EPA-6QO/5-77-014
December 1977
Socioeconomic Environmental Studies Series
REGIONAL MANAGEMENT OF AUTOMOTIVE
EMISSIONS: The Effectiveness of
Alternate Policies for
Los Angeles
Office of Air, Land, and Water Use
Office of Research and Development
U.S. Environmental Protection Agency
Washington, D.C. 20460
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development, U.S. Environmental
Protection Agency, have been grouped into nine series. These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology. Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are:
1. Environmental Health Effects Research
2. Environmental Protection Technology
3. Ecological Research
4. Environmental Monitoring
5. Socioeconomic Environmental Studies
6. Scientific and Technical Assessment Reports (STAR)
7. Interagency Energy-Environment Research and Development
8. "Special" Reports
9. Miscellaneous Reports
This report has been assigned to the SOCIOECONOMIC ENVIRONMENTAL
STUDIES series. This series includes research on environmental management,
economic analysis, ecological impacts, comprehensive planning and fore-
casting, and analysis methodologies. Included are tools for determining varying
impacts of alternative policies; analyses of environmental planning techniques
at the regional, state, and local levels; and approaches to measuring environ-
mental quality perceptions, as well as analysis of ecological and economic im-
pacts of environmental protection measures. Such topics as urban form, industrial
mix, growth policies, control, and organizational structure are discussed in terms
of optimal environmental performance. These interdisciplinary studies and sys-
tems analyses are presented in forms varying from quantitative relational analyses
to management and policy-oriented reports.
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
EPA-600/5-77-014
December 1977
REGIONAL MANAGEMENT OF AUTOMOTIVE EMISSIONS:
THE EFFECTIVENESS OF ALTERNATIVE
POLICIES FOR LOS ANGELES
Contract No. 68-01-2235
Project Officer
Roger Don Shull
Office of Air, Land, and Water Use
Office of Research and Development
U.S. Environmental Protection Agency
Washington, D.C. 20460
Prepared for
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
EPA-RIP LIBRARY
-------
DISCLAIMER
This report has been reviewed by the Office of Research and Development,
U.S. Environmental Protection Agency, and approved for publication. Approval
does not signify that the contents necessarily reflect the views and policies
of the U.S. Environmental Protection Agency, nor does mention of trade names
or commercial products constitute endorsement or recommendation for use.
ii
-------
ABSTRACT
This study has two objectives: first, to develop pro-
cedures to evaluate policies for controlling automobile
emissions; and second, to use these procedures to evaluate
specific pollution control strategies for Los Angeles.
The first objective is achieved by developing a rela-
tively quick and reliable method for estimating the cost
effectiveness of travel related policies. The methods used
include application of a behavioral demand model for auto-
mobile travel by mode, purpose and destination, and a model
which predicts the size of the auto stock and its age distri-
bution. These models are used to compute the costs to
society and individual travelers of various policies, and to
compute the emission reduction effects of various policies.
In applying these procedures to Los Angeles, the fol-
lowing specific strategies were evaluated:
• increased gas taxes;
• taxes on vehicle emissions per mile based on odometer
readings and emissions tests;
e nonresidential parking surcharges;
• extensions of route miles by conventional bus;
• annual taxes based on vehicle model, make and year.
The report's findings indicate that implementation of these
policies could significantly decrease pollution. Emission
taxes and gasoline taxes are particularly effective strate-
gies; parking taxes are a less effective but still viable
policy. Tax-induced decreases in pollution are reinforced
by improvements in conventional bus service.
iii
-------
TABLE OF CONTENTS
Chapter 1. SUMMARY ................. 1
Policy Instruments ................ 2
Effects of Policies on Travel Behavior ...... 4
Effect of Tax Strategies on Auto Stock ...... 7
Cost-Effectiveness of Travel-Related Policies. . . 10
Models and Data .................. 13
Travel Demand ................. 13
Auto Stock ................... 15
Cost of Improved Transit ............ 15
Data ...................... 15
Conclusions .................... 16
ATTACHMENT TO CHAPTER 1 ............... 18
Tax Levels .................... 18
Chapter 2. POLICY ANALYSIS FRAMEWORK AND
INSTRUMENTS ............... 21
Policy Analysis .................. 21
Travel Behavior ................ 21
Travel Demand ................ 22
System Performance ............. 23
Analysis of Policy Effects ........... 24
Analysis of Policy Costs .......... 24
Probability Choice Models ..... ..... 27
Auto Stock Adjustments ............. 29
Auto Stock Dynamics ............. 29
New Car Sales ................ 30
Used Car Scrappage ............. 32
Policy Instruments ................ 36
Parking Tax .................. 39
A Priori Analysis of the Effects of a
Parking Tax .............. . . 39
Review of Other Research . . ........ 42
Conclusion ................. 46
Gasoline Tax .................. 46
A Priori Analysis of the Effects of a
Gasoline Tax ................. 47
Review of Previous Research ......... 50
Conclusion ................. 53
Emissions Tax ................. 53
A Priori Analysis of the Effects of an
Emissions Tax ............... 54
Review of the Literature .......... 58
iv
-------
TABLE OF CONTENTS, Continued
Chapter 2. (Continued)
Transit Improvements 61
List of References, Chapter 2 65
Chapter 3. EFFECTS OF POLICIES ON TRAVEL BEHAVIOR. . 67
VHT and Trip Reductions with the Current Transit
System: Work and Shop Trips 70
Specific Tax Levels 70
Model Predictions 73
Predicted Changes in Travel Patterns 76
Effects of Policy by Distance of Trip. ... 76
Effects of Destination Changes . 79
Effects of Transit Availability 80
VMT and Trip Effects with Improved Transit System:
Work and Shopping Trips 80
The Aggregate Short-Run Travel Effects 84
Uncertainty of the Results 90
Model Structure and Estimates 92
Data 93
Model Application 93
Conclusions 96
List of References, Chapter 3 98
Chapter 4. EFFECTS OF POLICIES ON THE AUTO STOCK . . 99
Effects of a Model, Make and Year Tax 99
Effects of a Mileage-Emissions Rate Tax 109
Ownership Costs Ill
Retrofitting 113
Uncertainty of the Results 119
Effects of an MMY Tax 119
New Car Sales 119
Used Car Price 119
Aggregate Scrappage Rate 120
Scrappage Rate by Age of Auto 120
1975 Price of New Cars 120
1975 Price of Used Cars 120
Used Car Price by Age of Car 120
The Effects of Taxes on the Size of the
Auto Stock 122
v
-------
TABLE OF CONTENTS, Continued
Chapter 4. (Continued)
Relative Scrappage Rates by Age of Auto. . . 122
Effects of an MER Tax 122
Chapter 5. POLICY COST-EFFECTIVENESS
Individual Costs 124
Cost to Individuals of Gas/Emissions Tax. . . . 125
Cost to Individuals of Parking Tax 125
Cost to Individuals of Gas/Endssions Tax with
Transit Improvements 12"?
Individual Cost-Effectiveness 130
Resource Costs 133
Bus Costs 133
Emissions Test 135
Resource Cost-Effectiveness 137
Fuel Consumption Effects 139
List of References, Chapter 5 142
Chapter 6. CONCLUSIONS 143
Nonquantified Factors 144
Incidence of Costs 144
Feasibility and Phasing of Policies 145
Land Use Effects 147
Short-Term Research Recommendations 148
Additional Strategies 149
Research Design 151
Appendix A. TRAVEL DEMAND MODEL 152
The CRA Disaggregated Demand Model 153
Application of the Model to Zonal Aggregates ... 158
Approximation of Zonal Frequencies 159
Estimation of Variance-Covariance Terms .... 161
Density Function for Trip Distances 162
Constraints on Variance-Covariance Terms . . 166
Estimated Formulas for Variance-Covariance
Terms 168
Application and Tests of the Travel Demand Model
with 1967 LARTS Data 179
vi
-------
TABLE OF CONTENTS, Continued
Page
Appendix A. (Continued)
Work Trip Mode Split 180
Auto Versus Transit 182
Auto Versus Carpool 185
Auto Versus Serve Passenger 187
Auto Versus Walking 190
Mode Probabilities 192
Estimated Trips by Mode 195
Aggregate Work Trips 200
Evaluation of Work Trip Model 203
Shopping Trip Mode Split 204
Auto Vs. Transit 206
Auto Vs. Carpooling 206
Auto Vs. Serve Passenger 209
Mode Probabilities 209
Estimated Trips by Mode 211
Shopping Trip Destination Choice 216
Shopping Trip Frequency 218
Evaluation of Shopping Trip Model 221
List of References, Appendix A 227
Appendix B. AUTO STOCK ADJUSTMENT MODEL 228
Calibration of the Model 228
New Car Sales 228
Used Car Price 230
Scrappage Rate 232
Identities 234
Scrappage Rate by Age of Auto 235
Application of the Model to Policy Scenarios . . . 235
New Car Sales 235
Stock of Used Cars 236
Scrappage Rate by Age of Car 236
List of References, Appendix B 239
Appendix C. BUS COST MODEL 240
Estimates of Operating and Capital Costs for
Bus Service 24J
Operating Costs 241
Express and Conventional Bus 245
Minibus and Conventional Bus 247
vn
-------
TABLE OF CONTENTS, Continued
Appendix C. (Continued)
Capital Costs 248
Equipment Costs 248
Shop and Yard 250
Changes in Bus Service Output Variables Resulting
from Changes in Service Characteristics ..... 251
General Approach for Conventional Bus Service . 251
Ratio of Active to Peak Buses 254
Calculation Method of Bus Outputs for
Conventional Bus 255
Estimates of Incremental Costs of Changes in
Bus Service 257
General Method for Conventional Bus 257
Calculation of Incremental Costs of Bus
Service Changes 258
List of References, Appendix C 259
Appendix D. MODELS AND DATA 260
Travel Data 260
Zonal System 263
Trip Counts 264
Trip Purpose and Coding of Round Trips . . . 264
Mode Categories 270
Automobile System Performance Variables .... 272
Auto Line-Haul Time 272
Auto Trip Distance 272
Auto Costs 273
Transit System Performance Variables 273
Access Time 274
Identifying Interzonal Routes 276
Quantification of Characteristics of
Interzonal Bus Travel 277
Line-Haul Time 277
Wait Time 277
Schedule Delay 281
Total Line-Haul Time 282
Number of Transfers 282
Transit Fares 282
Socioeconomic and Descriptive Variables 283
Area 283
Number of Households. 283
Vehicle Availability 283
Income 284
Retail Employment 284
List of References, Appendix D 285
viii
-------
LIST OF FIGURES
Page
1. Framework for Analyzing 'Policy Effects 1
2. Costs of Policies to Individuals 25
3. The Distribution Function for Distance
of Trips Between Zones 165
4. Excerpt from LARTS Travel Interview Form 261
ix
-------
LIST OF TABLES
1. Estimated Effects of Pollution Control
Strategies on VMT' s and Auto Trips 6
2. Effects on Auto Stock of Annual Emissions
Tax on All Pollutants: Los Angeles
and Orange Counties 8
3. Example of Effects of Tax Scenarios or. Age
Distribution of Autos Over Time 9
4. Cost to Individuals and Resource Cost per
VMT Reduced 12
5. Estimate of Effect of Policies on Average
Percent and Absolute Cost Increases of
Average 1974 Round Trio ..... 15
6. Gasoline and Emissions Taxes Associated with
Each Level of Increased Variable Costs
per Mile 20
7. Elasticities of Number of Trips with Respect
to Auto Parking Price -13
8. Estimated Variable Costs per Mile of
Operating an Automobile 48
9. Estimate of Average Percent and Absolute
Cost Increases of Average 1974 Round Trip
Incurred by Policy Scenarios ... 71
10. Gasoline and Emissions Taxes Associated with
Each Level of Increased Variable Costs
per Mile 72
11. Effects of Taxes on Work Trips 74
12. Effects of Taxes on Shopping Trips 75
13. Mode Split Estimates on Work Trips 77
14. Mode Split Estimates on Shopping Trips ...... 78
15, Effect on Work Trips of Gas or Emission Tax
Combined with Transit Improvements 82
16. Effect on Shopping Trips of Gas or Emission
Tax Combined with Transit Improvements 83
17. Mode Split Estimates with Transit
Improvements for Work Trips 85
x
-------
LIST OF TABLES (Continued)
18. Mode Split Estimates with Transit
Improvements for Shopping Trips 86
19. Estimated 1974 Trip Purpose Shares 88
20. Estimated Effects of Pollution Control
Strategies on VMT's and Auto Trips 91
21. Base Case Estimates of 1975 Auto Stock
Characteristics for Los Angeles and
Orange Counties 10]
22. Estimates of 1975 Mid-year Auto Stock
Characteristics by Age of Car:
Los Angeles and Orange Counties 102
23. Auto Stock Effects of Annual Tax on Carbon
Monoxide: Los Angeles and Orange Counties . . . 105
24. Auto Stock Effects of Annual Tax on
Hydrocarbons: Los Angeles and Orange Counties . 106
25. Auto Stock Effects of Annual Tax on
Nitrogen Oxide: Los Angeles and Orange
Counties 107
26. Auto Stock Effects of Annual Emissions
Tax on All Pollutants: Los Angeles and
Orange Counties 108
27. Example of Effects of Tax Scenarios on
Age Distribution of Autos Over Time 110
28. Effects, by Age Category of Auto, of HER
Tax Equivalent to Average 25 Percent
Increase in Variable Costs per Mile 112
29. Retrofit Costs for CO and HC with
Catalytic Reactor 114
30. Retrofit Cost for NOx with Exhaust
Recycle Device 116
31 Costs to Individuals of Gasoline or
Emissions Tax 126
32. Costs to Individuals of a Parking Tax 128
33. Effects of Parking Tax on Auto Trips
Exclusive of Driver Serve Passenger Trips .... 129
xi
-------
LIST OF TABLES (Continued)
Pac
34. Costs to Individuals of a Gasoline or
Emission Tax with Transit System
Service Improvements 131
35. Co-st to Individuals per VMT Reduced 132
36. Costs of Bus Service Improvements for
Travel Within Los Angeles and Orange
Counties 136
37. Resource Cost per VMT Reduced 138
38. Fuel Consumption Changes as a Result of the
Policies: Los Angeles and Orange Counties . . . 140
39. Estimated Relationships in CRA Disaggregated
Demand Model 154
40. Nonzero Variance-Covariance Terms for
Bus (a) and Auto (b) Choice 169
41. Nonzero Variance-Covariance Terms for
Walk (u) and Auto (a) Choice 170
42. Nonzero Variance-Covariance Terms for
Car Pool with h Passengers (oh) and
Auto (a) Choice 171
43. Nonzero Variance-Covariance Terms for
Driver Serve Passenger (s) and Auto (a)
Choice 172
44. Variance of Several Comparison Functions .... 177
45. Variables and Sources for Auto vs. Transit . . . 183
46. Carpooling Variables 188
47. Driver Serve Passenger Variables 191
48. Walking Variable 192
49. Estimated Mode Probabilities 194
50. Variance of Comparison Function Between
Auto and Other Modes 196
51. Actual Work Trips 198
52. Estimated Work Trips 199
53. Estimated vs. Actual Trips by Mode for
172 Zonal Interchanges 201
xii
-------
LIST OF TABLES (Continued)
Page
54. Estimated vs. Actual Mode Shares for
LARTS Region 202
55. Auto vs. Transit Variables 207
56. Carpool Variables 208
57. Serve Passenger Variables 210
58. Actual Shopping Trips by Mode for
15 Zonal Interchanges 213
59. Estimated Shopping Trips by Mode for
15 Zonal Interchanges 214
60. Summary of Estimated vs. Actual
Shopping Trips by Mode for 15 Zonal
Interchanges 215
61. Actual vs. Estimated Destination Shares
for Shopping Trips from 15 Zonal
Interchanges 219
62. Actual vs. Estimated VMT's for Shopping
Trips Including Mode and Frequency
Choice for 15 Zonal Interchanges 220
63. Actual vs. Estimated Shopping Trip
Frequency for 5 Zone Origins 222
64. Actual vs. Estimated VMT's, Including
Mode Choice, Destination Choice and
Frequency of Trip, for 15 Zonal
Interchanges 223
65. Actual vs. Estimated Mode Shares for
LARTS Region 225
66. Equations and Identities in the
Auto Stock Model 229
67. Cost Estimation Equations for 1973 244
68. Approximate Purchase and Annual Capital
Costs in 1973 Dollars 249
69. Allocation of One-way Trips Between Land
Uses (LARTS Categories) to Round Trip
Purpose Categories 268
xiii
-------
M2QOCEDGEMENTS
We wish to acknowledge the helpful critical review of the original
EPA project officer for this study, Marshall Rose. Other helpful reviewers
included David Syskowski and Joel Horowitz, both of EPA, and Louise Skinner
of the Federal Highway Administration, U.S. Department of Transportation.
A number of individuals affiliated with southern California
transportation organizations gave freely of their time and information.
Their efforts contributed materially to the accuracy of the study, though
they are not, of course, responsible for any errors. These people include:
Jerry Bennet, Huong Kim, Karl Wellisch, Sterling Sorenson, and Bob Blythe,
of CALTRANS; Dave McCullough and John Curtis of SCRTD; Murray Goldman of
SCAG; David Schilling, John Mauroidis, Doublas P. Blankenship and Nancy
Coss of OCTD; Chris Ferrel of Long Beach Transit; and Vince Desimone of AAA.
None of the above people or organizations necessarily hold opinions
supporting this report or concur with results forwarded in it.
This study was completed by Charles River Associates Incorporated,
Cambridge, Massachusetts, under EPA Contract No. 68-01-2235, Frederick C.
Dunbar, Principal Investigator. The research team included Jon Press,
Shulamit Kahn, Eleanor Maxwell, Joyce Mehring, David Blau and
Carol Freinkel.
xiv
-------
1. SUMMARY
Among the most important policy issues in controlling
mobile source air pollution is determining whether a system
of incentives to decrease auto travel is preferable to direct
regulation of auto use. Most plans for meeting federal air
quality guidelines have emphasized regulation — for example,
rationing gasoline, restricting parking availability, placing
quotas on numbers of auto trips, and making retrofitting
devices mandatory. More recently, attention has been given
to the effectiveness of taxes on auto use and of transit
improvements which, when implemented, allow more individual
freedom of choice in travel behavior, while rewarding those
choices which lead to pollution abatement.
The study reported here had the following two objec-
tives: first, to develop procedures for evaluating policies
which create incentives for controlling automotive air pol-
lution; second, to use these procedures to evaluate a number
of specific pollution control strategies for Los Angeles.
The first goal was achieved by developing a relatively
quick and reliable method for estimating the cost-effective-
ness of auto disincentives and transit improvements in
changing travel and auto ownership patterns to meet air
quality goals. The methods used in this study are based on
the application of behavioral demand models to existing urban
transportation data bases.
-------
To achieve the second goal, policy evaluations were per-
formed by comparing the cost-effectiveness of various
pollution control strategies. The findings indicate that
incentive measures can cause major decreases in pollution.
Emission taxes and gasoline taxes are particularly effective
strategies; parking taxes are a less effective but still
viable policy. Tax-induced decreases in pollution are rein-
forced by improvements in conventional bus service.
In the remainder of this section, we provide additional
detail on the policy findings and outline the policy evalua-
tion methodology.
POLICY INSTRUMENTS
There are a large number of potential automobile pollu-
tion control strategies, and a full evaluation of all of them
is beyond the scope of most regional planning agencies. A
small group of pollution control strategies was selected for
in-depth analysis on the basis of an a priori determination
of their probable cost-effectiveness. The necessity for this
selection of policy instruments motivated an investigation
of the normative aspects of pollution and its control.
Automobile pollution should be viewed as an external
social cost of the t -'chnology and individual behavior which
cause it. An individual car owner has no incentive to
control his emissions because his contribution is a small
percentage of the whole; thus the individual does not pay
the costs borne by others as a consequence of his actions
and there is no noticeable effect caused by his sacrifice.
Some form of institutional action or adjustment is called
for when the problem becomes a social burden. An optimal
form of control would be for motorists to pay the marginal
social cost of their pollution constrained by other factors
-------
such as distributional justice or political feasibility. If
it was known what benefits society as a whole receives from
air pollution abatement, an optimal level of air pollution
would occur when the marginal social cost of pollution
abatement equaled the marginal social benefit derived from
that decrease in air pollution. A tax could be set at a
level which would induce motorists to reduce their contribu-
tions to air pollution by driving less, buying cleaner cars,
retrofitting, or other methods. The level of emissions that
resulted would be the level at which the cost of reducing
emissions by one more unit equaled the benefit gained by that
reduction.
However, due to the inadequacy of conventional policy
evaluation tools, and possibly because of the difficulty in
measuring the various social damages caused by air pollution,
attempts to improve air quality have focused on regulation
of technology and behavior rather than the imposition of
incentives for pollution abatement. Nonetheless, the notion
of applying disincentives rather than regulations to auto
travel as abatement policy is appealing on the grounds of
social welfare and personal freedom. Typically, travelers
will be better off and scarce resources (including clean air)
will be allocated more efficiently when tripmakers have more
choice in travel options. Thus, taxation policies and
transit improvements designed to meet ambient air quality
constraints are generally preferable to direct regulation.
These considerations indicated that the following poli-
cies were worthy of in-depth evaluation:
• increased gasoline taxes;
• taxes on vehicle emissions per mile estimated from
periodic odometer readings and emissions tests;
• nonresidential parking surcharges;
-------
• extension of route miles by conventional bus to re-
place auto trips foregone as a result of pollution
policies;
• annual taxes based on vehicle model, make and year.
The effects of these policies in reducing emissions are
measured in terms of the following:
• reductions in vehicle miles traveled (VMT's) in
personal autos;
• reductions in number of auto trips (cold starts);
• reductions in auto fleet size;
• changes in the age distribution of cars;
• incentives to retrofit.
For the purpose of exposition, the effects of the poli-
cies can be divided into two types: effects on travel beha-
vior and effects on the auto stock. These are discussed in
the following two sections. A third section compares the
cost-effectiveness of the various strategies. Taken as a
whole, these sections summarize the policy evaluation
findings of the study. The models and data developed and
used in the study are briefly described at the end of this
chapter.
EFFECTS OF POLICIES ON TRAVEL BEHAVIOR
The effects of pollution control strategies on travel
behavior are simulated using the disaggregate travel demand
model. The travel-related policies under consideration
include: a tax on the variable costs per mile of autos
related to either gasoline consumption or vehicle emissions;
a surcharge on all non-residential parking; and an extension
*
of route miles of conventional bus service. The demand
*An attachment at the end of this chapter presents the
actual levels for the various taxes.
-------
model predicts travel behavior for work trips and, sepa-
rately, for shopping trips. After the effects of the
policies have been simulated for these trip purposes, the
trip and VMT reductions are extrapolated to include all trip
purposes. The base year for the forecasted effects is 1974;
that is, the travel demand models were used to predict travel
behavior in 1974 in the absence of any of the proposed pollu-
tion control policies. For each tax strategy, four levels
of tax were simulated and their effects on VMT's and auto
trips estimated in terms of the percentage change from the
1974 base forecast. The effect of additional bus service
combined with a selected number of auto disincentives was
also simulated.
The simulated effects of the pollution control strate-
gies are presented in Table 1. A detailed account of these
effects on travel behavior, broken down by trip purpose and
mode of travel, is given in Chapter 3. All strategies reduce
auto travel significantly, though there are variations in
efficacy.
The gasoline and emissions taxes are more effective in
reducing VMT's than parking taxes when compared on a cost per
trip basis. The reasons for this are as follows: (1) taxes
which increase variable costs per mile have a greater effect
on long trips than do parking taxes, thereby inducing more
carpools and diversions to bus for those trips which contri-
bute most to VMT's (see the attachment); (2) parking taxes
can be avoided by driver serve passenger (chauffeured) auto
trips which increase the auto mileage associated with indi-
vidual trips.
The addition of bus service to serve auto tripmakers
when a tax is imposed substantially increases mode diversion
and further reduces VMT's. This effect is especially
-------
Table 1. ESTIMATED EFFECTS OF POLLUTION
CONTROL STRATEGIES ON VMT'S AND AUTO TRIPS
Gas or emissions tax:
variable cost
per mile increase3
25?
50
75
100
Parking tax:
parking cost
increase
$0.25
^ cr
^ . J V
0.75
1 . 00
Transit system
improvements
with variable cost per
mile increase3
25?
100
% Change
in VMT's
- 7.40?
-13.96
-19.58
-24. 13
- 5.04?
- 5.55
-13.07
-15.43
-20.21?
-35.77
% Change in
auto trips
- 5.45?
-iO.50
-15.27
-20.39
- 7.55?
-14.46
-19.15
-21.33
-16.43?
-29.63
Total VMT's,
millions oer
weekday^
57.940
53.635
50.319
47.472
59. 4>6
56.576
5^.392
52.915
49.925
40. 129
Table 5 in the attachment converts taxes into absolute and percentage
~-r..~"s "'or j/e';~e tr;^ lengths. Table 5 in t^e attac^enr also conver"t
variaoie cost per mile increases into equivalent taxes en ga.-iciine anc
•^r'. -_B\ ;--s .
"Based on an estimated total of 62.570 million VMT's per average week-
day in 1974 for the Los Angeles Air Quality Control Region.
-------
prominent for long trips where transit has a substantial cost
advantage over auto travel after the imposition of taxes
which increase auto variable costs per mile, such as gasoline
or emission taxes.
EFFECT OF TAX STRATEGIES ON AUTO STOCK
Several emission tax strategies were simulated to
determine their effects on the size of the auto stock, the
age distribution of cars, and the incentive for motorists to
retrofit. The design of such taxes is rather complex and
will not be described here (see Chapter 4), except to note
that they are higher for older cars.
The effects of two emission tax levels representing a
high and low range are summarized in Table 2. it can be
seen that the auto stock would decline in size as a result of
the tax. More importantly, from the standpoint of reducing
pollution, the scrappage rate of older cars increases sub-
stantially.
The impact of emissions taxes on the age distribution
of cars is cumulative over time. This can be seen from the
example given in Table 3 where the effect of the imposition
of a tax in 1975 is traced through to 1976. The accelerated
scrappage of older cars in response to the tax would have a
significant impact on the average emissions of the auto
fleet.
Emissions taxes which are also based on mileage
(odometer readings) provide incentives to retrofit. The
results presented in Chapter 4 indicate that voluntary
retrofitting would become widespread at emission tax rates
on the order of 50 percent of the average auto variable
cost per mile. In 1974 this tax rate would have been about
$0.03 per mile.
-------
Table 2. EFFECTS ON AUTO STOCK OF
ANNUAL EMISSIONS TAX ON ALL POLLUTANTS:
LOS ANGELES AND ORANGE COUNTIES
Indicator
Total auto stock (in vehicles)
New car sales (in vehicles)
Used car price, average
Aggregate scrappage rate
Average present value of tax,
al ! cars
Base case
4,240,053
405,896
$1272
0.0779
— — «_ —
Low tax3
4,149,052
373,260
$908
0. 1015
$392
High taxb
3,971 ,173
346,366
$574
0. 1 506
$784
Model year
1975
1974
1973
1972
1971
1970
1967-1969
pre-1967
Base case
scrappage
rate
0.0013
0.0028
0.0058
0.0120
0.0239
0.0449
O.I 132
0.1898
Low tax3
Annual
tax
S 56
62
68
81
105
122
158
176
Scrappage
rate
0.0013
0.0028
0.0059
0.0125
0.0266
0.0557
0. 1783
0.4398
High taxb
Annual
tax
$1 13
125
135
162
2! !
245
317
350
Scrappage
rate
C.OOI7
0.0037
0.0079
0.0171
0.0378
0.0314
0.2808
0.6290
Low tax equals $7/gm HC + $l/gm CO + $8/gm NOx.
High tax equals Sl4/gm HC + S2/gm CO + SI6/gm NOx.
8
-------
Table 3. EXAMPLE OF EFFECTS OF TAX SCENARIOS ON
AGE DISTRIBUTION OF AUTOS OVER TIME
Model year
1976 (assumed)
1973-1975
1970-1972
1967-1969
pre-1967
Change in
auto stock
1975 age
distribution
(assumed)
—
0.30
0.25
0.20
0.25
1976 age distribution
No tax
0.10
0.29
0.24
0.17
0.20
+2%
Low tax
0.10
0.32
0.26
0.17
0. 15
-5%
High tax
0.10
0.35
0.28
0.16
0. 1 1
-13$
-------
COST-EFFECTIVENESS OF TRAVEL-RELATED POLICIES
In order to compare policies, it is useful to quantify
their cost-effectiveness. This procedure normalizes the
various pollution control strategies in terms of their policy
objectives. Two cost-effectiveness measures were computed.
The first uses costs to individual households, while the
second uses resource (or social) costs. The common base for
these costs is the degree of VMT reduction. This, of course,
is only a proxy for the variable of most interest — the
decrease in emissions. The cost per VMT foregone is nonethe-
less useful as a method for comparing policies because VMT's
are strongly correlated with emissions.
The costs of the policies to individuals include two
components: first, the increased money costs associated
with travel; second, the opportunity cost associated with
travel foregone by the auto driver alone mode. The first
type of cost is simply the sum of the taxes paid as a result
of the policy. The second type of cost requires more elab-
oration.
As the cost of travel increases for auto driver alone
trips, the number of such trips declines owing to people
choosing alternative modes and destinations which were
formerly less desirable. Sometimes the change in modes
results in longer commute times and lower costs, as in the
cases of individuals switching to carpools or transit for
long trips. Some individuals will choose these alternatives
at relatively low increases in automobile travel costs,
reflecting relatively little discomfort in making the switch;
others require a high auto disincentive, indicating that the
opportunity cost of foregoing auto drive alone modes is rela-
tively great. Calculation of these costs requires estima-
tion of the changes in consumer surplus (which is, using
some simplifying assumptions, the change in the area under
the travel demand curve).
10
-------
The resource costs, that is, the costs of resources
which have alternative uses for society, must be calculated
differently. Briefly, the resource costs include the oppor-
tunity costs to households plus the costs of administering
the tax policy. These social costs do not include the taxes
paid by individuals because such transactions reflect trans-
fer payments rather than utilization of resources.
The results of computing individual and resource cost-
effectiveness of policies is presented in Table 4. Several
conclusions emerge from these calculations:
1. Parking taxes are considerably less cost-effective than
other policies because they induce driver serve pas-
senger trips and, compared to per-mile charges, the
longer the trip, the less their relative cost to the
tripmaker.
2. Tax collections from individuals are considerably higher
than resource costs, indicating that the taxes can be
used for administering the programs (including the test
facilities for an emissions tax) with enough remaining
to help alleviate adverse income distribution effects;
note that these tax schemes will be regressive unless
a redistribution program is incorporated.
3. Because the cost-effectiveness in terms of VMT's differs
only slightly between a gas tax and an emissions tax
near the high range of taxes (around $0.05 per mile),
the cost-effectiveness for actual reduction of emissions
may be significantly greater for emissions taxes,
because it will cost relatively more to drive higher-
polluting vehicles, and we can predict these vehicles
will be used relatively less frequently.
4. The resource costs per unit of pollution reduction are
greater with improved bus service, but the cost to
11
-------
Table 4. COST TO INDIVIDUALS AND RESOURCE COST PER VMT REDUCED
Policy
Gas tax:
Variable cost per mile increase
25%
50
75
100
Emissions tax:
Variable cost per mile increase
25%
50
75
100
Parking tax:
Parking cost increase
$0.25
0.50
0.75
1 .00
Transit system
improvements with gas tax:
Variable cost per mile increase
50$
100
Transit system
improvements with emissions tax:
Variable cost per mile increase
50%
100
Cost to individual
per mile reduction
$0.1852
0 . 1 886
0.1935
0.2612
$0.1852
0. 1886
0. 1935
0.2012
$0.4033
0.3922
0 . 392 1
0.3985
$0.1208
0. 1 149
$0. 1208
0. 1 149
Resource cost
per mile reduction
$0.0063
0.0123
0.0177
0.0217
$0.0216
0.0205
0.0235
0.0264
$0.0199
0.0377
0.0535
0.0675
$0.0229
0.0292
$0.0285
0.0324
12
-------
individuals is less due to increased mode diversions. More-
over, it appears that the tax revenues would be more than
adequate for covering the cost of additional bus service.
MODELS AND DATA
The procedures used to evaluate the various policies
included the application of a series of behavioral models
to existing transportation planning data. Figure 1 indicates
the role of models in policy analysis. As can be seen,
policy instruments are quantified in terms of model inputs.
Model relationships are exercised to determine variables
(model outputs) which are used in the cost-effectiveness
computations. The models and data are briefly defined below.
Travel Demand
The demand models used in the study are estimates of
individual probabilities of making specific travel-related
choices. The theoretical and statistical development of
these models, termed disaggregate demand models, is treated
*
in other studies. The models predict the following choices:
• whether bus or auto will be used for work trips;
• whether bus or auto will be used for shopping trips;
• which destination will be chosen for shopping trips.
To apply these models, they were generalized in the
following ways:
• other mode choices were added, including walking and
a variety of shared auto rides;
• the model was adjusted for application to sketch plan
zones;
*The particular model used in this report is developed in:
Domencich, Thomas A. and Daniel McFadden. Urban Travel
Demand. Amsterdam, North-Holland, 1975. 215 p.
+The travel demand models are developed for application in
Appendix A.
13
-------
Figure 1
FRAMEWORK FOR ANALYZING POLICY EFFECTS
Policy
Instruments
Gasoline Taxes,
Emissions Taxes,
Parking Taxes
Model
Inputs
Money
Cost
Line-Haul
and Wait
Time
Walk
Time
Model
Relationships
Mode Choice,
Destination Choice
Model, Make
and Year
Tax
Ownership Costs
of New and
Used Cars
New and Used Car
Safes, Scrappage
Rates
Model
Outputs
VMT's,
Number of Trips
Auto Stock Size,
Age Distribution
of Autos
Effect on
Pollution
I
Emissions
14
-------
• the model application results were expanded to include
trip purposes other than work and shopping.
Travel demand by mode is forecast as a function of trip
costs, trip time and socioeconomic variables. Policy instru-
ments affect travel demand, and consequently VMT's and auto
trips, by causing changes in the relative trip costs and
times of various modes and destinations.
Auto Stock
The model used to estimate the effects of policy on auto
ownership is a multi-equation system which predicts new car
sales, scrappage, total auto stock, and the age distribution
of cars. Emission control policies affect the auto stock
size and age distribution by changing the cost of owning
*
cars of different vintages.
Cost of Improved Transit.
To determine the costs of additional bus service, an
econometric bus cost model was estimated. This model is
linked to a bus network with an algorithm which computes
bus hours, bus miles, and number of vehicles for each level
of service. These variables are the inputs into the cost
model.
Data
The data used for analysis of pollution control strate-
gies were derived from a number of sources. Principal among
these is a household survey of trips performed by the Los
Angeles Regional Transportation Study (LARTS). The trip
records of a 24 hour period are tabulated in a trip table
at the sketch plan zone level of aggregation. The travel
*The auto stock adjustment model is developed in Appendix B,
+The bus cost models are developed in Appendix C.
