul
o
g
c
JT COMPUTED WITH
Washington Tabli
s: s
t_ p
S ft
a js
LU *»
(\L AND ESTIMAT
el Compared wi
=> T>
§ *
UJ
UJ
ABLE 3-5. COMPARISON BETI
I>H
f-l
n
& *O i-< j -o
p.
tn
n
n to
n tn
n n
II O =0
11 *-' (N tO 00 tO f>»
II "9 OQ\DtO*3-«4- t^in
C? 110100 oor^iv^C't}- ooto
"^ii t** oo Oicn>Oi-fknto*H-H o
SEn o^^>-4tnaoto to
uuii'n o r-jGoor*- ^*-
n -< o <* oo
« tn to -< ^ IA ~i
-w .12 -o^-«»**\oin ^CM
EC u)O >o rO ^t (N* sO (N \o
2^ <7»r^ o>cnS S^SSr^S ^
< ' » 00 \O 00
i-* *" , y ^3 »fl O »O
[£. *J C lO "*
1 f-t >< (ft » . * . O M i C *°
&vtn tu 4> *j j-i *o or oo*
t> P ' . PCW4-'~O1.C4->"*
^» C IH *«o-H*WC CW^W^"1
u:*it3 o i i *4-i - < *J m .M oo
o > a M Q c *o ' ) o j^ ^
« -M W .. 0 -0 -0 0 4> ^ £ In ^
^* h +J H4J+J4J .£ 4J ^ 3
cc3tn uJrtrt3ctuio3 cu
4> *- « ES'-iO-'-IVQtM4J
O U .rj t> *J O 5
H|[ f-i 0 O ,13*J*-'«4JSC3 "
*>« O-rH - , (SJ 1 4-> tf) «^ rt OOl/l
(A &. LO (H Oil tfl C/) tlj LJ < E U-. r-. JH i-1
^ rt ^t ,o a> tn o a
*-*"KCt>- CO tMtfjijP
n3 ii r: rt c . rt MH E ^ o u ^
^ 1 ^ » lj 0^3 .^SSo
*
o
t-4
B
CO
3
0.
4J
a
(O
81
-------
comparative terms indicate good estimates. Mean percentage error (and its
standard deviation) drops from 161.95% (239.28%) to 32.79% (53.59%) to
16.63% (17.17%), from the NAIVE to OUTPLQ estimates, an improvement of
almost 90%. The mean similarity index of OUTPLQ's estimates reaches .9256
on a scale of zero to one. Probably the most significant result is the fit
of the regression line for the estimated output as a function of the
actual. Ideally, we wish R2 = 1, slope of line = 1, and y-intercept = 0,
to indicate perfect estimation. The NAIVE model predicts gross outputs
with R2 = .9314, but its estimated y-intercept is 149 and slope is 1.126,
resulting in a consistent overestimate of values, with a standard error of
171. An error of 41% is present between actual and predicted total gross
output. The EMPLQ dramatically improves R2 to .9768, standard error of
estimate of 101, and estimated y-intercept of 21.5. The slope again
approaches 1, and total output for the 50-sector is overstated by only 22%.
All other statistics show great improvement. Finally, the OUTPLQ model
results in a mean percentage error of only 16.63%. A value of RZ = .9882
is achieved by a line of y-intercept -11.5 and slope 1.15. The estimated
total output is only 12% from the actual value for the 50-sectors, and all
other measures are similarly dramatically encouraging.6 These results
indicate that, despite some inaccuracy in the coefficients matrices
estimated with location quotients, these estimates can provide quite
accurate estimates of total gross outputs X.
Estimates of Other Portions of the Table --
While the derivation of the A and (I-A)"1 matrices and the estimated
total gross outputs vector which follows directly were the primary goal
of our study, other portions of the regional input-output table can be
derived as discussed in Section IV of this chapter. The ability of the
A-matrix to estimate the vector of intermediate outputs was tested, and
exports and imports by sector were derived.
Intermediate outputs were calculated as AX, where A was the estimated
matrix and X was the true vector of gross outputs. The results in Table
3-6 indicate continuous improvement as we progress from NAIVE to EMPLQ to
the OUTPLQ model, except for a small dip in R2. in fact, the improvement
Because all three estimates overstate total output while a regression line
indicates high correlation between actual and estimated values, the possi-
bility of some type of bias in the procedure was raised. However, no
evidence to explain such a bias could be determined. The NAIVE model
clearly must overstate predicted output because all national coefficients
must be greater than or equal to those derived using location quotients.
The location quotient adjustments (reductions) of the national coefficients
can be expected to reduce predicted gross output. No direct or implicit
causes of bias could be determined in the estimation procedure, and in
fact, the evidence provided'by only the two cases is probably not a valid
one upon which to conclude the existence of bias. Further, the close fit
of a least squares regression line coupled with a slight overstatement of
estimated values suggests that a possible adjustment factor could be easily
used to adjust other estimates derived from the model with little additional
computation if such cases were to occur consistently.
82
-------
o
I
UJ
vD
00 rH O> f»J *C
(N OO O*l ^ t^
rt O IS. Ol ,H
fs in ^t C4 o m
I r-l O CTl
O 10 in
to
in
\o
o
(M
o
in
to
vO
in
oo en
rsi MS
00 1
^ r-
g. -
rt to
O 00
in
Ol
in
to
o>
00
g
It
oc
2
5
a)
o>
«
«
*J
§
8
O
o-
X
c
-0
e
in
u
^
rt
3
a
to
a
o
.
oil
tu
o
4J
en
«>
g
O
4>
V
M
in
15
*->
g
H
VI
Cl
Ul
41
HH
-------
is quite substantial, with Mean Percentage Error decreasing by over a factor
of two-thirds. However, even the value of 156.14% MPE for OUTPLQ must be
considered too high to be acceptable. Mean Absolute Difference decreases by
over one-half, and the overestimate of total intermediate output decreases
by almost 70%, from 4958.4 to 1503.4 x $10& in moving from the NAIVE to the
OUTPLQ model. But again, even this OUTPLQ estimate overstates actual inter-
mediate output by almost 40%, a sizable error. This set of statistics thus
confirms our previous findings on the A-matrix coefficients in showing a
substantial degree of discrepancy in interindustry relationships. The good
predictive performance of our estimates of (I-A)-l in estimating total out-
puts, however, remains as an encouraging sign for our main goal of estimating
regional production by modifying national input-output tables.
As described previously, while (I-A)"^ can be used to determine X, given
final demands F, and AX can be used to find intermediate outputs, exports
and imports can also be found as residuals using the precise interpretation
of location quotients. (Exports are total less intermediate output less
local final demand, if LO^ >_ 1, and imports are zero; or exports are zero
and imports are national technology requirements not met by local
coefficients in the case of LQ. < 1.)
As can be seen in Table 3-7, exports calculated as residuals in all
three models are highly correlated {R2 = .8537, .9146, .9294 for NAIVE,
EMPLQ, and OUTPLQ, respectively), but they greatly underestimate the true
values.
2
NAIVE gives R = 0 for imports, because its implicit LQ. = 1 presumes
that zero imports are required- (see Table 3-7). However, once again
EMPLQ (R2 = .9308) and OUTPLQ {.8364} are highly correlated with the true
Washington values. They too are understated compared to the true values.
Replacing the given Washington exports component of final demand with
those calculated above, and recalculating X = (I-A)-lp, the NAIVE model, as
expected, now perfectly estimates X, because the calculation of exports for
each row (since LQ^ = 1, i = 1, ..., 50} as a residual balances the table
so X = (I-A)~ F is true by definition. The EMPLQ and OUTPLQ estimates now
underestimate X (by approximately the same amount previously overestimated),
as all exports are set to zero in rows with LQ. < 1.
Modifications: Balancing the Tables
Balanced Tables with Zero Exports in Rows with LQ. < 1 --
The balancing procedure, described earlier, was performed to achieve
consistency of results by removing negative exports and balancing where
rows of LQ. < 1. As explained previously in Section TV, this procedure, by
either ]
1) computing exports where all intraregional interindustry re-
quirements (national technology) are met, as are the given
final demand and total output; or
84
-------
I
>-
UJ
^>
o:
a
ui
co
O
IS)
&
I
LU
CO
a.
(j
c
^4
£
I
M
it
II
Ii
II
tl
M
M
II
A« II CN
£3 I) »O
On in
II
*-n
II
II
1 (
Ii
!?!! f?
ft 1 *
£ II CO
W H *T
II
II
"" n
1 1
ii
if
it
U ii
p* 1
S£ I O
I
I O
1 1
»J I -4
M vO
II
II
1 1
II
M CM
Ed f* CO
> |> Os
HI! rH
§ M *"*
1
*
1
[
1
1
1
1
II
1
U I
«H t
JJ |I
n ii
CO CO *H U*l 1^
£ £ ££o!££tn£ co
00 -d- r-l rs( O "i?
f4 ^ f^l m N CM
\O t-<
(no O o i i i r*
* i i i * i i
N-' © o o in r i
cr*
m> t*)
IH 1 OS W
i-l i-l CS
vw/ r-|
CM O
o% iDtMiosiN,in ,( i i
O ONO\rHr^OSCOI** CO
<* oo m in o CM
0s ^^ ^ P^ vD Os 00
in i r-* sf"
«^ ^ tn
tO CM
o !"* os. co in
n CD JJ « r
4) ItfA 3 O
fc 0) D en
q «-t u w M
g v " S '*
o a c w ^
u tJ o M j:
03 Irf f-* 1 DC W
« W CO > »< Cl
01 01 CJ
KJ ... 5 e
5 JJ AJ 0>
^ M evI'Mcntfl m «MM
n et\ ei\ v) in u> a o 01
O 0* O *M
HI B Q fe 6 >w
»l H -H 2 E 5 -H
H SI -) £ M co a
o
in
in
0
o
«n
(O
CN
CM
0
O
*
m
o
CM
n R
0 4J JJ
H H M
0 O
« a. a.
§ w &
u
a 5 3
«l U tl
u u u
U < g
"sll
vt- u
o> -d
C M 0)
H 0> *J
e u
n "H
^ oi *o
i-4 0)
3 5 a.
c. «
4J n r-i
§£ <
« ^ o
85
-------
2) with given final demand and total gross output met, distri-
buting the remainder of production across the row, with
exports as zero,
must result in the EMPLQ, OUTPLQ, and NAIVE models becoming identical. As a
consequence, the results in Table 3-8 hold for all three models in comparison
to the true WASH table. The revised A and (I-A)-l matrices can be seen to
be a clear improvement over the originally-derived NAIVE model. Some
statistics improve, some worsen compared to the original EMPLQ, but the
balanced tables are still inferior in most tests to the accuracy of the
original OUTPLQ. Exports were better estimated than before balancing where
negative exports were permitted. Because of the balancing, X must again be
perfectly estimated using the revised A when final demands include inferred
exports. With all final demands (including exports) used as given in the
survey-based table, estimated output, although overstated, has very high
correlation with the actual values.
Again, however, it is to be remembered that the adjustments used in
balancing nullify modifications made by the location quotients and distort
the initially derived estimated A-matrices.
Balanced Tables Allowing Positive Exports in Rows with LQ-, < 1 --
As explained in section IV, the originally-derived tables were also
balanced in a similar way except that positive exports were allowed for rows
with LQ. < 1. As a result, the NAIVE model must be unchanged. The minor
alterations to EMPLQ and OUTPLQ did provide a small" measure of improvement
over the other balancing procedure, and a slight improvement over the
original tables, but at the expense of the added manipulations to recreate a
full given table rather than the simple location quotient estimation of
the input-output matrix.
Summary of Results of Balancing --
As we have tried to emphasize, our main goal was to estimate the A-
matrix (and therefore (I-A)-l and total gross outputs X, given final
demands F). The balancing procedure, while an interesting modification,
nullifies the effect of location quotients in its attempt to estimate a
complete consistent table, and complicates our simple short-cut method. In
the estimation of the A and (I-A)~l matrices, the improvement due to the
use of the balancing procedure was rather small compared to the initial
estimates -based on location quotients. However, it does permit estimates
of other portions of a regional table, in cases where such info?rmation is
desired.
VII. CONCLUSIONS AND SUMMARY
In this description of our effort to examine the ability of two types
of location quotients to adjust national input-output tables to reflect
intraregional interindustry transactions, we have attempted to highlight
our efforts at improving the methods used in previous studies.
86
-------
10
Ul
£3
LLj
cc
a.
(S
0
CO
i Is
» -g
it* (2
S !!
II
II
II
II
II
II
II
VI II
4J H II
^ 0 II
0 4J »
C, O ||
x a ii
u: > u
ii
ii
n
n
a o f-
4^ 4J ||
K U II
»»- O ii
»r W |l
I* ii
C *j ii
" g,i!
C -4-> ii
M* 3 it
0 H
II
IT
li
!'
11
J" !!
1 II
i II
M
n "
]]
it
H ||
It
< fl
1 ft
*"-* II
|[
Jl
»
H
j]
H
II
-. II
< tl
II
11
11
]1
11
11
II
II
rH ||
U II
10 II
.H M
4-1 IP
cd H
4-> II
V) II
I/) n
M m *o PM ^ o*
tor- t^iyi^v ^*i«>
«-HK> O^Cit*»^± i/)K) 1,0
o^ co f^- to oo r^ o ff> *H ^^
tn oo c* o* ^o c*
^-» lO
41
r-» ^) o ^o '10 o
Srt to r*> o un ^N \D
r- *Hin(MQi r^i-ioi
O>O» or-fOr^iftOCM «
\O K) ^ %o f1*- r»-
1 *~r CM
«-H
CO O O CM fM h*
Ort Oiaoooo ,-,^
lOrO oo tn *vo O f-i/>^-
r*-m 1/ior^t^.^-oift oo
OO ** I** *^ N£> P- i K)
^ Kl ««T \ \& QO «N
CM V3 (s.
' ' 00
lO
U) O VD Is* O l/>
1NI/1 C00>«frK> r »fO
O>CO O «-H O 00 COO) **
Chen iitfHi'i!roo^"f"Hio o
^ ^ O >D G% r* l/l
«> CM sO ?S (M Ol
S / 1x114
i-H
^.
cs >o M eo h*
»~* P* Ol \O /-*> ' f O
O* 00 l/> O -! Ol Kl
01 a* ^ o -sr o
. . » . (S|
O) 00 CM
i^»
1/1
to cn tn oo
eo in o> ^ ^-, o
r-H ^H 00 O O O TT
Is- l/> ^ O *t »-l 00
* * r
^ if> CO
in (N r*.
»H ^O
jj
CQ (M^tHr-Hi -a .^
C3 OS CC U3O tfl X I-H O t-J. E-CStJ
'3 & E§E^co(S'^-'
IA
o
»o
0\
r*
^<
to
ul
r*.
in
M
in
. . .. i
a
(N
r*-
(HI
1A
(N
i~4
M
'O
\fy
-)
^
NO
t-H
N
t
1
^B
r^ o to
vO 00
to »»
in cn
O> tD
rH 00
CN «-H
1 |
1 |
1 |
t 1
f 1
-JJ-
« ,^-g
E-l rH
C8 X
n 3 w
« w >
2U ^
« 3
U '' «
.
3
l/>
(]^
^
8 £
a, S
*J
^J ^j
fl> CT
t.i 3
> O. i/>
5 3 "~
,Z O u.
m .a o
87
-------
First, the input-output matrices estimated using location quotients
were shown to be a significant improvement over the naive assumption of
national interindustry relationships within the region.