15
-------
data from the survey were enriched with socioeconomic
data from the 1970 Census and with a bus network developed
from maps and schedules. Auto stock effects of policies were
analyzed with auto registration data from a private service.
The area covered by the data includes Los Angeles County
south of the San Gabriel mountains and Orange County. This
area corresponds approximately to the Los Angeles Air Quality
*
Control region.
CONCLUSIONS
The approach to aggregate urban transportation policy
evaluation developed in this study holds considerable promise
for planners who must make quick analyses of a wide ranqe of
travel-related policy options. The predictive capability of
the methods and models was checked at several stages of
their development and found to be relatively accurate. The
resources and time necessary for using the approach are small
by conventional urban transportation planning standards.
As for the policies themselves, all were found to be
valid strategies for pollution control, though they vary in
cost effectiveness. As a general rule, the more direct a
disincentive the more effective it will be. If one desires
to reduce emissions, then those taxes imposed to abate
pollution are of most use when they relate as directly as
possible to the actions and technology which cause emissions.
In the instant case, emissions and gasoline taxes are more
effective than parking taxes because the former tax pollution
more directly than the latter.
In the longer run, emissions taxes become the more
powerful instrument. Though costlier to implement than gas
taxes, they cause changes in the age distribution of cars
*Data issues are described in detail in Appendix D
16
-------
and induce retrofitting, thus causing a significant decline
in emissions from the auto fleet. Also, as emissions
decrease, the taxes themselves decline. They are self-regu-
lating in ways that gasoline taxes, parking taxes and bus
improvements are not. It is especially pertinent to note
that high capital transit improvements, such as fixed rail
systems, would be a costly solution to the pollution problem
compared to either low capital bus improvements or changes
in auto technology induced by emissions taxes.
Several nonquantifiable factors also need to be con-
sidered in order to fully evaluate the various policies.
For example, the incidence of the tax policies is, in all
likelihood, regressive; some strategies for redistributing
the income lost in taxes should probably be considered.
Also, if a combination of strategies is used (such as taxes
and bus line extensions) then there should probably be
joint planning and phased implementation to ensure that
sufficient bus capacity will exist when taxes are imposed.
Finally, permanent changes in the transportation network
will have long-range effects on land use, and these should
be examined in advance to determine whether they conform to
existing land use plans and preferences. Most of the poli-
cies considered here would tend to cause more clustered
development.
Future research should probably focus on alternative
low capital system strategies rather than methodology
development. Promising policy options include wide area
transit using integrated paratransit feeder and express bus
line-haul, and improved traffic management. However, little
is now known about the effectiveness of these strategies,
and it will be some time before they can be thoroughly
evaluated. In the meantime, policies such as those consi-
dered in this report may be the best short-term solutions to
mobile source pollution problems in urban areas.
17
-------
ATTACHMENT TO CHAPTER 1
TAX LEVELS
Four tax levels were simulated for each policy. For
emissions and gasoline taxes, the four levels represented
percentage increases in auto variable costs per mile.
Parking charges, which are currently insignificant for most
trips in Los Angeles, were increased in absolute amounts.
In order to compare these sets of scenarios with each other,
Table 5 presents the average absolute and percentage cost
increase for work and shopping round trips. This table
allows comparison and conversion between percentage and
absolute terms of reference for the tax policies. It can
be seen that the average cost per trip incurred by an
increase in variable costs per mile is much greater for work
trips than for shopping trips. Conversely, equal increases
in parking charges will affect shopping trips more than
work trips on a percentage basis.
It is likely that an emissions tax will not be placed on
any single pollutant but will instead be based on a formula
which includes all of the emissions. Therefore, it is worth
noting that any weighted average of the emissions tax rates
in Table 6 is equivalent to the increased variable cost per
mile. For example, emissions taxes based on the formula of
50 percent from CO (carbon monoxide), 25 percent from HC
(hydrocarbons), and 25 percent from NOx (nitrous oxides)
at a combined rate of a 50 percent increase in the variable
cost per mile would have the following three components:
0.50 x 0.0873 = 0.0437C/gm of CO;
0.25 x 0.7936 = 0.1984
-------
Table 5. ESTIMATE OF EFFECT OF POLICIES ON AVERAGE
PERCENT AND ABSOLUTE COST INCREASES OF
AVERAGE 1974 ROUND TRIP
Policy
Variable cost per
mile increase:
25%
50
75
100
Parking cost
increase:
$0.25
0.50
0.75
1 .00
Work trip3
$ Increase
$0.20
0.40
0.60
0.80
% Increase
31$
62
94
125
Shopping trip
$ Increase
$0.07
0. 13
0.20
0.27
% Increase
9456
187
281
374
Based on an average work round trip length of 14.04 miles.
Based on an average shopping round trip length of 4.68 miles.
19
-------
Table 6. GASOLINE AND EMISSIONS TAXES ASSOCIATED WITH EACH
LEVEL OF INCREASED VARIABLE COSTS PER MILE
Variable cost
per mile
increase
25%
50
75
100
Gasoline tax increase3
%
35?
71
106
141
$/gal
$0. 19
0.39
0.59
0.78
Emissions tax
CO,
-------
2. POLICY ANALYSIS FRAMEWORK AND INSTRUMENTS
This chapter is divided into two main sections. The
first section introduces the conceptual framework which will
be applied in analyzing the effects and costs of alternative
auto pollution control strategies. The second section de-
scribes the policy instruments for controlling automobile
pollution which will be evaluated in this study.
POLICY ANALYSIS FRAMEWORK
Because it is assumed that individuals are not charged
for the social costs of pollution under current practice,
the policies are designed to raise the costs of pollution
production. The policies are effective to the extent that
they alter the incentives of those actions which cause
pollution or its abatement.
The first part of this section shows how the short-
run policy effects on auto travel are analyzable with the
use of a behavioral model of travel demand. The second
part outlines an approach for determining the effects of
emission control strategies on the composition of the stock
of automobiles.
Travel Behavior
The short-run response to the policies considered in
this report is limited to changes in personal travel
behavior; longer-run effects include adjustments in the
stock of automobiles and the regional pattern of land use.
Personal travel decisions — mode choice, destination of
21
-------
trips and frequency of travel -- are the result of equilibrium
between household travel demand and the performance of the
system. As policy instruments affect system performance, new
equilibrium travel patterns are formed. We explain this
process below and introduce the concept of travel demand
which will be used in the analysis of policy effects and costs.
Travel Demand — The theory of travel demand has been devel-
oped in its essentials for some time and is becoming more
widely utilized by transportation planners and other policy
analysts. Because more complete treatments exist elsewhere,1
we will only sketch the fundamentals here.
The first step in analyzing travel behavior is defining
trips in a way amenable for analysis. For our purposes we
will consider a trip to be a round trip from an origin to a
destination by a specific mode. We will consider separately
trips which have different purposes (work, shopping, etc.).
Typically, we will only be considering single-purpose trips
made by households; thus, the origin of a trip will be the
residence of the tripmaker and the destination will be a
land use which reflects trip purpose (employment center,
retail outlet, etc.).
The quantity of travel by a household depends on the
costs, speeds, comfort and other performance characteristics
of the available means of travel, as well as on the relative
attractiveness of alternative destinations and the charac-
teristics of travelers. Thus demand is treated as a function
of several variables where the number of trips demanded is
dependent on a comparison by each traveler of the expected
benefits to him of the contemplated trips and the costs and
inconveniences of making them. A demand model is a set of
functions which indicate the demand for the various alter-
natives available to a household. The full range of choices
includes decisions on whether and how often to make a trip,
the time of day, destination, mode and route. For our
22
-------
purposes, we will mostly be concerned with those relation-
ships which describe mode choice, destination choice and
frequency of trip per unit of time.
In making these choices it is assumed — and evidence
contained in transportation studies verifies this assumption
— that individuals make time and cost comparisons in a con-
sistent and empirically deducible pattern. In selecting the
mode of transportation, for example, where a traveler can use
either transit or auto, he appears to evaluate the trade-off
between travel time and cost in a consistent manner. Simi-
larly, the choice of where to travel is systematically influ-
enced by a comparison of the relative travel times and costs
of the available alternative destinations. Finally, the num-
ber of trips made by a household or individual for a given
purpose is influenced by the cost and inconvenience of travel.
System Performance — The short-run notion of the performance
of a system refers to the relationship between the generalized
costs of travel and the volume of travel. Some components of
the generalized cost, such as transit fares and fuel costs, do
not vary significantly with the level of use but rather are
determined exogenously. However, as a general rule, the
greater the volume of traffic on any mode and origin-destina-
tion pair, the more time-consuming and inconvenient the indi-
vidual trip. This volume-performance relationship is deter-
mined by the physical characteristics of transportation
facilities, the way in which they are controlled, and the exo-
genously determined prices which affect the costs of a trip.
The interaction of demand and system performance deter-
mines the equilibrium number of trips that will be taken and
the generalized cost or level of service associated with
these trips. Exogenous changes in the system, such as those
caused by policy instruments, can be evaluated by comparing
the system equilibria before and after the changes.
-------
Analysis of Policy Effects
Building on the analytical concepts presented above, it
is possible to measure the short-run effects of various
pollution control strategies on automobile trips and VMT's.
To do so, the analyst must be able to determine both how the
policy will change the system performance and the consequent
effect on travel demand.
Analysis of Policy Costs — The policies considered in this
report, excluding transit service improvements, increase the
money costs of travel to the individual. Thus, the cost of
the policy to individuals includes the increased money costs
of trips made after the cost increase. Because the cost
increase is usually equal to the tax per trip, the cost to
individuals includes all taxes paid as a result of the policy.
As the cost of travel increases for, say, auto driver
alone, the number of such journeys declines due to people
choosing alternative modes and destinations which were for-
merly less desirable. The cost to individuals for making
these less desirable choices is not determined by the new
taxes alone. Sometimes the change in mode results in longer
commute times and lower costs as in the case of individuals
switching to carpooling or transit for long trips. Some
individuals will choose these alternatives at relatively
low increases in automobile travel costs, reflecting minor
discomfort in making the switch; others would change modes
only at much higher increases, indicating that the cost of
forgoing auto drive alone modes is relatively great.
Figure 2 illustrates how equilibria between demand and
system performance are used to approximate the costs of
policies to individuals. The demand function is represented
by the curve DD; the system performance relationship before
a-policy instrument is applied is represented by SS. The
initial equilibrium occurs at price per trip P and with a
24
-------
AVERAGE MONEY
COST PER TRIP
P'
P
Figure 2
COSTS OF POLICIES TO INDIVIDUALS
c I
1ST N
AGGREGATE NUMBER OF AUTO DRIVER TRIPS
25
-------
volume of traffic equal to N. A policy increasing the cost
of travel, such as a tax on gasoline or parking, shifts the
system performance curve upward to S'S'. At the new average
cost per trip, P', the equilibrium number of trips has de-
clined to N1 .
Under reasonable assumptions, the area of the hypothetical
triangle connecting points a, b and c gives a reasonable
estimate of the aggregate opportunity policy which shifts
the money cost of travel from P to P1. This cost would be
added to the area formed by the rectangle with corners b, c,
d and e (the total increased cost of trips by single motorists
making trips after the policy) to estimate the total cost of
the policy.
The above discussion covers only the increased costs to
households of a system change. The resource costs, that is,
the costs of resources which have alternative uses for
society, must be calculated differently. Briefly, these
costs include the opportunity costs to households, plus the
incremental transportation system costs, plus the costs of
administering the tax policy. It should be noted that there
are also nonmonetary benefits to transit riders when transit
system performance is upgraded. Except to acknowledge that
such benefits should be included in both the calculation of
individual and resouice costs, we do not attempt to estimate
or monetize these benefits.
Individual Cost = Change in Money Cost (taxes)
+ Opportunity Cost
Social Cost = Opportunity Cost to Individuals
+ Change in System Costs
+ Administrative Costs
The actual calculations of policy costs are performed in
Chapter 5.
-------
Probability Choice Models — The demand models used in this
study use estimates of individual probabilities of specific
travel related choices. The theoretical development of
individual techniques is presented in another study to which
the interested reader is referred.2 In this section we place
the model in the context of the demand framework presented
above.
The estimated relationships which form the basis of
the models are ratios of the probabilities of selection for
two alternatives to a given choice. Equation 1 gives the
general form of these estimates:
P.
J
ln—~- = £,(X .-X .) (1)
^ J
where P. = probability of an individual choosing alternative i;
P. = probability of an individual choosing alternative j;
J
X. = vector of attributes, including costs and attrac-
1*
tiveness of alternative i;
X. = vector of attributes, including costs and attrac-
3
tiveness of alternative J;
B = estimated vector of coefficients-
If there exist a number of alternative choices, say n, then
the several estimates of pairwise probability ratios of the
form of Equation 1 can be transformed to get an estimate of
the probability itself by means of the following formula:
P. = J = i (2)
i n p, ^ o/x -X )
V / 6 It. *it
k=l PI k=2
Equation 2 is a generalized logit function,
27
-------
In analyzing individual travel choices it is useful to
separate the decisions involved into the following sequence:
1. Whether to make a trip;
2. Given a trip is to be made, where the destination
will be;
3. Given a trip will be made and the destination is
known, what mode will be taken.
Using this artifice, the probability that an individual
will make a trip to a given destination by a given mode can
be viewed as the product of three conditional probabilities:
P.. = P(t\i.) •P(j\t)i)'P(m\t1i,j) (3)
1, J TU
where P. . = probability of an individual at location i
Ijrn * 1
making a trip to j by mode m for a given
purpose over a specified unit of time, such
as a 24-hour period;
P(t\i) = probability that an individual at location i
will make a trip for a given purpose over a
specified unit of time, such as a 24-hour
period;
P(j tti) = probability that an individual at location i
will make a trip to location j, given that a
trip will be made for a given purpose;
P(m\ttitj) = probability that an individual at location i
will make a trip by mode m to location j,
given that a trip to location j will be made
for a given purpose.
*See Ben-Akiva, Moshe. Structure of Passenger Demand Models.
Ph.D. thesis, Massachusetts Institute of Technology, 1973,
for a discussion of this issue.
28
-------
In order to estimate the demand for travel from a zone, the
above probability choice functions are summed across all
individuals, or households, in the zone of location i.
(With most urban transportation data bases, this exercise is
not really possible. However, the model can still be applied
to zonal aggregates after certain adjustments to replicate
the effect of summing the probabilities of individual choice
across all households in a zone.)
The demand model used in this study is developed in
more detail in Appendix A. It contains estimates of the
probabilities of mode choice for work trips and estimates of
the probabilities of mode choice, destination choice and
frequency of trips for shopping trips. Chapter 3 gives the
results of applying these estimates to determine the effects
of pollution control strategies on VMT's and number of trips.
Chapter 5 uses these models for calculating individual and
social costs of the policies.
Auto Stock Adjustments
The policies considered in this report will affect the
relative costs of owning and operating cars of different
vintages. As a consequence of these policies, the number of
automobiles owned will be fewer, causing less auto travel,
and the median age of the auto stock will be decreased, so
that the emissions per car will on average be less. This
section presents the theoretical foundations for the models
which will be used in estimating these effects.
Auto Stock Dynamics -- The stock of cars in a region changes
from year to year as new cars are purchased, old cars are
scrapped, and households migrate with their personally owned
autos. These flows to and from the previous period's auto
stock determine the fleet size at a given subsequent time.
The process is depicted in the identity:
29
-------
Tt = Tt-l + Nt - Jt + Jt
where T = stock of cars at the end of period ft in a given
t
region;
T. , = stock of cars at the end of the period previous
t " -i
to t (or, at the beginning of period t) in a
given region;
N = total new car purchases during the period t
t
in a given region;
J+ = total cars scrapped (junked and abandoned)
if
during the period t in a given region;
It = net inflow or outflow of cars owing to house-
hold relocation into or out of the given region
during period t.
In analyzing the effect of pollution control strategies
on the stock of cars, it is useful to isolate the impacts
these policies will have on those components of auto stock
which change the fleet size from year to year. Such an
approach entails considering new car sales and used car
scrappage separately; it is presumed that the effect of the
pollution control strategies on net migration of households
is negligible.
New Car Sales — The purchase of new cars in a region is
largely the result of the interaction of demand by house-
holds and supply by auto manufacturers. Empirical evidence
suggests that there is a substantial amount of regularity
*
in the sensitivity of consumers to new car prices. Equation
5 presents a stylized demand function for new cars:
n = /pff,*0; (5)
*For a review of research to date, see Dewees, D. N. Economics
and Public Policy: The Automobile Pollution Case. Cambridge, MIT
Press, 1974.
30
-------
where .V = new car sales at time t in a given region;
PN, = price index of new cars at time t in a given
region;
KN. = other factors relating to households and sub-
stitute products of new cars (e.g., level of
income and availability of transit) which
affect new car sales.
We have purposely isolated price in this equation because
it mediates the impacts of increased ownership and operating
costs on new car purchases. Other variables, amalgamated
into the term KN^ , can be ignored in analyzing the effects
of policies considered in this study.
If Equation 5 accurately represents the demand for new
cars, then an increase in the ownership costs which is not
related to any increased consumer benefits would change new
car sales according to the following relationship:
N = f2(PNt + ^Nt,KN ) (6)
where hCN = the cost of increased ownership of a new car
£
during year t for a given region where the
increased cost is not related to increased
consumer utility.
The effect of increased ownership costs is identical
to the effect of an equivalent price increase. Most policies
will increase auto ownership costs on a yearly basis, as in
an annual tax; the increase in ownership costs at the time
of new car purchase decisions will be the discounted
present value of these costs over the lifetime of the car.
This proposition obtains even if the household does not
plan to own the car over its operating life. As will be
31
-------
shown below, the resale value of the car declines by an
amount approximately equal to the capitalized increase in
cost of ownership over its remaining life.
The effect of a policy to increase car ownership costs
will be to lower demand at the equilibrium price, but it
will not cause a change in the long-run equilibrium price.
The result will be fewer new auto sales.
The actual quantitative assumptions about the new car
demand function are given in Appendix B where a technical
description of the entire auto stock adjustment model is
presented. The results of pollution control strategies on
new car sales are presented in Chapter 4.
Used Car Scrappage — Automobiles tend to exit from the
fleet of cars when their costs of repair and reconditioning
are greater than the price of existing, operative vehicles.
Used car dealers, and households to a lesser extent, consign
such cars to wreckers or scrap dealers. Other factors which
affect scrappage are the rate of turnover of cars in the
market and the age of the auto stock. Higher turnover
gives dealers more cars on which to make scrap vs. resell
decisions. The older the existing stock of cars, the more
*
expensive are the costs of keeping the cars operative.
Pollution control strategies will have an effect on
scrappage by their impact on used car prices. As the costs
of owning a used car increase, the price of used cars will
decline (in a manner described in more detail below). Lower
market-clearing prices for used cars provide less incentive
for used car dealers to make costly repair and reconditioning
efforts in order to market the vehicles. The result is a
decrease in the stock of used cars.
*For a somewhat more extended discussion, see Walker, Franklin
V. Determinants of Auto Scrappage. Review of Economics and
Statistics. 503-506, November 1968.
32
-------
The scrappage relationship, in a functional form implied
by the above considerations, is as follows:
Jt = f2(PUt'CRt>Rt>Tt-l>At-l) (7)
where J = total cars scrapped during the period t for a
c
given region;
PU = price index of used cars in period t for a given
t
region;
Cft = cost index of repairing and reconditioning used
u
cars in period t for a given region;
P. = rate of turnover of the auto stock during period
t
t for a given region;
T = the stock of cars at the end of the previous
t ™~ J.
period — the beginning of period t — for a
given region;
A. 7 = the age distribution of cars at the beginning
t "• J.
of period t for a given region.
The price of used cars is determined by the interaction
between demand and available stock. The latter is in turn
affected by the scrappage rate. Thus there is a multi-equa-
tion system which adjusts used car demand and stock so as
to reach equilibrium. We consider first the demand relation-
ship.
As with new cars, the household demand for used cars is
considered to be sensitive to the existing market prices for
used cars. Thus, a general demand function for used cars
would be of the form:
33
-------
where U = number of used cars demanded in period t for a
given region;
PU. = price of used cars in period t for a given
region;
KU = other factors relating to households and sub-
stitute products (e.g., level of income and
availability of transit) which affect used
car sales.
Used car supply is a concept quite different from the
supply of new cars, or most other manufactured products.
The supply of used cars is the total stock of cars in any
given region. Except for scrappage, it is a fixed amount.
Though, obviously, all used cars are not sold, it can be
presumed that the market-clearing price of used cars repre-
sents their market value. Households most often decide to
keep their car rather than take the opportunity to sell it
to someone else at the market-clearing price. Referring
back to Equation 8, we see that equilibrium occurs when:
t ~ t-1
By substituting Equation 9 into Equation 8 and solving
for price, price determination in the used car market can
be seen to be a function of the used car stock and the other
factors in demand:
(10)
We are now in a position to illustrate how the system
will adjust to the increase in ownership costs of used cars
caused by the imposition of pollution control strategies.
34
-------
As in the case of new car demand, the added costs of used
cars will cause a downward shift in the demand. In equa-
tion form, the process is determined by adding the increased
costs to price in Equation 8 to yield:
y- J? +• +• } 4- \ " «*• /
where ACi/, = capitalized increase in ownership costs of
u
used cars in period t for a given region,
where the increased cost is not related to
increased consumer utility.
The new market-clearing price is Pi/i; it can be computed
using the following formula derived from from Equation 10:
Put =
The lower price would in turn cause the scrappage rate
to increase and, consequently, the stock of used cars would
dwindle. This would shift the supply of used cars and cause
a corresponding increase in the market-clearing price of
used cars. The process would continue until a new equili-
brium is established for the price of used cars, the scrap-
page rate, and the stock of used cars.
The system of equations used to simulate the above
process is developed in Appendix B of this report. There,
an econometric estimate of Equation 7 is presented together
with the assumptions used to determine Equation 8. The
results of applying this model to the actual policies
considered in this study are given in Chapter 4.
35
-------
POLICY INSTRUMENTS
This section presents the a priori case for, and prob-
lems with, each policy alternative along with the nature and
incidence of costs. Also, in each case, a brief review of
the literature is included so that the results of this study
can be checked for consistency against previous research
efforts.
Public action to control air pollution is necessary
because an individual owner of an emission source has no
incentive to reduce its emissions since its contribution
is a small percentage of the whole. The individual does
not pay the costs borne by others as a consequence of his
actions, and there is no noticeable effect from his sacri-
fice. Some form of institutional action or adjustment is
called for when the problem becomes a social burden. An
optimal form of control would be for polluters, in this case
motorists, to pay the marginal social cost of their pollution.
If knowledge existed about the benefits that society as a
whole receives from air pollution abatement, an optimal level
of air pollution would occur when the marginal social cost
of pollution abatement equaled the marginal social benefit
derived from that level of air pollution.
Again assuming such knowledge existed, a tax could be
set at a level which would induce motorists to reduce their
contributions to air pollution by driving less, buying cleaner
cars, retrofitting, or other methods. The level of emissions
that resulted would be the level at which the cost of reducing
emissions by one more unit equaled the benefit gained by that
reduction.
The major problem with applying an optimal policy is
that the benefits of pollution abatement are ill-defined
and, to date, largely indeterminate. Because the notion of
social damages caused by air pollution is quite ambiguous,
the theoretical approach of equating marginal social cost
36
-------
with marginal social benefit does not provide, in and of
itself, an operational guideline to policy development.
Typically, the desired level of air quality is given in the
political decision-making process in a manner which is beyond
the scope of this study. For our purposes, the benefits of
air pollution abatement are given; that is, the level of air
quality is an absolute constraint which is met by specific
individual and social costs. The purpose of this research
effort is to inform the political decision-making process of
the costs of various air quality constraints.
Many of the policy recommendations to date have focused
on regulation of technology and behavior rather than the
imposition of incentives for pollution abatement. Regulatory
proposals have included such things as gas rationing, parking
prohibitions, and restrictions on residential location and
place of employment. Aside from the not inconsiderable dis-
ruption and administrative costs of such an approach, direct
regulations such as these do not cause individuals to pay
for their pollution. Thus, even in the absence of exact
knowledge about the social costs of pollution, regulations as
a solution to the pollution problem are pr-ima facie less
satisfactory alternatives than a system of taxes and incen-
tives which allows scope for voluntary individual decisions
on whether to act so as to pollute while paying the social
penalty involved.
The policy instruments considered in this report include
the following:
• parking tax
• gasoline tax
• emissions tax
• transit system improvements
These are each covered in detail in separate sections.
37
-------
The effects of these policies are to decrease the
following factors which contribute to air pollution:
• vehicle miles traveled (VMT's)
• trips (the number of cold starts)
• the emission rate of the stock of automobiles
These factors are further influenced by policy effects which
alter the following:
• size of the auto stock
• age distribution of automobiles
• the extent of retrofitting
Vehicle miles traveled in a region is the most important
contributor to mobile source air pollution. However, it has
been demonstrated that a significant portion of total emissions
(depending upon the pollutant under consideration) does not
vary with mileage, but simply depends upon whether a car was
started while the engine was cold.3 The number of cold starts
is assumed to be proportional to the number of trips taken.
In order to consider policies designed to reduce auto
emissions, it is also necessary to know about the rate of
emissions from the auto stock in some detail. The rate of
production of a particular pollutant for a given vehicle at
a given moment depends upon the rate of emission of the
vehicle at low mileage, the amount of use the vehicle has had,
and its state of repair with respect to functional elements
which affect emissions. Two vehicles which have identical
emission rates coming off the assembly line may have different
rates after much use if one receives better maintenance than
the other. Two vehicles which have received identical main-
tenance may have significantly different rates depending on
their original low mileage (or optimal) rates. Since
increasingly strict emission controls have been mandatory on
new vehicles for some years now, especially in California,
low mileage emission rates generally vary with the model year
of the vehicle, although there is usually a high variance
38
-------
within each model year." Also, it is possible to derive
average amounts of use and average maintenance for each model
year, with which one may calculate deterioration rates.5
The primary factor which affects the current level of
emissions per mile from a given stock of cars is the age
distribution of the fleet. This is due to the large dif-
ferences in emissions between precontrolled and controlled
vehicles. It is estimated that in 1975, precontrolled
vehicles will account for about one-quarter of the vehicles
on the road, both nationally and in California. However,
as these vehicles emit pollutants at 5 to 100 times the rate
of controlled vehicles, their contribution to the total
pollution problem is greatly out of proportion to their
numbers.6 Finally, the policies under consideration may
affect the emission rate from the stock of cars by inducing
maintenance and retrofitting, causing older autos to decrease
their deterioration rates as well as their initial rate of
emissions.
In the remainder of this section, we shall discuss the
policy instruments evaluated in the study.
Parking Tax
There are many possible designs of parking taxes,
including those which vary by time of day and localized
conditions. However, as will be shown below, some parking
tax strategies are clearly preferable to others in satis-
fying the objectives of decreasing automobile miles traveled
and trips in the region as a whole.
A Priori Analysis of the Effects of a Parking Tax — A parking
tax must be defined in terms of spatial coverage and land
use before it can be analyzed. Most experience with parking
taxes has focused on the effects of increased restrictions
on parking in the central business district or even smaller
areas. The problem with a parking tax of this nature is
39
-------
fairly obvious — it fails to deter a significant number of
trips in the region. This is especially true in Los Angeles,
where the central business district has relatively minor
importance. (Less than 5 percent of the total regional
employment is in the Los Angeles central business district,
and an even smaller percentage of shopping trips have their
destination in the CBD.) Thus, the parking tax leaves most
trips unaffected, as there are enough alternative destina-
tions to the CBD to prevent substantial deterrence to auto
trips as a whole.
Similarly, a parking tax designed to raise the cost of
travel for a particular trip purpose, such as shopping,
would generally be difficult to administer and enforce. As
long as there are cheaper alternative parking places avail-
able, then auto trips for any purpose would use these alter-
natives unless the effort at monitoring the trip makers was
highly developed, which would be prohibitively expensive.
Thus, a parking tax over the region as a whole and for
all trip purposes presents itself as the most reasonable
scenario. The parking tax may be varied, however, by time
of day or climatic condition depending on when diversion of
auto trips would have the most benefit in abating air pol-
lution. Even this policy can present administrative diffi-
culties, owing to the difficulty of distinguishing between
residential parking and other parking.
The imposition of a blanket charge on all nonresidential
parking increases the out-of-pocket expenses for auto trips
when the vehicle is parked at the trip destination. The
percentage increase in the cost of short trips (in money
terms) is much higher than the percentage increase in the
cost of long trips. Hence, this tax applies unequally by
length of trip.
40
-------
Counteracting this effect is the tendency for a parking
tax to induce chauffeured trips in instances where the trip
previously had been made by a driver alone. For relatively
short trips there is an incentive for members of the same
household to ferry trip makers in order to avoid the parking
surcharge at the trip destination. Each trip of this nature
results in twice the vehicle miles traveled when compared to
a trip made by a driver alone. It also substantially in-
creases the amount of time spent in travel because of the
increased efforts made by a driver serving a passenger.
A priori, due to increased incidence in chauffeured
trips, it is likely that the number of VMT's reduced by a
parking tax will be less than the amount reduced by an
equivalent increase in the variable cost per mile for auto
travel. Also, as is discussed later, other tax strategies
which increase the variable cost per mile of auto modes will
have a greater effect on VMT's as opposed to number of trips.
It can also be assumed that a parking tax will have some
effect on the fleet size for the region. The increase in
auto operating costs will provide an incentive for households
to have fewer cars. However, it will not necessarily cause
the age distribution of automobiles to be lower, nor will it
cause the emission rates from the stock of cars to be less.
The parking tax provides no incentive for households to buy
automobiles which are less polluting per mile of travel.
Indeed, if the new car sales are more sensitive to ownership
cost changes than used car sales, then the median age of the
existing stock of autos may in fact increase from what it
otherwise would have been, thereby having counterproductive
effects on the average emission rate.
The long-run effects of a parking tax are somewhat
difficult to assess but are probably not very significant.
Some change in employment trip frequency may be seen and,
if the tax varies by hour, the work day may be adjusted in
41
-------
some cases so that work trips would occur when the tax was
at lower levels. Though shopping trips can be expected to
decline, total retail sales would be reduced only negligibly
as a direct result of lower discretionary income owing to
the tax; the adjustment which one would expect is that house-
holds would buy more at each shopping stop and frequent those
stores which are within walking distance or easy access by
transit. Thus, there may be some redistribution of retail
land use as shopping centers and small local retail outlets
become more attractive. Of course, if the parking tax was
applied to particular locations, then retail sales would
move from areas where the parking tax was highest. Parking
taxes on other trip purposes would cause similar adjustment?.
Review of Other Research -- Several cities have instituted
parking taxes whose effects have been recorded to a limited
extent. Kulash has compiled a number of calculations of the
elasticity of demand for auto trips with respect to auto
par1':-.;: ~nce. His r^s'i] ts *re summarized in Table 7.
For a number of reasons these elasticities are almost cer-
tainly too high. In every case where the effects of a
parking tax have been analyzed, the tax was limited to a
relatively small locality, usually within the central busi-
ness district. Hence, the tax affected only a small percen-
tage of the trips in a region, so any elasticity would have
to be applied to the percentage of non-through trips in the
area under consideration in order to uncover the regional
impacts. Kulash theorizes that this could be done by con-
sidering an area about four times the size of the typical
central business district, since it would have a low percen-
tage of through trips. However, this, approach would certainly
be invalid for the Los Angeles region because it has virtually
no strong core. Other reasons why the elasticities are
improperly calculated include:
42
-------
Table 7. ELASTICITIES OF NUMBER OF TRIPS WITH
RESPECT TO AUTO PARKING PRICE
City
San Francisco
L.A. Civic Center
Washington, D.C.
Liverpool (survey)
Auto driver
-.35 to -.43
-.29
-.41
-.3
Auto passenger
.26
Bus passenger
.47
.38
Work trips
SOURCE: Kulash, Damian. Parking Taxes for Congestion Relief: A Survey
of Related Experience. Urban Institute. Washington, D.C. Workino Paper
1212-1. May IQ73.
43
-------
- drivers who switch to being chauffeured to work
- illegal parking
- changes in destination
- drivers switching to taxis
- increased through trips because of people who begin
driving as a result of reduced congestion after
imposition of a parking tax.
The individual cases are briefly discussed below.
San Francisco — The elasticity for San Francisco is based
on a before and after study of the number of cars parked in
off-street facilities with the imposition of a 25 percent
parking tax. In addition to all the factors listed above
which contaminate these calculations of elasticity, transit
services were being improved at the same time, and their
effect on traffic was not isolated in the study. The report
on the San Francisco episode indicated a consensus opinion
that there was little effect on traffic, implying that the
elasticity in the range of -.35 to -.43 for parked cars means
very little as far as predicting the actual effects on auto
trips.8
Los Angeles civic center — The elasticities in this study
are based on a cross-sectional mode split calculation for
work trips. The elasticity of auto passenger trips with
respect to parking price does not indicate whether some
chauffeured trips are included or not. The sample utilized
has a much higher level of bus service than the region as a
whole; as a consequence, the elasticity derived for auto
diversion to bus would be too high if it were applied to
other destinations in the region. It might be noted that
44
-------
this consideration affects the other elasticities calculated
*
for CBD oriented trips as well.
Washington, D.C. — The format of this study was also cross-
sectional mode split, with results similar to those for Los
Angeles. The original source is not available, but it is
apparent from the data which Kulash presents that none of
the factors mentioned above have been accounted for in the
estimation of the elasticity of demand for auto trips with
respect to parking price.9
Liverpool -- This study is based on a survey of what people
would do in response to some hypothetical situation in which
the price of parking changes. A factor called "frustrated
demand" is explicitly taken into account. It consists of
people who would make trips into (or through) the area under
consideration but do not do so because of congestion. To
the extent that a parking tax would relieve congestion, these
people would start satisfying their "frustrated demand," thus
offsetting the effect on the reduction of trips through the
area due to an increased parking charge. As in the other
cases, the elasticities derived do not indicate what the
real effect on auto trips would be if a parking tax were
imposed, because of lack of consideration of other factors
such as chauffeured trips, etc.10
Pittsburgh -- One event reported by Kulash appears to cast
considerable doubt upon the elasticities calculated in his
report. In Pittsburgh during August 1972, a rather drastic
reduction in the availability of offstreet parking occurred
when 23,600 of 25,400 spaces were shut down by a strike.
*See Grominga, C. L., and W. E. Francis. The Effects of the
Subsidization of Employee Parking on Human Behavior. Course
Paper, University of Southern California. May 1969.
45
-------
The effect on shopping was limited, as retail sales were down
only 6 to 8 percent, but entertainment trips were estimated
to have declined by 70 percent. As to work trips, auto
volume was off 24 percent in the morning rush hour, while
bus ridership increased 12 percent. While these are not
insignificant results, a tax with effect equal to the Pitts-
burgh strike would be enormous, and Kulash's conclusion is
that short-run economic incentives have a "limited range
of impact" on auto trips.