Second, location quotients based on output were shown to provide better
adjustment factors for the national table than did those in which employment
was the measure of importance of an industry.
Third, the application of a large number of previously-used tests of
the accuracy of the estimated table allowed some degree.-of comparison to be
made with previous tests. In this way, for example, the'effects of a -large
number of sectors {and hence more homogeneous industry) and the use of out-
put as opposed to employment in computation of location quotients could be
examined. It was found in virtually all tests that output data provided
better results than employment. And, while in general the results of the
present study indicated improved estimation ability over previous studies,
contrary conclusions based on some comparisons or other tests may'be ..
explained by the increased cumulative error in the summary comparative
statistics due simply to the increase in data points Tnumoer of sectors,
squared).
Finally, in the desire to estimate accurately regional outputs using a
short-cut method to regionalize national input-output tables, the most
important goal is to be able to/accurately estimate regional putputs. Our
application of location quotients achieved quite good results using the
given Washington final demand vector.
By removing some handicaps suffered by other studies, and achieving
improved results, it is tempting to suggest that further improvements may be
attained if the problems we could not remove were eventually eliminated.
Specifically, definitional differences in sectors and conventions between
Washington and national.tables may account for additional error. The
transfers of secondary products and imports within the national table in
particular could be expected to influence the effect of location quotient
adjustments. Further,-the consistent overstatement of all three estimates
suggest some possible bias. It may be feasible to apply some sort of
adjustment factors to account for this bias. While these problems have not
yet been solved, the success achieved in the present study suggests that
location quotients can indeed be viable means of regionalizing national
input-output tables.
When an individual state or region wants to explore the effects of
national economic trends, environmental standards or energy development on
the economy and environment of its state or region, the preparation of a
state or regional 1-0 table by the location quotient method may be
warranted. While the resulting 1-0 table may not accurately reflect a
survey-based regional table, it may be sufficiently close to permit approxi-
mate estimates of the direct and indirect effects of exogeneous developments
on the regional economy. Thus the location quotient method may provide an
additional tool that allows a first order estimate of effects that an
88
-------
elaborate interindustry, interregional model (along with current and pro-
jected data bases) could provide if it were available. In other words,
location quotients would be, in our view, an acceptable method for regional
analysis from the perspective of a regional or state administration, until
such time as a satisfactory regional model becomes available.
VIII. APPENDIX: A REVIEW OF OTHER METHODS OF REGIONALIZING NATIONAL
INPUT-OUTPUT TABLES
Short-cut methods other than the use of location quotients have been
employed to derive estimates of regional input-output tables from national
tables. A survey of those methods is presented below.
Commodity-Balance Method or Supply-Demand Pool Technique (Morrison and
Smith, 1974T
The commodity balance method, or "Supply-Demand Pool Technique" (SDP)
uses the balance b. = xR - d-, where xR = local output of good i, and di =
local requirements of good i (d. = z x1? a }. If b. < 0,
then local supply covers local demand, and j J ""J ] the
national technical coefficients are used in row i of the regional trade
coefficients matrix. If b. > 0, then the national technical coefficients
are reduced by a factor n of xR / d.. Finally, using these results,
imports or exports are computed. ] 1
RAS Iterative Technique
The RAS iterative technique (RAS) was developed to project future
input-output tables from a base period matrix, but can be applied to project
regional tables from national tables. Survey data jjs. required to obtain
intermediate input and output data by sector. Regional intermediate output,
intermediate input, and gross output vectors, u, v, and x constrain input-
output coefficients-matrix A iteratively in the following manner, where
i = the identity vector and a circumflex indicates a diagonalized vector:
A x, to first approximate regional intermediate
1 output with national coefficients;
Jl
A = u u A , to satisfy row constraints;
v = i x A , to estimate intermediate inputs;
-
A = A v v
to satisfy column constraints;
and the new estimate of A is used to repeat the process until convergence is
reached and both sets of constraints are fulfilled (Morrison and Smith, 1974)
89
-------
Czamanski and Malizia'--
Czamanski and Malizia (1969) used the RAS method in this manner to
derive a regional input-output table for Washington state (1963) for which a
survey-based table was available. In order to test which sectors could be
satisfactorily estimated by RAS and which require surveys, the authors tested
six cases of various sector aggregations and relative price adjustments.
Convergence in the computations was so rapid that the authors concluded that
no iterations need be made after the first one.
Using as measures of deviation between the estimated and survey-based
models standard deviation and mean percentage error of the coefficients and
information theory "I" values^, the results were examined. In all six cases,
there were large deviations between real and estimated values. Reducing the
number of sectors from 43 to 36 by grouping the tertiary sectors yielded
some improvement, as Czamanski and Malizia had hypothesized that major errors
could have resulted within them due to differences in routing practices.
The wide range of activities in the primary sectors was also expected to
cause poor results. In fact, the errors in these sectors were more than two
standard deviations larger than the mean errors of the entire matrix. High
regional specialization was the third factor predicted to cause errors. In
fact, six of the 36 sectors had location quotients (using employment) of 2.0
or more. Regional specialization was observed to appear to imply a different
technology than the national average, with an exception of the aerospace
industry of which Washington has a large proportion of the national total.
The measures of error, mean percentage error (and its standard
deviation), ranged from 58.65 (and 2.211) to 80.81 (and 6.314), while "I"
values were 6.279 to 54.169. Only when the 36-sector matrix was reduced by
four primary industries and four regionally specialized industries to 28
sectors, did the "I" value reduce to 0.779, which the authors claim is
"certainly an acceptable level by any standard" (Czamanski and Malizia, 1969,
p. 73). From this they conclude that with considerable adjustments of
national input-output tables, "acceptable results can be achieved by the
methods tried on the Washington state table." However, the mean percentage
error was 38.93 (with standard deviation 2.160). Miernyk, in commenting on
this paper, notes that the error is "uncomfortably high" (Miernyk, 1969,
p. 82). He also points out that in steps performed to reach this level of
accuracy, six sectors were dropped (which were equivalent to twelve in the
Washington table), and ten sectors were aggregated into two, transforming a
54 sector Washington table into 34 sectors. This means that by Czamanski
and Malizia's proposals, field surveys would be required for 20 sectors
7 *
Where a.. = true coefficient and a.. = estimated coefficient,
UA
*
A , -
T
this expression is undefined when a.. = 0 and a
be arbitrarily resolved in 1J "
0, a problem which must
computations.
90
-------
(accounting for 13% of Intermediate sales, 12% of final demand, and 28% of
gross output). Further, Miernyk points out that the sectors in the combined
coefficients- are important analytically, and that if surveys are needed for
28% of output, the other key sectors should be identified and surveyed as
well. Miernyk's conclusion is that "Czamanski and Maliz.ia performed a
signal service by measuring the large errors that can result from short-cut
methods. They have provided the empirical evidence which has been missing
up to now to support a priori arguments that adjusted national coefficients
do not reflect true structural differences among regions. (Miernyk, 1960,
p. 82).
McHenamin and Haring --
McMenamin and Haring (1974) describe other applications of the RAS
method. The Belgian national economy was projected from 1953 to 1959 by
Paelinck and Waelbroeck. Bates and Bacharach explained errors from three
causes. Sectors were highly aggregated, a fault of the study and not RAS.
(It is interesting to note that Czamanski and Malizia were forced to aggre-
gate some sectors to improve their results.) A second cause, attributable
to RAS, is the effect of uniform substitution assumed over all sectors upon
a change in the requirements of one. The third problem with RAS is the
ripple effect -- an error in one estimate generates errors in the rest of
the table. Paelinck and Waelbroeck obtained a more accurate table by
removing cells seen to be troublesome in advance, recomputing with RAS, and
inserting the survey values for the 1959 table. McMenamin and Haring point
out several facts about the Belgian test of RAS. Few changes occurred in
the coefficients between 1953 and 1959. Of the 270 nonzero coefficients,
238 had an absolute difference of less than .005 between the estimated and
survey tables. The actual values for total gross output, total intermediate
input, and total intermediate output from the 1959 table were used in the
iterations, avoiding testing the effect of the alternative use of indepen-
dent estimates. Finally, some cells were replaced by 1959 survey values
rather than independent estimates. Thus, the test represents an upper
limit on the method's accuracy.
Haring and McMenamin constructed a regional model of Southern California
using RAS, but it could not be tested, due to the lack of a survey table.
"McMenamin has partially tested.the method on Washington State data and
found the results to be marginal at best " (McMenamin and Haring, 1974,
p_._l93_). They also devised an alternate iterative procedure (H-M) but
"l_ i_/t is only applicable to the problem of estimating a regional table
at one date based on a similar table from an earlier point in time." (1974,
p. 194) This was tested, along with the RAS method and the naive method
(no change from the original table) to estimate the 1967 Washington table
from that of 1963. No adjustments were made for price changes. Results
showed large errors in absolute percent deviation. By columnwise chi-square
8
*2=?
/ a.
this expression is also undefined when a. . = 0 and a. .
resolved arbitrarily. 1J 1J
Q and must be
91
.
-------
calculations for coefficients, of the 27 sectors, only 10 in the naive model
and 12 in the RAS and H-M were acceptable at the .05 level. Comparing their
results to those of Schaffer and Chu, import averages and Types I and II
multipliers were much better than Schaffer and Chu's location quotient,
cross-industry quotient, supply-demand pool, and iterative procedure, but
the authors are careful to point out that they began with a full-survey
regional table, while Schaffer and Chu used a national table. McMenamin and
Haring conclude their paper by stating that none of the methods they tested
provide dramatic improvement over the naive model, that accuracy depends on
the amount of structural change, and that it is worthwhile'to estimate ~
future regional tables if previous survey-based tables are available.
Malizia and Bond
Malizia and Bond (1974) used the RAS method to update the 1963
Washington State table to 1967. They point out that the method should be
restricted to the case where only one table exists, as Lecomber has shown,
because other methods exploit the additional information of more tables.
Basically, state Malizia and Bond, RAS deals with changes occurring when
relative intersectoral relationships change between intermediate demand and
final demand and between intermediate inputs and primary inputs. Its main
assumption is that, given an initial change, the degree of change is pro-
portional for each row and column. Their method of procedure was as follows.
Making no price adjustments, interindustry flows were adjusted, using the
1963 table, so that each row sum gave 1967 intermediate purchases. Flows
were then adjusted by 1967 intermediate sales and this process was repeated
until residual errors were less than three percent (within five rounds).
The results of their test, comparing the 1967 table estimated from the
1963 survey table against the 1967 survey table, indicated high average
coefficient error (104.7% - 132.7%) for various sector configurations, and
relatively high information theory I-values (1.0 - 7.9). Further, they
found large standard deviations (209.7% - 319.9%) and right skewness in the
distributions. They state that "/_ t_/hese data support the conclusion
that the RAS method, used without additional exogenous information on
interindustry flows for the projection year, is not powerful enough to
generate satisfactory forecasts of interindustry coefficients." (p. 360)
Comparing other aspects, Malizia and Bond found that forecasts of regional
coefficients were more accurate than those of technical coefficients,
aggregated tables were more accurate than disaggregated tables, and no
relationship was found between degree of error by sector and per capita GNP,
contrary to the claim of Chenery and Taylor, or between degree of error by
sector and the location of the sector in a triangulated table.
Malizia and Bond also tested the predictive accuracy after aggregating
to 18 sectors, using both RAS and actual tables. For total output and
intermediate demand, "the predictions were quite correct;" for value-added,
predictions were correct for the total and a few sectors. "However, in
each case the synthetic table gave predictions that matched the accuracy of
the actual table." (p. 362) Malizia and Bond conclude that the RAS method
produces "large and theoretically unsystematic coefficient errors" so it
cannot be recommended for forecasts of interindustry coefficients. "How-
92
-------
ever, the method appears to be a useful way to adjust interindustry coeffi-
cients when short-term predictions of main aggregates are needed for
regional planning and analysis." (p. 362)
Schaffer and Chu's Iterative Procedure
Schaffer and Chu (1964) developed another iterative procedure, the
Regional Input-Output Table (R-I-O-T) Simulator. After using estimated
regional output x- for each industry to compute required inputs r. . = x,a..
and estimating J regional final demand as a proportion of J J 1>3
national final demand, regional sales for each industry are allocated
initially using the national sales-distribution pattern. Then local sales
are reallocated row-by-row using an iterative process until regional pro-
duction and consumption are satisfied as much as possible. Exports are
calculated as any positive differences between regional output and regionally
required sales for an industry. The regional transactions matrix contains
the positive differences between regional production requirements and
available local output for each cell.
Comparisons of the Nonsurvey Methods
Schaffer and Chu --
Schaffer and Chu (1969| tested LQ, POLQ, CIQ, Pool technique, and
R-I-O-T, concluding that "/_ u_/sing the 1958 transactions table for the
United States and survey-determined industry outputs as program inputs,
these simulations are comparable to the 1963 Washington input-output table
and provide a limited test of acceptability." (p. 94) However, both
Washington and national tables were aggregated to 23 industries before any
calculations were performed. In addition, balancing for exports was required
for LQ and POLQ. By the chi-square tests, LQ and CIQ were most successful,
followed by R-I-O-T and the pool method.9 The iterative procedure yielded
the closest estimates of total imports into the state, while CIQ gave the
best estimates of imports by industries alone and of exports. Imports to
industries and exports are lowest for the iterative procedure, but since it
attempts to allocate maximum possible local output to local trade, low esti-
mates were expected. Mean income multipliers were high for all methods: for
Type I multipliers, LQ gave results 21%, higher, CIQ 38% higher; for Type II,
LQ gave 47% higher, iterative 79% higher.
Schaffer and Chu conclude (p. 96) that the methods may be useful as
supplements to survey studies, "/_"T>_7ut it seems that, at the moment, there
is still no acceptable substitute for a good survey-based study." Final
notes based on their acknowledgedly limited test, include the observations
that they avoided grossly mechanical computation and that their aggregatior
y
At a = .05, the number of acceptable sectors was 6, 6, 5, and 4, respec-
tively. (See Section V of this chapter.)
93
-------
procedure ignored the Washington product mix. On an optimistic note they
find that "it is surprising to us that our techniques yielded acceptable
regional production coefficients for over 25 per cent of the industry
categories," and if further tests yield evidence of consistent differences,
they can be used as an adjusting factor.
Morrison and Smith --
In a more comprehensive evaluation, Morrison and Smith (1974) estimated
the 1968 Peterborough table from the national table using SLQ, POLQ, CIQ,
RND, CMOD, RMOD, SDP, and RAS. The estimates were compared with a survey-
based table by means of five tests: mean absolute difference, correlation
coefficient, mean similarity index,10 information content, and chi-square.
Their results are presented .qualitatively by the six techniques' ranking in
each test. The RAS technique was best in all cases, with a coefficient of
correlation of 0.501. This was not considered surprising, as the RAS
method does use some survey material. The SLQ (employment rather than
output was used) was second best, resulting as both the simplest and best of
the totally nonsurvey approaches (but coefficient of correlation only
equalled 0.160). Considering the other methods, POLQ was inferior to its
unmodified version (SLQ) in all cases, but the modifications made to CIQ
and RND (yielding CMOD and RMOD) did improve them. SDP gave the second
best coefficient of correlation, but ranked poorly in the other tests.