Conclusion — For the reasons already mentioned, little of
a precise nature can be said about the effects of a parking
tax. The research to date has simply not determined the
true elasticity of vehicle miles traveled and auto driver
trips with respect to a change in parking charges. However,
it is likely that the figures presented above represent the
maximum elasticity which one would expect when the effects
of a parking tax are simulated in the following chapter.
The estimated elasticity should fall within the bounds of
.1 and -.3. A positive elasticity would occur if the number
of chauffeured trips and extra miles traveled to search for
cheaper parking overwhelmed the other effects of the parking
tax.
Gasoline Tax
A tax on gasoline sales would result in an increase in
the variable costs per mile of a trip and would consequently
be expected to induce a reduction in VMT's and, to a lesser
extent, in the number of auto trips. The legal and admini-
strative framework for gasoline tax collection has been in
existence and functioning for many years. Unlike some of the
other policy alternatives considered in this report, imple-
mentation of a gasoline tax would not require a large
46
-------
investment in equipment, manpower, or planning. The long-run
effects of a gasoline tax are somewhat difficult to determine
— for example, gas mileage among autos may be inversely
related to low emissions, implying that increased gasoline
taxes might decrease the rate of scrappage of older autos.11
A Priori Analysis of the Effects of a Gasoline Tax — The
effect of a gasoline tax on the output of automobile air
pollution depends, in the short run, upon the elasticity of
auto trips and VMT's with respect to a change in the vari-
able costs per mile of making a trip. Table 8 presents a
breakdown of variable costs per mile which indicates that
gasoline presently accounts for about 70 percent of the
marginal operating costs of an auto for a trip. This assumes
a retail price of $0.55 per gallon (including taxes). Thus,
a 10 percent tax on. the pump price of gasoline ($.055 per
gallon) would be equivalent to a 7 percent increase in
variable costs per mile. It is important to keep this
conversion factor in mind when discussing the effects of a
gasoline tax because travel demand models typically utilize
variable costs per mile rather than gasoline costs as an
argument in the demand function.
Gasoline taxes have the attractive feature of being
relatively well-correlated with emissions for any given
automobile, since gasoline consumption over a trip cycle
is directly proportional to the amount of emissions. However,
the rate of gas mileage across automobiles may be inversely
correlated with the rate of emissi.ons. Autos with emission
controls tend to have decreased mileage. The imposition of
a gas tax offers an incentive for households to keep older,
more economical, automobiles instead of trading them for
newer, cleaner cars.
47
-------
Table 8. ESTIMATED VARIABLE COSTS PER MILE
OF OPERATING AN AUTOMOBILE
(cents/mile)
Item
Cost
Gasoline (excluding taxes)
Gasoli ne taxes
OMC
Repairs and maintenance
Q
Replacement of tires
TOTAL
5.71
.Based on price (including taxes of 55C/ga! and 13.6 miles per gallon).
Federal tax = 4
-------
The flexibility of a gasoline tax is relatively limited
when compared to, say, a parking tax. It cannot be realisti-
cally varied over small locations because of the ability of
motorists to buy gasoline at those stations where the tax is
lowest. Even for an area of the size covered by the Los
Angeles Regional Transportation Study, this will be something
of a problem, although not of major proportions. Similarly,
a gas tax cannot be varied by time of day, because households
can easily arrange to buy gas during that period of the day
when it is least expensive. Even day-to-day variations in
the gasoline tax would be of little effect if the changes in
the gasoline tax could be accurately predicted by households.
Seasonal variations are possible, and such an approach may
be seriously considered in the case of Los Angeles, where
pollution is greatest during late summer and early fall.
As a percentage of the total cost of a trip, a gasoline
tax is roughly proportional to the length of the trip. In
absolute terms, long trips would incur a greater penalty
than short trips. As a general proposition, the higher the
cost of an activity, the more elastic its demand, other
things being equal. Thus, it can be presumed that a gasoline
tax would have greater effects on the longer auto trips;
correspondingly, one would expect to see greater impact of a
gasoline tax on VMT's as opposed to number of trips.
The long-run effects of a significant increase in the
price to households of gasoline would be to redistribute some
land uses to take account of the higher costs of travel. One
pattern that might emerge from the increased cost of trips
is the clustering of higher density residences. Commerce
would localize around these centers, which in turn would be
located with easier access to employment centers. Employers
49
-------
would be more likely to locate near these high density
labor markets.
Review of Previous Research -- There have been two unrelated
lines of research which can be reviewed in order to determine
what the expected effect of a gasoline tax would be. The
first of these is in the realm of travel demand forecasting
for planning purposes; the second is the recent body of
econometric estimates of the elasticity of demand for gaso-
line on an aggregate level. We discuss the recent results
in each of these types of studies.
Direct demand model of travel — Aside from the disaggregated
demand model of individual choice used in this study, the
only other estimated relationships between the variable
cost per mile of auto trips and the demand for travel are in
what have become known as direct demand models. The most
widely applied direct demand model was estimated by CRA on
Boston data at the traffic analysis zone level. A full
description of the model is presented elsewhere. The
relationships in the model estimate the number of trips by
mode between origin and destination pairs for a given purpose.
The number of trips so defined are functions of the costs
and time of the given mode, the costs and time of alternative
modes, the socioecor.omic characteristics of the zone, and
trip attractiveness of the zone or destination.
*See any of the following: A Model of Urban Passenger Travel Demand
in the San Francisco Metropolitan Area. Charles River Associates
Incorporated. Cambridge. Prepared for the California
Division of Bay Toll Crossings. 1967; Kraft, Gerald, and
Thomas A. Domencich. Free Transit. A Charles River Associates
Incorporated Report. Cambridge, Lexington Books, 1970;
Domencich, Thomas A., Gerald Kraft, and Jean-Paul Valette.
Estimation of Urban Passenger Travel Behavior: An Economic
Demand Model. Highway Research Board Record. 238, 1968.
50
-------
From the model, it is possible to calculate the elasti-
city of travel demand with respect to an increase in the
variable cost per mile of auto trips. These elasticities
have been computed at the mean of the Boston data and were
approximately -.4 for work trips and -.8 for shopping trips.
Some care must be taken in interpreting these figures.
They were estimated when the price of gasoline was much less
than currently obtains — both in absolute terms and in terms
of the percentage of variable costs per mile. Also, the
model may be subject to a long-run bias owing in part to its
having been estimated on cross-sectional data. That is, the
elasticities presented may be higher than would have been
obtained if purely short-run behavior had been isolated. As
the model now stands, there is some reason to believe that
it contains some aspects of locational behavior on the part
of households and the employers or retailers as well as the
effects of the costs of travel on automobile trips alone.
If the bias is indeed significant, then it is likely that
the elasticities presented above are relatively accurate
when applied to VMT's over the long run, but may be spurious
if applied to number of trips.
The two problems cited above -- relatively low operating
costs at the time of estimation and the long-run bias in the
estimates themselves — have effects in different directions
if the model is to be used to give a priori estimates of
the effects of a gasoline tax in Los Angeles today. It is
not known to what extent these will cancel each other in
interpreting the elasticities. However, taking the numbers
as they are given, gasoline costs accounted for 70 percent
of the variable cost in the Boston data, and consequently,
the model would predict that a 10 percent tax on gasoline
at current costs to the customer would entail a 2.8 percent
decline in work travel and a 5.6 percent decline in shopping
travel.
51
-------
Econometric estimates of the demand elasticity for gasoline —
Largely as a result of the oil embargo in the fall of 1973,
there have been a number of attempts to estimate the demand
for gasoline. Usually these models are estimated on a time
series or pooled time series/cross-section of data aggregated
to fairly large geographic units such as states or nations.
Because of errors in variables owing to the aggregation of
data to these levels, and the problem of multi-collinearity
among independent variables in the demand function over time,
most of the results must be approached with a high degree of
caution. It is remarkable that most of the estimates have
clustered around -.2 as the elasticity of demand for gasoline
in the United States.
Even if this figure is accurate, there are two important
caveats before it is applied to the conditions current in
Los Angeles households. First, the sample periods have
always been pre-oil embargo; therefore, the price of gaso-
line was lower relative to other goods and services, and we
may presume that the elasticity of demand was correspondingly
less. Recently, an attempt was made to measure the elasti-
city of demand for gasoline in Europe, where the prices are
significantly higher than those which exist in the United
*
States. Secondly, a significant portion of gasoline is
consumed by non-household trip makers. Gasoline consumed in
this way, usually for freight transportation, is a factor
input into other goods or services. To the extent that
gasoline is a relatively small portion of the total cost for
a given good or service, it can be presumed that the elasti-
city of demand for gas used in these trips is less than the
household elasticity of demand for gasoline. All things
considered, it can be presumed that -.2 represents a lower
region for the household demand elasticity of gasoline.
*See Charlotte Chamberlain, unpublished report, Transpor-
tation System Center, 1970.
52
-------
Conclusion — The estimates of gasoline demand give us a
lower bound on the expected effect of a gasoline tax on
VMT's in Los Angeles. The estimates of travel behavior as
a function of variable costs per mile from the direct demand
model give us a region in the upper end which we can check
for consistency against the disaggregated demand model
applied in the next chapter. The shopping trip purpose in
the direct demand model was most sensitive to automobile
operating costs. If other household-based non-work trips
were assumed to be as sensitive to costs of travel as shop-
ping trips, then this would give an approximate high end to
the range one would expect for the elasticity. In the Los
Angeles region, 25 percent of all trips were of the home-
based journey-to-work type; weighting the total number of
trips taken by an elasticity of -.4 for work trips and -.8
for all other trips gives a total elasticity of -.7 for all
auto travel with respect to the variable costs per mile.
This in turn implies a price elasticity of gasoline of about
-.5. In summary, we would expect to see the elasticity of
VMT's with respect to a change in the price of gasoline to
be on the order of -.2 to -.5.
Emissions Tax
There are a relatively large number of proposals for
basing a tax on motorists which is proportional to the
amount of pollution produced in their travel. A tax or
charge of this nature is based upon an estimate of the
emissions of a particular automobile over a given period of
time. Theoretically, an emissions tax should correspond
most closely to the actual marginal social cost of pollution
of any of the policy alternatives considered in this report.
The effects of such a tax would be to increase the costs of
owning and operating vehicles in proportion to the pollution
potential of the vehicle, thus making dirtier cars more
53
-------
expensive relative to cleaner ones. This should cause the
age distribution of vehicles in use to shift more rapidly
away from older, pre-controlled autos, as people would prefer
vehicles with low initial emission rates and deterioration
factors. A tax based upon the actual emissions per trip
would cause increases in the variable costs per mile of the
vehicle in much the same way as a gasoline tax. This could
induce a decrease in VMT's and number of trips, and, through
the incentive to perform maintenance and retrofit older autos,
cause a decrease in deterioration rates and initial emission
rates.
A Priori Analysis of the Effects of an Emissions Tax -- The
various proposals for emissions taxes can be initially
screened by judging a priori, their effectiveness in meeting
their pollution abatement goals. The objective of reducing
automobile emissions through a tax would be achieved by
inducing motorists to take measures to reduce the contribu-
tion of their vehicles to air pollution. These measures
include the following:
1) driving less;
2) purchasing cleaner cars;
3) retrofitting older cars;
4) performing maintenance; and
5) using a less polluting fuel.
Bearing in mind that an emissions tax should induce
motorists to take one or more of these steps, the following
proposals are considered:
1) Tax on the sale of new vehicles. This is a one-time
tax levied when the automobile is purchased, and graduated
relative to the projected lifetime emissions of the vehicle
(or to some parameter of the vehicle which is correlated
with the emissions rate).12
54
-------
2) Annual tax on vehicle based on the projected life-
time emissions. This tax would be collected annually on all
registered automobiles and would vary according to model,
make and year (MMY) based on tests on a sample of vehicles
which estimate the projected lifetime emissions r.-.te.13
3) Annual tax based on actual emissions. In this .ropo-
sal, each auto would be inspected annually to determine its
mileage and emissions rate (MER) for the year.11*
4) Monthly tax based on actual emissions. This propo-
sal differs from number three only in that the tax is
collected monthly, based on average emissions either for
the whole year or for that month and on actual miles driven
during the month.15
Of the above proposals, we find that the first and
fourth are either redundant or counter-productive. The first
proposal, a tax on the sale of new vehicles, would penalize
car owners for purchasing the newer and hence less polluting
automobiles. Such an approach may, in the long run, reduce
pollution after most currently owned automobiles have been
scrapped, but the interim result would surely be to recon-
dition and maintain the precontrolled automobiles much
longer than would have otherwise been the case. The fourth
proposal, an MER tax paid monthly rather than annually, is
based on the somewhat dubious proposition that a monthly
charge is more visible than an annual payment and would
prove to be a greater incentive to retrofit and reduce trips
than an annual payment equal to 12 monthly installments. We
find no evidence to support this supposed behavior; indeed,
there is as strong an argument in the other direction --
namely, that an annual charge will be so large relative to
12 roughly equal payments that it will be more visible.
For these reasons, we will further consider only the MMY and
MER taxes, both on an annual basis.
55
-------
In order to implement either of these proposals, certain
steps would have to be taken. For the MMY tax, proposal two,
the necessary steps include:
1) Derivation of reliable estimates of average projected
lifetime emissions for the models, makes and years of auto-
mobiles owned. A first approximation of emission rates can
be made by testing the sample of vehicles as they come off
the assembly line. More precise estimates are gained by
actually conducting tests in the field of vehicles after
specified time periods in order to gain information about
deterioration rates in on-the-road use. Field tests should
also be made for determining the effectiveness of retrofit
devices on precontrolled cars.
2) Provision for collection of the tax through some state
agency, most probably the automobile registration agency.
The principal requirements for the implementation of
an MER tax, proposal three, would be the following:
1) An efficient, easy to administer, reliable emissions
test capable of implementation on a massive scale.
2) The facilities, equipment, and trained personnel
necessary to administer the test.16
3) A relatively inexpensive, tamper-proof odometer
capable of installation on older cars and part of standard
equipment on new cars.
4) Provision for collection of the tax at the time of
registration or at the time of testing.
We consider later the results of research into whether
either of these proposals is administratively feasible. At
this time, it appears that emissions tests on a massive scale
and the monitoring of mileage on the same scale would be the
primary barriers to the imposition of an MER tax. It is thus
worth noting that many states have annual inspections of
registered vehicles for nominal fees, typically at service
stations designated for the purpose. The scale of such an
56
-------
enterprise is therefore not as much an issue as the cost of
technology for administering the tests and the organization
of a relatively fool-proof tax collection system.
The effectiveness of each tax on controlling pollution
is similar in the long run but divergent in the short run.
The MER tax would have many of the same effects as a gasoline
tax because both affect auto variable costs per mile: diver-
sion of auto trips to other modes, a stronger impact on VMT's
than on trips, and a reduction in discretionary travel,
particularly that to distant destinations. An effect which
would be different from those caused by the gasoline tax,
however, would be for multi-car families to utilize their
less polluting vehicles more intensively; it will be recalled
that a gasoline tax may have the alternative effect of
causing older, precontrolled vehicles, those with better gas
mileage rates, to be used more than would have otherwise
have been the case. To the extent that an MER tax increases
the variable cost per mile the same amount as a hypothetical
gasoline tax, it would have similar effects on driving
behavior overall, even though there would be differential
impacts according to the age of the car. If either an MMY
or MER tax provided an incentive for retrofitting, then this
form of pollution abatement would increase as the taxes them-
selves increased.
The auto stock would be smaller owing to the imposition
of either tax because of the effective increase in ownership
costs. Moreover, because the tax would be higher relative
to the value of older cars, the median age of the auto stock
would decline, since scrappage rates relative to new car
purchases would increase.
Similar to the other proposals considered above, the
emissions taxes would make transportation more expensive and
57
-------
would cause a realignment of land use. One would expect to
see, in the long run, a greater clustering of housing, employ-
ment centers and commerce serving households.
Review of the Literature — The effectiveness of any emissions
tax will be determined to a large extent by its administrative
and technical feasibility (i.e. , can the knowledge and
resources needed for implementation on a large scale be chan-
neled into the project at a reasonable cost). A related
problem is the degree of certainty with which average emis-
sions (and mileage) for classes of vehicles can be estimated.
One study, which attempted to predict mileage and emissions
from vehicle and engine characteristics, came to the pessi-
mistic conclusion that an "indirect" tax, such as that
considered in proposal two, is not feasible because the vari-
ation among supposedly identical vehicles is too high.17
Furthermore, since very little is known about the deteriora-
tion rates of emissions control devices under varying road
maintenance conditions, this introduces more uncertainty into
average emissions calculations.18 Stated in a positive way,
the conclusion from these results is that continual testing
of samples of vehicles is necessary in order to give accurate
figures for relative taxes among vehicles if the MMY option
is instituted.
On the other hand, there is considerable disagreement
as to whether an emissions tax of the type in proposal three,
an MER tax, is technically and financially feasible. Exten-
sive field experience may be required before it is possible
to decide whether an inexpensive, relatively tamper-proof
odometer is practical. To measure emissions, a large scale
inspection system is now feasible, according to d'Arge and
Northrup;19 Bellomo and Dewees have reached different con-
clusions ("the technology for measuring emissions quickly
and cheaply does not yet exist").20 One possible basis for
58
-------
the difference in opinions is that both d'Arge and Northrup
were considering the specific case of California, while
Bellomo and Dewees were basing their conclusions on consi-
derations of a national policy.
A cost function with which to measure the cost effective-
ness of different types of emission test systems was developed
by the Northrup Corporation in their study of the possibility
of implementing a mandatory inspection system in California.
Using what is known as the key mode test in state-operated
inspection stations, they estimated that inspection cost per
vehicle would be $1.05, with an initial investment of
$19,830,000. These costs are, of course, considerably dated
insofar as secular inflation has caused them to increase since
the study was completed in 1971.
There has been virtually no work done on what the effects
of an MER tax would be. It can be assumed, however, that
traveler reaction to an MER tax would be similar to the
effects of a gasoline tax. Both taxes increase the variable
costs per mile of an auto trip. Therefore, from our discus-
sion of the effects of a gasoline tax, we can conclude that
the elasticity of automobile trips (measured in terms of
VMT's and/or actual trips) would be between -.25 and -.7
when the tax is measured in terms of a percentage of the
variable costs per mile.
In analyzing the effects of a tax on emissions, it is
useful to make a comparison to the effectiveness of a policy
requiring cars to meet a mandatory standard. In a 1971 study
for California, it was estimated that for standards causing
a 50 percent rejection rate, an inspection system would
bring about an initial reduction on the order of 25 percent
in carbon monoxide and hydrocarbon exhaust emissions; these
reductions would decline as implementation was delayed
because of the change in the age distribution of automobiles.21
59
-------
In a somewhat less rigorous study, it was estimated that an
initial reduction of 10 percent to 25 percent in aggregate
emissions for a given state or region would be the upper bound
that could be expected, given a 20 percent to 40 percent
rejection rate. It was emphasized that the 10 percent figure
was more likely than the 25 percent figure, and that these
estimates did not take into account deterioration of emission
control systems over time.22
Little quantitative work has been done on estimating
motorist reaction other than travel response in order to
determine the appropriate annual level for a tax. One
California study concludes that a charge of $200 or more per
"pollution unit" would begin to induce motorists to take
measures to reduce pollution; a "pollution unit" is defined
as the total emissions from a non-controlled auto driven
10,000 miles per year.23 Moreover, if the charge is based
on an average emissions measure rather than a measure of the
emission of each individual auto, the incentive to perform
maintenance, or possibly even retrofit, is lost, and the
charge would probably have to be higher in order to induce
purchase of newer, cleaner cars.
In summary, most research on the effects that an MER or
MMY tax would have on changes in the auto stock, including
retrofitting, maintenance, and adjustments in the age distri-
bution, are highly speculative and qualitative. Much of the
commentary must be put in the perspective of the proportion
of ownership or operating costs which would be due to the
taxes under consideration. For example, an annual tax rate
of $200 is equivalent to 17 percent of the total average
ownership cost of $1,200 per year. However, taxes of the
MER type of sufficient quantity to induce retrofitting
would be closer to $100 a year for most vehicles; such
taxes would increase the variable cost per mile of a trip
from 2 percent to 40 percent, depending upon the particular
60
-------
pollutant being controlled and the age of the automobile.
Chapter 5 gives rough, quantitative estimates of the effects
of these tax strategies on retrofitting and changes in the
age distribution of the regional fleet.
Transit Improvements
For the purposes of this study, transit improvements are
a passive response to the other policies. As pollution
control taxes on auto travel increase costs to motorists, it
is expected that transit improvements will become a more
effective substitute for the auto. Such an approach to ana-
lyzing the impacts of changes in transit somewhat constrains
the number of potential transit service improvements. In
particular, we will consider only the effects of an addition
to transit which substitutes trips which would otherwise not
be made.
Even this constraint allows for a wide variety of types
of transit service improvements. Some of the options include
rail rapid transit, express bus, conventional bus, dial-a-
ride, etc., to name only those for which technology presently
exists. Within each type there are further issues of route
structure, frequency of service, and other options for net-
work design. In order to further screen the potential
service improvements, we need to consider which appear to be
most cost-effective on a priori grounds.
Of the several types of transit service improvements
which are technologically feasible, we need only be concerned
with those which can be implemented relatively quickly and
do not have high initial capital costs. The reason for
applying these criteria is that the benefits of auto trip
diversion are themselves a relatively short-run phenomenon;
as new cars with increasingly strict controls are introduced
into the fleet, the emissions from cars will decline toward
acceptable levels. The result of this consideration is to
61
-------
rule out rail rapid transit, since it typically takes a
decade to implement and requires many more decades to recover
the initial capital costs.
The remaining options include express bus service, dial-
a-ride and conventional bus. In Los Angeles, conventional
bus and express bus merge into the same category unless ex-
press bus is considered to be only those systems where bus
operations are put on separate right-of-ways. However, sepa-
rate grade right-of-way is also a high capital cost improve-
ment, though not as expensive as rapid rail. Separate right-
of-way of existing freeways would improve service somewhat
during rush hours, but a cursory look at the existing bus
network indicates that the most important components of trip
times are access time to the bus and schedule delay. For
these reasons, express bus is not examined in this report
though it is recognized that such systems should be evaluated
in order to determine their cost-effectiveness (see Chapter
6 for a discussion of future research needs).
Given that access time and schedule delay are the most
important considerations in designing service improvements,
one would expect that dial-a-ride transit (DRT) systems may
be an effective candidate for transit policy. However,
little is known at this time about the service characteristics
of a. DRT system of the size necessary to serve the Los Angeles
area. Most of those currently in operation are on modest
scales with limited area coverage. Though conceptually the
demand relationships used in this report can be utilized to
forecast the patronage for DRT under assumptions about the
level of service, the equilibrium system performance of a
DRT network cannot be determined. As a consequence, neither
the equilibrium travel nor the costs for the service can be
estimated without more research into the existing systems in
order to infer system performance. Again, additional research
in this area would be valuable (see Chapter 6).
62
-------
The above mentioned gaps in the state of knowledge on
DRT, in conjunction with the considerations on express buses,
suggest that the relevant system improvement to focus on is
changes in the conventional bus network. This is still an
ambiguous task, however. One approach is route extension;
another is adding buses to existing routes.
To compare the efficacy of these two methods, it is useful
to determine whether bus patronage is more sensitive to walk
time or to ride line-haul time. The major research on this
topic includes the previously mentioned Charles River Asso-
2 4
ciates estimate of direct demand models for urban travel.
The results of this model show that the elasticity of transit
trips with respect to an increase in access time (walking) is
over -.7; the elasticity of transit trips with respect to an
increase in line-haul time (riding) including schedule delay
is about -.4. The calculations of the cross-elasticities on
auto trips with respect to increases in access time and line-
haul time for transit also demonstrate that travel behavior
is more sensitive to access time. The results indicate that
decreasing access time by extending route miles is more
effective in drawing passengers to buses than increasing
frequency of service on existing routes.
In order to determine a priori the effects of increased
route coverage, we have to be highly speculative because the
evidence is sparse. However, the following calculations
will help give some idea of the expected effect on VMT's and
auto trips. Consider the elasticity of auto trips with
respect to an increase in variable costs per mile to be -.20
(the lower bound presented above). Then a 50 percent tax on
auto variable costs per mile would decrease auto trips by
10 percent. Further assume that the current mode split in
Los Angeles is 70 percent auto and 6 percent transit, with
the remaining 24 percent being auto passengers and walking.
If the transit network is to increase in order to substitute
63
-------
for auto trips foregone, it would need to about double its
capacity in route miles, thereby about halving the average
access time throughout Los Angeles. Using the average cross-
elasticity of transit access time on auto work and shopping
trip demand from the CRA direct demand model (.149), this
would decrease further auto trips by 8 percent. Thus, the
cumulative effect on auto trips of the increased variable
cost per mile and the assumed improvements in transit would
be a decline of 18 percent. The cumulative elasticity in
terms of the increased variable costs per mile would be -.36.
This is admittedly a rough estimate and there is consequently
some error associated with our a priori expectations of the
effects of transit system improvements on auto trips.
64
-------
List of References, Chapter 2
^omencich, Thomas A., and Daniel McFadden. Urban Travel Demand.
Amsterdam, North-Holland, 1975. 215 p.
Kraft and Wohl. New Directions for Passenger Demand Analysis
and Forecasting. Transportation Research 1. 205-230, 1967.
2Domencich and McFadden, op. cit.
3Horowitz, Joel, and Lloyd Pernela. Analysis of Urban Area
Automobile Emissions According to Trip Type. Transportation
Research Record. p. 492, 1974.
^Automotive Exhaust Emission Surveillance Study., A Summary. Calspan
Corporation. EPA Report #APTD-1544. May 1973. p. 40.
5Kircher, D. S. Light Duty Gasoline Powered Vehicles. In: Com-
pilation of Air Pollutant Emission Factors. Second Edition. Environ-
mental Protection Agency. AP-42. April 1972. pp. 3.1.2-5.
6Infeld, D., et al. Governmental Approaches to Automobiles Air
Pollution Control. Institute of Public Administration.
Washington, D.C. 1971. pp. 1-2.
7Kulash, Damian. Parking Taxes for Congestion Relief: A
Survey of Related Experience. Urban Institute. Washington,
D.C. Working Paper 1212-1. May 1973.
8Kulash, Damian. Parking Taxes as Roadway Prices: A Case
Study of the San Francisco Experience. Urban Institute.
Washington, D.C. Working Paper 1212-9. December 1973.
p. 17.
9Kulash, Damian. Parking Taxes for Congestion Relief... op.
cit. pp. 32-34.
10 Roth, G. J. Parking Spaces for Cars: Assessing the Demand.
Cambridge, University of Cambridge Press (Department of
Applied Economics), 1965.
^Dewess, Don. Chapter 5. In: Economics and Public Policy: The
Automobile Pollution Case. Cambridge, MIT Press, 1974.
12Dewees, D., op. cit. , p. 20.
13Dewees, D., op. cit. , p. 20.
14Dewees, D., op. cit. , pp. 20-21.
15 Lester, Lees, et al. Smog: A Report to the People. Pasadena,
California Institute of Technology, 1972. pp. 105-106.
16Jacobie, H. J., J. D. Steinbruner, et al. Clearing the Air:
Federal Policy on Automotive Emissions Control. Cambridge, Ballenger
Publishing Company, 1973. See especially Chapter 5 for the
many technical problems related to the measurement of
emissions.
65
-------
List of References (Continued)
l7d'Arge, R., T. Clark, and 0. Bubik. Automotive Exhaust
Emissions Taxes: Methodology and Some Preliminary Tests.
University of California at Riverside. Project Clean Air
Research Project #5-12. September 1970.
18Revis, J. S. Short Term Transportation Control Strategies
for Air Pollution Control. Highway Research Board Record.
465:4, 1972.
19 Mandatory Vehicle Inspection and Maintenance, Part A — A
Feasibility Study, Volume I: Summary. Northrup Corporation,
in association with Olson Laboratories. Prepared for State
of California Air Resources Board. June 1971.
2DBellomo, S. J., et al. Providing for Air Quality and Urban
Mobility. Highway Research Board Record. 465:4, 1973.
21Mandatory Vehicle Inspection and Maintenance... op. ait.
pp. 2-1, 2-2, 8-1.
22Infeld, D., et al. op. cit. , Chapter 1.
23Dewees, D. op. cit. pp. 138-139.
2k See: A Model of Urban Passenger Travel Demand in the San Francisco
Metropolitan Area. Charles River Associates Incorporated.
Cambridge. Prepared for the California Division of Bay Toll
Crossings. 1967; Kraft, Gerald, and Thomas A. Domencich.
Free Transit. A Charles River Associates Incorporated Report.
Cambridge, Lexington Books, 1970; Domencich, Thomas A., Gerald
Kraft, and Jean-Paul Valette. Estimation of Urban Passenger
Behavior: An Economic Demand Model. Highway Research Board Record.
238, 1968.
66
-------
3. EFFECTS OF POLICIES ON TRAVEL BEHAVIOR
This chapter presents the estimated effects of pollution
control policies on VMT's and auto trips. The models for
predicting these effects are developed in Appendix A (Travel
Demand). Also included in this chapter are the effects of
the policies on mode split. The uncertainty associated
with the models and their applications is discussed in
order to determine the robustness of the predicted results.
Each of the policies under consideration imposes costs to
individuals and society which must be considered. Using the
results of the predicted effects in this chapter and Chapter
4, the cost-effectiveness of the various policies is esti-
mated in Chapter 5.
The effects of pollution control strategies on travel
behavior were simulated using the demand model developed in
Appendix A. Chapter 2 covered the conceptual framework of
the model and the types of pollution control policies to be
analyzed, along with their a priori, expected effects. The
travel-related policies under consideration include:
• a tax on the variable costs per mile of autos related
to either gasoline or vehicle emissions;
• a surcharge on nonresidential parking;
• improvements in the conventional bus system which
substitutes for auto trips.
The demand model predicts travel behavior separately for
work trips and shopping trips. The results of these simula-
tions are extrapolated to include all trip purposes.
67
-------
The base year for the forecasted effects is 1974. The
travel demand models were used to predict travel behavior
in 1974 in the absence of any of the proposed pollution
control policies. The model itself was calibrated and tested
on 1967 data for Los Angeles and Orange Counties; thus some
changes in the level of service variables were necessary in
order to forecast 1974 travel behavior. Discussion with
transportation officials in the Los Angeles region revealed
that the only significant changes in overall system level of
service between 1967 and 1974 have been the increased variable
costs per mile of auto travel and a decrease in transit fares.
As noted in Table 8, the 1974 variable cost per mile is
$.0571 compared to $.0300 in 1967. The new fare policy,
introduced in the spring of 1974, is a flat fare of $.25
for a one-way transit trip with a $.05 charge for transfers.
These prices replace the 1967 fares which were based on
distance traveled. The above changes in trip costs were sub-
stituted into the travel demand model to predict the base
case 1974 travel behavior. (The base case estimates of mode
split are presented in Tables 13 and 14 for work and shopping
trips.)
Other changes in the intervening years affect the number
of trips in the zonal interchanges for the samples being
analyzed. Most notably, land use changes would cause adjust-
ments in trip generation and distribution. However, the model
is applied to determine the percentage change in trips and
VMT's rather than the absolute level of change. These per-
centages can then be applied to the regionwide totals of
1974 trips and VMT's to determine the absolute level of
effect of the policies. Thus, it is appropriate to use
zonal interchanges as samples of observations even though
the number of trips within any interchange is not the same
in 1974 as it was in 1967.
68
-------
It should also be noted that the travel demand model
*
was estimated on Pittsburgh data from 1967. Though the
model is behavioral, there are regional peculiarities in the
form of excluded variables which may bias its application to
Los Angeles. Chief among these would be climatic conditions
which would have the effect of making access time more onerous
in Pittsburgh than in Los Angeles. However, during the
process of estimating the model, it was found that weather-
related variables were not statistically significant in
explaining travel behavior. Moreover, the model does not
appear to underpredict 1967 Los Angeles transit patronage,
as would be the case if a significant bias was present (see
Appendix A).
More serious is the problem of applying 1967 estimates
to 1974 conditions, especially in the face of increases in
household income which have occurred over that time. Though
income is not an independent variable in the demand models,
it probably has an effect on the relative weight travelers
put on costs as compared to time. In order to determine the
significance of this potential bias, the average 1970 income
(from census data) for households in the Los Angeles sample
was calculated and compared to the average household income
in the Pittsburgh sample used for model estimation. For
the Los Angeles sample, average household income was $7939
per year? for the Pittsburgh sample, average 1967 income
was $8800 per year. Even allowing for inflationary increases
in income since 1970, the Los Angeles sample will not have
income levels greatly different from those obtained by house-
holds included in the Pittsburgh estimation sample. Thus,
it can be presumed that the model's implied relative trade-
offs between travel cost and time will be reasonably accurate.
See Chapter 2 for a brief discussion of the model.
See Domencich and McFadden. op, oit. Chapter 7. However,
day to day response to weather may not capture all the travel
effects due to climate.
69
-------
VMT AND TRIP REDUCTIONS WITH THE CURRENT TRANSIT SYSTEM:
WORK AND SHOP TRIPS
The specific policies considered in this section are the
increases in automobile variable costs per mile caused by an
emissions or gasoline tax and the increase in automobile out-
of-pocket expenses caused by a parking surcharge. For each
type of policy, four levels of tax were simulated, and their
effects on VMT's and auto trips were estimated in terms of
the percentage change from the 1974 base forecast.
Specific Tax Levels
For emissions and gasoline taxes, the four levels repre-
sented increases in auto variable costs per mile of 25 per-
cent, 50 percent, 75 percent and 100 percent. Parking
charges, which are currently insignificant for most trips in
Los Angeles, were increased in absolute amounts of $.25,
$.50, $.75 and $1.00 per trip in the simulations. Table 9
presents the average absolute and percentage cost increases
for work and shopping round trips. This table allows compari-
son and conversion between percentage and absolute terms of
reference for the tax policies. It can be seen that the
additional cost per trip caused by a given percentage
increase in variable costs per mile is much greater for work
trips than for shopping trips. Conversely, increases in
parking charges will affect shopping trips more than work
trips on a percentage basis.
The percentage increases in auto variable costs per mile
are converted to gasoline tax increases and emissions taxes
in Table 10. For example, a 50 percent increase in the
variable cost per mile for autos is equivalent to the
following:
• an increase in the pump price of gas of $0.39 per
gallon, or about 71 percent;
70
-------
Table 9. ESTIMATE OF AVERAGE PERCENT AND ABSOLUTE
COST INCREASES OF AVERAGE 1974 ROUND TRIP INCURRED
BY POLICY SCENARIOS
Policy
Variable cost per
mile i ncrease:
25%
5Q%
15%
100?