CIQ yielded the poorest results. A modification which resulted in estimates
closer to the survey^based table was one in which zeroes were inserted in
all principal diagonals of manufacturing and construction Industries, as
such interindustry_transactions were theorized to be small. Morrison and
Smith conclude, "/_o_/n the present evidence at least it would seem that
nonsurvey methods can only produce an approximation of a full survey based
table, and the adequacy of this approximation must depend on the objectives
of the particular exercise." (p. 13)
10
* *
$.. = 1 - |a.. - a. .| / (a.. + a. . ),
94
-------
CHAPTER 4
A REVIEW OF ALTERNATIVE REGIONAL
AND INTERREGIONAL MODELS
Following the rapid rise In energy costs, there has been both a public
and private effort to locate new sources of energy. New energy development
programs will obviously have an impact on the levels of economic activity
and environmental quality, and hence should be evaluated in order to obtain
the maximum benefit at the lowest cost from any program.
National models, either input-output or macro-econometric models, have
been used to explain and project the effect of rising energy prices and
energy supply constraints on the national economy.
Recently, a number of regional models have been developed to explain and
project regional economic activity. Due to the complexity of the problem,
and the not too infrequent lack of reliable data, most regional models are
"linked" in some fashion to a national model. Examples of these models are
the Shift-Share Models, Location-Agglomeration Models, Regional and Inter-
regional Input-Output Models, and Regional Macro-Econometric Models. Each of
these models was designed to explain and project regional economic activity
given an exogenous change in public expenditure policy. At the original time
of their formulation, however, these models were not designed for the purpose
of analyzing the regional economic and environmental impact of alternative
government energy policies.
In this chapter we will review and examine the different types of
regional projection models. The existing regional forecasting models can be
classified into four relatively distinct categories. These are:
I. The 'Shift-Share1 Approach,
II. The 'Location-Agglomeration' Approach,
III. Regional and Interregional Input-Output Models, and
IV. The Regional Macro-Econometric Models.
Our discussion of the different types of models will be based on an
examination of representative models. As an example of the first model we
shall use the "OBERS Projections of Economic Activity in the U.S." The
Curtis Harris, Multi-Regional, Multi-Industry Forecasting Model will serve
as an example of the second type of model. The third, Polenske's Multi-
Regional Input-Output Model, and fourth, Glickman's Regional Econometric
Model of the Philadelphia SMSA, will be examined respectively.
95
-------
Finally, we will provide a review of two other models describing the
regional economies and energy use that are best described as "other models."
I. 'SHIFT-SHARE' APPROACH
OBERS Projections of Economic Activity in the U.S. (U.S. Water Resources
Council, 1974, pp. 1-35}
Purpose
The OBERS projections were constructed to serve two purposes. These
purposes are: (l) to estimate the economic effects of specified water and
land resource constraints and of alternative programs for the development
and management of the nation's water and land resources at the regional
level; (2) to estimate regional demand for water and related land resources.
Geographic Region
Two geographic units were chosen for the OBERS study as the focal point
for analysis. The BEA Economic Area was chosen in order to analyze the
economic impact of a public investment project, such as water resource
development. The Water Resources Council Water Resources Subarea was
chosen to project agricultural production and land use.
Characteristics of the Model
Long-Run Forecasts -r
The OBERS projections are based on long-run or secular trends and
ignore the cyclical fluctuations which characterize the short-run path of
the economy. The effects in terms of personal income, employment, and popu-
lation changes are used to determine the effect on regional industrial and
agricultural production.
Full Employment
Nationally, reasonably full employment represented by a 4% unemployment
rate is assumed to prevail at the points for which projections are made.
Unemployment is disproportionately distributed regionally. Given this
assumption, an increase in activity by a primary user of the proposed
development project requires at least a partial offsetting reduction, or
foregone expansion, elsewhere.
Regional Migration --
Workers will migrate to areas of economic opportunity and away from
declining or slow growth areas. Consequently, regional earnings per worker,
income per capita, and regional employment/population ratios will tend to
move toward the national norm.
Technological Progress
Continued technological progress and capital accumulation will support
a growth in private output per man hour of 2.9% annually.
96
-------
Regional Export Industries --
Most factors that have influenced historical shifts in regional 'export'
industry location will continue into the future with varying degrees of
intensity.
New Products --
New Products will be accommodated within the existing industrial
classification system.
Variables Modeled into the OBERS Projections and Data Requirements
The effect on regional industrial and agricultural production in the
OBERS system is calculated in terms of personal income, employment, and
population changes.
Personal Income and Earnings in Constant Dollars --
Since there is no direct counterpart to GNP at the regional level, the
personal income series and its earnings of persons component were used as an
alternative to GNP. According to the OBERS projections, Volume 1, the choice
rested on three considerations: 01 personal income has a close and
comparatively constant relationship to GNP; (2) its regional location is
clear; and C3). it could be measured from available data sources, and the
methodology for preparing local area estimates of personal income had already
been developed. Estimates of regional personal income were obtained at the
local level by industry.
Industrial Output in Constant 1958 Dollars --
Projections of national industrial totals were similarly made for 37
SIC groupings. Projections of regional industrial output were obtained
using a 'shift-share1 estimation approach. The methodology for preparing
the economic area projections will be described in the next section.
Employment and Population --
The employment series used in the projections of the national industrial
structure, "persons engaged in production" is not available on a subnational
basis. The only industrially complete employment series available for local
areas is t&at from the decennial census of population. Therefore, the
projected national employment "persons engaged in production" concept was
converted to the census employment concept. Estimates of regional employment
and population were obtained at the local level by industry.
Projection Methodology for Economic Areas
The OBERS projections were made in two major steps. First the national
economy was projected in industrial detail. Secondly, these projected
national totals were distributed regionally in accordance with projected
trends in the regional distribution of economic activities. The projections,
except for the determination of food and fiber, are calculated mainly from
the supply sTde of the economy; therefore, there is an implicit assumption
that sufficient demand will be generated by the private and public sectors in
the production and distribution processes to maintain a full employment
level of economic activity nationally.
97
-------
The methodology for preparing the economic area projections utilizes
four models. The first pertains to basic industries except agriculture and
armed forces. The second projects the residentiary industries, with a
modification of this model used to project property income, transfer payment
and personal contributions to social insurance. The third relates to
agriculture. The fourth model derives population from projected employment.
Basic Industries --
For the basic, or export-oriented, industries, the projection model was
derived from the 'shift-share' technique for regional industrial analysis.
Shift-share analysis is designed to discern regional departures from national
industrial growth rates. In its simplest form, this technique distinguishes
a proportional growth element and a differential growth element between a
region and the nation.
t t x x x-t
E = (E /E ) E + C
ij io io ij ij
where the subscripts i, j refers to the ith industry and jth region, and the
subscript o refers to a summation. Superscripts t, x refer to the projected
time point and the base point respectively. Finally C-j-j* refers to the
difference between the hypothetical level attained at the national growth
rate of the industry over the period x to t and the regional level actually
attained in the industry over the same period.
The C. - term was projected by fitting a curve to each region's percent
of the J national total income and employment (separately) for the
selected years for which data was available. The curve was then extended
into the future and the values of the region's projected percentage in the
target years read from the curve. This technique yields a trend extension
of a region's historic percent of the national total of earnings and employ-
ment in a given industry. It was accomplished by using linear regression
analysis and applying substantial judgment to dispose of erratic or cyclical
observations in the historical time series including apparent discrepancies
between the earnings and employment series.
Residentiary Industries --
Local-service or residentiary industries primarily serve local businesses
and households with commodities and services which do not enter into inter-
regional trade in substantial amounts. The outputs of these industries,
therefore, are determined by the size of population and personal income in
each region.
Agriculture
Projections of agricultural production and land use did not go through
the economic area stage, but rather went through the projection methodology
for water resources regions, subareas and states. The projections are
based on the same framework of assumptions as the other industrial sectors,
and on the extension of historical trends and the use of a land availability
check to ensure the adequacy of resources within the regions to produce the
projected output.
98
-------
Projections of agricultural output for the water resources regions and
subareas were developed through a three-stage process: .First, the.projected
national output was distributed among states." Second, fhe state totals were
disaggregated to subareas and regions. Finally, the output projections in
percentage terms were converted to quantity and value estimates.
Population
It is assumed regional variations in the rate of natural increase in
population in an area are small, therefore, local area population projections
were based on interarea migration. Assuming the major motivating factor in
migration is economic opportunity, area population for the most part was
projected as a function of area employment. To account for non-economic
causes of interarea migration, the population was divided into three
categories: the pre-labor pool (ages 0-14), the labor pool (ages 15-64),
and the post-labor pool (ages 65 and over). Population in the labor pool
was projected as a function of area employment. The pre-labor pool popu-
lation was projected as a function of the total labor pool, while the post-
labor pool population was projected in terms of population aged 55 and over
at the preceding decade using the following formula:
65+
65+
5
t-1
55+
35+
5
t-1
This formulation holds constant the relationship between the 55 and over age
group at a given time and the 65 and over age group ten years later.
II.
1973)
Purpose
1 LOCATION-AGGLOMERATION1 APPROACH1
C. Harris1 Multi-Regional, Multi-Industry Forecasting Model (Harris,
The second type of model we considered in our investigation was the
Curits Harris Multi-Regional, Multi-Industry Forecasting Model. This model
is designed to predict the short-run impact and project.the long-run effect
of an exogenous change in public expenditures. In order to maintain
consistency between regional and national forecasts, the Harris model, like
the OBERS system, projects regional shares which are then applied to the
projected national values to obtain regional values.
1
We have classified the Harris Model as the "location-agglomeration" approach
due to the multitude of variables used in his equations to explain economic
activity.
99
-------
Geographic Region
The geographic unit chosen as the focal point for analysis is the
county, though in a later study (Harris 1974) Harris aggregates the 3111
counties into the 173 QBE Economic Areas.
Characteristics of the Model
The following assumptions characterize the Harris model.
(1) The estimates of national coefficients can be applied uniformly at
the regional level.
(2) The Harris equations assume uniform production and price functions
among regions (Miernyk 1973, p. 281).
(3) Changing snares in regional production resulting from changes in
transportation costs also reflect the dynamic character of the model. These
changing shares underlie the shifting location of regional industries,
(4) There are no abrupt changes in any regional economy, but rather
the changes that do take place result from the gradual adoption of innovative
technology.
Variables in the Harris Model and Data Requirements
The Harris equations are equipped to forecast industrial and construc-
tion output, employment, population, personal income, personal consumption,
government and defense expenditures, gross investment for a number of
industries, and marginal transport costs.
All variables, except the marginal transport costs, were estimated
using OLS regression analysis. Values for the transport costs were obtained
from a linear programming solution to a transportation cost minimization
problem.
P.r_ojectton___Metho_dol_pgy_
The model is recursive in that supply and demand data in year t are used
to forecast variables in the year t+1. The forecasts are then used as data
to make forecasts for the year t+2, and so on. In specifying the model,
Harris used a modification of a procedure referred to by Theil as "extending
the set of explanatory variables," According to Theil, this procedure
involves establishing a critical set of explanatory variables in the initial
regression; then additional independent variables are introduced in order of
their theoretical importance in a step-wise regression program. Variables
are added until a coefficient becomes insignificant, at which point the
procedure stops. Harris uses this approach: the least squares regression
routine was set up to preserve the theoretical order of variables entering
the equation, but was not terminated when one variable became insignificant.
In order for a.variable to be retained in a particular equation of the
model it had to pass two tests in addition to the standard "t" test of
statistical significance. First, the sign on the coefficient had to have the
postulated theoretical sign, the entering variable itself could not have
severe multi-collinearity with other variables in the equation. The decision
100
-------
rule for dropping a multi-col linear variable from an equation is that a
variable was rejected if the coefficient of determination (R?) of the inde-
pendent variable to be entered with those independent variables already in
the equation exceeded 0.70. The decision rule used for the test of signi-
ficance was that the 't' ratio had to be greater than one (the standard
error could not exceed the coefficient).
The forecasting model starts
area for the first year after the
output of each industry sector is
that firms face in each location.
payrolls, employment, population
demand sectors are also forecast,
are related to income; investment
are determined exogenously.
by forecasting output by industry in each
base year (,1965). The output or change in
explained by the marginal costs or prices
After output has been determined, then
and personal income are derived. The final
Consumption and governmental expenditures
is related to output; and, foreign exports
The parameters for each equation used in the forecasts were estimated
using Ordinary Least Squares in a cross-sectional regression of approxi-
mately 3,111 observations. 1965 was selected as a base year because it was
the only year for which a complete set of data for all counties was
available. In some equations some of the observations were eliminated if
the value of the dependent variable was zero. In the location equations,
observations were eliminated in some industries because of other data
problems. Data for the marginal cost variables were obtained using a
linear programming transportation algorithm.
The Harris model also uses dummy variables for each of the regional
observations to capture some of the variation in the dependent variable
not fully explained by the explanatory variables. Harris assumes there
is some unique constant in every county that is not being explained by a
particular equation. Therefore,there is a separate, though parallel
regress Ton line for each county.
III. REGIONAL AND INTER-REGIONAL INPUT-OUTPUT MODELS
K. Polenske's Multi-Regional Input-Output Model (Polenske, 1973,
pp. 1-37)
Purpose
A third model considered is K. Polenske's Multi-Regional Input-Output
Model (MRIO). This model, like Harris1 system of equations, is designed to
analyze the short-run impact and predict the long-run effect of an exogenous
change in government expenditures on a region's economy. MRIO goes one step
further, however, in that it is also designed to measure the interregional
repercussions resulting from a change in government policy. The inter-
regional economic repercussions are examined in terms of the changed inter-
regional trade flows.
101
-------
Geographic Regi on
- -.''.'. " ' " . , T . .
The focal point of regional economic analysis in MRIO is the state. The
complete regional and interregional accounting system consists of an 1-0
table for each of the states and the District of Columbia and 2550 trade flow
matrices illustrating the commodity flows between the regions.
Characteristics of the Model. >.
Unlike the previous two types of models, the Polenske model does not
assume uniform regional production functions {except in the trade and
service sectors).
Technical Coefficients of Production --
Separate technical coefficients were calculated, by state, for agri-
culture, mining and new construction. In the manufacturing sectors, national
coefficients were adjusted to allow for differences in product mix. Unad-
justed national coefficients were used for the trade and service sectors.
Final Demand and Trade Patterns
At any given point in time final demand and the coefficients are assumed
exogenous. With variation in final demand and the coefficients over time,
the Polenske model allows for regional shifts in production, consumption, and
trade.
Price Sensitivity of Output --
Given the nature of the Polenske model, output is not responsive to
changes in relative prices. Regional production may vary only as the result
of a change in demand for that output.
Variables in theMRIO Model andData Requirements
The complete multi-regional Input-Output accounting system consists of
a regional 1-0 matrix for each of the 51 regions and 2550 trade flow
matrices reflecting the interregional commodity flows between regions.
Therefore for the complete multi-regional 1-0 accounting system 2601
regional blocks of data are required. The regional 1-0 matrices are
designed to give purchases by the location of the purchaser, while the trade
flow matrices are designed to give the total inputs of a commodity i required
by industry j located in region g from all regions (including region g);
g 51 hg
e.g., X = i X v
ij h=l ij i,j,g
where h specifies the region in which the good was produced.
The regional 1-0 table consists of a matrix of inter-industry flows
||at.||, a vector of final demand, and a vector of value added. The
nJ technical coefficient, a?., is defined in value terms as the
"total amount of the input of 7J industry i in region g required per
unit of output of industry j. Since each.element of the regional table
gives the purchase of a commodity without regard to the location of pro-
duction, each row shows the total distribution of a commodity to the inter-
102
-------
mediate and final users within that region. The rows specify the producing
industries but do not designate the region in which the goods were produced.