Policy
Parki ng cost
increase:
$0.25
0.50
0.75
1 .00
Work trip,
$ increase
$0.20
0.40
0.60
0.80
Work trip,
% increase
J>\%
62
94
125
Shopping trjp,
$ increase
$0.07
0.13
0.20
0.27
Shopping tnp,
% increase
94%
187
281
374
Based on average work round trip length of 14.04 miles.
Based on average shopping round trip length of 4.68 miles.
71
-------
Table 10. GASOLINE AND EMISSIONS TAXES ASSOCIATED WITH
EACH LEVEL OF INCREASED VARIABLE COSTS PER MILE
Variable cost per
mile increase:
25?
50
75
100
Gasoline tax3 increase
% $/qal
35?
71
106
141
$0. 19
0.39
0.58
0.78
Emissions tax
-------
• a tax on carbon monoxide emissions equivalent to
.0873 cents per gram per mile;
• a tax on hydrocarbon emissions equivalent to .7936
cents per gram per mile;
• a tax on nitrous oxide emissions equivalent to .8390
cents per gram per mile.
It is likely that an emissions tax will not be placed
on any single pollutant but will instead be based on a formula
which includes all of the above emissions. Therefore, it is
worth noting that any weighted average of the emissions tax
rates in Table 10 is equivalent to the increased variable
cost per mile. For example, emissions taxes based on the
formula of 50 percent from CO, 25 percent from HC and 25
percent from NO at a combined rate of a 50 percent increase
A.
in the variable cost per mile would have the following three
components:
.50 x .0873 = .0437t/gm of CO
.25 x .7936 = .1984^/gm of HC
.25 x .8390 = .2098t/gm of NO
2C
Model Predictions
The effects of the tax policies are summarized in Tables
11 and 12. The implied elasticity of work trip VMT's with
respect to auto variable cost per mile is -.38; the implied
elasticity of shopping trip VMT's per auto variable cost per
mile is -.17. The former figure is well within the bounds
implied by previous travel demand studies (see Chapter 2).
The shopping trip elasticity is rather low, but it can be
explained by the tendency for shopping trips to be relatively
short (4.68 miles per round trip) and to offer comparatively
few transit options in the Los Angeles area, assuming no
change in tripmaking frequency. (This assumption is relaxed
at the end of the chapter.) The implied elasticity for
73
-------
Table 11. EFFECTS OF TAXES ON WORK TRIPS
Gas or emissions tax:
Variable cost per
mile i ncrease:
25%
50
75
100
Parking tax:
Parking cost
increase:
$0.25
0.50
0.75
1.00
% Change in VMT's
-10.073*
-19. 154
-27.547
-35.303
% Change in VMT's
- 3.496?
- 7.099
-10.618
-14.000
% Change in auto trips
- 7.226$
-13.709
-19.341
-24.442
% Change in auto trips
- 5.845$
-1 1 .583
-17. 109
-21 .998
Table 9 converts taxes into absolute and percentage terms for average
length trips. Table 10 converts variable cost per mile increases into
equivalent taxes on gasoline and emissions.
74
-------
Table 12. EFFECTS OF TAXES ON SHOPPING TRIPS
Gas or emissions tax: j
Variab le cost per
mile i ncrease:
25%
50
75
100
Parking tax:
Parking cost
i ncrease:
$0.25
0.50
0.75
1.00
% Change in VMT's
- 4.541$
- 8.726
-12.477
-15.872
% Change in VMT's
- 6.685%
-12.319
-15.938
-17.215
% Change in auto trips
- 4.498$
- 8.749
-12.939
- 1 7 . 006
% Change in auto trips
- 8.626$
-15.958
-20.209
-21.010
Table 9 converts taxes into absolute and percentage terms for average
length trips. Table 10 converts variable cost per mile increases into
equivalent taxes on gasoline and emissions.
75
-------
gasoline for work trips is -.27; for shopping trips, -.12.
Again, these results are within acceptable bounds, especially
considering the relatively low level of alternative transit
service.
As expected, the effects of a parking tax on VMT's
were significantly less than an equal increase in average trip
costs caused by a gasoline or emissions tax. For example,
extrapolating from the results for the work trip, a parking
surcharge of $.80 would cause an 11.29 percent decline in
VMT's; yet an equal increase in variable costs per mile (100
percent increase on average, from Table 9) causes a 35.30
percent decline in VMT's. A similar calculation for shopping
trips shows that a parking surcharge of $.27 causes a VHT
decline of 7.14 percent; a comparable increase in variable
costs per mile would induce a 15.87 percent decline in VMT's.
A more detailed breakdown of the effects of the pollution
control strategies is presented in Tables 13 and 14. From
the mode split changes it can be seen that the parking charge
significantly increases the number of driver serve passenger
trips, which consequently boosts VMT's, since each such trip
doubles the number of miles per trip which would have been
the case for a driver serving himself.
Predicted Changes in Travel Patterns
The model simulates over zonal interchanges and, as a
consequence, there is information additional to the above
tables about the effects of policies on travel behavior.
This detail on travel patterns helps one to understand how
the model works in giving predicted effects. Most of the
issues presented briefly below are discussed at greater
length in Appendix A.
Effects of Policy by Distance of Trip — The model indicates
that long distance auto trips are the most sensitive to
variable cost per mile increases whereas short distance
76
-------
Table 13. MODE SPLIT ESTIMATES ON WORK TRIPS
For Gas or Emissions Tax:
Mode
Auto
Transi t
Passenger
Driver
serve
passenger
Wa 1 ki ng
Base - 1974
72.984$
7.903
16.613
1 .452
1 .048
Variable cost per mile increase:
25% 50% 75% 100%
69.343$
10.787
17.924
0.730
1 .216
65.313$
14.216
18.684
0.325
1 .462
61.233$
18.086
18.816
0. 162
1 .703
57.456$
22.204
18.314
0.081
i .945
ft.:
For Parking Tax:
Mode
Auto
Transi t
Passenger
Or i ver
serve
passenger
Wa 1 ki ng
Base - 1974
72.984$
7.903
16.613
1 .452
1 .048
Parking cost increase:
$0.25 $0.50 $0.75 $1.00
66.295$
10.518
19.602
2.151
1 .434
59. 184$
13.265
22.528
3.061
1 .962
52.006$
15.972
25.309
4.090
2.623
45. 1 17$
18.622
27.782
5.224
3.255
77
-------
Table 14. MODE SPLIT ESTIMATES ON SHOPPING TRIPS
For Gas or Emissions Tax:
Mode
Auto
Transit
Passenger
Dri ver
serve
passenger
Base - 1974
67.505$
3.923
27.719
0.853
Variable cost per mile increase:
25% 50% 75% 100%
65.240$
4.966
29.238
0.556
62.709$
6. 129
30.776
0.386
60. 155$
7.342
32.246
0.257
57.542$
8.638
33.648
0. 172
For Parking Tax:
Mode
Auto
Transi t
Passenger
Dr i ver
serve
passenger
Base - 1974
67.505$
3.923
27.719
0.353
Parking cost increase:
$0.25 $0.50 $0.75 $1.00
59. 129$
6.551
32.545
1 .775
50.187$
9.571
36.954
3.288
41 .667$
12.601
40.244
5.488
34. 123$
15.522
42. 101
8.254
78
-------
trips are the most sensitive to parking tax increases. The
incentive to take a bus or share auto costs through a car-
pool is greater the larger the total cost of the trip; thus,
if costs increase with distance, incentives for mode diversion
also increase with distance. On the other hand, parking
taxes are a higher percentage of total trip cost for shorter
trips. Thus they penalize short trips more than long trips.
The effect is that for taxes of equivalent average size,
parking taxes will have less impact on VMT's because they
have less impact on longer trips. This effect is reinforced
by the driver serve passenger option. This mode is virtually
never chosen in long trips because of the inconvenience for
the driver who must make a round trip twice a day. However,
it is a viable alternative for short trips and is therefore
especially attractive as a way to avoid high parking charges.
If one increased variable costs per mile, however, the
driver serve passenger mode is penalized because it doubles
the additional cost of a trip compared to auto drive alone.
Effects of Destination Changes^ — The model allows for changes
in destination patterns as a result of travel cost increases.
Because the average length of a shopping trip is relatively
short in Los Angeles, this did not prove to be a major effect.
Most of the change in VMT's in the shopping trips came about
as a result of mode diversions from auto drive alone. This
is partially a result of using sketch plan zone data which
places an effective minimum (of about two miles on average)
on the round trip distance of shopping as the average
distance of an intrazonal trip. More refined data may
have somewhat increased the elasticity of VMT's owing to
destination changes. Long trips are the most susceptible
to destination changes but there were relatively few in the
Los Angeles sample.
79
-------
Effects of Transit Availability — The higher the level of
service of transit (shorter access time, shorter line-haul
time and fewer transfers) the greater the effect of auto
disincentives. Stated another way, the elasticity of auto
travel demand is higher the better the level of service of
the alternate modes. Thus, the policies have their greatest
impact on those origin/destination combinations which are
best served by transit. This result is partially confirmed
in the next section.
VMT AND TRIP EFFECTS WITH IMPROVED TRANSIT SYSTEM:
WORK AND SHOPPING TRIPS
This section considers the effects of increasing con-
ventional bus operations in order to provide alternative
service as a substitute for policy induced decreases in
auto trip travel. Model predictions showed that if a par-
ticular number of bus routes was increased in a zonal inter-
change when 100 or more auto trips had been diverted with-
out improved bus service, then the combined effect of a tax
on variable costs per mile and improved bus service would
be an insignificant change in number of trips — the net
effect of the strategy was an intermodal shift from auto
driver to bus passenger. Thus the bus service provides an
adequate alternative for those wishing to avoid the tax
penalties.
The assumed bus service improvements were all exten-
sions of route miles rather than higher frequencies on
existing routes. The number of bus routes added depended
on the total decline in automobile trips, a decline due to
increased taxes in the absence of changes in the bus network.
Generally, for every decline of 100 auto work trips within
a zonal interchange, a new direct bus route with 2.67
loadings (or scheduled buses — round trip) was instituted
at peak hours; in addition, for every decline of 100 auto
80
-------
shopping trips within a zonal interchange, a new direct bus
route with four loadings (round trip) was instituted. (A
more complete discussion of the assumed network changes is
presented in Appendix C.)
This rule for assigning bus capacity is motivated by
a need to accommodate diverted motorists rather than an
attempt to optimize transit investments. Placing new bus
routes where the most auto drivers are affected automatically
increases the level of service on the highest density routes.
However, as the above discussion indicated, these are also
likely to be the routes where the best service exists before
the improvement. An alternative strategy which locates
additional bus service where there was less impact on auto
travel but a large number of auto trips may be more effective
in reducing VMT's.
Tables 15 and 16 present the effects on VMT's and trips
due to the combined auto variable cost per mile increase and
improved bus service. (The combination parking tax and bus
service improvements scenario was not simulated. One reason
for not examining this set of scenarios is that gasoline and
emission taxes are the dominant policy on cost-effectiveness,
as the previous section has shown. Another reason is that
similar conclusions about the incremental cost-effectiveness
of providing substitute bus service would seem to apply in
either case.) As can be seen from the tables, the net effect
of these policies is a less than 1 percent change in auto
plus transit trips. The percentage changes are with respect
to the 1974 base case projections with no new taxes and no
changes in bus service.
It can also be seen from comparing Tables 15 and 16 to
Tables 11 and 12 that the effectiveness of gasoline and
emissions taxes in reducing VMT's is enhanced by providing
increased opportunity for travel by bus. This is especially
pronounced for shopping trips, where the percentage reduction
81
-------
Table 15. EFFECT ON WORK TRIPS OF GAS OR
EMISSION TAX COMBINED WITH TRANSIT IMPROVEMENTS
Variable
cost per
mile increase
50%
100
% Change
in VMT's
-24.497?
-43.038
% Change in
auto driver
alone trips
-16.796?
-28.177
% Change in
transit trips
+147.959?
+268.367
% Change in
auto and
transit trips
-0.698?
+0.798
82
-------
Table 16. EFFECT ON SHOPPING TRIPS OF GAS OR
EMISSION TAX COMBINED WITH TRANSIT IMPROVEMENTS
Variable
cost per
mile increase
50$
100
% Change
in VMT's
-15. 636$
-28.810
% Change in
auto driver
alone trips
-16.235$
-30.701
% Change in
transit trips
+270.652$
+521.739
% Change in
auto and
transit trips
-0.478$
-0.358
83
-------
in VMT's is nearly doubled. These results are somewhat
expected, since the additional bus routes offer significantly
reduced access time (and some reduced line-haul time, usually
owing to direct service being substituted for transfers), and
this is a particularly elastic component in mode choice
models. Note that the additional bus service did not typi-
cally have a large impact on wait time and that the assumed
fares were the same as currently obtain in Los Angeles.
Tables 17 and 18 show the mode split estimates which
result from the combined strategies. (The 1974 base esti-
mates were presented in Tables 14 and 15.) It can be seen
that the mode split under the assumed scenarios more
closely replicates the experience of older, CBD
oriented cities which have a more highly developed
transit system.
The aggregate elasticity for work trip VMT's with
respect to a change in both auto variable costs per mile
and improved transit is -.4304. The similar elasticity for
shopping trips is -.3004. The implied price elasticities
for gasoline are -.3251 for work trips and -.2123 for
shopping trips. All of these results satisfy a priori
expectations and indicate that in situations with higher
levels of transit service, VMT's are more sensitive to auto
penalties.
THE AGGREGATE SHORT-RUN TRAVEL EFFECTS
This section gives the results of extrapolating the
above simulations to include all trip purposes. Using 1974
*
adjustment factors to trip purpose categories, work trips
presently account for about 24 percent of total Los Angeles
region trips and shopping trips for about 17 percent. The
remaining trip purposes have been allocated to either the
*These were given to CRA by Jerry Bennett of Caltrans,
District 07 (LARTS District) .
84
-------
Table 17. MODE SPLIT ESTIMATES WITH TRANSIT IMPROVEMENTS
FOR WORK TRIPS
Policy Share of
alternative auto
Bus improvements
and auto operating
cost increase:
50$
100
61 .319$
52.888
Share of
transit
19.788$
29.374
Share of
passenger
17.427$
16. 1 II
Share of driver
serve passenger
0.326$
0.801
Share
walk
1. 140$
1.546
85
-------
Table 18. MODE SPLIT ESTIMATES WITH TRANSIT IMPROVEMENTS
FOR SHOPPING TRIPS
Policy Share of
alternative auto
Bus improvements
and auto operating
-cost increase
50%
100
56.886$
47. 122
Share of
transit
14.629$
24.570
Share of
passenger
28. 185?
28. 179
Share
serve
0.
0.
of driver
passenger
300$
129
86
-------
work trip elasticities or the shopping trip elasticities
depending on which category was deemed most appropriate.
Table 19 presents the estimated trip purpose shares
for the Los Angeles region in 1974. Trip purposes of similar
attributes in terms of average trip length and oJ sic 'ficant
destination choice have been grouped together. Using these
criteria, it can be seen that about 35 percent of total trips
are most closely associated with the attributes of work trips,
and the remaining 65 percent can be assumed to be associated
with shopping trip attributes. When the trip purpose shares
are weighted by the average length of trip, the relative
contribution to VMT's from each of the major categories is
roughly equal.
To determine the aggregate effects of the pollution
control strategies on auto trips and VMT's, weighted
averages were applied to the simulation results on work and
shopping trips. These weights depend on the relative shares
by purpose of region-wide auto trips or VMT's, whichever is
appropriate, in each of the two major categories. The
weights are changed at each tax level for simulations at the
next tax level to reflect the change in relative trips or
VMT's caused by the policy.
The area covered by the sample of work and shopping
trips includes Orange County and Los Angeles County south
of the mountains. This area corresponds roughly to the Los
Angeles Air Quality Control region. However, it does not
correspond to the original and existing Los Angeles Regional
Transportation Study (LARTS) area, which is the source of
all aggregate data for the region. LARTS includes all of
Los Angeles County, Orange County and parts of Ventura, San
Bernardino and Riverside Counties. It is estimated that the
Air Quality Control region contains about 65 percent of all
87
-------
Table 19. ESTIMATED 1974 TRIP PURPOSE SHARES
Purpose
Home-work
Work-other
Work-work
Home-education
Subtotal share ("work")
Home- shop
Home-social, entertainment,
recreation
Home-other
Other-other
Home- home
Subtotal share ("shop")
Total trip
shares
20.93$
3.35
6.23
4.23
34.74$
16.71?
14.49
14.64
17.07
2.36
65.27$
VMT
share
30.02$
4.80
8.94
4.55
48.31$
7.57$
18.25
10.57
12.32
2.97
51.68$
88
-------
LARTS VMT's (it will be seen that our estimate is consider-
ably less). The 1974 estimate of weekday VMT's for the
*
entire LARTS region is in the 160 to 170 million range.
It is, in general, difficult to get a consistent esti-
mate of total trips and VMT's for the auto trips which take
place solely within Los Angeles and Orange Counties. CRA
arrived at the figure of 62.570 million miles per weekday
using the following approach: first, we received from the
Southern California Rapid Transit District (SCRTD) an esti-
mate of total round trip bus riders per weekday for Los
Angeles and Orange Counties (412.5 thousand); using this,
we computed total household trips by all modes and for all
purposes by dividing the number of bus trips by the esti-
mated 1974 base case bus share. (The mode split estimate
for transit was checked with SCRTD planners and found to be
consistent with their own estimates.)
number of bus trips , , ,
total trips * = base case bus share
_ number of bus trips
total trips - base case bus share
Total auto trips were determined from the estimated 1974
base case mode split for auto trips (5.479 million round
trips). The aggregate number of trips was multiplied by
the average auto round trip length (estimated by CRA to
be 11.420 miles) to get the daily total VMT's of 62.570
million.
It should be noted that this estimate is significantly
less than would be expected from the Caltrans estimate of
total VMT's for the LARTS region. Obviously, significantly
*Gerry Bennett of Caltrans District 07 supplied CRA with
much of this information. It should be noted that the
estimates on VMT's are preliminary.
89
-------
fewer trips are covered by examining only Orange and Los
Angeles Counties. Also, our estimates do not consider
VMT's associated with vehicles other than household-owned
passenger cars and pickup trucks. It is somewhat distur-
bing that CRA uses an estimate of average trip length
that is significantly lower than that used by Caltrans.
For example, data from the 1967 Household Survey show that
the average work round trip was about J.4.5 miles; the 1974
CRA base case estimate is about 14 miles. LARTS and Caltrans
tend to use 18 miles as the average work round trip distance
from 1967 through 1974.
Table 20 represents a summary of the estimated region-
wide effects of the pollution control strategies. The
following conclusions emerge from the estimates:
• the elasticity of VMT's with respect to increases
in the variable costs per mile is -.27 with the
current transit system and -.38 when combined with
the assumed improved transit system;
• the implied elasticity on the price of gasoline is
-.19 with the present transit system and -.27 when
combined with the assumed improved system;
• the elasticity of VMT's with respect to increased trip
costs due to a parking tax is -.12 with the current
transit system.
UNCERTAINTY OF THE RESULTS
Though the approach to estimating the effects of
pollution control strategies on travel behavior represents
a considerable advance in the state of the art, there are
several sources of potential error which should be noted.
Some sources of error lead to greater randomness in the
predictions; others may bias the results in a particular
direction. We briefly discuss each potential source of
error and its effects on the estimated results.
90
-------
Table 20. ESTIMATED EFFECTS OF POLLUTION
CONTROL STRATEGIES ON VMT'S AND AUTO TRIPS
Gas or emissions tax:
variable cost
per mile increase9
25%
50
75
100
Parking tax:
parking cost
increase3
$0.25
0.50
0.75
1 .00
Transit system
improvements
with variable cost per
mile increase3
25%
100
% Change
in VMT's
- 7.40?
-13.96
-19.58
-24. 13
- 5.04$
- 9.58
-13.07
-15.43
-20.21?
-35.77
% Change in
auto trips
- 5.45?
-10.50
-15.27
-20.39
- 7.66?
-14.46
-19. 18
-21 .33
-16.43?
-29.83
Total VMT's,
millions per
weekday"
57.940
53.835
50.319
47.472
59.416
56.576
54.392
52.915
49.925
40. 189
-T^hie Q converts taxes into absolute and percentage terms for average
trip lengths. Table 10 converts variable cost per mile increases into
•'- . - • . j' 4'it tnx.es on casoiine ana emissions.
b6ased on an estimated total of 62.570 million VMT's per average week-
day in 1974 for the Los Angeles Air Quality Control Region.
91
-------
Model Structure and Estimates
As discussed elsewhere,1 the disaggregated demand
model has strong theoretical foundations as a represen-
tation of travel choice behavior. The structure and speci-
fication of the model are relatively sound.
Unfortunately, estimation techniques for the model do
not allow easy interpretation of most commonly used test
statistics. However, some information about the confidence
one may have in the estimates is available, including the
following:
• The parameter estimates in all relationships are
generally significant within a 2.5 percent level
of confidence using a one-tailed t-test.
• A sensitivity test of parameter estimates by
estimating alternative specifications indicates
that the mode choice models are quite robust.
• The predictive ability of each relationship is
relatively good when applied to individual house-
holds in the estimation sample, although no easily
interpretable test statistic for this characteristic
of a probability choice model has been devised.
• The parameter estimates were consistent with a
priori expectations and other studies of travel
demand and value of time.
The predictive ability of the model has been validated
in this study with one important exception: the frequency
of shopping trips relationship is overly sensitive to the
cost of travel. For this reason, the choice of whether
to make a shopping trip during a 24 hour period was not
*See Appendix A of this report.
92
-------
simulated under the alternative policies. The effects of
this omission on the estimates are discussed in more detail
in a later section.
Data
Underlying errors in the data will affect estimation
results in two broad categories: errors in observation and
sampling error. These are each discussed in turn.
Errors in observation — The data are discussed in detail
in Appendix D. The primary source of data for this project
has been the 1967 LARTS Household Survey. The survey itself
can be presumed to have yielded reasonably accurate obser-
vations of the variables included in the model simulations,
and considerable effort has gone into augmenting this source
with other data on interzonal distances, transit level of
service and socioeconomic characteristics of zones. It can
be concluded that these efforts have been regarded with
minimal errors in the data.
Sampling error -- Although only a fraction of the total
zonal interchanges were simulated, the samples are repre-
sentative of the regional population. Aggregate mode
shares from the samples were very close to those obtained
from the regionwide survey.
Model Application
For this study, the area of model application gives
rise to the ^:iost serious sources of error. Six potential
problems are discussed below.
Generalizability -- As mentioned before, the model para-
meters were estimated on 1967 Pittsburgh data, and it is
possible that this would make its estimates inapplicable
to a different region for a different year. However, the
behavioral nature of the model implies interregional gener-
alizability. The major influence of time on the model
93
-------
structure — secular increases in the value of time owing
to rising money income — has been previously discussed.
It was concluded that this is not a major factor in
biasing the results.
Application beyond the range of observations for estimates --
As a general rule, forecasts employing estimated equations
become more uncertain as the values of independent variables
used in the forecast exceed the range of independent vari-
ables used in estimating the model. The auto variable
cost per mile is the only variable for which values in
the simulations substantially exceeded the values found in
the Pittsburgh data set. The policies themselves are most
responsible for this problem insofar as one of their pri-
mary impacts is to increase significantly the variable cost
per mile for automobile trips. In this regard, the use of
a priori information helps to validate the predictions of
the model.
Application to other modes — The model was originally esti-
mated on only two modes, auto drive alone and transit, but
is here applied to several others including carpooling,
driver serve passenger and walk. Some uncertainty may be
attached to the estimates which are obtained for the other
modes.
Aggregation to zonal levels — Some adjustments in applying
the model were necessary for it to be an appropriate tool
for prediction with zonal data rather than data on indivi-
dual households. The results of these adjustments, as
documented in Appendix A, add some error to the simulated
effects, but the forecasts are well within reasonable
bounds of accuracy. This issue is of some importance
because the effects of the policies in reducing VMT(s
partially depend upon diversions from auto drive alone to
carpooling. Also, part of the reason parking taxes are
94
-------
found to be less effective (at the same average cost per
trip) compared to other strategies, lies in the inducement
to chauffeured trips. The decisionmaking process for these
other modes is more complex than that represented by the
rather ad hoc adjustments to the basic model performed in
Appendix A. However, accurate models of carpool behavior
do not exist at this time — nor do the data exist to esti-
mate such models. More importantly, the relative ranking
among policies is probably not affected by errors in the
estimated mode shares of carpool, walk and driver serve
passenger as long as the direction of predicted effect is
correct. In this regard, the model performs approximately
as one would expect traveler behavior to change in response
to auto trip cost increases. Thus, though there may be
some error in the elasticity estimates, the ranking of
policies remains sound.
Application to other trip purposes -- The ability of the
unadjusted work or shopping trip model to predict mode
split for other purposes is discussed briefly in Appendix A.
The general conclusion is that uncertainty is added to the
projected policy effects for aggregate travel, but it is
difficult to tell whether these projections are biased.
Again, one must rely on a priori reasoning in making a
judgment about the confidence with which aggregate pro-
jected VMT's and auto trips can be used.
Frequency of trips — It was assumed that the aggregate
number of trips remains constant, although mode and desti-
nation choices may change as the cost of auto travel
increases. Even if such an assumption is appropriate when
analyzing the short-run behavior of non-discretionary travel
(work and education related trips), it is not necessarily
plausible for some other trip purposes. To put quantitative
95
-------
bounds on the underestimation which may occur because of
this assumption, the following exercise was performed:
• Home-shop, home-social/entertainment/recreation, and
home-other were all determined to be trip purposes
for which the frequency of travel may decline owing
to an increase in auto trip costs.
• Their contribution to regionwide VMT's was calculated
from Table 19 to be about 40 percent.
• It was assumed that the effort to cut back on VMT's
by taking fewer trips was equivalent to the effects
of choosing other modes and closer destinations; that
is, the elasticity of VMT's with respect to variable
costs per mile is the same for changes in trip fre-
quency as it is for changes in the combination of
destination choice and mode choice. This assumption
has the effect of, for example, doubling the elasti-
city of shopping trips with respect to a policy
scenario.
• The effect on aggregate changes in VMT's was calculated
using the 1974 base case weights and elasticities from
Table 20; the change in VMT's implied by these assump-
tions was a decrease of approximately 25 percent in
the changes predicted in Table 20.
The above computations place a reasonable lower bound
on the predicted effects of the previous section.
CONCLUSIONS
Though no confidence intervals have been explicitly
calculated, an upper bound on the predicted changes in VMT's
and auto trips has been implied by making a strong assump-
tion about the effects of trip frequency choice on discre-
tionary trips. This upper bound augments the conclusions
about short-run travel response to pollution control policies
in the following ways:
96
-------
• The elasticity of VMT's with respect to increases
in the variable costs per mile is -.27 (with a bound
of -.34) with the current transit system and -.38
(with a bound of -.48) when combined with an improved
transit system.
• The implied elasticity on the price of gasoline is
-.19 (with a bound of -.24) with the current transit
system and -.27 (with a bound of -.34) when combined
with an improved transit system.
• For equivalent incremental costs per average trip,
the gasoline and emissions taxes are over two times
more effective than parking taxes in reducing VMT's/
a conclusion unaffected by the frequency of travel.
Subject to the qualifications discussed in this chapter,
these results are confirmed by a priori expectations and
can be considered to be reasonably reliable.
97
-------
List of References, Chapter 3
'Domencich and McFadden. op. cit.', McFadden, Daniel. Condi-
tional Logit Analysis of Qualitative Choice Behavior. In:
Frontiers of Econometrics, Zarembka, Paul (ed.). New York,
Academia Press, 1974.
98
-------
4. EFFECTS OF POLICIES ON THE AUTO STOCK
This chapter presents the estimated effects of pollution
control policies on age distribution of cars and size of auto
stock. Appendix B (Auto Stock Adjustment) develops models
for predicting these effects. This chapter also discusses
estimates of policy effects on retrofitting. The confidence
to be placed in predicted results is examined. Chapter 5
uses the results presented below in determining the cost-
effectiveness of various policies.
The effects of emissions taxes on the size and age
distribution of automobiles was determined by simulating
the model presented in Appendix B. This chapter considers
two separate forms of emissions taxes, both of which have been
discussed in Chapter 2: first, we present the effects of an
annual tax based on the estimated average emission rate per
mile for automobiles of specific models, makes and years
(MMY tax); next.we describe the impacts of a tax which is
based on the estimated amount of emissions as a result of
the car's emission rate and miles driven (HER tax). Among
the impacts considered for the MER tax is the incentive for
car owners to retrofit with emissions control devices.
EFFECTS OF A MODEL, MAKE AND YEAR TAX
The age distribution of cars affects the aggregate
emission rate from the total stock of cars in that a fleet
with a higher proportion of old cars will have, on average,
greater pollution rates per vehicle. The age distribution
99
-------
of cars is determined by the scrappage rate of cars of dif-
ferent vintages and the proportion of the auto stock which
is made up of new cars. This section gives estimates of
the effects of various MMY taxes on scrappage rates by age
of car and on new car sales.
Table 21 gives the base case estimates for the charac-
teristics of the 1975 auto stock in the Los Angeles area.
These estimates were produced by simulating the model
described in Appendix B with assumptions, also discussed in
Appendix B, about the average price of new cars (assumed to
be $3,850) and the average price of used cars. Scrappage
rates by age are calculated by using Equation 53.
Although only Los Angeles and Orange Counties are repre-
sented, they account for 86 percent of the autos owned in
all five counties which have part of their area in the LARTS
region. These autos undoubtedly account for an even higher
percentage of the pollution in the Los Angeles region. It
can also be assumed that the characteristics of the auto
stock in Los Angeles and Orange Counties are reasonably
similar to those in relevant areas not covered by the data.
The characteristics of automobiles by age are further
described in Table 22. It shows that although there are sig-
nificant numbers of higher polluting cars in the fleet, their
contribution to vehicle miles traveled is much less than
their numbers would otherwise indicate. Pre-1970 cars,
largely uncontrolled, were grouped into two categories
because they have similar emissions rates within the cate-
gories. The decline in emission rates since 1970 has been
fairly continuous for all pollutants. Also to be noted is
that each dollar of an annual tax puts a greater burden in
absolute terms on newer cars because their longer life expec-
tancy increases the present discounted value of the annual
100
-------
Table 21. BASE CASE ESTIMATES OF 1975 AUTO STOCK
CHARACTERISTICS FOR LOS ANGELES AND ORANGE COUNTIES
Total auto stock (in vehicles)
New car sales (in vehicles)
Used car price, average
Aggregate scrappage rate
Model year
1975
1974
1973
1972
1971
1970
1967-1969
pre-1967
4,240,053
403,896
$1,272
0.07786
Scrappage rate by age of auto
0.0013
0.0028
0.0058
0.0120
0.0239
0.0449
O.I 132
0.1898
101
-------
Table 22. ESTIMATES OF 1975 MID-YEAR AUTO STOCK
CHARACTERISTICS BY AGE OF CAR:
LOS ANGELES AND ORANGE COUNTIES
Model year
1975
1974
1973
1972
1971
1970
1969-196?
pre-1967
Model year
1975
1974
1973
1972
1971
1970
1969-1967
pre-1967
Proportion
of total3
0.078!
0.1018
0. 1031
0.0972
0.0818
0.0808
0.2030
0.2552
Discounted
present
cost of $1
annual taxc
$5.6502
5.3282
4.9676
4.5638
4. i 1 14
3.6048
2.6136
2.4018
Average mileage
per vehicle^
15,000
13,000
1 1 ,000
9,600
8,400
7,600
4,900
3,700
Proportion of
total VMT's
0. 1585
0.1696
0.1228
0. 1083
0. 1071
0.0815
0. 1395
0. 1 1 16
Emission rates
(gm/mi)d
CO
19
22
25
26
48
52
72
83
HC
2.70
2.83
2.97
3.05
3.35
4.21
5.62
8.66
NOx
2.30
2.55
2.71
4.20
4.23
5.10
6.14
4.21
Table continues on following page.
102
-------
Table 22. ESTIMATES OF 1975 MID-YEAR AUTO STOCK CHARACTERISTICS BY AGE
OF CAR: LOS ANGELES AND ORANGE COUNTIES (Continued)
Based on estimates from model simulation, including scrappage rate by
age of car, starting with July I, 1973 data on Los Angeles and Orange
counties from Polk Statistics, National Vehicle Registration Service.
See Appendix B for auto stock model and assumptions.
Source: Lees, et al. Smog: A Report to the People. Pasadena, Cali-
fornia Institute of Technology, 1972. Table 18, p. 128.
Based on assumption of 12 percent discount rate of interest. Also,
cars are assumed to have a 10-year life span, except that 3;! cars
are considered to have at least three years cf life rema;ning.
Pre-1970 rates are actja! rates, measured by California State Air
Resources Board, cited in: Downing. Benefit/Cost Analysis of Air
Pollution Control Devices for Used Cars. University of California,
Riverside, September 1970. pp. 3-9.; or predicted rates based on
data in Compilation of Air Pollutant Emissions Factors. Environ-
mental Protection Agency. Report #AP-42. February 1972. pp.
3.1.1-6 through 3.1.2-9.; the higher of either is presented. Other
years are base:: upon precisions in Ccr&ilation of Air Pollutant
Emissions Factors, op. oit.
103
-------
tax stream; however, older cars have the greater burden when
the present discounted value of the tax is viewed as a pro-
portion of the market value of the car.
It is important to note that in the long run nearly all
of the higher polluting cars will exit from the fleet even
in the absence of an emissions tax. National standards on
new cars will cause older autos to be replaced by lower
polluting vehicles as age takes its toll on the existing
stock. Thus, the effects of emissions taxes must be viewed
as providing only interim benefits in terms of lowering the
average emission rate per vehicle. Also, a separate effect
of an emissions tax — a reduction in new car additions
to the auto stock — may only have short-run impact if
currently proposed national standards for future automobiles
are maintained. The projected emission rates from cars in
the late seventies implies that any but the most massive
emissions tax on current cars would be negligible on future
cars. Thus, the reduction in travel induced by less car
ownership would be transitory. (This report does not consider
the feedback effects of automobile ownership on travel
demand, though such an analysis is conceptually possible.)
As a consequence of these effects, the model simula-
tions in this section must be interpreted as providing
interim benefits only. In particular, the questions asked
were:
• whether an MMY tax will significantly increase the
rate at which older cars are scrapped;
• whether an MMY tax will lead to a short-run decrease
in the size of the auto stock.
The results of the model simulations are presented in
Tables 23 through 26. Tables 23 through 25 focus on taxes
based on particular pollutants: CO, HC, and NO . Table 26
Jt
shows the effects of a combination tax on all three pollu-
tants for both a high and a low level of taxation.