Similarly, each column of the table shows the total purchases of goods,
services and value added by the particular intermediate or final user
located in that region.
Usually, in a closed economy, the sums of corresponding rows and columns
of an input-output table will be equal. This, however, is not true when
dealing with relatively open regional economies. The discrepency between the
sums of corresponding rows and sums is due to the volume of interstate trade
flows. If a specific commodity is not traded, then it can be expected that
the sura of the corresponding row and column will be equal. Using inter-
regional 1-0 tables, each total purchase made by an intermediate or final
user in a particular region can be traced back to its source. In the inter-
regional trade tables, each row sum indicates the total production of each
commodity in each state, while each column gives the total purchases of each
commodity by each region.
For the MRIO analysis, detailed state input requirements for 1963 were
assembled for three large sectors of the economy: agriculture, mining, and
construction. Most of the research effort on regional differences in
technology was concentrated on these sectors because locational factors are
likely to cause significant state-to-state variations in their input require-
ments. For the remaining industries, namely the manufacturing and service
industries, either variations in the state coefficients were obtained
through the product mix method, or for few of those industries, where the
components of the industry could not be separated, the national coefficients
were used directly.
Methodology
In estimating the regional technical coefficients, K. Polenske made use
of the national primary and secondary matrices. The primary product matrix
is a product-by-establishment matrix, in which the rows show the total amount
of a given product distributed by each establishment regardless of which
establishment produced the product and the elements in the columns show the
total amount of a particular good purchased by establishments to produce the
total output of the establishments their primary and secondary products.
The secondary product matrix is an establishment-by-product matrix that
provides information on the composition of secondary-product output for each
industry. The elements in each row give the value of each of the secondary
commodities produced by that establishment according to the industry of
which they are the primary commodity. The elements in each column, on the
other hand, give the value of commodities of the given establishment that are
^
Product Mix Method: To obtain state inputs, the national direct primary
input coefficients for the manufacturing and service industries are
weighted by the state establishment outputs. The resulting regional input
requirements for a given industry varied from state to state, reflecting
regional variations in the composition of goods produced within different
states.
103
-------
produced by other industries. By definition, all diagonal elements,are zero,
since each element s.. represents the quantity of goods of industry i that
is produced by "' industry i, which cannot be secondary products of
industry i.
Since regional secondary product data were not readily available, all of
the regional technology data assembled for the MRIO study were reconciled
with the QBE 1963 national primary-product matrix, while the regional
secondary products were estimated using the national primary and secondary
matrices and state outputs as a basis.
Having estimated the elements in the regional secondary product matrices
they were combined Into a single row and column and inserted as an extra row
and column along with those of the regional primary matrices to obtain the
total transactions: matrix [|a?. [(for each region.
' J
IV. REGIONAL MACRO-ECONOMETRIC MODELS
N. Glickman's "Philadelphia SMSA Regional Econometric Model"
(Glickman, 1971}
Purpose
The fourth general type of model we considered is a regional macro-
econometric model. This type of model, theoretically, assumes one output
per region, but Glickman relaxes this assumption in order to forecast output
by various manufacturing and non-manufacturing industries.
The Glickman model has two principal purposes. First, it can be used
for macro-econometric forecasting analysis of regional economies; and second,
it can be used for multiplier analysis of policy-related phenomena.
Geog raph i c Reg i on
The Glickman model was constructed-,to. analyze the regional economy of
the Philadelphia SMSA. «.;
Characteristics of the Model
The Glickman Model was designed as part of a satellite system to be
attached to a system for the national economy as a whole. Unlike the
Polenske model, it is a "top-down" rather than "bottom-up" model, implying
the national economy 'drives' the regional economies. This approach is
taken because of the lack of regional data.
Due to the data constraints only annual data can be used in the
regressions, and even then relatively few economic variables are used.
104
-------
Vanables Included in the Mode] and Data Regui rements
The Philadelphia SMSA Regional Econometric Model consists of fourteen
blocks of equations, aimed at providing a fairly comprehensive view of the
economy. These blocks are categorized as follows:
manufacturing output
non-manufacturing output
manufacturing employment
non-manufacturing employment
wages, prices, and income
Federal and local government
manufacturing investment
demographic
retail sales
banking
the Ci'ty of Philadelphia Accounts
the suburban subregion accounts
consumption
a quarterly submodel
Of these equations, forty-seven equations summarize the important private
and public sector activity in the City. Given the variables determined there
and for the region as a whole, a residual block for the 'suburban1 portion
of the region is calculated. Consequently, three separate area units are
considered; the City of Philadelphia, the suburbs, and the entire SMSA.
In order to specify many of the model's equations, 'a priori1 informa-
tion was used from the Philadelphia input-output table. For example, the
import and export columns of the table were ased to delineate local from
export-oriented industries. The specification of equations in the manu-
facturing output and employment blocks was also facilitated by noting the
important interindustry relationships in the transactions matrix.
Methods of Estimation
The blocks in the Philadelphia model are related primarily in a
simultaneous rather than a recursive manner. To the extent the equations
are recursive, they were estimated by Ordinary Least Squares; otherwise,
the equations were estimated by Two-Stage Least Squares -- Principal
Components and Iterative Instrumental Variables.
"In the Philadelphia model, national market-oriented
output is related to national variables such as GNP and
national trends in those industries and to local inter-
industry partners. The model is designed to be 'plugged in'
to the Wharton Annual and Industry Long-Term Model of the
United States, taking as exogenous many of the variables
forecasted by the Wharton model such as Gross National
Product and U.S. output and wages in various industries.
On the other hand, output which serves a local market is a
function of Philadelphia's personal income and other measures
of the local economy. Thus, a modified economic base method
105
U.S EPA Headouarters Library
Ma-! coot b J-*T
1200 h-i-.iitivivr/:';.- -svtnue NW
Washington. UC 20460
202-56b-0556
-------
is employed. Employment by industrial sector is related to
sector output as in an inverse production function.
There ts a rather large Federal and local government
section. In addition to determining Federal personal income
tax accruals from the Philadelphia region, the model
estimates tax revenues for both municipalities and school
districts; certain governmental expenditure variables are
endogenously determined. A large number of policy instru-
ments on the national and local level are contained in the
model, including inter-governmental revenues, the Federal
personal income tax rate, defense spending, Philadelphia
City tax rates and Federal government employment in the
region. These policy variables can be manipulated to
examine the impact of government policy changes in the region:
some of these simulation experiments are described in pages
below.
In the income block, wages by sector are related to
national wages and the local unemployment rate. Proprietors'
income is related to local retail sales and population and
national trends in proprietors1 income. Transfer payments
are determined by the corporate bond rate (representing
private transfers), and total U.S. transfer payments (proxying
public transfers]. Personal income is the sum of wages and
all non-wage income. Consumer prices are a function of
local unit labor costs, national consumer prices and regional
profits.
Retail sales for the region are related to personal
income and the employment block helps determine the demographic
block; the latter includes the unemployment rate, among other
variables. The output and personal income block equations
impinge on the banking sector where demand and time deposits
for the local- Federal Reserve district are determined.
A small quarterly submodel predicts important regional
variables such as manufacturing and non-manufacturing
employment, unemployment and the consumer price index. Thus,
there is a mixture of annual and quarterly data within the
model.
In addition to time-series data, cross-section information
is employed in the construction of the model. Eleven com-
ponents of household consumption expenditures are estimated
from a 1960-1961 BLS sample survey." (Klein and Glickman, 1977)
106
-------
V. SUMMARY OF FIRST FOUR TYPES OF MODELS
We have reviewed a variety of existing regional models with the
objective of determining which of these models can be incorporated, with
or without modification as the Regional Module of SEAS. Due to the
complexity of the problem, and the not too infrequent lack of reliable
data, most regional models are 'linked1 in some fashion to a national model.
Examples of these models are the (1) Shift-Share, and (2) Location-Agglom- ,
eration Models, (3) Regional and Interregional input-Output Models, and-
(4) Regional Macro-Econometric Models. Each of.these.models, is designed to
explain and project regional economic activity. At the original time of
their formulation, however, these models were not designed for the purpose
of analyzing the economic and environmental impact of alternative govern-
ment energy policies.
Existing Regional forecasting Models
The existing regional forecasting models can be classified into four
relatively distinct categories and these are: 1) the OBERS "shift-share"
type model; 2} the Harris "Location-Agglomeration" type model; 3) the
Klein-Glickman type Macro-regional econometric models; and 4) the Polenske
MRIO model. These categories are not, however, entirely distinct since
some of the models utilize,a number of different techniques to conduct
forecasting analysis. For example, the Harris model and the OBERS pro-
jections combine regression analysis.with the regional shares, and 1-0
approaches to determine the effect of alternative transportation systems
and water resource development on a regional economy, respectively. MRIO
is strictly an input-output technique developed to project regional and
interregional economic activity, and the Glickman model is strictly an
econometric type model using regression analysis to project economic
activity in the Philadelphia region, i
Energy and the Environment
None of the models, however, contains explicit formulations to project
the environmental impact and the effect on energy demand of alternative
energy scenarios. The 'Glickman model and MRIO are 'pure' regional economic
impact models. The OBERS model tries to project the environmental impact
of water resource development on land use, and the Harris model introduces
environmental quality variables to explain regional variation in output.
The Multiregional Economic & Energy Demand Model (MREED) seeks to
integrate energy and environmental impact variables w.ith some of the. better
features of the models studied. (The MREED model is presented in Chapter 5.)
Structure of the-ModeIs
All of the models, except MRIO, consider the national economy as the
'driving force behind the regional economies.' The OBERS projections and
the Harris model project regional shares of national values -- output,
employment, income into the future. Changing shares over time reflect
the dynamic character of the model and the shifting location of regional
107
-------
industries. The Glickman regional model of the Philadelphia SMSA was
designed to be part of a satellite system with the national economy in the
center of the system. MRIO, on the other hand, was designed to estimate
each region's level of output, employment, income and expenditures, and
then obtain the national totals as the sum of all the regional values.
The MREED model will serve as the Regional Module of SEAS; therefore,
it will project, along the OBERS and Harris lines, regional shares of
national totals.
Geographical Unit of Analysis
To project economic activity levels OBERS bases its analysis on the
BEA Economic Area. The Harris model has also been applied to the 173
Economic Areas, but was initially intended for the county subdivision.
The MRIO region is the state, a political unit. The Glickman model is
based on the Philadelphia SMSA. Of these modular regions, the MRIO state
is probably the most open since SMSA's extend beyond political boundaries.
The BEA economic areas are the most self-sufficient in the employment
sense since they usually contain labor's place of residence and place of
employment in the same area.
The MREED model, partially for this reason, will be based on the BEA
Economic Areas.
VI. OTHER MODELS
THE FEDERAL ENERGY ADMINISTRATION'S READ MODEL
As must be expected with models currently under development, a com-
prehensive evaluative description of the Federal Energy Administration's
Regional Energy, Activity and Demographic Model (READ) was not available
for external review. As such, the internal workings of READ could not be
subjected to as great a detailed examination as would have been preferred.
However, a partial description of its structure (consisting of general
flow charts and two memoranda) did permit a review of some of the model's
conceptual underpinnings and linkages.
General Structure of the READ Model
The Federal Energy Administration's Regional Energy, Activity and
Demographic Model (READ) generates forecasts of county-level output, by
industrial sector, based on exogenous energy specifications. The county
forecasts are then aggregated to a regional level for a structural demand
model which assigns the forecasted regional outputs to the least-cost
technological process, determined by the structural model. In this way,
The document on which this review is based came to us from ERDA at one
of the SEAS design meetings.
108
-------
the model's projected demand for energy use is partially a function of the
READ forecasts.
From the model structure depicted in Figure 4-1, data bases from various
federal government agencies are used as inputs for such types of parameters
as commerce data, demographic information, employment, income, government,
and weather. Direct inputs to READ include regional energy prices generated
by Project Independence (PIES), national macroeconomic, employment, income,
and population forecasts, the previous year's regional supply, demand,
population, and employment, and governmental structure.
READ then is used to project energy demand by sector by region.
Location equations estimate regional output which leads to agricultural,
industrial, and commercial energy demand. Demographic estimates are used
to determine energy demanded by residential and commutation activity, with
the latter in turn influencing transportation energy demand, subject also to
transportation requirements and shadow prices. Energy required by govern-
ment is a function of government expenditures.
Industrial Structural Demand
The READ model also will be used to forecast regional industrial
capacity, and thus estimates of annual equipment expenditure and construction
purchases by industry. These forecasts interface with the structural energy
demand equations in the following manner (the model will be utilized for the
residential (Hirst Model), commercial, and industrial sectors, but we will
discuss only the industrial sectors).
The industrial structural demand for an industry is calaculated as:
Q = f(Capacity , Technology Penetration ,
ijr r j,r
Short-Run Utilization , Fuel Usage )
j>r i,j,r
for fuel i, technology j, region r. Initially, structural models are
being developed for the ten highest energy-using industries. The total
capacity of the industry in region r is the maximum economically feasible
output level with existing capital stock. (The measurement of this para-
meter is considered highly feasible, because the energy-intensive industries
are primarily continuous process ones.) The capacity is not specific to
technology (e.g., "steel mill output").
The technological penetration parameter uses the alternative energy-
using industrial process and their individual capacities. A least-cost
basis is used to assign technological process. The Demand Analysis
Division (DAD) is in the process of determining the most probable future
technological processes in use in 1990, including estimates of technological
shifts over time. The READ equipment expenditure forecasts and the
technological projections will enable DAD to forecast levels of techno-
logical penetration. (The residential and commercial sectors will be driven
109
-------
o
^
JJ
«
«o
c
o
a:
"4-
O
0)
J-
3
cn
no
-------
CfMrgy Prices
Population
bicome
Industrial Activity
Aluo Efficiency
Standards
Alriiiie Load Factors
Conservation
Savings
Household Sector
Total
Energy Demand
Index
Specific Fuel
Demand Indices
Commercial Sector
Total Energy
Demand index
Specific Fuel
Demand indices
Industrial Sector
Fuel ana Power
Total Energy
Demand Index
Fuel and Power
Specific Product
Demand Indices
Raw Materials
Demands
T
ransportation Sector
Automobile
Simulation Model
Other Vehicles
Natural Cos
Transporudon
Energy Product
Demands by Sector,
by Region
Figure 4-2. Energy demand in READ.
(source: personal communication)
m
-------
V
as well by READ'S forecasts of population and residential and commercial
construction expenditures,) An example of the model's hypothetical
technology penetration for the rtfl region with technology j is "electric
blast furnace."
The short-run utilization variable indicates the fraction of process
capacity currently being used in the production process. Industrial output
6y region is obtained 5y multiplying utilization by capacity. Short-run
utilization would typically be a function of fuel prices, output prices,
and possible prices of other factors. For the jtfl technolpgy in region r,
the parameter can include the number of shifts and number of hours.
The component of fuel demand in the equation allows technological
transportation from a particular process's level of output to its required
amount of fuel i.
While an alternative source of capacity and output forecasts can be
used instead of READ, two advantages of using READ to drive the structural
demand models are that (1) both models share a common data base consistent
with PIES, and (21 the industrial location and population sectors of READ
are energy-dependent.