104
-------
Table 23. AUTO STOCK EFFECTS OF
ANNUAL TAX ON CARBON MONOXIDE:
LOS ANGELES AND ORANGE COUNTIES
Variable
Total auto stock (in vehicles)
New car sales ( n vehicles)
Used car price, average
Aggregate scrappage rate
Average present value of tax,
all cars
Model
year
1975
1974
1973
1972
1971
1970
1967-
1969
p re- 1967
$1/qm
4,207,293
392,974
$1, I 18
0.0863
$163
$l/qm
Annual
tax
SI9
22
25
26
48
52
70
81
Discounted
present
value of
cost
$107
1 17
124
1 19
197
187
183
195
Scrappage
rate
0.00 3
0.0027
0.0057
0.0120
0.0248
0.0487
0. 1400
• 0.2732
$2/gm
4, 65,515
382,627
$969
0.0971
$326
$2/qm
Annual
tax
$ 38
44
50
52
96
104
140
162
Discounted
present
value of
cost
$2 4
234
248
237
395
375
366
389
Scrappage
rate
S\ (*. f , -7
u . Gu i 3
0.0027
0.0057
0.0120
0.0260
0.0528
0. 1678
0.3581
105
-------
Table 24. AUTO STOCK EFFECTS OF
ANNUAL TAX ON HYDROCARBONS:
LOS ANGELES AND ORANGE COUNTIES
Variables
Total auto stock (in vehicles)
New car sales (in vehicles)
Used car price, average
Aggregate scrappage rate
Average present value of tax,
all cars
Model
year
1975
1974
1973
1972
1971
1970
1967-
1969
pre-1967
$7/gm
4,219,864
392,974
$ 1 , 1 64
0.0830
$1 14
$7/gm
Annual
tax
$19
20
21
21
23
29
39
61
Discounted
present
value of
cost
$107
106
103
97
96
106
103
146
Scrappage
rate
0.0013
0.0027
0.0057
0.01 19
0.0241
0.0465
0. 1269
0.2507
$14/gm
4, 95,762
382,627
$1 ,057
0.0892
$228
$14/gm
Annual
tax
$ 38
40
42
43
47
59
79
121
Discounted
present
value of
cost
$2 4
21 1
207
195
192
212
206
291
Scrappage
rate
0.0012
0.0027
0.0057
0.01 19
0.0244
0.0486
0. 1419
0.3136
106
-------
Table 25. AUTO STOCK EFFECTS OF
ANNUAL TAX ON NITROGEN OXIDE:
LOS ANGELES AND ORANGE COUNTIES
Variables
Total auto stock (in vehicles)
New car sales (in vehicles)
Used car price, average
Aggregate scrappage rate
Average present value of tax,
all cars
Model
year
1975
1974
1973
1972
1971
1970
1967-
1969
p re- 1967
$8/gm
4,219,462
393,272
$1,163
0.0831
$1 5
$8/ gm
Annual
tax
$18
20
22
34
34
41
49
34
Discounted
present
value of
cost
$104
109
108
153
139
147
128
81
Scrappage
rate
0.0013
0.0027
0.0057
0.0120
0.0243
0.0476
0.1309
0.22 7
$16/qm
4 . 1 94 , 9 1 0
282,194
$1,056
0.0895
$230
$16/gm
Annual
tax
$37
41
43
67
68
82
98
67
Discounted
present
value of
cost
$208
2 7
2 5
307
278
294
257
162
Scrappage
rate
0.00i3
0.0027
0.0057
0.0121
0.0251
0.0509
0. 1503
0.2562
107
-------
Table 26. AUTO STOCK EFFECTS OF
ANNUAL EMISSIONS TAX ON ALL POLLUTANTS:
LOS ANGELES AND ORANGE COUNTIES
Variables
Total auto stock (in vehicles)
New car sales (in vehicles)
Used car price, average
Aggregate scrappage rate
Average present value of tax,
al 1 cars
Model
year
1975
1974
1973
1972
1971
1970
1967-
1969
pre- 1 967
Low tax3
4, 149,052
373,260
$908
0. 1015
$392
Low tax
Annual
tax
$ 56
62
68
81
105
122
158
176
Discounted
present
value of
cost
$316
330
338
370
432
440
413
423
Scrappage
rate
0.00(3
0.0028
0.0059
0.0 25
0.0268
0.0557
0.1783
0.4398
High taxb
3,971,173
346,866
$574
0.1508
$784
High tax
Annual
tax
$1 13
125
135
162
21 1
245
317
350
Discounted
present
value of
cost
$633
661
676
739
863
880
826
845
Scrappage
rate
0.0017
0.0037
0.0079
0.0171
0.0378
0.0814
0.2808
0.6290
Low tax equals $7/gm HC + $l/gm CO + $8/gm NOx.
tax equals $!4/gm HC + $2/gm CO + !$!6/gm NOx.
108
-------
Generally, the taxes have a large impact on the scrappage
of older cars and on used car prices. Over time, this effect
could be expected to speed considerably the exit of older cars
from the fleet. For taxes on the order of those presented in
Table 26, the result over time will also be a decline in the
auto stock.
To give an example of the effects the taxes in Table 26
may have, the scrappage rates were applied in an exercise
presented in Table 27. This exercise is based on the fall
1975 age distribution of automobiles; the percentages given
in the four age categories approximate the results that would
be predicted by the model in Appendix B for Los Angeles and
Orange Counties. It is further assumed that 1976 auto sales
are 10 percent of the auto stock in all tax scenarios. The
scrappage rates from Tables 21 and 26 are then applied to
determine the resulting age distribution of cars.
Generalizing from Table 27, it can be seen that the
effect of age alone on redistribution of the auto stock is
significant and that the emissions taxes would further
deplete the stock of older cars. Moreover, the taxes would
cause a decline in the auto stock: the present discounted
value of the "low" tax is about 25 percent of the market value
of the weighted average of new and used cars and causes a
net change — 7 percent in the auto stock on a one to two
year period. The change incurred by the higher tax is
roughly proportional.
EFFECTS OF A MILEAGE-EMISSIONS RATE TAX
There are a number of potential reactions to an MER tax,
including: higher scrappage rates for older autos; fewer
VMT's and trips by auto; switching trips to the lower pol-
luting auto in multicar households; retrofitting presently
owned cars with pollution control devices or tuning cars
more often to lower the measured emission rates. Because
109
-------
Table 27. EXAMPLE OF EFFECTS OF TAX SCENARIOS ON
AGE DISTRIBUTION OF AUTOS OVER TIME
Model year
1976 (assumed)
1973-1975
1970-1972
1967-1969
pre-1967
Change in
auto stock
1975 age
distribution
(assumed)
—
0.30
0.25
0.20
0.25
—
1976 age distribution
No tax
0.10
0.29
0.24
0.17
0.20
+2%
Low tax
0.10
0.32
0.26
0.17
0.15
-5$
Hiqh tax
0. 10
0.35
0.28
0. 16
O.I 1
-13?
110
-------
of the multiplicity of responses, it is not possible to
determine with any precision what the impact of an HER tax
on the age distribution of autos would be; nor could much
meaning be attached to a change in the age distribution as
it relates to aggregate emissions because of the incentive
to retrofit and switch trips among cars in multicar house-
holds. Nonetheless, this section does present some compu-
tations which are indicative of the effects of an MER tax.
Ownership Costs
A tax divided equally, on average, over emissions per
mile of CO, HC and NO is briefly analyzed in this section.
X
The tax is equivalent to, again on average, a 25 percent
increase in the variable costs per mile. Table 28 shows
how the tax would be distributed across age categories of
automobiles. Though the average annual MER tax per car
would be roughly equivalent to the average annual low MMY
tax considered in Table 26, it would be distributed across
age categories quite differently. An MER tax affects the
auto age distribution significantly less than an MMY tax.
The reason is that older cars are driven much less than
newer cars; hence, the annual tax charges on older cars are
less for the MER tax relative to the MMY tax.
Though it is difficult to predict the effect of the MER
tax on the distribution of VMT's across model years, Table 28
presents the results of an exercise using the elasticity of
VMT's with respect to variable costs per mile. This elasti-
city was computed to be -.27 in Chapter 3. Applying the same
elasticity to the percent change in variable cost per mile
for each age category yields an estimate of the average
reduction in annual mileage as a result of the tax. If VMT's
are reapportioned after the tax (see the last column in
Table 28), assuming no change in the age distribution of
autos, there will be a small shift in the proportion of
111
-------
Table 28. EFFECTS, BY AGE CATEGORY OF AUTO, OF MER TAX
EQUIVALENT TO AVERAGE 25 PERCENT INCREASE IN
VARIABLE COSTS PER MILE
Model year
1973-1975
1970-1972
1967-1969
pre-1967
Model Year
1973-1975
1970-1972
1967-1969
pre-1967
Cents per
mile tax3
1 .05
1.92
2.65
2.94
Percent
change in
variable cost
per mile
18.39
33.63
46.41
51 .49
Average
annual
mileage
pre-taxb
13,000
8,600
4,900
3,700
Average
annual
mileage
post-tax
12,354
7,819
4,286
3, 186
Annual cost
at pre-tax
mileage
$136
164
129
108
Annual cost
at post-tax
mileage
$130
150
1 14
94
Proportion of
VMT's pre-taxb
0.45
0.30
0. 14
0. 1 1
Proportion of
VMT's post-tax
0.47
0.30
0. 13
0.10
Based on 0.0145 gm/mi CO, 0.1323 gm/mi HC, and 0.1398 gm/mi NOx.
3Table 22.
112
-------
auto travel away from older to newer cars. These results do
not take into account the ability of multicar households to
shift trips to those vehicles with lower operating costs per
mile. Thus, the actual impact would probably be greater than
the estimated shift shown in Table 28.
Retrofitting
Retrofitting is one of the ways a motorist could reduce
an HER tax bill. However, the incentive to retrofit only
exists if the tax rate is high enough to justify the initial
expense (and higher future maintenance costs) of retrofitting.
Moreover, retrofit devices vary in effectiveness of control
*
for given components of emissions.
In order to determine how large an incentive to retrofit
is implied by the MER taxes considered in this report, the
costs of retrofitting are computed on an average per mile
basis for each age of car. It can be assumed that tax rates
greatly exceeding this cost provide a strong inducement for
retrofitting.
The data on retrofitting costs and effectiveness are
scanty or, in the case of recent models of autos, nonexistent.
Tables 29 and 30 use what data is available to determine
retrofitting cost/effectiveness. These data take account
of neither the deterioration rates of retrofit devices nor
the effects of regular tuneups. It is assumed that a cata-
lytic reactor is applied to reduce carbon monoxide and
*For studies of mandatorv retrofitting schemes, see:
Evaluating Transportation Controls to Reduce Motor Vehicle Emissions in
Major Metropolitan Areas. Institute of Public Administration.
Washington, D.C. Report KAPTD-1364. Environmental Protec-
tion Agency. November 1972. Chapter One; The Automobile and
the Regulation of Its Impact on the Environment. Legislative Drafting
Research Fund of Columbia University. Draft Report. 1974.
Chapter 6; and California Air Resources Board for information
on California's mandatory retrofit plan.
113
-------
Table 29. RETROFIT COSTS FOR CO AND HC WITH
CATALYTIC REACTOR
Model
year
1975
1974
1973
1972
1971
1970
1969
1968
1967
1966
-------
Table 29. RETROFIT COSTS FOR CO AND HC WITH
CATALYTIC REACTOR (Continued)
Model jear
1975
1974
1973
1972
1971
1970
1969
1968
1967
1966
pre-1966
Per mile costc
Cents/mile
0.53
0.60
0.70
0.81
0.94
1 .08
1 .59
1 .81
2.29
2.39
2.86
h
-------
Table 30. RETROFIT COST FOR NOx WITH
EXHAUST RECYCLE DEVICE
Model Year
1975
1974
1973
1972
1971
1970
1969
1968
1967
1966
pre-1966
Emissions
reduction,
gm/mia
Max Win
0.77
1.02
1 .18
2.67
2.70
3.57
4.84
4.25
3.78
3.54
2.47
0.22
0.47
0.63
2. 12
2.15
2.30
4.84
4.25
3.78
3.54
1 .92
Annual costs of retrofit"
Annual i zed
initial
l_ cost
$ 7.45
7.45
8.00
8.50
9.50
10.82
i2.53
15.10
19.40
19.40
19.40
Maintenance
$6.00
6.00
6.00
6.00
6.00
6.00
6.00
6.00
6.00
6.00
6.00
Total annual
cost
$13.45
13.45
14.00
14.50
15.50
16.82
18.53
21 . 10
25.40
25.40
25.40
Pre-1970 listings based on data in: Downing. Benefit/Cost Analysis
of Air Pollution Control Devices JOT Used Cars. University o* Cali-
fornia, Riverside, September 1970. Maximum reductions occur when
levels are achieved equal to the lowest levels occurring in pre-1970
autos (Downing); minimum reductions are to highest levels achieved in
pre-1970 autos (Downing).
Retrofit costs estimates derived from data in Downing, op. ait.
Annual!zing procedures are the same as were used in Tab's 29.
Table continued on following page.
116
-------
Table 30. RETROFIT COST FOR NOx WITH EXHAUST
RECYCLE DEVICE (Continued)
Model year
1975
1974
1973
1972
1971
1970
1969
1968
1967
1966
p re- 1966
Retrofit
cost
per mile,
-------
hydrocarbons and that the costs of the reactor can be allo-
cated equally between the control of CO and HC. An exhaust
recycle device would be necessary for reducing nitrogen oxide.
The last columns in Tables 29 and 30 give the estimated
average cost of retrofitting per gram/mile reduction.
Motorists will tend to benefit from retrofitting when the
HER tax is greater than this figure. For example, an MER
tax which increased auto variable costs per mile by 25 percent
and which would provide only a marginal inducement for most
car owners to retrofit would be divided equally among the three
pollutants as follows: .0145 cents per gm/mi on CO; .1323
cents per gm/mi on HC; .1398 cents per gm/mi on NO . However,
X
under a tax rate which had the same proportions among pollu-
tants but was double the rate, making it 50 percent of auto
variable operating costs, most motorists would benefit from
retrofitting.
It should be noted that these figures are quite approxi-
mate. Moreover, within any model year there will be signifi-
cant variation owing to differing characteristics of various
auto models, variation in the personal discount rate on
initial expenses for retrofitting, variation in subjective
estimates of the remaining years left on an auto, variation
in the miles driven by individual car owners, etc. Obviously,
not every automobile will be retrofitted even if the average
tax reduction is greater than the cost of retrofitting.
These qualifications notwithstanding, it appears that
voluntary retrofitting would become widespread at emission
tax rates on the order of 50 percent of the current average
auto variable costs per mile; this would entail an additional
cost of about $0.03 per mile on average. Tax rates per mile
on auto travel would be decreased in proportion to the extent
of retrofitting, and there would be a feedback effect on
auto VMT's and the auto scrappage rates by age of car.
118
-------
The incremental effects of emissions tax rates on auto travel
would decline as more retrofitting occurred. Also, as retro-
fitting would prevent higher ownership costs due to MER taxes,
the impacts of these on the auto stock size would decline at
higher rates. As an approximation, retrofitting would prob-
ably begin to become an option as taxes were raised to 25
percent of variable costs per mile; the effects? of er^^a
taxes beyond 50 percent of variable costs per mile on travel
and auto ownership would be quite small.
UNCERTAINTY OF THE RESULTS
As in the estimates of policy effects on travel behavior,
we examine the confidence which can be placed in the results
obtained in this section. The discussion is divided into
two major sections corresponding to the MMY tax and the MER
tax.
Effects of an MMY Tax
The estimated impacts of an MMY tax on the auto stock
derive from a model of auto stock adjustments. Some of the
relationships in the model are econometric estimates while
others are assumptions. Each relationship is considered
briefly below.
New Car Sales — The estimates of the effects of annual
taxes on new car sales derive from an assumed supply-demand
structure. The key assumption is that the price elasticity
of demand for new cars is -1. This assumption has been
verified by careful econometric studies, but there have
been estimates of automobile demand which yielded other
measures of the price elasticity. Virtually all estimates
fall between -,5 and -1.5.
Used Car Price — There is less verification for the assump-
tion made here that the elasticity of demand for used cars
is -1. The one major study of the demand for cars (by
Gregory Chow, op, sit.) yields this result.
119
-------
Aggregate Scrappage Rate — The equation used in this report
is based on an ordinary least squares estimate from national
data. Parameter estimates were highly significant; the
coefficient of determination (R2 corrected) was .6346, and
the standard error was 11 percent. The equation was further
adjusted to be applicable to the Los Angeles area where the
scrappage rate due to age alone appears to be less than the
national average.
Scrappage Rate by Age of Auto — The equation for scrappage
by age of auto due to age effects alone was estimated by
Franklyn Walker on national data. All parameter values
were highly significant (1 percent level of confidence
with t-tests) and the coefficient of determination (R2
corrected) was .9994. However, the scrappage rate was
adjusted to be applied to Los Angeles, which probably
increases the error associated with the equation.
Besides these relationships there several key assump-
tions which should be noted:
1975 Price of New Cars — It was assumed that the 1975 price
of new cars would be $3,850 on average. In fact, the price
of the new models would be very difficult to predict due to
the countervailing impacts of inflation and recession.
1975 Price of Used Cars — It was assumed that the 1975
price of used cars would be $1,272 on average. However,
the price of used cars is dependent to some degree on the
price of new cars, which is itself largely indeterminate.
Used Car Price by Age of Car — It was assumed that the price
of used cars by age is inversely proportional to the scrap-
page rate by age. This assumption is somewhat arbitrary, but
not unlikely. The reason older cars are scrapped at higher
rates is that their value is less when repaired and main-
tained. Thus, there is a direct relationship between the
120
-------
value by age of auto and its scrappage rate. Of course, the
relationship need not be one of direct proportionality as
assumed here.
The quality of data also affects the confidence associ-
ated with the results. On the whole, data on the auto stock
characteristics of cars in Los Angeles and Orange Counties are
quite reliable; they are the result of an independent survey
made annually by Polk Associates. These data are only avail-
able through July 1, 1973; the 1975 auto stock characteris-
tics (Tables 21 and 22) are based on model simulations with
1973 data as inputs. The model as a whole was calibrated
on data from Polk statistics over the period July 1, 1969
through July 1, 1973.
Data on recent emission rates and average VMT's per
auto year are reasoned predictions which may be open to
question. Pre-1970 emission rates are the result of actual
samples taken in California.
With two exceptions, the model is applied to values
of variables within the ranges of observations which were
used for estimation. The exceptions are areas for concern
in interpreting the results: first, the current high prices
of new cars and the added ownership costs of an MMY tax may
strain the assumption that the price elasticity of demand
for new cars is -1; second, the aggregate scrappage rates
predicted by the model become less certain as they increase
past 10 percent of the auto stock. Thus, for example, the
estimated effects of the high tax scenario presented in Table
26 should be viewed with less confidence than the estimates
based on low taxes.
In general, the sum of the various sources of error
indicates that there is significant randomness in the esti-
mated results. However, two important results can be
presumed to be reasonably robust:
121
-------
The Effects of Taxes on the Size of the Auto Stock — The
model itself is relatively reliable in predicting auto stock
size because the estimated or assumed relationships predict
flows which are small proportions of the total auto stock
size. That is, new car sales are generally less than 10
percent of the existing auto stock, and aggregate scrappage
is an even smaller proportion. Also, they work in opposite
directions, causing the net change to be less than their
relative proportions would indicate.
Relative Scrappage Rates by Age of Auto — The estimated
relationship for scrappage owing to age alone gives very
good fits to the actual data. Thus, the relative scrappage
rates among age categories can be presumed to be reasonably
accurate even though the predicted net aggregate scrappage
rate is less reliable.
Effects of an MER Tax
As indicated previously, it is difficult to determine
the effects of an MER tax on automobile ownership and VMT's
by age of car. All that can reasonably be said is that the
impact on auto ownership will be less than that of an equi-
valent MMY tax and that there will be some increase in the
proportion of VMT's from newer relative to older cars.
Quantitative estimates of these effects were not attempted,
except to the extent that Table 27 puts an upper bound on
the auto stock effects and Table 28 puts a lower bound on
the VMT effects.
This uncertainty is attributable to the estimates of
costs and effectiveness of retrofit devices. Though the
assumptions are conservative, any single number which pur-
ports to describe the effectiveness of retrofitting for a
model year is subject to considerable variability for the
122
-------
reasons already cited. Nonetheless, it appears reasonable to
conclude that retrofitting would not be widespread at emission
tax rates less than 25 percent of auto variable costs per
mile, but would become prevalent before taxes reached 50
percent of auto variable costs per mile.
123
-------
5. POLICY COST-EFFECTIVENESS
Each policy discussed in the previous chapters imposes
a cost on individual travelers and a cost on society. As
discussed in Chapter 2, these costs will differ because
taxes on individuals are transfer payments rather than
social costs, whereas transit improvements and program
administration impose social costs. These costs are examined
in this chapter. The last section of the chapter considers
the effects of alternative policies on fuel consumption.
INDIVIDUAL COSTS
Individual costs due to taxes on travel (gasoline,
emissions and parking taxes) include both the increased cost
of auto trips from the tax and the opportunity cost associ-
ated with choosing other alternatives. The opportunity cost
includes the value of additional time spent on other modes
and the value of choosing less preferable destinations.
Although the theory of computing policy costs is con-
ceptually the same for all policy scenarios (see Chapter 2
for an explanation), there are differences among the policies
themselves which require that the costs be calculated in a
somewhat different manner. Each type of policy (gas/emis-
sions tax, parking tax, and gas/emissions tax with improved
transit service) is considered in a separate section, and
the cost-effectiveness of each is compared in a final section,
124
-------
Cost to Individualsof Gas/Emissions Tax
With a gasoline or emissions tax, the good being taxed
is vehicle miles traveled. Thus, the cost of this tax to
motorists is that added cost per mile times the miles traveled
after the imposition of the tax, plus the opportunity cost
associated with fewer VMT's. For each vehicle mile traveled
after the tax, the two components of the individual's costs
are calculated as follows:
1) cost per vehicle mile driven = tax per mile
2) opportunity cost per vehicle mile driven =
Jj(tax per mile) [change in VMT's]
[VMT's after tax]
As described in Chapter 2, this is an approximation of
the actual individual cost. In particular, relationship 2)
assumes that the demand curve is linear. In the model used
to compute changes in VMT's, the demand is nonlinear. How-
ever, the potential error of using this approximation is
small compared to the estimated costs. It is a common pro-
cedure for computing consumer surplus.
The aggregate costs for the region are determined by
multiplying each of the above components by the aggregate
VMT's after the tax. Table 31 presents the costs of four
tax levels, the travel effects of which were presented in
Chapter 3. The data on percentage change in VMT's and
aggregate VMT's for the area, both as a result of each tax
level, are from Table 20. The variable cost per mile before
the addition of a gasoline or emissions tax is $0.0571.
Cost to Individuals of Parking Tax
With a parking tax, the good being taxed is the auto
trip {except for serve-passenger trips) rather than vehicle
miles 'traveled. The aggregate costs to individuals in the
125
-------
Table 31. COSTS TO INDIVIDUALS OF
GASOLINE OR EMISSIONS TAX
Tax rate:
percentage
of variable
costs per mile
25%
50
75
100
Tax rate:
percentage
cf variable
costs per mile
25%
50
75
100
Cost per vehicle mil
Added cost of
trip taken,
$/mi
$0.0143
0.0286
0.0428
0.0571
Opportunity cost
of vehicle miles
not traveled,
$/mi
SO. 0005
0.0020
0.0043
0.0069
e
Total ,
$/mi
SO. 0148
0.0306
0.0470
0.0640
Aggregate cost of weekday travel within
Los Angeles and Orange Counties
Added cost of
trips taken,
$ thousands
$ 828.5
1,539.7
2,153.7
2,710.7
Opportunity cost
of vehicle miles
not traveled,
$ thousands
$ 29.0
107.7
216.4
327.6
Total ,
$ thousands
$ 857.5
1 ,647.4
2,370.1
3,038.3
126
-------
region is arrived at by multiplying the added cost per trip
times the auto trips (exclusive of serve-passenger trips)
taken after the imposition of the tax, plus the opportunity
cost of trips foregone by this mode. The two cost components
are calculated as follows:
1) cost per auto trip made = parking tax
2) opportunity cost per trip =
^(parking tax)[change in number of auto trips]
[number of auto trips after tax]
Again, this is an approximation based on the assumption of a
linear demand curve.
Table 32 presents the individual costs associated with
a parking tax of various levels. The effects of such a tax
on travel were presented in Chapter 3. It is estimated that
there were 5.397 million auto trips, other than driver serve
passenger, per weekday within Los Angeles and Orange Counties
in 1974. This estimate is arrived at by multiplying the
share of vehicle trips which are not driver serve passenger
(.985) times the estimated total of household vehicle trips.
It should be remembered that this figure would include neither
travel which crosses into other counties nor truck and bus
travel.
The percentage change in aggregate auto trips exclusive
of driver serve passenger for all trip purposes is given in
Table 33. The percentage changes were calculated the way
changes in VMT's and all auto trips were calculated in Chap-
ter 3.
Costs to Individuals of Gas/Emissions Tax with
Transit Improvements
The net cost of this scenario has four components:
• the increased cost per mile owing to the tax;
• the opportunity cost associated with VMT's foregone;
127
-------
Table 32. COSTS TO INDIVIDUALS OF A PARKING TAX
Parking tax:
parking cost
increase:
$0.25
0.50
0.75
1.00
Cost per trip
Added cost of
trip taken,
$/trip
$0.2500
0.5000
0.7500
1 . 0000
Opportunity
cost of
trips foregone,
$/trip
$0.0130
0.0532
0. 1 184
0.2040
Total,
$/trip
$0.2630
0.5532
0.8684
1.2040
Aggregate cost of weekday travel
within Los Angeles and Orange Counties
Parking tax:
parking cost
increase:
$0.25
0.50
0.75
1.00
Added cost of
trips taken,
$ thousands
$1 ,209.0
2,124.5
2,769.4
3,195.5
Opportunity
cost of
trips foregone,
$ thousands
$ 62.9
226.0
437.2
651.9
Total ,
$ thousands
$1,271 .9
2,350.5
3,206.6
3,843.3
128
-------
Table 33. EFFECTS OF PARKING TAX ON AUTO TRIPS
EXCLUSIVE OF DRIVER SERVE PASSENGER TRIPS
Added parking cost
Percentage change from 1974 base
$0.25
0.50
0.75
1.00
-10.39?
-21.27
-31.58
-40.79
129
-------
• the benefits of improved service to bus patrons who
would have taken transit in the absence of any tax
on auto travel or transit improvements;
• the benefits to people who become bus patrons
solely because of the transit improvements.
Extra computation is also necessary to monetize the benefits
of decreased access and line-haul time associated with
transit.
The individual costs per vehicle mile traveled in this
scenario are the same as those calculated for the case of a
gasoline or emissions tax in the absence of transit system
improvements (Table 31). The aggregate costs to auto drivers
are equal to aggregate VMT's after the imposition of the
policies (from Table 20) times the cost per vehicle mile
traveled, plus the opportunity costs associated with VMT's
foregone.
It was not possible, given summary information on
travel demand, to calculate the benefits of improved transit
service to new and continuing riders. Also, it should be
noted that the demand estimates assume no changes in fares
as a result of the service improvements; thus, resource
costs are passed on to individuals only to the extent that
they are covered by the existing fare structure. Table 34
presents the results of computations based on the costs to
motorists.
Individual Cost-Effectiveness
In order to make a comparison among the policies, the
aggregate dollar costs were divided by the aggregate reduc-
tion in VMT's. The results of these computations are
presented in Table 35. As expected, the parking tax is the
least cost-effective method of reducing VMT's. Also, the
cost-effectiveness of a gas or emissions tax shows a slight
tendency to decrease as the tax is raised.
130
-------
Table 34. COSTS TO INDIVIDUALS OF A GASOLINE OR EMISSION TAX
WITH TRANSIT SYSTEM SERVICE IMPROVEMENTS
Aggregate cost to auto drivers of a
weekday of travel within
Los Angeles and Orange Counties
Transit system
improvements with
a tax on variable
cost per mile of:
50$
100
Additional
costs for
trips taken,
$ thousands
$1,427.9
2,294.8
Opportunity cost
of miles not
traveled,
$ thousands
$ 99.9
277.3
Total ,
$ thousands
$1 ,527.8
2,572. 1
131
-------
Table 35. COST TO INDIVIDUALS PER VMT REDUCED
($/mile)
Gas
variable
or emissions tax:
cost per mile increase
25%
50
75
100
Parking tax:
parking cost increase
Transit
variable
$0.25
0.50
0.75
1.00
system improvements with
cost per mile increase
50$
100
Cost per mile reduction
$0.1852
0. 1886
0. 1935
0.2012
$0.4033
0.3922
0.3921
0.3985
$0.1208
O.I 149
132
-------
RESOURCE COSTS
The resource costs of the policies include opportunity
costs of trips foregone, costs of bus service improvements
and costs of administering tax programs. Opportunity costs
were calculated in the previous section. In this section
bus costs are calculated and administrative costs are dis-
cussed. The section concludes with an evaluation of the
cost-effectiveness of the policies from the standpoint of
resource costs.
Bus Costs
The bus network changes were specified in Chapter 3.
Briefly, they consisted of adding new bus routes to zonal
interchanges where more than 100 auto trips had been diverted
by the policy; for every 300 auto trips diverted a new bus
route was designed. For every 75 auto trips diverted in a
zonal interchange during peak hours, an additional bus
trip (with two loadings) was installed. For off-peak
travel, additional bus trips were instituted for every 50
auto trips diverted and a new route was designed for every
200 auto trips diverted. Peak travel was simulated with
work trips, and off-peak travel was simulated with shopping
trips.
For any policy which induces greater bus ridership,
there is the problem of determining whether additional
capacity will be necessary in order to serve the added pas-
sengers. Thus, though no bus system improvements were
specified in some of the auto tax scenarios, it is likely
that their effectiveness assumed added capacity in the
existing system. Similarly, in the scenario where added bus
service is coupled with an auto travel tax, the combination
of the two policies yields greater additional transit rider-
ship than the initial increase in capacity.
133
-------
The additional capacity which may be necessary to meet
induced demands is not considered in the calculations which
follow. The cost calculations are only based on the assumed
network changes specified above. If there is no excess
capacity in the existing system, then the costs estimated
below may underestimate somewhat the costs associated with
additional bus capacity to serve diverted motorists.
The bus costs were calculated as follows:
• The network changes in the samples of zonal interchanges
were put into terms of additional annual bus hours,
annual bus miles, and buses needed in order to provide
the service (see Appendix C); it is assumed that only
peak travel entails additional buses and, consequently,
off-peak travel uses the additional bus capacity needed
for the peak.
• The annual bus hours, annual bus miles, and buses needed
for peak travel were expanded to give the values for
serving all of Los Angeles and Orange Counties by multi-
plying the sample values times the total number of trips.
The latter is obtained from the sample size in the
following way:
sample size = number of trips sampled
^ total trips
total trips = number of trips sampled
^ sample size
• The annual bus hours and annual bus miles for off-peak
travel were expanded to the values for all of Los
Angeles and Orange Counties in the same way as for peak
hour travel.
• The aggregate annual bus hours, annual bus miles and
buses for peak and off-peak travel in Los Angeles and
Orange Counties were used in the incremental bus cost
function estimated in Appendix C which computes the
134
-------
additional 1974 annual cost of the improved bus service.
(Cost per weekday is determined by dividing annual cost
by the weekday factor of 351. This is a factor used by
the SCRTD.)
The sample size of peak-hour trips is determined by
dividing the number of trips in the CRA sample of work trip
zonal interchanges by the total estimated peak-hour trips
in Los Angeles and Orange Counties for 1974. Estimated in
this way, the peak-hour sample size is 4.55 percent of total
peak-hour trips. The off-peak sample size is similarly cal-
culated as 2.47 percent of total off-peak travel. Peak-hour
travel is assumed to be work and education-related; off-peak
travel includes all other purposes.
Table 36 presents the estimates of bus system changes
and resulting costs for Los Angeles and Orange Counties
resulting from the above procedures.
Emissions Test
As mentioned in Chapter 2, an emissions tax based on
actual annual pollution (an MER tax) requires testing
facilities and an administration to oversee the inspection
and tax collection system. (It can be presumed that the
added administrative expenses of a gasoline tax will be
negligible compared to the other costs in this chapter. A
parking tax would require some administrative and enforce-
ment expense, but we will not attempt to estimate these in
this chapter.) These expenses represent resource costs
which should be estimated in order to infer the cost effec-
tiveness of an MER tax.
Because there is no experience with testing facilities
of the scale necessary to monitor all cars in the Los
Angeles-Orange County area, actual figures on costs are
highly speculative. We will start with the 1971 estimates
for California (given in Chapter 2) of $1.05 per vehicle
135
-------
Table 36. COSTS OF BUS SERVICE IMPROVEMENTS FOR TRAVEL
WITHIN LOS ANGELES AND ORANGE COUNTIES
Annual Changes in System
Transit system improvements
with auto variable cost
per mile increase of:
50$
100
Bus miles
per year,
thousands
30,618
66,995
Bus hours
per year,
thousands
2,205
4,700
Number of
Buses
800
1,774
Costs of Changes
Transit system improvements
with auto variable cost
per mile increase of:
50$
100
Annual ,
$ thousands
$ 63,676
1 14,681
Weekday,
$ thousands
$181 .4
326.7
136
-------
inspection cost and an initial start-up cost of $19,830,000.r
The 1974 cost per vehicle inspection on a smaller scale for
Los Angeles-Orange Counties would be about $2.00 per test.
Start-up and administrative costs, annualized over the life
of the program, can be assumed to add another $2.00 per
vehicle per year in the area where the program is applied.
In sum, if two inspections per year are required, the
total cost of administering the program would be on the
order of $6.00 per vehicle per year. Using the CRA estimate
of 4,166,288 vehicles in Los Angeles-Orange Counties in 1974,
these assumptions lead to an annual program administration
cost of $25 million. Put on comparable weekday terms
(dividing by 351) of the other costs in this chapter, the
cost is $71 thousand.
Resource Cost-Effectiveness
To compare the policies in terms of social costs, the
aggregate resource costs of each were divided by the aggre-
gate reduction in VMT's. Table 37 presents the results of
these calculations. Several important caveats must be kept
in mind in order to interpret these results appropriately.
In particular, the following biases affect the figures
given in Table 37:
• In comparing the emissions tax and gas tax costs, VMT's
are an inappropriate measure of effectiveness; an
emissions tax of equal proportion to a gas tax will
cause significantly lower pollution levels because of
its greater impact on lowering VMT's of older, higher
polluting automobiles (see Chapter 4).
• The parking tax scenario does not include the costs of
administering and enforcing a parking tax program.
• The net costs of transit system improvements do not
include the benefits to ongoing patrons of improved
137
-------
Table 37. RESOURCE COST PER VMT REDUCED
Policy
Gas tax:
Variable cost per mile increase
25%
50
75
100
Emissions tax:
Variable cost per mi le increase
25%
50
75
100
Parking tax:
Parking cost increase
$0.25
0.50
0.75
1 .00
Transit system
improvements with gas tax:
Variable cost per mile increase
50%
100
Transit system improvements
with emissions tax:
Variable cost per mile increase
50%
100
Aggregate resource
cost per weekday,
$ thousands
$ 29.0
107.7
216.4
327 . 5
$100.2
178.9
287.6
398.8
$ 62.9
226.0
437.2
651 .9
$289. 1
654.3
$360.3
725.5
Cost per mile
reduction
$0.0063
0.0123
0.0177
0.0217
$0.0216
0.0205
0.0235
0.0264
$0.0199
0.0377
0.0535
0.0675
$0.0229
0.0292
$0.0285
0.0324
138
-------
service, nor do they include costs of additional
capacity which may be necessary to serve induced demand.