Other Components of READ
As previously explained, little description of the READ model's
structure was available for review, other than Figures 4-1 and 4-2
(.attached) which are self-explanatory. Various exogenous parameters serve
as inputs to the model, including demographic data, outside; agency data
and forecasts, and intra-FEA data and forecasts (or scenarios). The
absence of almost any detailed description of the source and acceptability
of data, and the structure of the READ model, prevents analysis of its .
potential usefulness (although the presence, in Figure 4-1 of a feedback
mechanism for output energy demands to be incorporated in new energy price
calculations, does provide evidence of one potentially favorable structural
aspect). The exception of the construction sector, for which a six-page
memorandum was available,1 does allow some detailed analysis.
In the construction sector, the straightforward computations of capital
stocks do not leave much room for argument, except for the fact that they
incorporate some rather gross assumptions. The Dodge Construction Potentials
data must be adjusted for "many" differences to conform to National Income
Accounts, a task found "impossible," other than by simply scaling county
data to aggregate National Income controls. Aggregate National Income
Accounts deflators will be used to convert county construction data to
constant dollars. Farm construction and oil and gas well drilling and
exploration data are not available from Dodge, and potential sources are
currently being investigated. State-level input-output multipliers for
all industries (including construction) are available in sufficient detail
for some construction sectors. However, others must use national coeffi-
cients. Capital stocks are unavailable at the regional level, and "(even)
national estimates of capital stocks are far from ideal," although
estimates "by considerable industrial detail...(and)...by types of plant
112
-------
and equipment" "can be constructed." National capital-output ratios or
capital-labor ratios will be applied to county output or employment.
National discard rates as well (historical and projected) will be applied.
The lack of much required regional data, at least in the construction
sector, does cast doubt on the confidence which can be placed on regional.
estimates.
Summary of READ Model
From the information on READ available to us, the model appears to
have a reasonable structure. However, several potential problems must be
confronted.
First, one structural aspect which can be questioned upon examination
of the flowcharts is the projection of output on a county level. The
potential error inherent in attempting such micro-level projections based
on national forecasts raises questions about the accuracy of s-uch estimates.
Further, such errors would be compounded in any aggregation up to a larger
regional level.
Second, as evidenced by the problems in obtaining regional data for the
construction sector, it would appear that estimation and calibration at a
county level would be subject to great data limitations in all areas. This
may result, as; in the construction sector case, in unfortunate reliance on
national parameters in many cases.
Third, if the projection of regional output depends on only capacity
and utilization (based on prices), as it appears, the regional allocation
process itself is suspect.
While other questions could be raised about the structure of READ if
more detailed information would be available (particularly the flowchart's
"location equations" which determine regional output), the ability of READ
to project confidently regional economic, environmental and energy outputs
would appear suspect, at least in its present working-stage form.,
OAK RIDGE NATIONAL LABORATORY'S MULTIREGION (OLSEN, 1976)
While other models examined in this study generate projections of
regional economic output, the stated purpose of MULTIREGION is to handle
"core" requirements of regional decision makers and energy planners --
"basically accurate and sufficiently detailed forecasts of regional popula,-
tion and employment." (01 sen, p. 1) Developed by the Regional and Urban
Studies Section of the Energy Division of Oak Ridge National Laboratory,
MULTIREGION is designed only for these "core" requirements, while tradi-
tional methods are to be used to obtain other types of^projections. As
such, the model is inappropriate, or at least incomplete, for incorporation
into SEAS as a regionalization component. However, we will describe its
structure as an alternative approach to regional modeling.
113
-------
Overview of MULTIREGION
MULTIREGION attempts to combine four forecasting methods - trend pro-
jection, cohort-survival analysis, export-base analysis, and shift-share
analysis - so that divergent forecasts typical in their separate applica-
tions can be avoided.
By viewing Bureau of Economic Analysis (BEA).economic areas as labor
markets, regional and interregional supply and demand processes can be
modeled. Progressing in five-year time steps, for each BEA economic area,
computations are made of:
(1) trial population (last period population + births - deaths + net
migration)
(2) trial labor supplies (population times labor participation rates)
(3) trial labor demand (sum of forecasted employment)
(4) trial labor market conditions (from labor supply and demand)
(5) final labor market conditions (reiterating (!) - (4)).
Factors in the Model
Little consideration la given to economic influences in the demographic
effects projected By MULTIREGION. Mortality and fertility vary over regions,
but without reference to labor market conditions, while migration is
affected by labor market tightness, interregional accessibility, and popu-
lation density. Finally, labor force participation rates depend on labor
market tightness and female participation in the labor force (a reflection
of the regional industry mix).
The employment sector of the model includes three components: manu-
facturing, service, and natural-resource-based employment.
The natural-resource-based sectors use a shift-and-share framework to
exogenously treat activity levels of those sectors "fairly exogenously."
In this way, labor market conditions do not affect these sectors' employ-
ment.
Employment in the manufacturing sectors takes into account initial
conditions (past locations of employment) and interregional market size as
the two most important influences on industrial locational change.
Regional service employment projections use initial conditions, total
employment growth, and to a lesser extent interregional market size,
population density, and labor market tightness.
Outputs of the Model
The outputs of MULTIREGION are intended to resemble those of the OBERS
in both form and potential applications. For both the nation and each of
the 173 individual BEA economic areas, historic and projected values are
given for employment by industry sectors (.37), age and sex cohorts (32),
and labor force cohorts by age and sex (16). Rates of growth and summary
measures such as interregional market potentials, industry location
114
-------
I
quotients, coefficients of specialization, and shift-and-share components
are also generated as output by MULTIREGION.
Summary and Conclusions
By design serving only a limited purpose, the demographic and employ-
ment projections estimated by MULTIREGION are inappropriate for the
regionalization module of SEAS. The existing.weakness of REGION can be
attributed to the use of employment .to allocate output. Further, alternate
development or policy scenarios cannot be tested in a model which deals
essentially as basically one of demographics and only in implicit economic
interactions. Finally, the dynamic effects of potential changes in the
economy (e.g. Western coal development, air emissions regulations, or
drastic energy price changes) cannot be reflected in a model relying heavily
on past trends and the "inertia" of "initial conditions." For these
reasons, MULTIREGION would be an inappropriate choice for incorporation
into SEAS.
115
-------
CHAPTER 5
A MULTI-REGIONAL MODEL OF REGIONAL
ECONOMY AND ENERGY DEMAND IMREED)
I. INTRODUCTION
In an affluent, spatially integrated economy such as the U.S., the
regional economies are intimately linked to one another and to the national
economy. The size and nature of the regional economy are largely determined
in the national markets and by the spatial configuration of activities in
national space.
What constitutes the national economy at any point in time is rooted
in the determinants of final demand income distribution and consumer
preferences, and the contemporary organization and technology of production.
As the latter change, so do the size and composition of the national
economy, and its spatial manifestation the regional economies. Location
theorists have identified several distinct phases, each of which is charac-
terized by the importance of different locational factors, in the history
of the spatial evolution of the American economy (Chinitz 1960, C.D. Harris
1954, Ullman 1958, Perloff and Wingo 1961, Berry and Neils 1969). These
locational factors stem from the multiple influences of natural resources
and technology (,of transportation) in the earlier periods and amenity and
environment in the contemporary period.
Following the tenets of classical location theory, Chinitz recognizes
two historical phases in the distribution of production in the U.S. largely
by the structural changes in transportation. In the 19th Century, the rail
network which reduced the cost of long distance travel relative to short
haul led to increased concentration of manufacturing industries in the U.S.
Early in the 20th Century, technological developments that reduced the cost
of short haul relative to long haul (e.g., truck transport) led to a
second phase of decentralization of industry.
The 20th Century has seen an increasing role of the market as a
determinant of location of industries in the nation. Harris and Ullman
have documented this by showing the juxtaposition of population and
economic activity in the industrial 'heartland' or the manufacturing belt
extending westwards from New York, in the area bounded by the Pennsylvania
coalfields, Lake Superior ores and the capital, entrepreneurial- experience
and the engineering trades of the North East (roughly Middle Atlantic,
New England, and Great Lakes Regions). Berry and Neils have shown that
this heartland early became the center of heavy industry in the country but
116
-------
has since remained the center of national demand, determining patterns of
market accessibility. Perloff and Wingo show further the historical
tendency for fabricating industries to concentrate in the heartland while
first-stage resource users or processing industries tend to dominate in the
hinterland. Perloff and Wingo suggest a spatial pattern of the national
economy as
a great heartland nucleation of industry and
the national market, the focus of large-scale
nation-serving industry...(Outside the heart-
land) are the resource-dominant regional
hinterlands specializing in the production of
resource and intermediate inputs for which the
heartland has need..J
Perloff and Berry and Neils have noted an emerging factor in regional
economic growth in the second half of this century. This is the amenity
factor and environmental quality the so-called "new natural resources that
count." In the contemporary affluent urbanized age, those new resources
are clean air and water, open space, and other amenities. Berry and Neils
trace the emerging relationships between these environmental amenity
features and regional growth outside the heartland in the South and
Southwest - the so-called Sun Belt. Further, rising energy prices in the
last decade have begun to affect regional purchasing power and energy
induced structure of transportation rates (Miernyk 1976(a), Miernyk 1976(b),
Giarratoni and Socher 1977). Since the energy resources are largely in the
national hinterland, significant regional economic and locational effects
are likely to ensue.
The regional models developed so far have not addressed the role of
environmental and energy factors in their locational analysis. The OBERS
model uses a 'shift share1 trend model that may capture only aggregate
past trends. The Curtis Harris Model uses a transportation cost variable
to reflect the locational factors in the regional economy (Equation 4 of
the Harris Model). Regional econometric models sidestep locational
variables and use a 'top down' approach of depicting regional economy as a
satellite system of the national model (Klein and Glickman 1977).
MREED represents an early attempt to incorporate in a regional
economic model (in addition to the traditional market and transportation
variables) the environmental and energy variables that appear to be
emerging as significant in location of regional economic activity. The
effects of the environmental (amenity) variables have been experienced in
the last two decades (Berry and Neils 1969, Perloff 1969). The effects
of rising energy prices while dramatic since the oil embargo of 1973 have
been evident for a longer period about a decade (Miernyk 1976(a),
Miernyk 1976(b)).
We hope to capture the effects of environmental and energy variables
on regional location by estimating MREEDrs parameters over the last
decade.
Perloff and Wtngo, op_. cit., pp. 211-12.
117
-------
The primary and manufacturing sector activities whose locations are
determined in the above fashion in MREED, are viewed as export activities
in each region determining the output levels in each region. Regional
income is determined in MREED by an integration of a Keynesian system
and economic base theory. Labor market and energy demand specification
round off MREED's view of a region's economy. The detailed rationale of
the model is explained next.
Table 5-1 lists the names and number of sectors used in the model.
These 37 sectors are identified by the relevant SIC categories. It must be
noted tfiat th.e first year's work proposed here is at the 37 sector detail,
to take advantage of directly available regional data. MREED,can be
estimated at a later phase at the 99 sector, detail (at which.'INFORUM
employment is detailed),. The notation and the model are presented in the
following pages.
The regional unit in the model is the BEA economic area, each of which
is a labor market area including the place-of-work and place-of-residence
of its labor force. The BEA areas 173 in all and covering the:.entire
U.S. -- are economic rather than political or administrative units,
permitting economic analysis in finer geographical detail than do.states
and in finer geographic and industrial detail than do'the SMSA's.^ The
BEA's do not divide SMSA's or counties and are usually named after the
largest SMSA. They range in size from area 14 (New York) with 18.2 million
inhabitants to area 101 (Scottsbluff) with-104,000 people.
II. THE MODEL EQUATIONS
The model equations (Figure 5-2) are grouped into "blocks" repre-
senting the different classes of production sectors, regional expenditures,
employment and energy. Special attention is paid to the demand .f,or energy,
since energy policy has come to occupy a central position in-public policy.
Energy as a composite good is consumed by households and"commercial
businesses as a final product and used by industry and transportation
sectors as an intermediate input. Furthermore, energy is produced by
combining the different sources of energy oil, gas, coal, electricity,
solar energy. Demand for each of these sources will" also be;examined
as part of the energy block of the model.
The major equations driving MREED-are equations (l);"and (3) that
describe output by each'of several primary and manufacturing industries.
The output in each industrial sector in a region j is viewed as a function
of input prices (W1it, EN-t), size of the market (YA-jt) a^rid. regional
environmental Jt J quality (El -t).. The J variable YA-t defines
the size and ..spatial array of markets ^available to the
industry located in region j at time t. It introduces space or the
'friction of distance1 into the model by using a fami-liar.sform of
operationaliz'ing' the concept of market potential (Stewart 1974, .Lakshmanan
2 - ' - , «7%«.i
See Regional Economic Analysis Division, "The BEA "Economic'Areas:
Structural Changes and Economic Growth, 1950-73," Survey of Current
Business, Nov. 1975, Vol. 55, No. 11, pp. 14-25.
118
-------
TABLE 5-1. SECTORS PROPOSED FOR MREED PHASE I
SIC NUMBERS
Primary
Agriculture, forestry mad fisheries
Agriculture 01, 07
Forestry and fisheries 08, 09
IMning
Metal ;. 10
Corf. - 11. 12
Crude petroleum and natural gas 13
Bonaetallic, except fuels 14
Hmmfacturing
Food tad kindred produces...... 20
Textile Bill products 22
Apparel and other fabric products 23
Laaber products and furniture 24, 25
Paper and allied produces..... 26
Printing and publishing 27
Chesrlcals and allied products 28
Petroleum refining 29
Priaary netals 33
Fabricated metals 34, 19
Machinery, excluding electrical 35
Electrical Machinery and supplies 36
Motor -vehicles and equipment. 371
Transportation equipment, excluding motor vehicles... 37 except 371
Otter Manufacturing 21,30-32,38,39
Service*
Contract Construction 15-17
transportstion, communications and public utilities
Railroad transportation 40
Trucking and warehousing 42
Other transportation and services 41,44,47
("mil in Inn i 48
Utilities (electric, gas, sanitary) 49
IboteMle and retail trade 50,52-57,59
Finance, insurance and real estate , 60-67
Other Services
Lodging places and personal services 70, 72
Business and repair services 73, 75, 76
Asnsenent and recreation services 78, 79
Private households 88
Professional services 80,81,82,84,86,89
ximGnt
Civilian government
Federal government 91 except Fed.
military
State and local government 92, 93
Forces Part of 91
r BEA (US Dept. of Commerce) 1972 PEERS Projections (itTtegi
Activities in the U.S. Vol. 2, U.S.G.P.O. April 1974, p.l
119
-------
x s>
I- n
** B
tn >-i
5 X
= «
0 W
as
J U,
-5
1
0
!
O
C
3
1,,
O 4->
&"
V
s
c:
0
C6
3
S.
O
C.
X
!&
4^ 'ft
n IA
i.S
o
CM
£
w
C
£7
4-> X
A*
'%\
I?
-o
O
s
' H-I
: £
ri
1 B
1
1
MANUFACTURING
i
V)
00
C0
-^
iJ
g.
I
f ^ tl.
O
>
a
,^ji
-------
Figure 5-2. The HREED Model.
Notation
AS.^ » Land assessment in region j in year t
Cj. a Generalized costs of transportation from i to j
C^ *> Personal consumption in region j in year t
» Cooling degree days in region j
COD.
Debt.
EC
kjt
ED
jt
ED
IE
jt
T
V
"j.
V
HDD.
Kit
a Total debt in region j in year t
* Energy input of k type of energy source into region
j in year t
a Energy demand in residential/commercial sector in
region j in year t
* Energy demand in energy intensive industrial sector
in region j in year t
« Energy demand in non-energy intensive industrial sector
in region j in year t
= Energy demand in transportation sector in region j
in year f
= Energy demand in region j in year t
« Environmental index of type V in region j in year t
Average energy costs in region j in year t
« Government expenditures in region j in year t
Heating degree days in region j
= Stock of capital in primary sector i in year t
» Labor force in region j in year t
= Employment in primary sector i in region j in year t
n Employment in manufacturing sector i in region j in year t
(Continued)
121
-------
Figure 5-2. (Continued)
Nit
p
PE
POPjt
POPD..
RE
it
UN
Wit
x11
V
xp
ijt
Employment in service sector i in region j in year t
Employment in sector in in the region at t
Total employment in region j at time t
National consumer price index
Relative price of energy
Population in region j in year t
Population densely in region j, in year t
Ratio of cost of energy inputs to total costs in an industry
Total tax receipts and'transfer payments in region j in year t
Total property tax receipts in region j in year t
Total sales and income tax receipts in region j in year t
Transfer payments in region j in year t
Unemployed in region j in year t
Wage rate in primary sector i in region j in year t
Wage rate in manufacturing sector i in region j in year t
National wage rate in sector i in year t
Total output of all energy intensive industries in region
j at time t
Total output of all non-energy intensive industries in
region j at time t
Output of primary sector i in region j in year t
Output of manufacturing sector i in region j in year t
Output of service sector i in region j in year t
(Continued)
122
-------
Figure 5-2. (.Continued}
*
Jt
Y
r
Uj
i
3
k
t
v
1.
Output of region j in year t
Income in region j at time t ;
Demand potential in (National income accessible to) region
j at time t
National income
Interest rate
Unemployment rate in region j at time t
Industrial sector (i = 1,...,75)
Region (BEA economic area) (j a 1.....173)
Type of energy source (k * 1,...,5)
Year
Type of environmental index
Average tax rate in the region
Accessibility index for region j
(Continued)
123
-------
Figure 5-2. [Continued)
THE MODEL
Output Block
Primary, Manufacturing, and Service Sectors
CD
(2)
(4) W?.
C5) Y*
s Y.. e-pcijt
Income Block
C7) Y4, -
C8)
(9)
Tit
f(cjt, rx..t, G.t)
f(T.t, Debt.t)
CU)
Tp + T°
T° , Y
' T
(13)
,
jt-1 jt-l . jt-1 jt-1 ' jt-l
(Continued)
124
-------
Figure 5-2, (Continued)
Labor Market Block
(14)
(IS) N».t *
(16)
(17) N.fc
Jt
(18) LF..
i"t' ^
!L. t)
£(N.t, t)
(19) POP
jt
(20) UN.t . LFjt -
(21) u.
UN.t/LF.t
Energy Block
Residential Energy Demand
122) ED^ = «(l-lj)Yjt. P^, CDDjf HDDj}
Industrial Energy Demand
(23) EDJJ = fCXJJ,
(24) ED" - fCx£.Wit
(Continued)
., PE)
»E)
125
-------
Figure 5-2. (Continued)
Transportation Energy Demand
(25)
>Jt - fCrjt. pqpDjt,
(26) ED
jt
126
-------
and Hansen 1965). Stated in equation (5), the demand potential or the size
of the national market accessible to industries in region j is a weighted
sum of demand in all regions j deflated by costs of interregional communi-
cation c.. .3 The larger the regional income and the more aggregate prox-
imity 1J it has to other regions, the larger the markets accessible to
it. The generalized communication costs c-jj^, include in theory money
costs, time costs, and risk costs of transportation. If these
costs are assumed to be proportional to distance, investments in transpor-
tation (e.g., Interstate System) reduce money and time costs. However,
because of increased traffic, risk costs go up, effects will vary by
sector, this will lead to larger market and supply potentials, increases
in interregional trade and a trend towards larger national markets. On the
other hand, increases in energy costs will cause increased transportation
costs, with the effects most severe among those industries where transpor-
tation of goods is relatively important. Market potentials and interre-
gional traffic will decline. Klassen has shown in such a case that smaller
regions will lose more than larger regions since regional economies will
move to some degree towards more autarky with less interregional trade
and a stronger concentration on their own market (Klassen 1976). However
the reduced traffic will also-lead to lower congestion and air pollution, '
thus leading to improved environmental quality. The operational measure
of Y?£ is described in the next section. In practice, a time lagged value
of Y?^ will be used to facilitate ease of estimation.
Jt
The environmental quality variable used in equations (1) and (3) is
intended not as one index but as a subset of indicators or measures that
collectively convey information about the quality of the broad and complex
environment in the regions. Some of the components of the environment
such as pollutant concentration and congestion can be measured directly
while many others cannot be (Thomas 1972, Hornback et al., 1974, Craik and
Zube 1976). Several indicators may be integrated into one index for the
more complex components where data are available (MITRE 1974, MITRE 1972,
Liu 1975, CEQ 1972). However, the weighting involved in the development
of indices is open to controversy (Thorn and Ott 1975, Thomas 1972, Liu
1975). While we are aware of the conceptual and empirical difficulties,
we hope to experiment and make a beginning with two or three key indicators.
Those indicators will be selected that appear to affect regional location
of firms and households and also respond to changes in environmental
policy. Though it is difficult to specify these variables in any greater
detail at this stage, we hope to use pollution indices and congestion
indices for which data series can be developed.
the reciprocal of the accessibility of region j to all
other regions.
127
-------
Regional wage rates and average energy costs -- the price variables of
labor and energy inputs will be other independent variables. While
national energy costs had been declining in the early and mid-sixties and
have been since turning upward in the nation, regional variations have
always been significant. We hypothesize that such regional variations will
explain part of the regional differences in industrial location.
The wages (equations (2) and (4)) in the several primary and manu-
facturing sectors are viewed as functions of the lagged wages in the
sector and capital/labor ratios. This form of wage equations was earlier
estimated at the SMSA level by the principal investigator and others
(Scully 1969, LaRs&manan,- Lo and Krishnamoorthi 1972).
The regional output in the market oriented service sectors (equation
6) is determined by regional income. An integration of the Keynesian
system and economic base theory appears in the income determination block
of the model. The primary and manufacturing sectors are viewed as export
oriented and their role in income determination is reflected by their
combined output (equation (7)). Regional consumption expenditures C.. and
local government expenditures G-t are other determinants of J
regional tncorae in equation J (7). Regional consumption is viewed
as a function of disposable income and lagged consumption (equation (8)).
Equations (9] through (13) describe the local government revenues and
expenditures. Equation (9) is merely a simple relation between local
government expenditure, G-t, and total local government revenue, T. ,
and the public debt. ' J Equation (10) is ah identity that Jt
treats local government revenue as the sum of three components: local
property tax, K , other local taxes and revenues, T? , and federal and
state transfer J payment.to the ..local government,,Jt T^ .Equation (11)
is a property tax equation that states that local J property tax
is a function of gross assessed property value, AS... Equation (12)
relates other local taxes and revenues to regionalJ income Y-. Finally,
a behavior equation of federal and state transfer payment to local
government is given in equation (13). Transfer payment to the local
government, is determined by lagged values of total local revenue effort,
per capita local tax, and regional income.
The equations predicting regional employment and population for a
given year are provided in the labor market block. The equations
determining employment in all sectors - primary, manufacturing, and
service - are inverse forms of production function relating employment
to output (equations (14), (15), and (16)). We follow Glickman in
using a time variable which is intended to capture productivity changes
(Glickman 1974). Equation (17) is an identity for total employment in
the region. Labor force is related to the level of employment and a
time variable (equation (18)). Labor force varies directly with employ-
ment*, opportunities. The effect of migration on the labor force is
provided by the time variable t. Population is then related to labor
force and the time variable which is intended to capture changes in the
labor force participation rate (equation (19)). The equations for
determining total unemployment and the unemployment rate are definitional
(equations (20) and (21)).
128
-------
The energy block describes the demand for energy by each major use
group. The demand for residential energy use in a region is related to
regional disposable income and the relative price of energy. In addition,
the number of annual heating and cooling degree days in the region, which
reflect the regional variations in space heating and cooling requirements
are also used as independent variables (equation (22)). Industries are
disaggregated into two groups -- energy intensive consumers (ED1!) and
non-energy Intensive consumers (EDlE). The idea behind this distinction
is that the, energy consumption behavior of the former group in response to
price changes is likely to be quite different from that of the latter
group whose energy inputs are a smaller proportion of total inputs. Each
group has a separate demand equation (in a later phase, we may estimate
energy consumption by each of the manufacturing industries listed in
Table 1, as was done by the principal investigator earlier Lakshmanen,
Lo and Krishnamoorthi 1972). For both industrial groups, the demand for
industrial energy is hypothesized to be related to output level, real wage
and relative price of energy (equations (23) and (24)). The demand for
transportation energy is a function of regional income and relative energy
price. In addition, population density in the region is introduced into
this equation as a proxy for level of use of public transportation
(equation (25)). Equation (26) is an identity describing the total
demand for energy.
III. DATABASE
Data requirements for calibration of the model's equations pose some
but tractable problems in that tittle data are presently available in fine
detail by BEA economic area. The only substantial data publication has
been the series^of OBERS historical data and projected estimates for
selected years.4 However, county-level data may be .aggregated up to BEA
areas. Also, data by SMSA can likely be extrapolated to BEA areas, as
the BEA classification essentially centers around one or more SMSA's and
surrounding counties. Also, allocation of state figures to BEA areas may
be permissible in some cases where reasonable assumptions can be presumed
to hold.
A preliminary exploration of available data reveals standard series
of federal government publications as the primary source of data. The
economic censuses of the Bureau of the Census at five-year intervals
provide county- and SMSA-level detail for many employment and production
data requirements. The Labor Department's Bureau of Labor Statistics
also provides detailed data on wages and employment, as well as price
indices. Federal Power Commission publications provide energy data in
great detail. The current intensive energy data gathering program in the
Federal government will likely provide additional detailed data.
In some cases, available BEA data may be used to scale unobserved data.
For example, earnings by industry may be used to scale INFORUM sector
output to BEA area level.
129
-------
CHAPTER 6
A MODEL TO ASSESS REGIONAL ECONOMIC EFFECTS OF CHANGES IN
FINAL DEMAND ON ENERGY PRICES
I. INTRODUCTION
This chapter seeks to present a model in which the effects of an
exogenous change in one region's final demand or set of prices on inter-
regional trade flows, regional production and value added can be. analyzed.
Before the model is presented, there are a number of assumptions
which need to be made clear. First, the model is a short-run model and
therefore assumes the regional production coefficients and interregional
trade coefficients are fixed. It is assumed that the period elapsing is
short enough so as not to allow production and trade coefficients to respond
to changes in input and output prices. In a longer-run, dynamic version
of this model explicit account must be taken of substitution effects in
production and consumption.
Initially, final demand is taken as exogenous. However, any changes
in production will normally lead to a change in income, and hence a change
in consumption and other elements of final demand. This aspect of income
induced effects on final demand will be examined in the third section of
this chapter. A review of some of the possible applications of this model
will be taken up in the fourth and final section.
II. THE MODEL
In discussing the effect on interregional trade flows and regional
production following an exogenous change in final demand or prices in one
of the regions, one must realize that each region comprises a number of
states which interact among themselves within the region and directly with
the states in any other region. Taking this into consideration, we see
that the two region model breaks down very rapidly to a complicated multi-
regional model. To avoid this problem, what follows is a presentation of a
two region model, subsuming for the moment that each region consists of a
number of smaller subregions.
130
-------
Output Determi nati on
A X + F
X = 1 jri
(*\
( /
X
1
JT
2
/
_ 1 '
/
m ^
"A
11
A
21
»
A
12
A
22 ,
X
1
r
2 m
+
'F
i
F
2
where A * , A *m refer to submatrices of coefficients showing the input of
region 1's products to regions 1 and 2 respectively.
^?i ' ^22 refer to submatrices of coefficients showing the input of
and 2 respectively.
X_, X? are outputs in regions 1 and 2 respectively.
F , F are final demands in regions 1 and 2 respectively.
The solution to this problem is:
X-
where [l-A ] " is the standard Leontief inverse showing the total
ultimate effects on production in the two regions following an exogenous
change in F-j or Fg.
Internal and External Multipliers --
Decomposition of the Leontief inverse using partitioned matrices yields
the internal and external matrix multipliers for the two regions. The
internal multiplier shows the inter-industrial propagation effects within
each region, while the external multiplier shows the repercussion process
in region 1 (2) due to the induced effects on output activities between the
two regions.
Derivation of the Leontief Inverse --
By solving the following set of equations we can derive the Leontief
inverse, which consists of the internal and external matrix multipliers for
each region.
(1) X = AX + AX + F
(2}
131
-------
(a) Solve for X and X partially in each equation:
1 2
-1 -1
(3) X = [ I - A ] A X + [I - A F
1
11
12 2
11
1
-1 -1
(4) X = r I - A I AX+rl-Ai F
rt *»/*-! o T i ^ oo J *
2 22J 21 1
(b) Substitute (4) into (3).
-1
(5) X = rl - A -i A
1 IT 12
22'
- A .-i
22
-1 -1
AX + rl-Ai P } + rl - A i F
21 1 22J 2 11 J 1
-1 -1
)X = J- I - rl - A 1 A r I - A
-1
-1
J
11 J 12 l 22 ] 21
-1
22 J 2 }
11 J 1 12 L
The analogous expression for X? is:
(6) X = r I - r I - A n A rl - A , A
2 L 22 J 21 L 11 J 12
-1 -1
r I - A i rA rI-A n F + F i :
L ?'?-|1'>ll- 11 J 1 O;t
To simplify notation, define the following:
B = 1 - A ^
11
J
22
2
11
-1
B
I - A
22
-1
I - B A B A
11 12 22 21
132
-------
22
[ I - B A B A ]
22 21 11 12
-1
Examining equation t3} or (4), we see that the partial derivative of X (X )
with respect to F (F ] yields B (B ): 12
12 11 22
9X
_L = B =[I - A ]
1
-1
1
2
-1
B = [I - A ']
22 22
B , B can therefore be interpreted as the internal matrix multipliers for
11 22 regions (1) and (2) respectively"] Since each internal
multiplier is not independent of the other region's industrial activity,
but interregional propagations accompany the operation of the internal
matrix multipliers, (£ and Cj can be interpreted as the external
muItipliers for
11
22 regions (!) and (2) respectively.
In matrix notation equations (5 '} and (6) become
(7)
j
where the diagonal term (C_B ) shows the ultimate total effect of reper-
cussions within each ii ii region separately of an exogenous
change in its own final demand, (F ).
i
X
1
IT
2
=
" 2
,C B
11 11
1
C B A B
22 22 21 11
2
C B A B
11 11 12 22
1
C B
22 22
F
1
('
2
j -1 -1
eg:CB = [ I - B A B A ] I I - A J
ii ii ii ij jj ji ii
-1
= {[I-A][I-BABA ]}
ii ii ij jj ji
= [ I - A -BABA +ABABA }
ii ii ij jj ji ii ii ij jj ji
-1
133
-------
A -ABA E B (I - A ) ]}
ii ij jj ji ii ii
-1
J
CB
={I_(A
+ A B A ' ) }
ii ij JJ ji
-1
The term in parentheses, A .. + A . £ . A .., is what Yamada and Ihara (1967)
call the "Augmented Input n U JJ Ji Coefficient." It reflects the
fact that production activity in region i, to begin with, needs its own
goods-input (A^). At the same time, it also needs a certain amount of
the jth region's goods as inputs (A..) to support its production,
This supply for region i from region j induces J the effect of production
activity on region j (B--). To realize this production activity, region j
in turn needs a JJ certain amount of region i goods-input (A. .)