In addition to the above qualifications, there are
uncertainties in the estimates owing to the predictions of
VMT reductions (see Chapter 3), the approximate nature of
the opportunity cost estimates, and the lack of validation
of assumed costs for administering an emissions inspection
program.
Given these sources of error in the cost-effectiveness
computations, it is somewhat difficult to draw definitive
conclusions comparing the policies. Nonetheless, it appears
that a parking tax is the least preferable policy on grounds
of cost-effectiveness. More information is necessary to
distinguish among the other policies (see Chapter 6), but
it should be noted that the cost of bus service improvements
is not inconsiderable — it adds probably about 50 percent
to the cost per VMT reduced when conjoined with other
policies.
FUEL CONSUMPTION EFFECTS
This section computes some of the impacts on fuel con-
sumption which can be attributed to the policies. Typically,
disincentives to automobile travel can be expected to
decrease fuel consumption. Table 38, using U.S average
miles per gallon figures, shows how much gasoline can be
saved in Los Angeles and Orange Counties as a result of the
various tax policies. The table also presents the increased
fuel consumption due to the assumed changes in the bus net-
work.
The effect of retrofitting on fuel consumption has two
impacts which cause the estimates in Table 38 of fuel savings
due to emissions taxes to be misleading:
139
-------
Table 38. FUEL CONSUMPTION CHANGES AS A RESULT OF THE POLICIES:
LOS ANGELES AND ORANGE COUNTIES
Policy
Gas or emissions tax:
Increase in auto vari-
ab 1 e costs per mile
25?
50
75
100
Parking tax:
1 ncrease i n parki ng
costs
$0.25
0.50
0.75
1 .00
Transit system
improvements with
increase in auto vari-
able costs per mile:
50?
100
Passenger car
gasoline consumption
decline weekdays,
103 gallons/day3
338.92
639.40
896.77
1 105. 17
230.87
438.76
598.63
706.75
925.61
1638.29
Percent
change
in VMT's
- 7.40?
-13.96
-19.58
-24. 13
- 5.04?
- 9.58
-13.07
-15.43
-20.21?
-35.77
Bus diesel fuel
consumption
increase weekdays,
103 gallons/dayb
13.43
29.39
Miles oer gallon for passenger cars equals 13.67, from Statistical
Abstract of the United States. Washington, D.C., Government Printing
Office, 1973.
Miles per gallon for buses equals 6.49, from U.S. Department of Trans-
portation. Characteristics of Urban Transportation Systems. Washington,
D.C., Government Printing Office, March 1974.
140
-------
If retrofitting occurs on a massive scale (say, after
a 50 percent tax on auto costs per variable mile) then
the VMT reduction estimated by the travel demand model
will be seriously overstated.
Retrofitting with catalytic reactors (for CO and HC)
causes a decrease in the fuel economy of cars on the
order of 3 to 4 percent.
141
-------
List of References, Chapter 5
1 Mandatory Vehicle Inspection and Maintenance, Part A — A Feasibility
Study, Volume I: Summary. Northrup Corporation, in association
with Olson Laboratories. Prepared for State of California
Air Resources Board. June 1971.
142
-------
6. CONCLUSIONS
This chapter reviews the major conclusions from the
preceding chapters, discusses other factors in assessing
the pollution control strategies, and recommends fruitful
areas for future research.
The results of Chapters 3, 4 and 5 suggest the following
conclusions:
• In the short run a gasoline or MER tax is more cost-
effective in reducing VMT's than a parking tax.
• Taxes based on emission rates of cars (HER and MMY)
will have a large impact on the age distribution of
cars and some impact on the distribution of VMT's by
age of car.
• Emission taxes between $0.015 and $0.03 per mile will
induce retrofitting.
• Improving transit lowers the individual's travel cost
for pollution control but increases the cost to society.
• For individuals, the cost per VMT reduced of all stra-
tegies except parking taxes varies from $.10 to $.20
per mile; for parking taxes, the costs are around
$.40 per mile, depending on the size of the tax.
• The resource costs per VMT reduced of all strategies
except parking taxes varies from $.015 to $.035 per
mile; for parking taxes, the costs are around $.05
per mile.
We next turn to a consideration of other issues which
are important in determining optimal pollution control
strategies.
-------
NONQUALIFIED FACTORS
We stress that the results of the foregoing analysis of
policy effects and costs should be placed in the context of
certain factors not quantified in this report to give a
balanced evaluation of the alternative strategies.
These factors are the incidence of taxes across income
groups, the feasibility and phasing of policies, and the
land use impacts of taxes on auto travel. Most of these
issues require more research in order to evaluate completely
the proposed pollution control strategies. However, the
qualitative discussion presented below will help clarify the
issues and show the interrelationships among the pollution
control strategies and the diverse impacts of transportation
system changes.
Incidence of Costs
Most tax schemes either explicitly or implicitly affect
income groups differently; the tax proposals reported in
this study are no exception. As a general rule, excise
taxes on nonluxury consumer goods are regressive -- the poor
pay a higher percentage of their income than the rich. This
is particularly true when the good bearing the tax has a
relatively inelastic price demand, as is the case with auto
travel. In the short run, VMT's can be reduced by indi-
viduals, but on average this does not recoup their increased
total cost of auto travel.
Though low income groups can, and often do, use their
cars less, their total cost of auto travel is greater in
proportion to their income than is the case for high income
groups. Thus, it can be expected that gas, emission and
parking taxes will consume a higher proportion of income for
low income households.
144
-------
Taxes which increase with respect to the age of cars,
such as emissions rate based taxes, would also tend to be
regressive. Higher income families owning more than one
car will be able to use the newer, lower-taxed cars more
intensively. Low income families will, on average, not
have as much freedom of choice because they will not be
multicar households and will probably tend to own older
models.
To ameliorate the regressive aspects of emissions
control taxes, policies can be introduced which either redis-
tribute income through other taxes or give more options to
families in their travel decisions. Generally, the former
approach is preferable, if it is feasible. However, the
ability of local governments to impose income redistribution
tax schemes is severely limited. An example of another
approach is to improve transit service, perhaps using the
ample revenues from emissions control taxes.
Feasibility and Phasing of Policies
There is some question as to whether certain policies
are feasible and, when more than one policy is considered
in the overall strategy, whether there should be a lag
between implementation of various policies. As noted in
Chapter 2, there may be considerable technical, enforcement
and administrative obstacles to certain strategies. Also,
because the effectiveness of the policies in reducing
emissions tends to decline as new cars meet stricter stan-
dards, there is a need for the policy to be quickly imple-
mented and then dismantled, perhaps gradually, in order to
capture short-run benefits without incurring long-run costs.
As noted in Chapter 2, parking taxes should cover all
trip purposes and land use categories, except residential,
in order to be effective. Otherwise, enforcement is imprac-
tical and trips can be made without incurring the tax
penalty, thereby leading to less (if any) reduction in VMT's.
145
-------
Yet such a broadly based tax would be difficult to administer
because of the dispersion and number of parking facilities,
including the potential for onstreet parking.
Also as noted in Chapter 2, there is some dispute as to
whether an emissions tax based on inspections of vehicles is
technically feasible. Current experience with safety inspec-
tions suggests that the scale of the system is not a problem.
The major issue seems to be the existence of relatively low-
cost, simple and reliable emissions tests. Some sources
claim that such tests now exist, but there is little experi-
ence with them in terms of the routine processing of 4 million
or more vehicles a year -- the current size of the Los Angeles
fleet.
Since improvements in the bus system may be a necessary
policy step before the imposition of taxes or other con-
straints on automobile travel, some attention must be given
to the delays entailed in adding to bus system capacity.
Owing in part to federal funding procedures, it now takes
about a year to obtain conventional buses. A large order
could cause considerably more delays. The additions to the
fleet contemplated in this report could not, in all prob-
ability, be made in less than 18 months. Some improvements
in bus service can begin earlier with certain steps, such
as mobilizing used bases, as the Southern California Rapid
Transit District did recently. Nonetheless, potential
bottlenecks in improved transit service would require inter-
facing among various agencies in order to implement policies
in an orderly fashion.
Finally, it is important to realize that the emissions
rate from the auto stock decreases over time, and in the
relatively near future (such as the early 1980's), it may
reach acceptable levels. If this is indeed the case, the
policies which might entail low annual costs but high initial
costs should be avoided. In particular/ transit system
146
-------
improvements with high initial capital expense and with
long recovery periods (such as rapid rail) are not cost-
effective in this scenario, even though they may bring con-
siderable auto trip diversion. Certain transit improvements
may not even be implementable in the time period which is
required for pollution abatement benefits to be gained.
This is not to say that these transit system changes should
not be made, but rather that in their evaluation their
pollution control benefits should be discounted heavily.
A similar problem arises with tax policies not directly
related to the rate of emissions from automobiles. There
would be a strong tendency to keep a gasoline or parking
tax for general public revenue purposes even after the pol-
lution control benefits of the tax had declined.
Land Use Effects
Increasing costs of travel, if permanent, may have sig-
nificant long-run effects on urban form. Unfortunately,
there is only limited understanding of the interrelationships
between transportation system changes and general urban
land use impacts; the small amount of theoretical and
empirical literature on the subject is oriented toward cities
with strong cores, unlike Los Angeles. Nonetheless, some
inferences about location decisions can be made based on the
assumption that households and businesses will tend to
reduce superfluous travel owing to increased costs of auto
trips.
In Los Angeles, work trips are typically longer than
trips for other purposes; for example, work trips average
about 14 miles, round trip, whereas shopping trips are
between four and five miles, round trip, on average. Taxes
on miles traveled, such as either a gas or emissions tax,
would tend to increase the cost of a work trip relative to
a shopping trip. This realignment in the relative costs of
147
-------
travel would cause households and employers to tend to move
closer together. Because households have more flexibility
in location, the tendency would be for higher density resi-
dential development located closer to the many employment
centers in the area. Such a migration would accentuate the
current trend in Los Angeles toward multifamily dwellings.
Retail centers tend to be as dispersed as households;
it is apparent from the data in the 1967 LARTS Household
Trip Survey that most families shop within a five-mile radius
of /their homes. This suggests that there would be some
relocation of retail land use if household dwelling patterns
change. Because the present land use system in Los Angeles
appears to be most beneficial for shopping trips — these
trips tend to be shorter, hence less expensive, than trips
for other purposes — the effects of the taxes on retail
location in the absence of any tax-induced residential
changes would be slight.
Finally, it should be noted that land use changes will
be lessened if households feel that the pollution controls
on travel are of relatively short duration. For example,
an emissions tax may increase the cost of auto travel for
most families for only a few years; after that, the new
and cleaner autos available will lower the cost of travel,
thereby causing travel patterns to return to those which
currently obtain. The locational advantage of dwellings
closer to employment centers will consequently decline.
SHORT-TERM RESEARCH RECOMMENDATIONS
In addition to the issues discussed above, there are
other areas of study which are potentially important in
equipping policy makers with the information necessary
to develop motor vehicle pollution control policies.
148
-------
These studies involve the evaluation of other strategies to
control mobile source emissions. We first examine the sub-
jects and then briefly describe issues related to research
design.
Additional Strategies
An important area of research involves the evaluation of
transportation other than the private automobile. As
mentioned above, these additional systems should involve
low initial capital outlays and be readily available for
implementation.
An integrated feeder and line-haul bus service rhere
the feeder portion is paratransit may have these character-
istics. The initial capital cost of a paratransit vehicle
is about half that of a conventional bus (the service life
is also about half that of conventional buses? see Appendix C)
Paratransit tends to be much more costly per passenger mile
than a conventional bus largely because of the lower pas-
senger-to-driver ratio. However, the service is also of
higher quality, and some of the increased cost can be
recovered by charging higher fares for the extra convenience.
Little is known about the characteristics of an inte-
grated feeder and line-haul bus system on the scale necessary
to serve and connect all the more densely populated portions
of Los Angeles and Orange Counties. Most existing systems
are on a demonstration project basis, and thus users view
them as a risky future alternative; consequently, the
service has had little impact on automobile ownership and
other factors which would increase patronage over time.
The line-haul portion of the trip for such a service
involves express buses. Such buses may have their own
facility (as does the El Monte express run by the SCRTD) and
their own right of way on existing freeways, or they may
simply compete with autos for freeway space. The first
149
-------
alternative may improve the commuting time of bus passengers
because the buses would not be slowed on portions of freeways
which are currently congested at peak periods.
An advantage of bus rapid transit is that it requires
relatively low initial funding unless stations and separate
graded right-of-ways are deemed necessary. An express bus
costs less than a conventional bus, though it has a shorter
service life (see Appendix C). It also has the benefit of
substituting for relatively long auto trips, which would
incur the greatest penalty from the imposition of a gas or
emissions tax, and offer the greatest potential for reducing
VftT's. The express bus would do little for shopping trips
or other trips of intermediate to short range distance.
Auto travel controls are another area of potentially
useful research. These are measures which would encourage
carpools and/or reduced auto travel. There is an intuitive
appeal in avoiding severe constraints, such as gas rationing
without white markets, because such controls in other markets
have tended to result in severe resource misallocation. The
effect of national gasoline rationing and regulations in the
winter of 1974 is a case in point. Yet it must be admitted
that there is very little research on the costs and effects
of such controls.
One travel control now in use in Los Angeles is freeway
monitoring at certain entry points. To the extent that such
a control is used to restrict access to freeways, it raises
the costs of auto travel and may reduce VMT's. However,
there are potentially counter-productive effects if auto
drivers choose to use unlimited access, more circuitous
routes in order to attain their destinations.
Policies overtly designed to increase carpooling have
not yet received thorough analysis. It appears from pre-
vious experience that simply attempting to market carpooling
through public awareness campaigns and passenger matching
150
-------
programs has little impact on overall auto occupancy rates.
Merely increasing the costs of auto travel through gas or
emissions taxes may provide the most effective incentive to
carpool.
Research Design
The conceptual approach for evaluating the above strate-
gies is similar to the one used in this report: travel
demand and system performance are equilibrated using assumed
effects of a policy on system performance variables (see
Chapter 2). Under the presumption that an appropriate
travel demand model exists, the important gap in our know-
ledge is the system performance characteristics of para-
transit, express bus, and travel controls, such as freeway
monitoring.
Demand and system performance models for policy evalu-
ation of these strategies have data requirements similar to
those utilized in this report. It is also good research
strategy to have information which validates the model
predictions.
151
-------
Appendix A. TRAVEL DEMAND MODEL
This appendix explains the travel demand model used in
the evaluation of policies in Chapters 3, 4 and 5. The
actual estimates of individual choice probability functions
are first presented. An approach is then developed to apply
these probability functions to zonal aggregates of travel
behavior and system performance; the resulting model predicts
interzonal travel patterns under assumed policy scenarios.
This model is tested for consistency against 1967 Los Angeles
Regional Transportation Study data to determine the level of
certainty which can be placed on the results of its appli-
cation.
As explained below, the estimates of individual choice
behavior need to be adjusted in order to apply the model to
data of the zonal interchange level of observation. This
is commonly called the aggregation problem — namely, that
variation of attributes among individual travelers in a zonal
interchange biases forecasts made with zonal data. The
approach taken to reduce this bias involves making a Taylor's
series expansion about the mean of the zonal values of inde-
pendent variables (system performance and socioeconomic
variables) in order to approximate the expected frequency
of travel by mode and destination for each zonal interchange.
The Taylor's series approximation requires estimation of
variances and covariances among the independent variables in
the model. These estimates are made using assumed
152
-------
probability distribution functions for system performance
variables and census data for socioeconomic variables.
The adjusted model is then tested for consistency
against 1967 LARTS Household Trip Survey data. Several new
modes are added to the original model to account for car-
pools, driver serve passenger trips and walking trips. The
model is used to predict round trips by mode for work and
shopping. This procedure yields accurate forecasts of auto
trips and vehicle miles traveled.
THE CRA DISAGGREGATED DEMAND MODEL
The estimated relationships in CRA's disaggregated
demand model are presented in Table 39. The numerical terms
and coefficients in the equations were estimated from data on
individual trips made in Pittsburgh in 1967. The model is
behavioral in the sense that the estimated parameters should
apply to individuals regardless of location. It is policy
sensitive in that demand for travel is a function of system
*
performance variables.
Each estimated relationship gives the ratio of proba-
bilities (or the odds) between choosing two alternatives.
This ratio is determined by a comparison of the attributes
of the alternatives. In the case of mode choice (Equation
13 for work and 14 for shopping), the relative level of
service associated with each mode (auto or transit) deter-
mines the odds. For choice of shopping destination
(Equation 15), a comparison is made between the costs of
travel to the destinations and between the shopping oppor-
tunities available at each destination. The odds of taking
one trip versus no trip (Equation 16) depend on a comparison
*0ther than the cursory comments contained in this section,
we will not describe the estimation of the model or many of
its properties. For a complete analysis, see Domencich,
Thomas A., and Daniel McFadden. Urban Travel Demand: A Behavtoral
Approach. Amsterdam, North-Holland, 1975. 215 p.
153
-------
Table 39. ESTIMATED RELATIONSHIPS IN
CRA DISAGGREGATED DEMAND MODEL
Work Modal Sp I i t
— 4<77 -
Shoppi ng Mode I Sp I i t
P(A-i i )
In v,'--i = - 6.77 - 4.1KC... - CD. .) - . 0654(T. . . - T_ . .)
,j) A^J B^J A^J B^J
-.374(5,. . - SB. .) + 2.24? ( 14)
Shopping Destination
ln jrj =-2- 06(xij - xik} + • 844 (EJ ~Ek} ( ' 5)
Shopping Frequency
(16)
Identities
. + .0654TA.. +.
..
J
(17)
(18)
E. = I P(j:i)E. (19)
Table continued on following page.
154
-------
Table 39 (continued). ESTIMATED RELATIONSHIPS IN
CRA DISAGGREGATED DEMAND MODEL
Variable Definitions
P(A:ijj) = probability that an individual will choose auto for a given
round trip between origin i and destination j for a given
purpose.
P(B:i3j) - probability that an individual will choose transit for a
given round trip between origin •£ and destination j for
a given purpose.
C . . . = variable costs per mile for a round trip by auto between
" i and j.
Cn . . = total fares for a round trip by transit between i and j.
"
T '. . = total travel time, excluding walk time, for a round trip by
auto between i and j.
T- . . = total line-haul and wait time for a round trip by transit
^ between i and j.
S . . . = total access (walk) time to auto for a round trip by auto
^ between i and j.
S_ . . = total access (walk) time to transit for a round trip by
" transit between i and j.
Y = number of autos per trip maker.
P(j:i) - probability that an individual will choose destination j
for a round trip by any mode from origin i. for shopping
purposes.
X,. = generalized cost of travel between i and j for an individual
making a round trip for shopping purposes.
P(0:i) = probability that an individual at i will make no round trip
for shopping purposes in a given 24 hour period.
P(l:i) - probability that an individual at i will make a round trip
to any destination for shopping purposes in a given 24 hour
period.
Table continued on following page.
155
-------
Table 39 (continued). ESTIMATED RELATIONSHIPS IN
CRA DISAGGREGATED DEMAND MODEL
Variable Definitions (Continued)
E. = proportion of retai I employment in zone j relative to total
retail employment in the region.
X. = generalized cost of travel to all destinations from i for an
individual making a round trip for shopping purposes.
E. = generalized availability of shopping opportunities for an
individuaI i n zone i.
156
-------
between the generalized costs and opportunities associated
with the trip as opposed to the zero costs and opportunities
which occur when no trip is made.
The probability of an individual choosing any given
alternative can be calculated from the following formula:
P(a) = - 1 - (20)
n
1 + le
b=l
where a is one among n alternatives and Ya~D is the function
comparing costs and attributes of alternatives a. and b.
One may view Y , as the right hand side of estimated
Equations 13 through 16, depending on whether the choice set
is mode, destination or frequency.
The model as estimated has three major limitations in
application to the policy options addressed in this report:
1. The mode split equation was estimated for only two
alternatives — auto driver and transit.
2. There are demand relationships for only two (round)
trip purposes — home-to-work and home-to-shop.
3. The model cannot be directly used to predict zonal
aggregate behavior unless data on individuals are
available .
The second and third problems are considered in later
sections of this appendix. The first problem is discussed
below.
Since mode choices are restricted to the two alter-
natives represented in Equations 13 and 14, the model will
not accurately predict the range of modes resulting from a
policy which significantly alters system performance. This
is especially pertinent in the case of work trips/ where
destination and frequency are assumed to remain unchanged
157
-------
in the short run. It is reasonable to suspect, for example,
that an increase in the variable costs per mile of an auto
trip will cause more walking trips and carpooling besides
diverting motorists to transit. Also, a parking tax would
probably cause more driver serve passenger (chauffeured)
trips.
To consider these new modes, it is necessary to construct
new comparison functions, Y ,. These functions are formed
by attributing to each new mode a variable cost per mile, a
time spent in vehicle and waiting, and a walk or access
time for round trips between each i>j zone pair. Each of
these trip system performance variables could then be substi-
tuted for their transit counterparts in Equations 13 and 14
to derive the odds between choosing auto drive alone and a
given new mode. Using Equation 20 one could derive the
probability of choosing a given alternative over all other
modes -- auto drive alone, transit, carpool, serve passenger
and walking. The actual calculations and assumptions used
in specifying the attributes of new modes are presented in
a later section of this appendix.
APPLICATION OF THE MODEL TO ZONAL AGGREGATES
The disaggregated demand model has been estimated on
individual household data and is most appropriate when used
to predict the individual probabilities of a household's
choice among alternatives. However, most existing trans-
portation data sets only maintain data at the aggregate or
zonal level. Even if household data existed, the attributes
of all alternatives are generally not specified at the house-
hold level. If one wishes to develop a technique for
analyzing the effects of pollution control strategies which
can be generalized over urban areas, it is advisable to
derive behavioral equations capable of being applied to
zonal aggregations of data.
158
-------
Approximation of Zonal Frequencies
Consider an individual t choosing among n alternatives
(mode, destination, etc.). The multivariate, multinomial
logit model estimate of t's probability of choosing alter-
native a (using notation similar to that presented in Equa-
tion 20) is the following:
(21)
t
n Y
1 + I e~ a
b=l
where Y , is the "comparison" function between choices a and
b. We will return to this function later. For our purposes,
it is necessary to have an estimate of the total number of
choices of a, such as auto trips per 24-hour period N in a
zone of T individuals:
T
/V = I P+(a) (22)
a t=l t
when the only information we have about the argument of the
y's is their means for the zone and, perhaps, the variances
and covariances of the terms in Y. There is no analytical
form which translates this information into an estimate of
V
Qualitatively, the problem is that the zonal means of,
say, system attributes cannot simply be used in Equations
13 and 14 in order to compute the average, or expected,
probabilities among mode choices for individuals in a given
zone. Such an approach would tell us the choice an indi-
vidual would make if he were confronted with that system
performance for alternative modes which is the zonal average.
However, individuals in a given zone face a great many
different levels of system performance, and their choices
159
-------
vary accordingly. When there is such intrazonal variation,
it is unlikely that the expected value of an individual's
choice will equal the choice made when confronted with the
average system performance throughout the system.
To take a pertinent example, consider the access time
to transit for a given origin zone. If bus routes through
a zone are relatively few, then the average distance of
households from a bus stop may be quite large, especially
in the zones in the LARTS region which are typically twenty-
five square miles. Average access time may easily be on the
order of thirty minutes each way. An individual who faces
a thirty minute walk to the bus has virtually zero probability
of taking transit. Thus the predicted mode split for such a
zone would be negligible. Yet there are a number of house-
holds who live relatively close to bus stops, say within a
ten minute walk, and for whom transit may consequently be
an attractive alternative. Thus the observed mode split for
transit, though small, would be greater than the predicted
mode split, unless the model were adjusted.
In order to develop an approximation to N , first note
that:
N =
a
(23)
where the E operator stands for the expected value of the
term in brackets. We use the following Taylor's series: x
1 * I
b=l
n
+ I
I o-l
o=l
R
n
(24)
160
-------
where Y is a vector of the attributes of all modes con-
fronting an individual and P(a\Y) is the probability of
choosing a. A bar over a variable indicates the mean; all
derivatives are calculated at the mean of the Y , 's. /?
ab n
stands for the remaining terms of the series.
The second order partials will thus be functions of
the probabilities evaluated at Y:
0=1
= P(a\Y)\2P(b\Y)z-P(b\Y)\ (25)
By taking the expected value of Equation 24 and trun-
g R (which will be considered
operational expression is derived:
eating R (which will be considered later), the following
_ r n - - i
ElP(a\Y)l = P(a\Y)'\2 + I var [ Y fc] • P (b \ Y) • (P (b \ Y) -%J (26)
Estimation of Variance-Covariance Terms
Consider the comparison function, Y
, .
There are
three distinct functional types corresponding to mode
choice, destination choice, and frequency of trips. Of
initial interest are the mode choice probabilities; for
these the functional form of Y is:
(27)
161
-------
where C. = operating cost of the trip;
T. - waiting and line-haul time of the trip;
5. = walk time for the trip;
0 = availability of an automobile;
a.,0 = estimated constants;
a,b = mode indices.
For this equation there are 28 possible variance-co-
variance terms. Fortunately, about half can be presumed to
be zero because of stochastic independence or constancy
over a zone. Of the others, one may assume that many are
proportional to, or simple functions of, the variance of
the distance traveled in a zone interchange.
One cannot be very precise in measuring the variance
in distance in zonal interchanges with the available infor-
mation. The approach here is to assume that distances (or
origin and destination points) are distributed over the
area of a zone pair according to a well defined probability
density function. This approach ultimately allows estimates
of the variance of distance as a function of the areas of
the two zones in a given zonal interchange.
Density Function for Trip Distances -- There are a large
number of potential probability density functions which one
can assume for this problem. Several of these — including
exponential distributions and the rectangular distribution
— were investigated in some detail, and the probability
density function described below appeared to give the best
results.
In deriving the appropriate density function, it was
presumed that, for a given zone pair, trips are distributed
over a range which reflects both the distance between the
zone centroids (geographic centers) and the sizes of the
zones. In symbols:
162
-------
D £
Y.+Y. Y.+Y.\
D' T^-2- , D' + _! i (28)
where D = a stochastic variable representing distance
between zone i and zone j for person trips;
D' = the distance between the geographic centers of
zones i and j;
Y.,Y. = some measure of the size of zones i and j (to be
i j
considered in more detail later.)
We now introduce another stochastic variable, X, which
takes on values in the range from 0 to Y. + Y.. The distance
i- j
for any trip can then be represented by the sum of two vari-
ables; one represents the distance between zones and is
nonstochastic, and the other distributes trips within zones
and is stochastic:
f Y. + Y.}
D = \D' V^-l + X (29)
Another way of viewing the above relationship is to consider
that trips must travel a minimum distance (the term in
parentheses) but that the rest of the distance varies
randomly between zero and Y.+Y..
"i- 0
The next problem is to determine the variance of D.
Note that:
uar [D] = E[D2] - (E[D]) 2
= E[X2} - (E(D})2 (30)
The distribution function which has been assumed for X is:
2(Y.+YJ ~ (Y.+Y J2
^ J ^ J
for 0 < X < Y.+Y.
i~ 0
163
-------
Though this formula is rather uncommon, it seems well
suited for distributing trip lengths. Its graph is depicted
in Figure 3.
The premise of the density function is that the distri-
bution of trips can be approximated by a linear declining
over the relatively short range bounded by Y . + Y .. The first
i j
two moments about zero of the distribution are:
s[y .+Y .}
E(x) = l i J} (32)
J. 6
(Y .+Y .}2
E(X2) = —£—£ (33)
From Equation 31 the variance of distance can be
calculated from the above moments:
11 (Y .+Y .}2
var(D] = var(X] = * (34!
The measures of zone size over which trips are
distributed should reflect the "length" of the zone. An
intuitive measure of length, appropriate to square-shaped
zones, is the square root of area. This leads to the
following formula for variance of distance:
var(D] - - 144 = - - j - -, its (35)
In the range of zone sizes generally found in Los
Angeles and Orange Counties — 12 to 40 square miles — the
range of variance within zones is about 4 to 12 square miles ,
164
-------
Figure 3
The distribution function for distance of trips between zones
f(X)
2 (Y.+ Y.)
i i
2 (Y. + Y.)
165
-------
For intrazonal trips, the above equation must be modi-
fied to account for the stochastic part of the range being
only half, on average, that of interzonal trips. This leads
to the following relationship:
11A .
var[D] = - J for intrazonal trips in zone j. (36)
In the range of zone sizes normally encountered, the variance
ranges from 1 to 3 square miles for intrazonal trips; the
expected trip length implied by making a similar correction
in Equation 32 is 1.44 to 2.64 miles, which is consistent
with the data.
Constraints on Variance-Covariance Terms — We now turn to
the remaining terms in the Taylor expansion (Equation 24).
These approach zero as n increases. The third order deriva-
tive of P(a\Y) reaches a maximum at P=0.5 (as do the cross
derivatives) with a value equal to -0.125. However, for
most mode splits in the Los Angeles region (those between
0.3 and 0 or between 0.7 and 1), the value of the third
order derivative will be between -0.05 and -0.04. The
central moments of the assumed distribution of distance
tend to increase (in absolute value), but at a rate which
is less than exponential. It is clear that the nth moment
divided by n factorial approaches zero. Thus we conclude
that the Taylor series can be truncated after the third
term with a fairly insignificant loss in accuracy.
This truncation raises the possibility that for values
of Y which are rather far from Y, expression 26 will not be
well-behaved in the sense of providing a measure of prob-
ability which increases monotonically with P(a\Y). Thus,
terms should be constrained such that the following three
conditions hold:
166
-------
I E(P(a\Y)} = 1 (37)
a=l
0 <. E[P(a\y)} for all a (38)
3E[P(a\J[)} ^ Q for all P(a|y; £ [0j2j (39)
dP(a\Y)
Conditions 37, 38 and 39 ensure that E[P(a\Y)] is a
probability measure. However, much stronger constraints
will be needed if E[P(a\Y)] is to have properties which are
plausible in terms of mode split or other individual choice
behavior. One such condition is that the elasticity be
greatest at E [P(a\Y)]=0.5. This will occur if the following
conditions are also met:
= 0 at E[P(a\y)} = 0.5 (40)
37^
\P(a\Y)
< 0 at E[P(a\y)} =0.5 (41)
Of the five above conditions/ 37 and 40 hold for all
values of the variances. The following constraints on the
variances are sufficient to ensure that the other conditions
are met:
n
I var[Y ,} < 16 (42)
b= 1 ab
var[Yab] < 1 for all b^a (43)
167
-------
The latter constraint also appears to be necessary for con-
dition 41 to be met. (This proposition was proved only for
the binary choice case. It was presumed to be true when the
number of nonzero probability alternatives was greater than
two.)
Estimated Formulas for Variance-Covariance Terms — Tables
40 through 43 present the expressions and assumptions
used in calculating the variances for mode split equations.
Only pairwise comparisons between auto and other modes are
presented because these probability function estimates are
sufficient for computing any mode choice probability (see
the following section for actual probability calculations).
However, with this approach the probability choice functions
differ somewhat from that presented in Equation 21.
If the subscript a denotes auto and the subscript o
denotes some other given mode, then individual t's prob-
ability of choosing the cth mode can be computed as follows:
P~ ac
Pt(c) = ~^L (44)
1 + le
b=l
where only pairwise comparisons between auto and other modes
appear in the arguments.
Computing probability choices this way entails a some-
what different Taylor's series approximation for mode c
than for auto choices. To see this, note that the second
order derivatives of P,(c) with respect to comparison
t»
functions are as follows:
= P(c\Y)'(P(b\X)'(2P(b\Y)-2)) for b*c (45)
168
-------
Table 40. NONZERO VARIANCE-COVARIANCE TERMS
FOR BUS (2?) AND AUTO (a) CHOICE*
b} = OL2(.05)2var[D]
a.2var[C ] = a.2(. 03) 2var [D]
Cat/
b] = a.zT(4)2var(D}
B2var[0] = &2(.086)
,yC ] = -2oi2(.05)(.03)var[D]
DO. 0
f. 05) (3)var> [D]
f. 03) (S)var [D]
*5ee notes on assumptions Cpp . 173-174).
169
-------
Table 41. NONZERO VARIANCE-COVARIANCE TERMS
FOR WALK (u) AND AUTO (a) CHOICE*
a*var[C ] = cdf. 03)zvar [D]
L> a o
[D]
a.2c(17.14)2var[D}
o
= B2(.086)
,S ] - -2ctJ*c(.03)(18.99)var[D]
CL W L b
•2a-,a.,oov[T ,5 ] = -2a~QLc.(3) (18. 99)var [D]
i o a. W 1 b
*For intrazonal trips only; also, see notes on assurr.pt ions, pp. 173-174,
170
-------
Table 42. NONZERO VARIANCE-COVARIANCE TERMS
FOR CAR POOL WITH h PASSENGERS (oh)
AND AUTO (a) CHOICE*
-T , } = *2(h2(3)2var(D]
*See notes on assumptions, pp. 173-174.
171
-------
Table 43. NONZERO VARIANCE-COVARIANCE TERMS
FOR DRIVER SERVE PASSENGER (s)
AND AUTO (a) CHOICE*
o.2var(C } = a.2(.05?l)2var[D]
L> O. Ls
] = a2(2)2 (. 0571)2var [D]
CSC
var(D]
] = a2(3)2(3)2var(D]
J. S J.
. 0571) (3)var [D]
f, 0571) (3)var [D]
2a.ra.^ov[C ,T ] = 4v.ra.J. 0572) (Z)var [D]
Lr J. S d Lf J.
= Q2(.086)
*See notes on assumptions, pp. 173-174.