The off-diagonal term in (7) (d.B. .A..B..), by analogy, shows the
ultimate total effect of J JJ repercussions within
region i of an exogenous change in final demand of region j. For example,
given a change in F-, there will be set off a process of production within
region j, J reflected in B... This will give rise to a
demand for import-inputs from JJ region i (A..). This, in
turn, will induce production in i (B..), which when J
premultiplied by C-j. yields the n ultimate total effect of
repercussions in region i caused by a change in the demand for
its exports.
Price Determination
The set of price formation equations pursuant to the output formation
equations is given by
T T
(8)P = A P +AP + V
1 11 1 21 2 1
T T
(9)P=AP+AP+V
2 12 1 22 2 2 '
where V-j and \/2 are vectors of value added in regions -1 and 2 respectively.
Along the same lines used to solve-for X, and X, we
can solve for P-j and P2 as functions of the value added in ^
each region.
T -1 T
(10) P = rl - A 1 A P
1 11 21 2
T -1
+ [I - A I V
11 1
134
-------
T -1 T T -1
(11) P = [I - A ] A P + [ I - A ] V
2 22 12 1 22 2
T -1 T T -1 T -1
(12) P = { I - TI - A ] P FI - A ] A }
1 11 ^ 22 12
T -1 T T
rl-Ai {V + A [I - A ] V }
11 1 21 22 2
T -1 T T -1 T -1
03) P = £1 - II - A ] A [I -A] A}
2 22 12 11 21
T -1 T T
[I -A] {A[I-A]V+V>
22 12 11 1 2
Letting
T T -1
B = [I - A ] i = 1,2 Tstranspose
it i i
1
we can rewrite
P
1
P
2
=
_ T
2 T
C B
11 11
1 T T T
C BAB
22 22 12 11
T
2 T T T
C B A B
11 22 21 22
T T
C B
22 22
m
V
1
V
2
or
-1
P = [ I - D* ] V
[ I - D* ] P = V
135
-------
where the relationship between prices and value added in the two regions is
analogous to the relationship between regional outputs and final demand.
From the perspective .of price determination, the problem is to determine the
effect on value added of a change in any element of the price vector in the
two regions.
III. INCOME INDUCED EXPENDITURES
In the Leontief 'Open End1 system, final demand is treated exogenously.
In a theoretical model of a closed economy, however, consumption and invest-
ment expenditures are hypothesized to be functions of income, or value
added. This section attempts to incorporate the income induced effects of
consumption and investment expenditure on output, and thus close the Leontief
system.
The introduction of the income multiplier process into the Leontief
system is largely the work of Miyazawa and Masegi (1963). We will first
present the model in a national or uni-regional setting, then, we will
present the model in the two region case with a revised interpretation from
that given in Section II.
Uni-regional case :
In the model presented in this paper consumption and investment expen-
ditures, domestic absorption, are assumed to be a linear function of the
level of income. Government expenditures and net exports, on the other hand,
are assumed to compose the autonomous part of final demand.
The equation determining output can, therefore, be written as follows:
(I) X = AX + Q + Z
where: Q is domestic absorption = C + I, a (n x 1) column vector.
Z is government expenditures plus net exports a (n x 1) column vector.
Q + Z - F = final demand
X is a column vector (n x 1) of final outputs
A is a (n x n) matrix of technical coefficients.
Since domestic absorption is a linear function of income, Q can be
written as follows:
(2) Q = EVX
= i.i e v x
ij i J J
where E is a (n x n) column vector of coefficient of domestic absorption.
For the itn element, e- = E./Y is the expenditure share in total
income of the
commodity
J.L
1 x n row vector of value added ratios. For the jin element,
V is a
v- = Y./x. is the income earned in industry j.
J J J
136
-------
"X is a (n x 1) column vector of outputs.
Substituting (2) into (1) we obtain:
(3) X = AX + EVX + Z
Solving (3} for x yields:
-1
(4) X = I - A - EV Z
An alternative way of writing (4) is:
"1
(5) X = (I-A)
where [l-EVCl-A)"1 ] " is what Miyazawa and Masegi refer to as the "Sub-
joined Inverse Matrix (1963, p. 92}." Formula (5) distinguishes the inverse
reflecting expenditure activity from the inverse reflecting production
activity.'
Two-Region Case
In the two-region case we must separate the income induced from the
autonomous components of final demand for both regions. The set of equations
for the two- region case is:
(6)X =AX +AX +EVX + Z
1 111 12 2 111 1
X =AX + A X +EVX + Z
2 21 1 22 2 222 2
where . A , A are respectively 1 x 1 and 1 x m coefficient matrices
11 21 showing the input of region Ts products to regions
1 and 2
. A , A are respectively m x 1 and m x m coefficient matrices
12 22 showing the input of region 2's products to regions
1 and 2.
Solving the set of equations (6) for X and X yields:
1 2
-1 -1 -1
(7) X = { I- [I-A -E V 1 A r I-A -E V I A }
1 1111 12 22 2 2 21
-1 r _ -1
. [ I-A -E V J {Z + A I-A - E V 1 Z }
11 1 1 1 12 L 22 2 2-1 2
T
Note, the expenditure coefficients and value added ratios are assumed to be
fixed during the period of analysis. Over time, however, they as the a 's
are subject to change. ij
137
-------
r V1 r r1
X = {I - I I-A - E V A I-A - E V A }
2 L 22 2 2 J 21 I 11 1 1 J 12
I-A - E V | {A
22 22
1 {A I I-A - E V 1 Z + Z }
J 21 L 11 11-1 1 2
To simplify notation, define the following:
-1
B = [I-A -E V ]
it ii t i
~2 -1
C = [ I-B A B A ]
11 11 12 22 21
1 = 1.2
C = [I-B A B A I
22 22 21 11 12
where . B.. is the internal matrix multiplier for each region incorporating
11 the effect of income induced expenditures in the
production process.
. C is the external multiplier for each region incorporating the
ii influence of income on each region's expenditures and
output and interregional trade flows.
In matrix notation (7) becomes:
C8)
X
1
x~
2
=
C B
11 11
~ |
C B A B
22 22 21 11
C B A B
11 11 12 22
~* 1
C B
22. 22
.
Z
1
Z
2
.
138
-------
IV. APPLICATIONS OF THE MODEL
As indicated in the previous sections of this chapter, this model can be
used to analyze the effects on regional output, income, private expenditures,
and interregional trade flows of an exogenous change in public expenditures
in either of the two regions.
Furthermore, the effect of an exogenous change in a region's vector of
prices will first be felt in that region's level of income, and will
influence output, private expenditures and interregional trade flows
through the income multiplier process.
This model, however, can also be used in an entirely different way.
The two regions can be viewed as two distinct industrial sectors - industrial
goods and agricultural goods producing sectors, energy-intensive and non-
energy intensive sectors, etc. As with the analysis in the two region case,
the effect on output, income, prices and intersectoral flows can be analyzed.
By categorizing the industries in a national 1-0 table into energy
producing and non-energy producing industries, for example, we can examine
the degree of dependence (independence) of non-energy producing industries
on the energy sector. This can be done by examining how non-energy output
and intersectoral trade flows respond to an output expansion in the energy
sector. Assuming, however, fixed production and trade coefficients, one
must realize the dependence of one sector on the other will be overstated.
Similarly, we can analyze the cost-push inflation caused by increased
energy prices. Again by dividing the national 1-0 table of prices and
wages into those pertaining to energy and non-energy producing sectors the
effect of energy price increases on other prices can be determined.
However, we must again realize that by ignoring the price sensitivity of
inputs, the extent of 'cost-push1 inflation caused by energy price increases
will be overstated.
V. CONCLUSION
As long as we are willing to accept the limitations of input-output
analysis and have the resources to collect national and regional data, the
model developed by Miyazawa and generalized by Yamada and Ihara will prove
to be an invaluable instrument to analyze national, regional, and inter-
regional economic repercussions of alternative energy and public expenditure
policies.
139
U.S EPA Headpi.'9rters Library
Mail GO-; <14T
1200 Pennsylv.-- . Avenue NW
Washington ;>C 2u460
-------
CHAPTER 7
APPENDIX--DATA SOURCES FOR CHAPTER 3
I. EMPLOYMENT DATA
Tables 7-1 and 7-2 indicate the sources of employment data us'ed in the
study of the use of location quotients to regionalize the national input-
output table (Chapter 3).
TABLE 7-1. SOURCES USED FOR EMPLOYMENT DATA FOR INDUSTRY SECTORS
AS = Agricultural Statistics 1969, U.S. Department of Agriculture, U.S.6.
P.O., Washington, 1969.
BLS-SA = Bureau of Labor Statistics, U.S. Department of Labor, Employment and
Earnings. States and Areas. 1934-68. Bull. 1370-6, 1969.
BLS-US = Bureau of Labor Statistics, U.S. Department of Labor, Employment and
Earnings. United States, 1909-75, Bull. 1312-10, 1976.
CA = Social and Economic Statistics Administration, Bureau of the Census,
U.S. Department of Commerce, 1969 Census of Agriculture. U.S.G.P.O.,
Washington, 1973.
CBP = Bureau of the Census, U.S. Department of Commerce, County Business
Patterns. 1967, U.S.6.P.O., Washington, 1968.
CG = Bureau of the Census, U.S. Department of Commerce, 1967 Census of
Governments, U.S.G.P.0., Washington, 1969.
CM = Bureau of the Census, U.S. Department of Commerce, 1967 Census of
Manufactures, U.S.G.P.O., Washington, 1971.
HFW = Wage and Hour and Public Contracts Division, U.S. Department of
Labor, Hired Farmworkers; A Study of the Effects of the $1.00
Minimum Wage Under the Fair Labor Standards Act. Submitted to the
Congress, 1968, January 1968, Table 1.
MY = Bureau of Mines, U.S. Department of Interior, Minerals Yearbook.
1967, U.S.G.P.O., Washington, 1968.
USN = Office of Civilian Personnel, U.S. Navy.
140
-------
TABLE 7-2. SOURCES USED FOR EMPLOYMENT DATA, BY INDUSTRY SECTOR5
1.
2.
3.
4.
5.
6.
7.
.8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
SECTOR
Field Crops
Vegetables & Fruitb <
Livestock
Fish, Forestry, &
Agriculture
Services
Meat Products
Dairy Products
Canning & Preserving
Grain Mills
Beverages
Other Foods
Textiles
Apparel
Mining
Logging
Sawmills
Plywood
Other Wood
Furniture & Fixtures
Pulp Mills
Paper Mills
Paperboard Mills
WASHINGTON
HFW
CA, v.2, ch.4, p. 442
,AS, p. 58
CBP, 49, p. 3
CG, 7, p. 47, p. 22
AS, p. 58
CA, v.2, ch.4, p. 442
HFW
CBP, 49, p. 4
CBP, 49, p. 4
CBP, 49, p. 4
CBP, 49, p. 4
CBP, 49, p. 4
CM, III, p. 48-7
CBP, 49, p. 5
CBP, 49, p. 5
CBP, 49, p. 4
CBP, 49, p. 5
CBP, 49, p. 5
CBP, 49, p. 5
CBP, 49, p. 5
CBP, 49, p. 5
CBP, 49, p. 5
CBP, 49, p. 5
CM, III, p. 48-8
UNITED STATES
HFW
AS, p. 58
,
CBP, 1, p. 6
CG, 3, p. 20
AS, p. 58
CA, p. 442, v.2, ch.4
HFW
CBP, 1, p. 7
CBP, 1, p. 7
CBP, 1, p. 7
CBP, 1, p. 7
CBP, 1, p. 7
CM, I, p.28C
CBP, 1, p. 8
CBP, 1, p. 8
CBP, 1, p. 6
CBP, 1, p. 9
CBP, 1, p. 9
CBP, 1, p. 9
CBP, 1, p. 9
CBP, 1, p. 9
CBP, 1, p. 9
CBP, 1, p. 9
CM, 1, p.32d
(cont'd)
141
-------
TABLE 7-2. SOURCES USED FOR EMPLOYMENT DATA, BY INDUSTRY SECTOR (cont'd)
SECTOR
WASHINGTON
UNITED STATES
22. Printing, Publishing
23. Industrial Chemicals
24. Other Chemicals
25. Petroleum Refining
26. Glass and Stone
27. Cement and Clay
28. Iron and Steel
29. Nonferrous Metals
30. Aluminum
31. Heavy Metal
32. Light Metal
33. Nonelectric Equip.
34. Machine Tools
35. Industrial Equipment
36. Electric Machinery
37. Aerospace
38. Motor Vehicles
39. Ship Building
40. Other Manufacturing
41. Transportation
42. Utilities
43. Communication
CBP, 49, p.6
CBP, 49, p.6
CBP, 49, p.6
CBP, 49, p.6
CM, III, p. 48-8
CBP, 49, p.8
CBP, 49, p.6
CM, III, pp.48-8,48-9
MY, III, p.834
CBP, 49, p.7
CBP, 49, p.7
CBP, 49, p.7
CBP, 49, p.7
CBP, 49, p.7
CBP, 49, p.7
CBP, 49, p.7
CBP, 49, p.7
CBP, 49, p.7
USN
CBP, 1, p.10
CBP, 1, p.10
CBP, 1, p.10
CBP, 1, p.10
CM, I, pp.34,36
CBP,!, p.11
CBP,1, p.11
CM, I, p.36e
CBP, 1, p.llf
CBP, 1, p.12
CBP, 1, p.12
CBP, 1, p.12
CBP, 1, PPJ2-13
CBP, 1, pp.12-13
CBP, 1, p.13
CBP, 1, p.13
CBP, 1, p.13
CBP, 1, p.13
USN
CM, III, pp.48-7,48-8,48-109CBP, 1, pp.7,8,10,13,14
CBP, 49, pp.6,8
BLS-SA, pp.538-9
CBP, 49, p.8
CG, v.7, 47, p.22
CBP, 49, p.8
(cont'd)
142
BLS-US, pp.592-601'
CBP, 1, p.151
CG, v.3, p.21
CBP, 1, p.15
-------
TABLE 7-2. SOURCES USED FOR EMPLOYMENT DATA. BY INDUSTRY SECTOR (cont'd)
SECTOR
WASHINGTON
UNITED STATES
44. Construction
45. All Trade
46. Finance
47. Insurance
48. Real Estate
49. Business Services
50. Personal Services
CBP, 49, pp. 4,10
CBP, 49, pp. 8-9
CG, v.3, p.78
CBP, 49, pp. 10-11
CBP, 49, p. 10
CBP, 49, p. 10
CBP, 49, pp. 11-12
CBP, 49, pp. 11-12
CG, v.7, 47, p.22
CBP, 1, pp. 7,17
CBP, 1, pp. 15-15
CG, v.3, o :>o
CBP, 1, p. 17
CBP, 1, p. 17
CBP, 1, p. 17
CBP, 1, pp. 18-19
CBP, 1, pp. 17-19J
Notes:
The full titles of the data sources are abbreviated as in Table 7-1.
Agricultural employment by sector was estimated for 1967 rather than
combined, in order to minimize aggregation of the table. For Washington,
estimates were calculated by crop, as the state's 1969 fraction of U.S.
hired workers for over 150 days (CA), times U.S. 1967 hired employment for
that crop (HFW), times the ratio of total 1967 Washington farm workers to
hired 1967 Washington farm workers (AS). The totals for each crop in a
sector were then normalized so that their total sum equalled the 1967
total. U.S. employment was derived, by crop, by multiplying national 1967
hired workers for that crop (HFW) by the 1967 national ratio of total to
hired farm workers (AS). Again these figures were summed over all crops
in a sector, then normalized.
d
"CM was used because of CBP disclosure problems.
CM was used because of CBP disclosure problems.
'CM was used because of CBP disclosure problems. For Washington CM also has
disclosure problems, so the mean value of the upper and lower limits of the
range was taken.
143
-------
CM was used because of CBP di
tire problems.
Because of disclosure problems, upper bound of range was used for SIC 31,
and lower bound was used for SIC 19. '
i
BLS was used in order to include railroad employment. State and local
government public transit services, airports, port operations, and highways
were included. Federal government transportation could not be included
because it is not separable in the national input-output table.
CG provided data on state and local government utilities. Federal
electric utilities were also included, with data obtained from Tennessee
Valley Authority, Southeast, Southwest, and Alaska Power Administrations
of the Department of Interior, and 1970 Bonneville Power Authority (BPA)
figures for Washington and all employers. The 1967 BPA figures were
estimated by multiplying the 1970 figures by the ratio of 1967 to 1970
operation and maintenance costs (with the 1970 figure discounted by a
private, non-farm industry productivity index).
CG provided data on state and local government services (health and
hospitals, local libraries, parks and recreation).
144
-------
II. POTENTIAL SOURCES OF REGIONAL GROSS OUTPUT ESTIMATES BY INDUSTRY SECTOR
The 1963 Washington State input-output report (Philip J. Bourque, e_t
aj_., The Washington Economy: An Input-Output Study, University of Washington
6raduate~SchooT of Business Administration and Washington State Department
of Commerce and Economic Development, 1967) provides a reference of data
sources used to estimate gross output in its 1963 table. Manufacturing
sector data relies primarily on the Census Bureau's Census of Manufactures.
Its Census of Business is also used. Various other state, federal, and
trade publications are cited for the other sectors, and similar documents
are likely available for most regions.
Similarly, Definitions and Conventions of the 1967 Input-Output Study
(Social and Economic Statistics Administration, Bureau of Economic Analysis,
Interindustry Economic Division, U.S. Department of Commerce, October 1974)
provides a list of publicly available data sources for the national level,
many of which/ contain regional data. The Census Bureau's Census of
Agriculture is highly disaggregated and detailed by region. Trade publica-
tions and the Interior Department's annual Mi nerajl s Yearbook a 1 so provide'
output data.
145
-------
REFERENCES
CHAPTER 1.
1.
Hitch, Charles J. (_ed.), Modeling Energy-Economy Interactions: Five
Approaches. Resources for the Future, Washington, D.C., 1977.
2. Hogan, William W. and S. Parikh, Comparison of Models of Energy and
the Economy, Stanford Energy Modelling Forum Working Paper EMr 1, 4,
February 1977.
3. House, Peter, Trading Off Environment, Economics and Energy, Lexington,
Mass., D.C. Heath, 1977.
4. Lakshmanan, T.R. and S. Krishnamoorthi, The SEAS Test Model: Design and
Imp!ementation, U.S. Environmental Protection Agency, WasliingtonT, Q.C.,
T97J:
CHAPTER 3.
1. Consad Research Corporation, Regional Federal Procurement Study.
Washington, D.C., Office of Economic Research, U.S. Department of
Commerce Contract 7-35211, October 1967.
2. Czamanski, Stanislaw and Emil E. Malizia, "Applicability and Limitations
in1 the Use of National Input-Output Tables for Regional Studies,"
Papers of the Regional Science Association, xxiii, 1969, 65-78.
3. Goldman, Morris R., "Comments on Czamanski and Malizia," Papers of the
Regional Science Association, xxiii, 1969, 79-80.
4. Isard, Walter, Methods of Regional Analysis, MIT1, 1960.
5. Malizia, Emil and Daniel L. Bond, "Empirical Tests of the RAS Method
of Interindustry Coefficient Adjustment," Journal of Regional Science.
14, no. 3, 1974, 355-65.
6. McMenamin, David G. and Joseph E. Haring, "An Appraisal' of Nonsurvey
Techniques for Estimating Regional Input-Output Models," Journal of
Regional Science, 14, no. 2, 1974, 191-205.
7. Miernyk, William, "Comments of Czamanski and Malizia," Papers of the
Regional Science Association, xxiii, 1969, 81-82.
146
-------
8. Morrison, W.I. and P. Smith, "Nonsurvey Input-Output Techniques at the
Small Area Level: An Evaluation," Journal of Regional Science. 14, no.
1, 1974, 1-14.
9. Round, 0. I.-, "Regional Input-Output Models in the U.K.: A Re-Appraisal
of Some Techniques," Regional Studies, 6, 1972, 1-9.
10. Schaffer, William and Kong Chu, "Nonsurvey Techniques for Constructing
Regional Interindustry Models," Papers of the Regional Science Associa-
tion, xxiii, 1969, 83-101.
11. Walderhaug, Albert, "State Input-Output Tables Derived from National
Data," Proceed!ngs, Amerlcan Stati sti cal Association, Business and
Economic Statistics Section, 1971.
CHAPTER 4. ,
1. Crow, Robert T. "A Nationally-Linked Regional Econometric Model,"
Journal of Regional Science. Vol. 13, No. 2, 1973, 187-204.
2. Glickman, Norman J., "Son of the Specification of Regional Econometric
Models," Papers of the Regional Science Association, Vol. 32, 1974.
3. Glickman, Norman J., "An Econometric Forecasting Model for the
Philadelphia Region," Journal of Regional Science, Vol. 11, No. 1, 1971.
4. Harris,. Curtis, Urban Economies, 1985: A Mult i regional, Multi-Industry
ForecastingModel,Lexington Books, (D. C. Heath), Lexington, Mass.,
1973.
5. Klein, Lawrence R., "The Specification of Regional Econometric Models,"
Papers of the Regional Science Association, Vol. 23, 1969.
6. Klein, Lawrence R., and Norman J. Glickman, "Econometric Model Building
at Regional Level," Regional Science and Urban Economics. 1977, 3-23.
7. Matilla, John M., "A Metropolitan Income Determination Model and the
Estimation of Metropolitan Income Multipliers," Journal of Regional
Science. Vol. 13, No. 1, 1973.
8. Miernyk, William H., "Regional and Interregional Input-Output Models:
A Reappraisal," in Perlman, Leven and Chinitz (eds.), Spati a1. Reg i ona1
and Population Economics, Gordon and Breach, Science PublTshers, Ltd.,
T973.
9. Moses, Leon N., "A General Equilibrium Model of Production, Inter-
regional Trade and Location of Industry," Review of Economics and
Statistics. 1966, 42, 373-397.
147
-------
10. 01 sen, R. J., "An Overview of MULTIREGION," A paper presented at Seventh
Annual Conference on Modeling and Simulation, Pittsburgh, PA, April
1976.
11. Polenske, Karen R., State Estimates of Technology, 1963, D.C. Heath and
Company, Lexington, Mass., 197T.
12. Polenske, Karen R., A Guide for Users of the U.S. Mujtiregional Input-
Output Model (revised"Version), PB-242-558, Department of Commerce,
March 1974.
13. U. S. Water Resources Council, Concepts, Methodology, and Summary Data,
Vol. I of the 1972 OBERS Projections, Economic Activity in the U.S.,
April 1974.
CHAPTER 5.
1. Adams, F. G., C. Brooking and N. J. Glickman, "On the Specification and
Simulation of a Regional Econometric Model: A Model of Mississippi,"
Discussion Paper No. 266, Department of Economics, University of
Pennsylvania, Philadelphia, 1973.
2. Berry, Brian J. L. and Elaine Neils, "Location Size and Shape of Cities
as Influenced by Environmental Factors: The Urban Environment Unit
Large," in The Qualjty of the Urban Environment. (Ed.) H. S. Perloff,
Johns Hopkins University Press, Baltimore, 1969, 257-302.
3. (CEQ) Council on Environmental Quality, The Third Annual Report ofthe
Council on Environmental Quality, U.S. Government Printing Off Tee,'
Washington, D.C. 1972.
9
4. Chinitz, B., "The Effect of Transportation Forms on Regional Economic
Growth," Traffic Quarterly, 14, 1960.
5. Craik, K. H. and E. H. Zube, Perceiving Environmental Quality. Plenum
Press, New York, 1976.
6. Crow, R. T., "A Nationally Linked Regional Econometric Model," Journal
of Regional Science. XIII, 1973, 187-204.
7. Dutta, M., and V. Su, "An Econometric Model of Puerto Rico," The Review
of Economic Studies. XXXVI, 1969, 319-333.
8. L'Esperance, W. L., G. Nestel and G. Fromm, "Gross State Product and an
Econometric Model of a State," American Statistical Association Journal,
LXIV, 1969, 787-807.
9. Giarratani, Frank and Charles F. Socher, "The Pattern of Industrial
Location and Rising Energy Prices," Atlantic Economic Journal. V, 1,
March 1977, 48-55.
148
-------
10. Glickman, N. J., "Son of 'The Specification of Regional Econometric
Models'," Papers of the Regional Science Association, 32, 1974.
11. Glickman, N. J., "An Econometric Model of the Philadelphia Region,"
Ph.D. dissertation, University of Pennsylvania, Philadelphia, 1969.
12. Glickman, N. J., "An Econometric Model of the Philadelphia Region,"
Journal of Regional Science, XI, 1971, 15-32.
13. Harris, Chauncey D., "The Market as a Factor in the Localization of
Industry in the U.S.," Annals of the Association of American
Geographers.
14. Harris, Curtis C., The Urban Economies, 1985. Lexington, Mass.: D. C,
Heath and Co., 1973.
15. Hornsback, Kenneth E., J. Guttman, H. L. Himmalstein, A. Rappaport
and R. Reyna, Studies in Environment, Vol. II. Quality of Life.
Prepared for U. S. EPA. EPA-600/5-73-0126, Washington, D.C., U.S. GPO,
February 1974.
16. Kendrick, J. W. and C. M. Jaycox, "The Concept and Estimation of Gross
State Product," Southern Economic Journal, XXXII, 1965, 153-168.
17. Klassen, Leo H., "Energy, Environment and Regional Development," in
Environment, Regional Science and InterregionaJ Model1ing, (eds.) M.
Chatterji and P! V. Rompuy. Springer-Verlag, Berlin,1976, 1-15.
18. Lakshmanan, T. R. and Fu-chen Lo, "A Regional Economic Model for the
Assessment of Effects of Air Pollution Abatement," Environment and
Planning, Vol. 4, 1972, 73-97.
19. Lakshmanan, T. R., F. Lo and Krishnamoorthi, The OAP Regional
Econometric Model: A Revised Version, Prepared for Erfvironmenta1
Protection Agency by CONSAD, NTIS Publication, Pittsburgh, September
1972.
20. Lakshmanan, T. R. and Walter G, Hansen, "A Retail Market Potential
Model," Journal of the American Institute of Planners, May 1965.
21. Liu, Ben-chieh, Quality of Life Indicators in the U.S. Metropolitan
Areas, 1970, Prepared for the U.S. Environmental Protection Agency,
T975~:
22. Mattila, J. M., "A Metropolitan Income Determination Model and the
Estimation of Metropolitan Income Multipliers," Journal of Regional
Science, XIII, 1973, 1-16.
149
-------
23. .Miernyk,. William H., "Some Regional Impacts of the Rising Costs of
Energy."'Papers, and Proceedings of the Regional Science Association,
1976(b), 201-212. T~T ~~^" : ^~
24. Miernyk, William H., "Regional Economic Consequences of High.Energy
. Prices in the U.S." The Journal of Energy and Development, Spring 1976,
213-239.
25. MITRE Corporation, National Environmental Indices: Air Quality and
Outdoor Recreation. MTR 6159. April 1972.;
26. MITRE Corporation, The PDI Index. Working Paper 7693, Washington, D.C.,
September 1974. ...
27. Perloff, Harvey S. and Loudon Wingo Jr., "National Resource Endowment
and Regional Economic Growth," in J. J. Spehgler {ed.), Natural
Resources and Economic Growth, Conference Papers published by RFF,
1961, 191-212.
28. Perloff, Harvey S., "A Framework for Dealing with the Urban Environment,"
in The Quality of the Urban Environment, (ed.) H. S. Perloff, The Johns
Hopkins University Press, Baltimore 1969, 3-25.
29. Polenske, K., A Hultiregional Input-Output Model for the United States.
Cambridge: Harvard Economic Research Project, 1970.
30. Scully, G. W., "Interstate Wage Differentials: A Cross-Section Analysis,"
American Economic Review. Vol. LIX, 1969, 757-773.
31. Stewart, J. Q., "Empirical Mathematical Rules Concerning the Distribu-
tion and Equilibrium of Population," Geograpjrjc Rev i ew, 37, 1947.
32. Thorn, Gary C. and Wayne R. Ott, Air Pollution Indices. Prepared for CEQ
and Environmental Protection A'gency, December 1975.
33. Thomas, William A. {ed.), Indicators of Environmental Quality. Plenum
Press, New York, 1972.
34. Ullman, Edward L., "Regional Development and the Geography of Concen-
tration," Papers and Proceedings of the^ Regional Science Ass_ociation_,
4, 1958.
CHAPTER 6.
1.
Polenske, Karen R., State Estimates of Technology. 1963, D. C. Heath
and Co., 1974.
150
-------
2. Polenske, Karen R., A Guide for Users of the U.S. Multiregional Input-
Output Model (revised versTonl, PB-242-558, Department of Commerce,
March 1974.
3. Miyazawa, K., "Internal and External Matrix Multipliers in the Input-
Output Model," Hitotsubashi Journal of Economics, Vol. 7, No. 1, 1966,
38-55.
4. Miyazawa, K. and S. Masegi, "Interindustry Analyses and the Structure of
Income Distribution," Metroeconomica. Vol. XV, Fas 2, 3 (1963), 89-103.
5. Yamada, H. and T. Ihara, "Input-Output Analysis of Interregional
Repercussion," Papers and Proceedings of the 3rd Far East Conference of
the Regional Science Association, Tokyo, Japan, 1967, Vol. Ill, 3-31.
U S OOVZBNMEHT PRINTING OFFICE: 1»8O 311-132/70
151
-------
-------
EP 600/7
79-119
EPA
OEMI
The SEAS region model; an
TITLE assessment of current status
and prospects
DATE DUE
GAVLORD 43
BORROWER'S NAME
-------
DATE DUE
BORROWER'S NAME
-------
DATE DUE
-------
United Stales
Environmental Protection
Agency
Olficial Business
Penalty for Private Use
$300
\
Fourth-Class Mail
Postage and Fees Pi
EPA
Permit No. G-35
Washington DC 20460
ERIC
MR H LUGER
LIBRARIAN
USEPA
TECH LIBRARY
EDISON. NJ O8817
012500004
------- |