172
-------
Notes on Assumptions for Tables 40-43
a. Bus fares are $0.05 per mile in 1967. (Los Angeles currently
has a flat fare system which will require putting to zero all variance
covariance terms involving C, when the model is applied to current
cond i tions .
b. Auto operating costs are $0.03 por mile in 1967.
c. Bus speeds on city streets are 15 miles per hour, implying a
rate of 4 minutes per mi le.
d. Auto speeds average 20 miles per hour, implying a rate of 3
mi nutes per mile.
e. The variability of wait time and schedule delay time within
a given zone interchange is negligible; (this assumption is verified
by the data) .
f. The distribution of access time to buses is uniform within a
zone — this implies a rectangular distribution; thus, the variance is
one-third the square of the mean:
g. The variance on auto availability is calculated as the variance
of households owning no cars — the average within-zone variance across
the zones in Los Angeles and Orange Counties is tightly clustered around
a mean of 0.086 (calculated from 1970 Census of Households data).
h. The speed of walking trips is assumed to be 3.16 miles per
hour, or a rate of 18.99 minutes per mile.
i. The variance of costs for carpooling compared to auto with a
driver only is zero. To explain, further, refer to Equation 29. The
expected costs of auto with driver only are:
ElC ] = (.03)0 = (.03)\(D' - % "; + E[X]
CL I "
whereas the cost per person of a carpool (equally shared) with h pas-
sengers would be: ^73
-------
-03
where it is assumed that the extra intrazonal travel incurred by a
randomly selected carpool is proportional to the number of vehicle
occupants. The above implies that:
var[C -C , ] = 0
a en
j. It is assumed that the extra time incurred in carpool ing is
directly proportional to the intrazonal travel time with the constant
of proportionality equal to the number of vehicle occupants. That is,
= (3)2var[X-(h+l)X] + var[SD]
where var[SD] represents the variance in schedule delay. An average
schedule delay of 15 minutes is assumed with a rectangular distribution
between 0 and 30 minutes. The variance in schedule delay is,
therefore, I/3U5)2. Schedule delay and travel time are assumed to be
stochast i caI Iy i ndependent.
The following chart gives the values of the estimated parameters
used in the preceding tables.
Parameter Work Tr i p Shoppi ng Trip
o.c 2.24 4.11
ay 0.04M 0.0654
a- 0.114 0.374
6 3.79 2.24
174
-------
(46)
Using this result, one can derive the expected value of the
truncated Taylor's series about F for P(o\Y):
E[P(c\Y)} = P(c\y)-
b=l
, } • [P(b\y)-&) (P(b\Y)-%)}
«here « - \J0 if 4% 07)
Examples of Calculations — The following is an example of
a variance calculation.
*
-------
By combining like terms from the preceding tables, the
following expressions for the variance of y are developed.
Auto vs. transit — work trip:
1. Add up all terms with var[D] from Table 40 to get
.00056 (A. + 2/JTJT + A .) for interzonal trips
^ ^ 3 3
.00056(A.
if
2. Add a*
S 6
from Table 40
,1,772
for intrazonal trips
chart p. 174
= .00433('Sb)2
3. Add &2(.086) from Table 40
= (3. 79)2 (. 086)
= 1. 2353
This yields the following equations:
var[Yn,] = .OOOS6(A.
^A~ + A.} + .00433(Sh)2 + 1.23531
* v v ®
n,
CLD if
(for interzonal trips, that is, for i^j
= .00056(Ai)
(for intrazonal trips)
1.23531
Table 44 lists var[Y ^'s obtained in this manner for other
types of comparisons such as auto vs. transit — shopping
trip and auto vs. walking — work trip.
176
-------
Table continued on following page.
Table 44. VARIANCE OF SEVERAL COMPARISON FUNCTIONS
Auto vs. Transit — Work Trip
var[Y J = .00056(A. + 2SA.A. + A.) + . 00433 (J,)2 + 1.23531
CLD 1' 1* J J D
var[Yab] - .00056(AJ + .00433(~§b)2 + 1.23531 i=j
Auto vs. Transit -- Shopping Trip
var[Y J = .00164(A. + 2/OT + A .} + .04663(J,)2 + 0.43151 i?j
dU If If J J D
var[Y ,} = .00 164 (A.) + .04663('S,)2 + 0.43151 i=j
ab ^ D
Auto vs. Walking (intrazonal trips only) — Work Trip
var[Y ] = .29651 (A.) + 1.23531
COJ 1,
Auto vs. Walking (intrazonal trips only) -- Shopping Trip^
var(Y } = 3. 51064 (A.) + .43152
aw t,
Auto vs. Carpool (with h passengers) — Work Trip
uor [Y , ] = h2 r. 00116) (A . + 2/4~X7 + A .} + . 12668 i/j
,
var[Yaoh] = h2(. 00116) (AJ + .12668 i=j
Auto vs. Carpool (with h passengers) — Shopping Trip
,] = h2(.00294)(A. + 2/4 .A . + A .) + .320775 i/j
CLCrl If 1, J J
= h2f. 00294) (AJ + .320775 i=j
177
-------
Table 44 (continued). VARIANCE OF SEVERAL COMPARISON FUNCTIONS
Auto vs. Auto Serve Passenger — Work Trip
.] = . 028S54(A . + 2/4 .A . + A .) + 1. 2353 ift
var(Y J = .028554 (A.) + 1.2353 i=j
CIS 7*
Auto vs. Auto Serve Passenger — Shopping Trip
var[Y ] = 1.04154(A. + 2SA .A . + A .} + .431514 ijfj
/7 Q ^ *7 "7 1 T ^ ' t/
6iO £• L> d J
var(Y } = 1.04154(A J + .431514 i=J
as 2-
Note: For choice types other than mode split, it was assumed that the
constraint of Equation 43 was binding, and all variances of comparison
functions were set identically equal to one. This assumption was made
after inspecting a Taylor's series expansion of the destination choice
model; the number of nonzero variance-covariance terms is quite high,
and this assumption appeared to be warranted.
178
-------
APPLICATION AND TESTS OF THE TRAVEL DEMAND MODEL
WITH 1967 LARTS DATA
This section tests the travel demand model for consis-
tency with the 1967 LARTS Household Survey results on inter-
*
and intrazonal trips by mode and purpose. Such a test of
the model is instructive because it shows, step by step,
how the model is applied in order to obtain forecasts of
travel for zonal interchanges. The approach for applying
the model can be summarized in the following steps:
• Home-Work Mode Split
-Odds functions for auto vs. other modes are estimated
for each zonal interchange using the zonal averages
for system performance data and assumptions about the
system performance of modes other than auto driver or
transit (carpooling, driver serve passenger, and
walking); these functions are based on the estimated
relationship in Equation 13 given in Table 39.
-Probabilities of each mode choice for each zonal inter-
change are calculated from application of Equation 21
or 44, whichever is appropriate to adjust for aggre-
gation.
-The mode shares for each zonal interchange are calcu-
lated using the probabilities from the previous step
and the calculated variance-covariance terms from the
formulas in the previous section in Equation 26 or 47,
whichever is appropriate to adjust for aggregation.
-The estimated mode shares are checked against the actual
mode shares to determine the reasonableness of the model
*The construction of zonal trip tables from the records of
the 1967 Household Survey is described in Appendix D. Data
for the independent variables generally came from other
sources, also as described in Appendix D.
179
-------
• Home-Shopping Mode Split
-The same procedures are applied for the home-shopping
model as are used for the home-work model except that
the estimated relationship in Equation 14 is used in
the first step.
• Home-Shopping Destination Choice
-Odds functions among destination choices are calculated
for zonal interchanges with Equation 15.
-The probabilities of choosing among alternative zone
destinations from a given origin are computed with
Equation 20.
-The shares of trips among alternative zone destinations
from a given origin are estimated from Equation 26,
assuming all variance terms are equal to unity.
-The estimated shares are compared to a sample of actual
shares to determine the reasonableness of the model.
• Home-Shopping Frequency
-The odds of a household in a zone taking a shopping
trip rather than no such trip over a twenty-four hour
period are computed using Equation 16 from Table 39.
-The probability of a household in a given zone taking
a trip is calculated from Equation 20.
-The average number of trips per household in a given
zone is calculated from Equation 26, assuming the
variance term is unity.
-The estimated trips per household are compared to the
actual trips per household in a sample of zones to
determine the reasonableness of the model.
Work Trip Mode Split
This section will compare the actual LARTS data for
1967 against predictions from the disaggregate demand model.
The data given are the number of person-trips between zone
180
-------
pairs by travelers surveyed in the 1967 Household Survey.
(See Appendix D for details on the LARTS Household Survey.)
The following modes are represented in the survey: auto
*
driver, transit, passenger, and driver serve passenger. In
order for mode choices to be more indicative of the actual
decisions made by households, seven modes were modeled:
auto driver, transit, carpooling with one passenger, car-
pooling with two passengers, carpooling with three passengers,
driver serve passenger, and walking. The LARTS survey did
not tabulate walking trips and driver serve passenger trips
(as defined here). The other estimates of mode choice could
be transformed into the LARTS mode categories.
The following equation from Table 39 is used for all
work trip mode split estimates:
= .4.77 -2. 24(C. . .-C. .J - 0. 0411(T . . .-T . J
-0. 114(5. . .~S0 . .) + 3.79Y (13)
A^J B^J
where P... = probability of an auto round trip from home i
n t' J
to work j;
P . . = probability of an alternative mode choice
" i* j
round trip from home i to work j-;
€...-= auto variable costs per mile for a round trip
A 1J
between i. and j;
-------
S . . • - access time for auto round trip between i and j;
A1> Q
SB- • - access time for alternative mode round trip
Dl J
between i and j ;
Y = number of vehicles per tripmaker.
The above equation estimates the odds function for two
modes of transportation. In the following sections, the
equation will be applied to the following mode pairings:
1) auto vs. transit;
2) auto vs. carpool;
a) with one passenger;
b) with two passengers;
c) with three passengers;
3) auto vs. serve passenger;
4) auto vs. walking.
The estimates of these six pairwise odds will be used to
estimate the mode split for the seven alternatives for each
zonal pair (i,j). A random sample of 400 zonal pairs was
taken from the 10,000 zonal pairs in the LARTS data. Of
these, 172 pairs were located exclusively in Los Angeles and
Orange Counties and were selected for further analysis
because of the availability of transit data. In order to
show the development of the model in the following sections,
12 zonal pairs for which transit trips were recorded are
examined in detail. The mode split estimates of the aggre-
gate sample of 172 pairs are compared to the actual data at
the end of this section.
Auto Versus Transit — Table 45 gives the variables and data
for the 12 zonal pairs. Subscript A is auto and subscript T
is transit. Transit is alternative mode B in the generalized
Equation 13. In the auto vs. transit comparison, the data
are substituted directly into the generalized equation,
giving ln(P ../P ..). This result will be used later in the
1 l> J
mode split calculations.
182
-------
Table 45. VARIABLES AND SOURCES FOR AUTO VS. TRANSIT
Zone I
2
2
3
3
4
12
12
15
23
30
33
45
Zone J
2
41
1
3
7
1
3
32
1
1
33
19
^Aij CTii
("in dol lars)
$0. 10
0.96
0.80
0. 1 1
0.30
0.80
0.24
0.30
0.88
I .98
0. 12
0.90
$0.64
1.65
i .42
0.62
0.72
1 .42
0.72
0.88
1 .32
2.46
0.62
1 .58
IAIJ TTii
(in minutes)
24.8
86.6
71 .0
23.2
30.8
75.5
31.2
37.6
72.4
125.6
21 .3
72.0
43.0
277.0
139.0
51 .0
103.0
145.0
1 13.0
102.3
1 13.9
196.3
93.8
176.0
!MI !m
(in minutes)
4
4
6
4
4
6
4
4
6
6
4
4
18
18
16
24
34
12
20
18
16
12
18
14
Y
0.7187
0.7187
0.9068
0.9068
0.9328
0.91 19
0.91 19
0.5918
0.8781
0.8800
0.9253
0.6843
Variable
Definition
Source
CTij
.03(20. .)
if Df < 5 — . 60 + TR
\ f 5 < D. . < 33 — TR +
1*3
2 [.30 + .08(D . ./3)]
1*0
\ f D.. > 33 — TR +
•z-J
2 [1.28 + ,0?(D. .-33)/2]
.03 is the assumed a'jto oper-
ating cost for one mi le.
D.. is the average one-way
mneage from zone i, to zone
j from Time and Distance
file (see Appendix D).
Based on zonal transit fares
which were in effect in 1967,
TR = 0.10 (Average number of
peak transfers).
Table continued on following
page.
183
-------
Table 45 (continued). VARIABLES AND SOURCES FOR AUTO VS. TRANSIT
Variable
T - •
AiJ
T
Tij
S • •
/4tJ
5, ..
IJ
y
-i
Definition
HVADT - HWADIWT - HVADJWI
2(peak line-haul + speak
headway)
4 if i / 1 and 3 i- 1
6 if i = 1 or j = 1
8 i f i - 1 and j = 1
Approximations of average
walk times for potential
bus riders to bus stops
in the zone.
A. + An + A,.
1 2 3
± (J U
HH
ill*
Source
LARTS data where:
HWADT - round trip line-haul
time for home-work
purpose in auto
driver mode;
HWADIWT = amount of walking
time in zone i;
HWADJWT = amount of walking
time in zone j .
Los Angeles bus schedules --
see Appendix D for a detailed
exp lanation.
Assumption — one minute of
access time at each end of
each trip except in Zone 1
(CBO) where there would be
two minutes.
Bus route maps — see Appendix
D for more deta i I s.
1970 Census data:
A1 = number of households in
zone i witlY one car;
A* - number of households in
zone 7' with two cars;
A _ = number of Households in
3
zone ^ with three cars;
HH = number of households in
zone i.
184
-------
Auto Versus Carpool — Because carpooling is a mode not used
in the estimation of Equation 13, the formula does not apply
directly. In particular, the mode specific constant and the
vehicle-per-tripmaker variable do not appear to be applicable
to this mode choice situation; therefore, the following
equation was applied:
In ~^~ = -2. 24(C . . .-Cr . .) - 0.041KT . . .-Tr . .)
pckij A^ Ck^ A^ cktj
where the subscript C denotes carpooling as the alternative
mode in Equation 13 and k denotes the number of passengers
in the carpool.
In addition to the above change, it is assumed that
S . . is equal to S. . .; that is, access time for the traveler
G -,1. J A 1 J
is equal whether he is in a carpool or is a driver alone.
Therefore, only time and cost will be factors in estimating
•p ,/p
- A^o C^o.
Several different methods for inferring the costs and
times of trips by carpools based on mileage between zones,
schedule delay and additional line-haul time based on number
of passengers were tried. The two approaches below seem to
give the most reasonable estimate of the modal split between
single passenger trips and carpools.
Both approaches take into account the number of people
in a car and the distance that will be driven. In Method I
the cost of a multipassenger trip is found by multiplying
the cost per mile ($0.03) of an auto trip times the round
trip distance between zones of origin and destination plus
an estimate of additional miles driven because of the pick-
ups and delivery of members in the carpool. For each
185
-------
person-trip, this cost will then be divided by the number
of people in the car, since they will be assumed to share
the cost equally.
In Method II line-haul time is based on the number of
passengers and the extra mileage driven for a carpool. Car-
pool line-haul time equals auto driver alone line-haul time
(HWADT) for the zonal interchange, plus the time, based on
intrazonal distances and 15 miles per hour driving speeds,
to pick up and deliver the additional passengers and a
schedule delay based on number of passengers.
The above descriptions are represented by the following
equations for Method I and Method II. Note that 20 minutes
schedule delay is added for each passenger to account for
additional wait times associated with carpool pickups and
deliveries. Zone areas originally in acres are divided by
640 to convert them to square miles.
Method I
. 03(20^ . + fcM^ + A . ..
k+1
A . . + A . .
= HWADT + ^^r;}^ + k(20)
Method II
.03(20 .. + 2k(A . . + A . .)}
I'd 1'1> 33
k+1
s/60
k = number of passengers
. _ 5 /Area of zone i _ 5 /Area of zone
Aii ~ IT/ 640 Ajj ~ T2/ 640
186
-------
Method I is used when the distance between zone i and
zone j is relatively short. The reasoning behind this is
that with short trips there are relatively more work trips in
the zonal interchange, and so there is more chance of a
convenient carpool. Method II is used with longer distances
because with fewer work trips in a zonal interchange, there
is less likelihood that a carpool can be formed. By testing
it was found that if the round trip distance of the trip
(2D . .) is less than or equal to 16 miles, then Method I is
i j
appropriate? otherwise Method II is used. (For trips of
great distances, such as zonal pair (30,1) in the sample,
Method II over-estimates carpooling because of the great
savings it appears to produce; in actuality, the opportunity
for carpooling is quite limited.)
Table 46 shows the estimates for C „ • • and T_ . . for
C^J C^i3
k = 1, 2 and 3. Substituting these estimates and the ones
fqr CA.. and TA .. listed in Table 39, P^/P,
Auto Versus Serve Passenger — The serve passenger mode
actually involves three round trips: one for the passenger
and two for the driver chauffeuring the passenger to a given
*
destination. These three trips are combined into one
(vehicle) trip for modeling purposes; i.e., the serve pas-
senger process is one trip with the cost elements of three
trips. After the number of such trips is predicted, they
are transformed to actual driver and passenger trips. The
estimating equation is:
*Note that our definition of driver serve passenger is sub-
stantially different from the LARTS definition. There is,
in fact, no LARTS data category which corresponds to this
type of trip though they are presumably captured under a
different purpose category. See Appendix D for more details
on this subject.
187
-------
Table 46. CARPOOLING VARIABLES
Zone I
2
2
3
3
4
12
12
15
23
30
33
45
Zone J
2
41
I
3
7
1
3
32
1
1
33
19
'V'j
$0. 10
0.60
0.49
0. 1 1
0.22
0.48
0. 18
0.21
0.52
1.08
0.12
0.55
CC2iJ
$0.10
0.48
0.39
0.12
0. 19
0.38
0. 16
0. 18
0.40
0.77
0.12
0.43
CC3ij
$0. 10
0.42
0.34
0. 12
0. 18
0.33
0.15
0. 16
0.35
0.62
0.12
0.37
V'j9
59.5
138.2
1 16.4
60.2
69.8
1 18.0
66.7
73.0
1 14.9
166.2
57.2
1 18.8
TA2iJa
94.2
189.8
161.7
80.2
108.7
160.5
102.2
103. 3
157.4
206.9
93. 1
165.6
TA3ija
129.0
241 .4
207. 1
100. 1
147.7
202.9
137.8
143.7
199.8
247.5
129. 1
212.4
All times are in minutes.
188
-------
-4.77 -2. 24 (CA^ .-C . .) - 0.0411(1 ..-
-0.114(SAij-Ssij) + 3.79Y (13.2)
where PS. • = probability of making a serve passenger trip;
CS'- = cost of a serve passenger trip;
TS . - = line-haul time of a serve passenger trip;
Ss-- = access time of a serve passenger trip;
and the other variables have been previously defined.
In order to find the cost and line-haul time for serve
passenger trips, the parameters of cost and line-haul time
for the home-work purpose and for the home-shopping purpose
were compared. The chart below shows these parameters.
Cost Line-haul Time
Home-work 2.24 0.0411
Home-shopping 4.11 0.0654
It is assumed that the driver in the serve passenger
mode places the same value on time and costs as shopping
trip makers (that is, they are non-work travelers more
closely resembling shopping trip motorists). The extra cost
of a serve passenger trip incurred by the driver will be
equivalent to;
2 (home-shopping cost-)* (C . . •)
home-work cost
and likewise the extra line-haul time will be:
2(home-shopping line-haul)* (T . .)
home-work line-haul "
189
-------
Thus,
2. 24
2. 24
,0411 + 2(. 0654)•(Tt . J
.0411
The assumption is made that total access time for a
serve passenger trip is 8 minutes for all zonal pairs,
except for trips originating in zone 1, where S .. equals 14
O 1, J
minutes. This is based on the prior assumption that 1
minute of access time per person occurs at each terminus
of the trip. For most trips, the driver will have 4 minutes
of access time and the passenger will have 4 minutes of
access time. If the trip originates in zone 1, the driver
will incur 8 minutes of access time and the passenger 6
minutes.
Table 47 shows the system performance variables for the
serve passenger mode and their estimates for the observations
used in testing the model. When these are combined with
the variables for C. . ., T, . . and 5. . . found in Table 39,
A^^ Atj A^3
(P.../?„..) can be calculated.
A^3 Stj
Auto Versus Walking — The LARTS data base has no information
on walking trips, but this mode of transportation should be
included when it is a reasonable alternative. Walking trip
related variables are estimated as follows: cost is zero;
line-haul time is zero; access time (walking time in other
modes) is calculated as a function of distance. Assuming
that the average person can walk 3.16 miles per hour or
0.0526 miles per minute, the formula for access time is:
-------
Table 47. DRIVER SERVE PASSENGER VARIABLES
Zone I
2
2
3
3
4
12
12
15
23
30
33
45
Zone J
2
41
1
3
7
1
3
32
1
I
33
19
CSij'
dollars
$0.47
4.48
3.74
0.51
1 .40
3.74
1 . 12
1 .40
4. 1 1
9.25
0.56
4.20
Tsir
minutes
103.7
362.0
296.8
97.0
128.7
315.6
130.4
157.2
302.6
525.0
89.0
301 .0
ssu-
minutes
8
8
8
8
8
8
8
8
8
8
8
8
191
-------
It is further assumed that S . . must be less than 90
minutes for a round trip by walking to be a legitimate
alternative.
The estimating equation for (P . ./P^^. _.) is:
ln _4ii , _4.77 -2.24CA.. -.0411TA.. -.114(5^-3,,^)
+ 3. 79Y
(13.3)
Only three of the observations being tested allow for
walking trips. S^. . is given in Table 48 for these obser-
vations, which are for intrazonal trips.
Table 48. WALKING VARIABLE
(minutes)
Zone I
2
3
33
Zone J
2
3
33
svij
60.84
69.96
68.82
Mode Probabilities — The odds functions for six alternative
modes of transportation have been estimated in relation to
auto driver trips. In order to combine all seven alter-
natives to find the modal splits for all trips between the
zonal pairs, the following equations are used:
192
-------
Auto probability = i + D -
ij
PBi i
Transit probability = i + n
Carpool 1 probability = 2 + D . .
P.
. .
C 2^J
Carpool 2 probability = 2 + D
Carpool 3 probability =
, + D
p /P
w w -i •*. Si j' Aij
Serve passenger probability - — — -—^ -
ij
P Wij Ai-j
Walk probability = ~ 2 + D -
where D =
P . PCij PC ij PC ij P
A^J A^J " Aij Aij Aij
Carpool 1 = carpool mode with one passenger;
Carpool 2 = carpool mode with two passengers;
Carpool 3 = carpool mode with three passengers.
It can be seen that auto share is calculated from an
equation of the form given in Equation 21, whereas for the
others Equation 43 is applied.
Table 49 shows the estimated choice probabilities for
the observations tested.
193
-------
Table 49. ESTIMATED MODE PROBABILITIES
(percentages)
Zone
I
2
2
3
3
4
12
12
15
23
30
33
45
Zone
J
2
41
1
3
7
1
3
32
1
1
33
19
Auto
68
76
72
72
76
68
74
70
63
32
73
72
Transit
13
—
1
3
—
2
1
4
6
1
1
I
Car-
pool
1
15
21
22
16
18
24
20
20
25
45
17
23
Car-
pool
2
4
3
4
4
4
5
5
5
6
17
4
4
Car-
pool
3
1
—
1
I
1
I
1
I
I
5
1
1
Serve
passenger
5
—
—
3
—
--
—
__
—
—
4
—
Walk
3
—
—
I
--
--
—
--
—
—
1
—
194
-------
Estimated Trips by Mode — In order to estimate the mode
shares for a zonal interchange, the figures from Table 49
must be adjusted for the variance-covariance of independent
variables within each zonal interchange. Equations 26 and
47 are applied to make these calculations.
The variance of comparison functions between auto and
each other mode pair is given in Table 50 for each zonal
pair. These figures are computed from the formulas derived
in the previous section.
By a method described in Appendix D, home-work and work-
other trips are combined to form the actual 1967 LARTS number
of residence-based round trips to work for any given zonal
interchange. Then:
Actual auto trips = HWADC . . + WOADC . . + HWASC.. + UOASC . .
IsJ Z-J 2-J 2-t7
Actual transit trips = HWBAC. . + WOBAC . . + HWBNC.. + WOBNC. -
t-J 7-J "Z-J 1>J
Actual passenger trips = HUAPC. . + UOAPC. .
TsQ I'd
where HWADC.• = total home-work round trips for auto driver
^•J
from zone i to zone j;
WOADC . . = total work-other round trips for auto driver
"Z-J
from zone i to zone j;
HWBAC. . - total home-work round trips for transit, car
•z-J
available, from zone i, to zone j;
WOBAC . . - total work-other round trips for transit, car
13
available, from zone i to zone j;
HWBNC. . = total home-work round trips for transit, car
"Z-J
not available, from zone i to zone j;
WOBNC . . = total work-other round trips for transit, car
T-J
not available, from residence zone i to
zone 3 ;
195
-------
Table 50. VARIANCE OF COMPARISON FUNCTION
BETWEEN AUTO AND OTHER MODES3
Zone
I
2
2
3
3
4
12
12
15
23
30
33
45
Zone
J
2
41
1
3
7
1
3
32
I
1
33
19
Transit
1 .00
1 .00
1 .00
1.00
1 .00
1 .00
1 .00
1 .00
1 .00
1 .00
1 .00
1 .00
Carpool
1
0.149
0.231
0. 194
0.156
0.277
0.179
0.227
0.225
0. 179
0. 171
0. 153
0.201
Carpool
2
0.217
0.542
0.396
0.246
0.727
0.338
0.528
0.521
0.336
0.305
0.233
0.425
Carpool
2
0.330
1 .000
0.732
0.396
1 .000
0.603
1 .000
1 .000
0.597
0.527
0.365
0.799
Serve
passenger
1.00
1 .00
1 .00
1 .00
1.00
I .00
1 .00
1 .00
1.00
1 .00
1 .00
1 .00
Walk
1 .00
1 .00
1 .00
Constrained to be between 0 and I.
196
-------
BWAPC .. = total home-work round trips for passenger
1*3
from zone i to zone j;
WOAPC . . = total work-other round trips for passenger
1*3
from residence zone i to zone j;
HWASC . . = total home-work round trips for LARTS-defined
13
serve passenger from zone i to zone j;
WOASC . . = total work-other round trips for LARTS-
1*3
defined serve passenger from residence zone
i to zone j .
Results of these computations are presented in Table 51.
In order to compare the estimates to the actual results,
the trips from the seven estimated modes must be distributed
among the five data categories. The following equations show
how this distribution is made, while also finding the esti-
mated round trip person count for each mode for a given zonal
interchange.
Est. auto trips = [Est. auto share + .5(Est. carpool 1 share)
+ .333(Est. carpool 2 share) + ,25(Est.
carpool 3 share)](Total trips)
Est. transit trips = (Est. transit share)(Total trips)
Est. passenger trips = [.5(Est. carpool 1 share) + .667(Est.
carpool 2 share) + .75(Est. carpool 3
share) + (Est. serve passenger share)]
(Total trips)
Est. serve passenger trips = (Est. serve passenger share)(Total
trips)
Est. Walk Trips = (Est. Walk Share)(Total Trips)
197
-------
Table 51. ACTUAL WORK TRIPS
Zone I
2
2
3
3
4
12
12
15
23
30
33
45
Zone J
2
41
1
3
7
1
3
32
1
1
33
19
Auto
Driver
173.2
5.5
32.0
139.9
54.7
13.8
28.9
20.7
6. i
1 .4
167.3
1 .6
Transit
20.2
1 .0
2.5
1 .0
2.0
2.0
2.0
3.0
0.6
0.5
0.5
1 .0
Auto passenger
30.9
0.5
6.0
16.0
5.6
2.5
2.7
4. 1
3.1
0.0
12.7
0.0
198
-------
Table 52. ESTIMATED WORK TRIPS
Zone
I
2
2
3
3
4
12
12
15
23
30
33
45
Zone
0
2
41
1
3
7
1
3
32
1
1
33
19
Auto
driver
152.4
6.0
33.5
124.5
53.2
14.7
28. 1
21.9
7.3
1 . 1
146.6
2.2
Transit
31 .4
0.0
0.8
6.5
0.2
0.5
0.3
1 .4
0.8
0.0
2.2
0.0
Auto
passenger
40.5
1 .0
6.2
25.9
8.9
3. 1
5.2
4.5
1 .7
0.8
31 .8
0.4
Driver
serve
passenger
16.7
0.0
0.0
7.5
0.2
0.0
0.2
0. 1
0.0
0.0
9.8
0.0
Walk
8.8
1 .2
1 .4
199
-------
where (Total home based trips).- = HWADC.. + WOADC. .
"Z-J tj i*,]
+ HWBAC . . + WOBAC . . + HWBNC. . + WOBNC . . + HWAPC . .
*L 7 "if "? "If 1 *L 1 *£• 7
+ WOAPC . . + HVASC . . + WOASC.. + WALK..
^3 13 ij ^0
The results of the estimations are given in Table 52.
A comparison of Tables 51 and 52 shows how well the
model predicts and highlights some of the biases which may
result in applying the model. These will be discussed in
more detail after the aggregate results of estimating trips
by mode and total VMT's for all 172 zonal interchanges are
presented.
Aggregate Work Trips — Another test of the performance of
the model is to compare its predictions of travel behavior
with both the observed travel in the random sample of 172
zonal interchanges and the modal split over the entire Los
Angeles region. The first such comparison is presented in
Table 53, where the actual number of person-trips by mode
and VMT's for passenger vehicles are compared to the esti-
mates given by applying the model. VMT's are calculated
as follows:
VMT = *[(2D ..(Driver trips.. + 2-Serve passenger trips..)
• • "Z-J 'Z-J "Z-J
T>J
+ (D.. + D ..)(Passenger trips)} (48)
'Z-'Z- 33
The first term in the above brackets is self-evident from
the definition of driver trips and driver serve passenger
trips. (Driver serve passenger trips are zero in the LARTS
data for work trips owing to the way their data is classi-
fied. See Appendix D for more details on the LARTS data.)
The second term accounts for extra distance which must be
traveled in order to pick up and deliver passengers within
zones.
200
-------
Table 53. ESTIMATED VS. ACTUAL TRIPS BY MODE
FOR 172 ZONAL INTERCHANGES
Mode
Auto dri ver
Trans it
Auto passenger
Driver serve passenger
Walk
VMT's
Actual
1040
47
123
—
--
15302
Estimated
960
54
196
40
10
I545I3
'includes driver serve passenger VMT's of approximately 240 miles,
201
-------
Table 54. ESTIMATED VS. ACTUAL MODE SHARES
FOR LARTS REGION
(percentages)
Mode
Auto driver
Transi t
Auto passenger
Actual
0.84
C.04
0. 12
Estimated3
0.79
0.04
0. 16
Excludes driver serve passenger and walk.
202
-------
Table 54 presents the modal shares estimated for the
sample of 172 observations as compared to the modal shares
for the region as a whole. This comparison will allow a
determination of whether predictions based on the sample will
be representative of the entire LARTS area.
Evaluation of Work Trip Model
In general, the model performs reasonably well, although
certain of its biases should be noted in order to better
evaluate its application to policy scenarios in Chapter 3.
A comparison of Tables 51 and 52 shows that there is a ten-
dency for auto driver alone trips to be underpredicted,
whereas passenger trips are somewhat overpredicted. It is
possible to "calibrate" the model further to decrease pas-
senger trips and increase auto driver trips. Unfortunately,
it is not known to what extent driver serve passenger trips
would be decreased relative to carpooling. In the absence
of extraneous information, it was decided not to proceed with
adjustments which would cause the model to give better pre-
dictions for 1967. However, it appears that the serve pas-
senger mode is overpredicted for short (typically intrazonal)
trips and that carpooling is overpredicted for long (say,
greater than 25 miles round trip) distances. The model also
predicts more variation in transit usage than appears in the
sample of 12 observations; this is to be expected, since a
criterion for including an observation in the sample was the
existence of at least one (one-way) transit trip.
Examining the aggregate results in Tables 53 and 54
shows that some of the above-mentioned errors tend to cancel
when the model is used to predict regionwide auto trips and
VMT's. From Table 53 it can be seen that total vehicle
trips, exclusive of driver serve passenger, are underpre-
dicted by 7.69 percent. The two figures are not comparable
because the predicted VMT's, which include miles traveled
203
-------
attributable to driver serve passenger trips, are intrazonal
with an average distance for each 'of four legs of approxi-
mately 1.5 miles; deducting the aggregate approximate total
of 240 miles from the estimated 15,451 yields 15,211. From
this rough calculation it appears that aggregate VMT's are
underestimated by about 1 percent.
Shopping Trip Mode Split
The shopping mode split model is part of a larger travel
demand model which also predicts destination choice and fre-
quency of shopping trips per household per day. It was found
that in order to test these latter relationships, the shop-
ping trip sample of zonal interchanges had to be augmented
to include observations where more than one destination zone
was associated with a given origin zone. In the sample of
zonal interchanges finally selected, each observation repre-
sents a destination-choice with another destination alter-
native in the sample. This constraint lowered considerably
the number of useful observations for determining the effects
of system changes on destination choice. Nonetheless, the
fifteen zone interchanges were selected to be highly repre-
sentative of the regional population. Also, because shop-
ping trips are clustered into fewer zonal interchanges, the
percentage sample of total shopping trips is comparable to
the work-trip sample (about 4 percent)..
In order to test the model, three series of computations
were made corresponding to modeichoice, destination choice,
and frequency of trips per household. Rather than give the
values of variables at every step, as was done in the work
trip model, we will present the formulas and assumptions at
each step and then give the predicted vs. actual travel
behavior. We shall.first consider mode split.
204
-------
As in the work trip mode split equations/ odds functions
for auto vs. other modes were first generated, and from these
probabilities of mode choice were calculated. The probabili-
ties were then adjusted for variation in order to forecast
modal shares for each zonal interchange. The following
sections present the equations and assumptions for each odds
function.
The basic home-shopping odds function from Table 39 is:
P
In •L = -6.77 -4. 11 (C . . - C . .) - .0654(1... - TD. J
B^J
-.374(8... -Sn..) + 2.243 (14)
where P. . . = probability of an auto driver home-shopping
n 1^ Q
round trip from zone -i to zone j, given a trip
will be made;
P . . - probability of a given alternative mode home-
D If J
shopping round trip from zone i to zone j,
given a trip will be made;
C . . = variable cost per mile of an auto round trip
f\. "2-^7
from zone i to zone j;
C . . - cost of an alternative mode round trip from
Dlr (J
zone i to zone j;
T. . . = line-haul time for auto round trip from zone
A^3
i to- zone j;
TK.. = line-haul time for alternative mode round trip
& I* J
from zone -i to zone j ;
5.. . = access time to auto for a round trip between
A I* (7
zone i- and zone j ;
= access time to alternative mode for a round
trip between zone -i and zone j;
J = number of autos per tripmaker.
205
-------
Auto vs. Transit — The variables defined in Table 55 are
substituted into Equation 14 for the home-shopping model.
The subscript B has been replaced by a T to denote transit
as the alternative mode. The source of each variable B is
given in the table. As in the work trip model, the result
?../?„.. will be used to find the mode splits.
n 7^(7 -t ^(7
Auto vs. Carpooling — The carpooling assumptions used in
the home-work purpose proved inappropriate for the home-
shopping purpose. New carpooling options were added based
on reasonable inferences about shopping trip behavior. C
and C- are equivalent to the home-work carpooling, in that
cost is shared by the occupants of the car and line-haul time
is auto time plus an additional time for pickup and delivery
of passengers. Only one and two passenger type carpools are
provided for. C ,. and C are similar to family carpooling.
That is, the carpool with one passenger replaces the alter-
native of two trips by the occupants. Likewise, c has two
passengers who would have the option of separate trips had
the carpool not been utilized. C is a home-shopping car-
1 D
pool trip in which the passenger only took the trip for
companionship.
As in the home-work model, access time will be equal
for each person, whether in a carpool or driving alone, and
the constant and availability of a vehicle are suppressed.
The carpooling equation will be as follows:
where P ,.. = probability of an individual making a round
trip carpool trip with k passengers.
Table 56 defines the variables for each of the five different
types of carpooling. The auto variables are the same as in
Table 55.
-------
Table 55. AUTO VS. TRANSIT VARIABLES
Variable
Definition
Source
.03(20^.)
.03 is the assumed auto operating
cost for one mi le.
D. . is the average one-way mileage
from zone i to zone j from Time and
Di stance file.
i f D.. < 5; .06 + TR
^J
if 5 < D. . < 33; TR
= W =
+ 2(.30 + ,08(D . ./3)}
13 J
if D. . > 33; TR +
^J
2(1.18 + ,07(D. .-33J/3]
Based on zonal transit fares which
were in effect in 1967.
TR = .10 (Average number of off-
peak transfers.)
Aij
HSADT - HSADIWI
- HSADJVT
LARTS data where (in minutes):
HSADT = home-shop purpose, auto
driver mode, round trip
Ii ne-hauI t ime;
HZADIWT = amount of walking time
in zone -i;
BSADJWT = amount of walkint time
i n zone j.
T-ij
2(Off-peak line-haul
+ off-peak schedule
delay)
Los Angeles bus schedules — see
Appendix D for a detailed explanation
(times are in minutes).
3 ...
A^J
6 minutes
Assumption — one and one-half
minutes of access time at each end
of each trip.
Approximations of aver-
age walk times for
potential bus riders to
bus stops in the zone.
Bus route maps — see Appendix D
for more deta i1s.
A- + 2. 1(AJ
o O
HE
1970 Census data:
Ag = number of households in zone i
wi th two cars;
A_ = number of households in zone i,
with three cars;
HH = number of households in zone i.
207
-------
Table 56. CARPOOL VARIABLES
Variable
Definition
Source
°CUj
'C2ij
CCUij
'C2Aij
'ClBij
Clij
ClAij
C2Aij
ClBij
.
OS (D . . + A . + A .} 2
1 u i .r
C. . .
A^J
4 (A . + A .) + 20 + T. . .
^ 3
8(A . + A .) + 40 + T. . .
^ J A^J
T...+ 20
T. . . + 40
20
See carpooling in home-work purpose.
See carpooling in home-work purpose.
Assumption (no extra cost for pick-
up and deli very).
Assumption (no extra cost for pick-
up and deIivery).
Assumption (no extra cost for pick-
up and delivery; no sharing of costs)
See carpooling in home-work purpose.
See carpooling in home-work purpose.
Assumption (additional time for
divergences in personal schedules).
Assumption (additional time for
divergences in personal schedules).
Assumption (additional time for
divergences in personal schedules).
208
-------
Auto vs. Serve Passenger — The auto vs. serve passenger odds
function required an adjustment to the constant term in order
to give reasonable results. The resulting equation is:
P . .
(14.2)
where P . . = probability of a round trip by a driver serving
o 1*3
passenger and a passenger from zone i to zone 3.
The reason for increasing the constant from -6.77 to
-4.77 appears to be that the original constant (in Equation
14) is mode specific for transit only. The adjustment is
necessary to prevent the serve passenger mode from dominating
other modes to an unreasonable degree. The definitions of
auto serve passenger variables are presented in Table 57;
their assumptions are based on the reasoning discussed in
the work trip section on the serve passenger mode. Auto
variables are the same as the ones used in Table 55.
Mode Probabilities — The seven odds functions described
above, auto vs. other modes, are calculated for each zonal
interchange. (It was found that the estimated walk prob-
abilities were negligible. Hence, they were ignored.) In
order to find the model probabilities for each of the eight
alternatives, the following equations were used:
Auto probability =
2 + D . .
p , ./p . .
Transit probability = —"-
209
-------
Table 57. SERVE PASSENGER VARIABLES
Variable
Definition
Source
Sij
S(T. ..)+ 40
2(S..J
Assumption (two vehicle round trips
for each serve passenger mode trip).
Assumption (three person round trips
for each serve passenger mode trip
plus an additional time penalty for
personal schedule divergences).
Assumption (access time for two
person round trips).
210
-------
p /p
r 7 v i
Carpool 1 probability = , ^
P . ./P
Carpool 2 probability = C*^ D Ai
id
P
Carpool 1A probability =
ij
P . ./P
Carpool 2A probability = C^J D
id
P • ./P
Carpool IB probability = C
1 + D . .
Pe • -/PA • •
S If 1 A 1r 7
Serve passenger probability = —•=—^—^—sL
where:
P P..P P . . P .P .P..
B^3 , C1^3 C2ij C1A^J , C2A^J L C1B^J _,_ S^J
D; i ~ p + p + p + ~D + ~S + ~~p + p
is tJ IT . • * IT , . • c . . . :r . . • i..» Jr..« i..**
/l^j ^^J 4^J ^tj Aij A^3 Ai,j
Estimated Trips by Mode — Several more steps are necessary
in order to estimate the number of trips by mode. First,
the mode probabilities calculated as above must be adjusted
for variance within zonal interchanges in order to determine
modal shares. This approach is the same as that used for the
home-work model. Then the carpool mode is divided into
drivers and passengers. The serve passenger mode is assumed
to have one passenger round trip and one serve passenger
round trip for each round trip estimated. The following
equations show how the estimates of modal trips for each
zonal interchange are derived.
211
-------
Est. auto trips = (Est. auto share + .5 (Est. carpool 1_ share
+ Est. carpool 1A share + Est. carpool IB
share) + .33(Est. carpool 2_ share + Est.
carpool 2A share)}(Total trips)
Est. transit trips = (Est. transit share)(Total trips)
Est. passenger trips = (.5 (Est. Carpool 1_ share + Est. car-
pool 1B_ share) + .667(Est. carpool 2_
share- + Est. carpool 2A share) + Est.
serve passenger share] (Total trips)
Est. serve passenger trips = (Total trips)(Est. serve pas-
senger share)
where Total trips^ . = HSADC. . + HSBAC. . + HSBNC. . + HSAPC. . + HSASC.
from the LARTS data
Actual auto trips = HSADC..
Actual transit trips = HSBAC.. + HSBNC..
Actual Passenger trips = HSAPC..
Actual LARTS-defined serve passenger trips = HSASC..
i> 3
The actual trips by mode for the 15 observations are
given in Table 58; the estimated trips by mode are given
in Table 59.
A summary of the performance of the mode choice rela-
tionship in the shopping trip model is presented in Table 60.
Actual and predicted VMT's are also shown; VMT's are com-
puted here as they were for the work trip using Equation 48.
212
-------
Table 58- ACTUAL SHOPPING TRIPS BY MODE
FOR 15 ZONAL INTERCHANGES
Zone I
2
2
2
3
3
12
12
14
14
14
31
31
31
31
31
Zone J
2
14
56
1
3
3
12
2
13
14
7
31
32
35
46
Auto
driver
183.9
4.5
0.6
0.7
210.7
21 .0
151 .6
16.3
37.2
62.9
1 .4
129.4
14.7
13.0
Transit
4.8
0.7
0.0
0.0
5.2
0.4
0.7
0.2
1 .4
1 .0
0.0
1 1 .0
0.9
3.5
3.6
Auto passenger
79,0
3.0
0.0
0.0
58. 1
10.5
63.0
6.4
18.2
35.9
0.0
44.7
2.6
10.5
9.3
213
-------
Table 59. ESTIMATED SHOPPING TRIPS BY MODE
FOR 15 ZONAL INTERCHANGES
Zone I
2
2
2
3
3
12
12
14
14
14
31
31
31
31
31
Zone J
2
14
56
i
3
3
12
2
13
14
7
31
32
35
46
Auto driver
181 .4
5.5
0.2
0.4
201 .0
22.4
156.4
15.7
40.3
72.1
0.6
1 16.9
12.7
18.5
12.1
Transit
18.0
0.0
0.0
0.0
0.8
0.0
1 .6
0.0
0.0
0.0
0.0
25.3
0.0
0.7
8.8
AutO
passenger
68. 1
2.7
0.3
0.3
72.2
9.5
57.2
7.2
16.5
27.7
0.6
42.8
5.5
7.8
5.4
Driver
serve passenger
6.3
0.2
0.0
0.0
3.6
0.0
3.4
0. 1
0.6
2.9
0.0
3.2
0. 1
0. 1
0.0
214
-------
Table 60. SUMMARY OF ESTIMATED VS. ACTUAL SHOPPING TRIPS
BY MODE FOR 15 ZONAL INTERCHANGES
Mode
Auto driver
Transi t
Auto passenger
Driver serve passenger
VMT's
Actual
861 .3
33.4
341 .2
4001 .0
Estimated
856. 1
55.3
324. 1
20.4
4065. Oa
Includes approximately 122 miles attributable to driver serve passenger.
215
-------
Shopping Trip Destination Choice
This section shows how the destination choice relation-
ship in the shopping trip demand model is applied to estimate
the shares of trips among alternative destinations from any
given origin. The results of predicting destination choice
using the approach described below are compared to the actual
shares for the 15 zonal interchanges used in simulating
shopping travel behavior. The odds function for destination
choice is:
p
In =Z. = (-2.06)1
p ik
aik-v
A
+ .844(E . - E )
J K
(15.1)
where P. . . = the probability of taking a shopping round
13
trip from zone i to zone j using any mode;
Pm ., = the probability of taking a shopping round
"V K
trip from zone i to zone k using any mode;
N . . = the number of shopping round trips from zone
0,1> J
i to zone j using the ath mode;
N .,= the number of shopping round trips from zone
i to zone k using the ath mode;
E . = retail employment in zone j as a percentage of
0
total regional retail employment;
E, = retail employment in zone k as a percentage of
total regional retail employment;
A' . . - IN . .
•ij La a^J
• ik ~ ^ aik
a
X , . = the generalized cost of a round trip between
zone i and zone j by the ath mode;
X ., - the generalized cost of a round trip between
zone i and zone k by the ath mode.
216
-------
In applications of this model, auto and transit are the
only modes used to compute the generalized cost of travel.
Using the estimated trips from Table 59, the mode shares for
weighting the generalized cost of travel are calculated as
follows:
. . , Est. auto trips
Auto share = ^
Est. auto trips + Est. transit trips
' '• li
m ., r _ Est. transit trips
Transit share = -=—- - - - - — : - — r - r^ - r- — - — : -
Est. auto trips + Est. transit trips
"TJ.J
where A = auto mode;
T = transit mode.
The generalized cost (X ) for each mode was computed
using the following formulas. For the auto driver mode:
X... = 4.1KC..J + .0654(T...) + .374(3...)
Aij Aij Aij Aij
while for transit (T) :
.... - -6.77 + 4.
+ 2. 242
where the variables are defined in Table 55.
The retail employment variables (E . and £,) are com-
puted as follows:
217
-------
,, _ retail employment in zone j ^nn
Ci . — ; • i - ; :— • J. U U
employment ^n region
p - retail employment in zone k t ,„.,
k retail employment in region
The probability of an individual in zone i choosing to
go to zone j, given a shopping trip will be made, is calcu-
lated using Equation 21. In order to estimate the share of
trips to destination j among all those made from origin i,
the probability must be adjusted for variation among indi-
viduals by using Equation 26. This is done by assuming that
the variance terms in Equation 26 are all unity.
Table 61 presents the actual and estimated shares for
the 15 zonal interchanges in the sample. For each origin
zone, the number of shopping trips to the alternative desti-
nations in the sample varied from 85 percent to 100 percent
of the total trips from the origin. Thus, the sample cap-
tures most of the shopping trips which would be affected by
a change in the transportation system.
Table 62 presents the estimated VMT's which result when
both mode split and destination choice are estimated from
the data. Comparing this figure to the actual VMT's is an
important test of the travel demand model.
Shopping Trip Frequency
The third stage of the shopping model estimates the
odds that a household will take one shopping trip vs. no
shopping trips in a twenty-four hour period. This section
describes how the model is applied to determine the number
of trips per household in a twenty-four hour period in a
given origin zone. The results of the application are
tested against the data for the 15 zonal interchanges which
comprise the shopping trip sample. The odds function for
frequency of trips is as follows:
218
-------
Table 61. ACTUAL VS. ESTIMATED DESTINATION SHARES FOR
SHOPPING TRIPS FROM 15 ZONAL INTERCHANGES
Zone I
2
2
2
3
3
12
12
14
14
14
31
31
31
31
31
Zone J
2
14
56
1
3
3
12
2
13
14
7
31
32
35
46
Actual share
0.964
0.032
0.004
0.004
0.996
0.364
0.636
0. 124
0.315
0.561
0.000
0.718
0.073
0. 108
0. 101
Estimated share
0.775
0.225
0.000
0.017
0.983
0.276
0.724
0.340
0.274
0.386
0.000
0.479
0.253
0.102
0. !66
219
-------
Table 62. ACTUAL VS. ESTIMATED VMT'S FOR SHOPPING TRIPS
INCLUDING MODE AND FREQUENCY CHOICE FOR 15 ZONAL INTERCHANGES
VMT's
Actual
4001.4
Estimated
3954. 3a
Includes approximately 122 miles of driver serve passenger VMT's.
220
-------
N. . . rN
f. /v. . . (/v . . •)
In (1 p = (-1. 72)\ ^—^ • ZLgt>^ 'aXa-i '}
+ (3.
90)1
j
N. . .
' i, '
• E .
J
(16.1)
where P. = the probability of making a shopping round trip
If
in a twenty-four hour period by any mode to any
destination from zone i;
N. .. = the number of shopping round trips in a twenty-
Lr
four hour period from zone i by all modes to all
other zones.
All other variables have been previously defined.
In order to test this function against the LARTS data,
the predicted values for all independent variables are cal-
culated for each origin zone. The probability of an indi-
vidual household making a trip is estimated using Equation
21. The frequency of trips per household is estimated by
adjusting the probability for variation using Equation 26;
the variance term is assumed to be unity.
Table 63 presents the actual vs. estimated frequency
for the five origin zones in the shopping trip sample. The
actual observations are the total number of shopping round
trips from the origin zone to all other destinations,
including those not in the sample of 15 shopping trip zonal
interchange^
Table 64 gives the results of predicting VMT's using
the frequency model on estimated VMT's from mode choice
and destination choice.
Evaluation of Shopping Trip Model
Inspection of Tables 58 through 60 indicates that the
mode choice model replicates travel behavior in the shopping
trip sample with a high degree of accuracy. Because these
221
-------
Table 63. ACTUAL VS. ESTIMATED SHOPPING TRIP
FREQUENCY FOR 5 ZONE ORIGINS
Zone I
2
3
12
14
31
Actual
0.426
0.606
0.848
0.538
0.330
Estimated
0.914
0.510
0.263
0.203
0.423
222
-------
Table 64. ACTUAL VS. ESTIMATED VMT'S, INCLUDING MODE CHOICE,
DESTINATION CHOICE AND FREQUENCY OF TRIP, FOR
15 ZONAL INTERCHANGES
VMT's
Actual
4001 .4
Estimated
4821 .2a
includes driver serve passenger VMT's.
223
-------
results will be extrapolated to the rest of the region, it
is also worth considering how the mode split projections
from the 15 zonal interchanges in the shopping trip model
compare to the mode share for the region as a whole. These
data are presented below in Table 65. It will be recalled
that the zonal interchanges in the sample were originally
preselected because the 1967 data suggested they were
somewhat representative; thus the comparison in Table 65
is not as strong a test as would have been the case had the
sample been randomly selected.
The evaluation of the two stage model — applying both
the mode choice estimates and the destination choice esti-
mates -- rests on the evidence presented in Tables 61 and
62. As in the work trip model, VMT calculations are not
strictly comparable because driver serve passenger VMT's
are included in the estimated results. Deducting the esti-
mated 20.4 serve passenger trips at an average 6 miles per
trip yields 3832 VMT's exclusive of serve passenger trips;
this approximation implies that the shopping trip model
underpredicts VMT's by about 4 percent. The one potentially
important bias in the model is that it tends to underpredict
somewhat the share of intrazonal shopping trips.
The results of applying the trip frequency model are
given in Tables 63 and 64. This part of the travel demand
model appears to be significantly more inaccurate than the
others. Part of the problem lies with the data — number of
shopping trips divided by households is not a good represen-
tation of what the model is actually supposed to predict.
If this were the only problem, the model would still be
acceptable in that predicted VMT's were only 17 percent
greater than the actual. There is, however, another aspect
of the estimated frequency equation which makes it use
inappropriate. The implied elasticity of trips with respect
224
-------
Table 65. ACTUAL VS. ESTIMATED MODE SHARES
FOR LARTS REGION
Mode
Auto driver
Transi t
Auto passenger
Actual
0.68
0.02
0.30
Estimated
0.69
0.04
0.27
225
-------
to a change in auto variable costs per mile is on the order
of -4. This is much too elastic, as the discussion on a
priori information about travel elasticities in Chapter 2
indicates. Application of this model to policy scenarios
of auto disincentives greatly overpredicts their effects.
In the policy simulations in Chapter 3, it is assumed
that frequency of trips do not change with respect to a
change in the auto variable costs per mile. Because shop-
ping trips are relatively short in Los Angeles, this assump-
tion may be relatively accurate for modeling the effects of
gasoline or emissions taxes. Reasonable assumptions about
changes in trip frequency are also applied in Chapter 3.
226
-------
List of References, Appendix A
Aggregate Travel Demand Analysis with Disaggregate or Aggre-
gate Travel Demand Models. Proceedings of Transportation Research
Forum Fourteenth Annual Meeting. 13(1) : 583-603, October 1973.
227
-------
Appendix B. AUTO STOCK ADJUSTMENT MODEL
This appendix presents the model used in determining
the effects of emissions taxes on the auto stock in the Los
Angeles region. The relationships in the model determine
the total fleet size, new car sales, used car price, aggre-
gate scrappage rate from the auto stock, and the scrappage
rate for each model year of autos in the stock. Table 66
presents the equations and identities used in the model.
An explanation for each follows.
CALIBRATION OF THE MODEL
New Car Sales
New car demand is assumed to have the following func-
tional relationship:
NC = PN~aKM (49.1)
where a is an assumed constant and KN is a variable which
takes account of all other factors determining new auto
demand. In order to determine a, the price elasticity of
*
demand, previous studies of new car demand were reviewed.
*See reviews of econometric models of automobile demand in:
The Effects of Automotive Fuel Conservation on Automotive Air Pollution.
Charles River Associates Incorporated. Cambridge. Prepared
for Environmental Protection Agency. 1976; and Dewees,
Donald N. Economics and Public Policy: The Automotibile Pollution
Case. Cambridge, MIT Press, 1974. 214 p.
228
-------
Table 66. EQUATIONS AND IDENTITIES IN THE AUTO STOCK MODEL
NC. E (PN.)~1(1.555)(109) (49)
~ ~
_.
NC '742' - 9121
r, = 302.6868- (PU) 'yitil (51)
Tt E Tt-l + NCt ~ St (52)
St " VW
r _ .3252
L. = fgr-f- (54)
* 3.8032 + 846. 36e~ ^^
where NC = new car sales for Los Angeles and Orange Counties in
£
period t
PN = average price of new cars in period t
~c
PU = average price of used cars in period t
~c
T - stock of cars in Los Angeles and Orange Counties in period t
"U ""*
r = rate of scrappage in period t
t
S, = number of cars scrapped in Los Angeles and Orange Counties
~c>
in period t
L. = rate of scrappage of autos of age •£ owing to the effects
u
of age a I one
229
-------
Estimated elasticities tend to vary between -.75 and -1.2.
Consequently, it is assumed for the purposes of this project
that a equals unity.
In order to calculate KN for 1975, the following proce-
dure was applied:
• KN was computed for the years 1970 through 1973 (the
years for which consistent data on Los Angeles and
Orange Counties is available) according to the formula:
NC.PN. = KN,
C "C £
which is derived from Equation 49.1 with a equal to one
• The average rate of growth of KN was calculated (equal
Z-
to 2.4 percent).
• This growth factor was applied to determine KN. in 1974
and 1975.
In order to determine the effect of new car demand on
sales, it was further assumed that the supply of cars to Los
Angeles at the market clearing price is perfectly elastic.
Thus, in the absence of any additional ownership costs
imposed by pollution control policies, Equation 49.1 is
sufficient to project 1975 new car sales in Los Angeles,
given a value for P:V.
Used Car Price
The following demand/supply relationships are used to
develop the used car price equation:
DU. = PU~*KU. (50.1)
SU. = T + . (50.2)
(• U — 1
where DU. = demand for used cars in period t;
£
230
-------
KU, = variable combining all other factors determining
t
used car demand in period t-,
SU - supply of used cars in period £;
t
6 = assumed constant equal to the elasticity of used
car demand.
As discussed in Chapter 2, the used car market is viewed
as one where all existing autos have a price or opportunity
cost to the owner. Some used cars are sold in a well-defined
used car market where the transactions determine the market
price for used cars, and consequently, the opportunity cost
associated with owners not selling their cars. The market
clearing condition for this market is:
DU. = SU.
Is ~C
Substition from Equations 50.1 and 50.2 gives:
PU~.BKU. = T. .
t t t-1
which upon rearrangement yields the following price deter-
mination identity:
T
t-1
In order to determine a value of 8, it was assumed that
the price elasticity of used car demand was equal to the
price elasticity of new car demand. This assumption is
partially verified by a study of the demand for new and used
automobiles by Gregory Chow.1 Thus, $ is set at unity.
KU for 1975 was calculated by solving the equation:
231
-------
putTt-l = Kut
T (the stock of cars in Los Angeles and Orange Counties
"C*"* JL
in 1974) was computed by simulating the model described by
Equations 49 through 52. Data on the first nine months of
1974 indicate that the average used car price increased by
5 percent over the 1973 price. A further increase of 8 per-
cent over 1974 was assumed for the 1975 price. With these
estimates and assumptions about the 1974 stock of used cars
and the 1975 price of used cars, KU was calibrated for 1975.
Scrappage Rate
A scrappage equation, originally estimated on national
data, was adjusted for application to the Los Angeles region,
The estimate of national scrappage rates was:
= L.( 5. 5234)
"C
NC
t
. 7424
PIU
t
PMU
t
-. 912142
(51.1)
where PIU . = national consumer price index for used cars
Is
in period t;
PMU, = national price index on auto maintenance and
"V
repair in period t',
L. = scrappage rate owing to the effects of aging
c
alone (discussed in more detail in the
following section).
*See The Effects of Automotive Fuel Conservation on Automotive Air
Pollution. Report for the Environmental Protection Agency.
Charles River Associates Incorporated. Cambridge, 1976.
The estimated equation is based on a specification which
appears in: Walker, Franklin V. Determinants of Auto
Scrappage. Review of Economics and Statistics. 503-506,
November 1968.
232
-------
All t-statistics on parameters are significant at the 1 per-
cent level of confidence; the standard error of the estimated
equation is 11.1 percent; the coefficient of determination
(R2 corrected) is .6346.
When data for Orange ;and Los Angeles Counties on new car
sales and stock of autos was substituted into Equation 51.1,
it was found that the resulting estimates were consistently
10 percent higher than the actual scrappage rate. it was
inferred from this that the estimate of scrappage owing to
age alone (L ) was higher for the rest of the country than
~c
for the southern California region. Though no research was
performed to determine why this should be the case, one
possible explanation is that both the climate and road con-
ditions in the Los Angeles area are more benign toward auto-
mobiles than would be the case in more northern regions of
the country.
This result led to an adjustment in the equation which
predicts scrappage owing to age alone. A multiplicative
constant factor was calculated which gave the best fit be-
tween observed and predicted scrappage rates over the four
year period covered by the data. The value of the constant
factor is .9182.
The indices for used car prices and maintenance/repair
costs were also changed so that the average price of used
cars (rather than the price index) was isolated in the equa-
tion. The following equation shows the relationship between
the 1975 average price of used cars and the indices used in
the estimated scrappage equation:
PU19?5
PIU19?5
233
-------
where b is the base year in formulation of the used car
price index. In order to isolate PU . in the scrappage
equation, it is necessary to have a value for PMU ,. In
1 y i o
the absence of any recent data, it was assumed that PMU
increases at the same annual rate as PU over the two year
period from 1973 to 1975. Using this assumption, the
following relationship was computed:
-.9121
(n,, ^-.9121\ 1 }-.9121
PMU2975
1975 1558.0817
Making the above substitution into Equation 51.1, and multi-
plying by the adjustment factor for scrappage owing to age
alone (.9182) yielded Equation 51 in Table 66.
Identities
The auto stock changes over time according to the
following identity:
T.=T..+ NC. -S, (52)
Z* t — J. C £
This equation assumes that there is no net change in the auto
stock owing to migration of auto owners or the import/export
of used cars from the region by other means.
The total scrappage of used cars in the region is
determined by:
St =- "t^t-j) (53)
234
-------
Scrappage Rate by Age of Auto
Using a logit specification, Franklin Walker2 estimated
a scrappage rate by model year equation on national data:
m. = 7/?77,- (54.1)
^ 3.8032 + 846. 36e ' ' °'J'
where i is the age of the auto model. All standard errors
on parameter estimates are less than 5 percent; the standard
error of the estimate is .29 percent; and the coefficient
of determination (R2 corrected) is .9994.
As discussed above, when the results of this equation
were used in scrappage rate Equation 51.1, the estimated
scrappage rates were higher than those observed in the Los
Angeles area. Consequently, it was assumed that the scrap-
page rate owing to age alone, calculated from Equation 54.1,
was overpredicting for Los Angeles, and an adjustment factor
was calculated. Applying this factor to Equation 54.1 yields
Equation 54 in Table 66.
APPLICATION OF THE MODEL TO POLICY SCENARIOS
Chapter 4 presents the results of simulating the auto
stock model under assumed policy scenarios. This section
of the appendix briefly discusses the use of the model under
various policy assumptions.
New Car Sales
It is assumed that the extra costs of ownership associ-
ated with emissions taxes do not add to the customer's
utility in owning the car. The consequences of this assump-
tion is that the demand for new cars can be represented by
substituting into the demand equation price minus the present
value of future taxes. In equation form, the sales of new
cars can be computed as follows:
235
-------
NC. = (PN. - PC,}~1(1.SSS)109 (49.2)
£ t C
where PC is the present discounted value of future costs of
c
ownership associated with pollution control strategies (and
is equal to zero in the base case).
Stock of Used Cars
The same reasoning applies to the demand for used cars.
However, because the stock, or supply, of used cars is
initially fixed, the first reaction to the increased costs
of ownership will be a decline in the market price of used
cars equal to the amount of the discounted present value of
future taxes. As explained in Chapter 2, this decline in
average used car prices will cause the scrappage rate to
increase and, consequently, the stock of used cars will
decline. The lower supply of used cars will then cause an
increase in the used car price. The system of changes in
used car price, scrappage rates, and used car stocks will
continue until a new equilibrium is reached.
In order to determine the new equilibrium, Equations
50 through 53 are iterated with PU -PC substituted into
£ ~t>
Equation 51 in the initial iteration.
Scrappage Rate by Age of Car
Emissions taxes will vary in amount depending upon the
age of the car. Generally, older autos will have higher
pollution rates and larger taxes. (This generalization is
not necessarily true for NOx. Precontrolled cars tend to
have lower rates than early vintages of HC and CO controlled
cars.) Also, the market value of older cars is less, and
hence the emissions taxes will be greater in proportion to
market price taxes for older cars.
236
-------
In order to distribute the effects of emissions taxes
over scrappage rates by age of auto, the following approach
was used. First, the following assumption was made:
± A, _ i PC. ,,5.
L. ~ p PU. ~ PU ( '
^ ^
where L . = scrappage rate due to age alone for auto of age £;
7"
r = average scrappage rate for whole auto stock;
PC . = present discounted value of taxes for auto of
I*
age i ;
PU. = average used car price for auto of age i;
Z-
PC ~ average present discounted value of taxes for all
used cars;
PU = average used car price.
In other words, the above relationship states that the
difference between the percentage change in scrappage rate
for an auto of vintage i, LL./L., and the percentage change
i, i,
in the aggregate scrappage rate, Ar/r, is equal to the dif-
ference between the policy costs as a percentage of used car
price for vintage i, PC./PU., and policy costs as a percen-
't' 7*
tage of the average used car price, PC/PU. Equation 55
gives quantitative form to the notion that scrappage rate
changes will be higher for those automobiles which carry
the greater tax burden.
Equation 55 cannot be used directly because data do not
exist on the market price of used cars by age of auto.
However, the theory behind the scrappage model suggests that
used car prices are inversely related to their scrappage
rates caused by age alone. Equation 56 incorporates this
assumption:
237
-------
PU .
That is, the ratio of used car prices by age (Pi/.) to average
Lf
used car prices (PU) is equal to the inverse of the ratio of
used car scrappage rates (L .) to the average scrappage rate
t'
(r).
Solving Equation 56 for PU . and substituting the result
Lr
into Equation 55 yields the following formula for calculating
the proportionate change of scrappage rates owing to an
emissions tax:
ALi Ar PC PCi
~T = ^ ~ j% + r (57)
Li P ^U -^-(PU)
Li ,
i
The resulting scrappage rate by vintage of car is then
calculated as follows:
L*. = L . (1 + AL./L J (58)
11 11
where L* is the new scrappage rate for autos of age i after
If
the imposition of a tax.
All the data necessary to calculate Equation 57 come
from simulation of the model represented by Equations 49
through 54. Ar is determined after all used car demand/
supply interaction effects have been calculated. The methods
for estimating PC and PC . are given in Chapter 4 where the
t»
effects of emission taxes on the stock of autos are presented.
238
-------
List of References, Appendix B
Gregory C. Demand for Automobiles in the United States.
Amsterdam, North-Holland Publishing Company, 1957.
2Walker, Franklin V. Determinants of Auto Scrappage. Review
of Economics and Statistics. 503-506, November 1968.
239
-------
Appendix C. BUS COST MODEL
This appendix derives the method used in Chapter 5 for
estimating the change in operating and capital costs of bus
services. The method uses the following:
• A cost model to estimate long-run operating costs of
conventional bus service and capital costs of conven-
tional bus, express bus and minibus service. The
operating cost model is defined as a function of the
number of bus miles and bus hours operated, as well as
the number of vehicles maintained for service.
• Functional relationships defining, for alternative
services, the number of bus hours and bus miles
operated as well as the number of vehicles maintained
for service. These output variables are defined as
a function of the headway, route length, and average
vehicle speeds along the routes.
The incremental cost due to a change in service is esti-
mated by using the cost model and the operating relationships
describing output. In general, the system costs before and
after the service change can be estimated using the cost
model, the system output variables before the service change,
and the change in variables due to the service change. The
cost model is described in the next section; output vari-
ables are defined in a separate section; and a method of
calculating incremental costs by using the model and vari-
ables is defined in the final section.
240
-------
ESTIMATES OF OPERATING AND CAPITAL COSTS FOR BUS SERVICE
To estimate the long-run incremental costs of alternative
bus services in this study, estimates of total operating
costs and capital costs of conventional bus are used. We
also present the costs of minibus and express bus, though
such options are not considered explicitly in the policy
scenarios in Chapters 3 and 5. The total operating costs
include the following cost categories, defined in the ATA
Transit Operating Report: equipment, maintenance and garage,
transportation, station, traffic and advertising, insurance
and safety, administration and general, operating taxes and
licenses. The capital costs for equipment and shop and yards
are represented. Estimates of the capital costs of CBD ter-
minals and stations are not estimated, since the service
improvements considered do not affect these costs.
Operating Costs
The total operating cost of conventional bus service is
defined as a function of annual bus miles, annual bus hours,
and the average number of vehicles available for service.
All components of operating costs except fuel and oil for
conventional and express bus are estimated using the same
cost model. As a result, the average cost per bus mile will
differ among modes principally as a result of differences in
the speed of operating the various services.
The operating cost model for the conventional bus was
calibrated using data on conventional bus operations. The
operating cost model in 1973 dollars is as follows:
TOC77 = 1.15845(7.14788H + .00000109H2 + .091647M + 2739.02U)
f O
(59)
where TOC^,7 = total operating costs in 1973 dollars;
/ o
H = annual bus hours operated;
M = annual bus miles operated;
U = number of buses owned available for service.
241
-------
The method used to estimate the cost model parameters for
conventional bus service is described below.
The costs of bus service are modeled in this study using
the functional form of the model for conventional bus service
*
selected in a prior CRA analysis of bus costs. In particu-
lar, for cost categories analyzed in the previous study,
cost/output relationships selected in that study were used.
The parameters of the relationship were estimated for this
study using regression analysis and 1973 data of a cross-
section of bus services. This approach reflects the quite
different rate of cost increase in 1967 and 1973 for various
cost categories. For cost categories not analyzed in the
previous study, simple cost/output relationships were
selected and the parameters estimated using 1973 data.
The data used in this study to estimate parameters for
the operating cost model were taken from the 1973 ATA Transit
Operating Report. Data for a sample of 24 firms reporting
in Part II of the report, Motor Bus Transit System, were
used. This sample included all firms reporting in that sec-
tion whose data had the following three properties: the
cost categories were ICC rather than ATA cost categories;
diesel fuel was at least 99 percent of all fuel used in the
operation; all cost and operating statistics categories used
in the analysis were reported.
Ten cost categories and three types of operating sta-
tistics were used to define the cost model. The ten cost
categories are combinations or subdivisions of the cost
Conventional bus costs for 1967 were analyzed by CRA as
described in: Mack, Ruth, et al. Chapter 4. In: Urban
Transportation and Recreation. Summary and Import. New York/
Institute of Public Administration, July 1970.
242
-------
categories presented in the ATA Transit Operating Report.
The functional form used for each of these cost categories
and the parameters estimated using 1973 data are shown in
Table 67; t-statistics are presented in parentheses.
The cost/output relationship for cost categories 1, 3,
4, 5, 8 and 10 in Table 67 are identical to those selected
in the CRA analysis of 1967 bus costs. Of the remaining
cost categories 2, 6, 7 and 9, categories 2 and 6 were ana-
lyzed in the previous study. The forms used for categories
2 and 6 in this study are simpler than those used in the pre-
vious analysis. They were estimated satisfactorily without
a weighted refression by assuming a proportional relationship
between cost and hours.
Traffic and advertising expense plotted against any of
the output variables show considerable random variation.
Bus mileage was chosen as the output variable in this study.
The proportional relationship shown in Table 67 was selected
since the constant term in a regression using a linear form
was not significant.
Operating taxes and licenses were expected to vary with
the number of active buses (buses maintained in condition for
service) chosen as the output variable. For this category,
*The operating cost categories were: ATA Line Entry (1973)
1. Repairs to revenue equipment 5
2. Tires and tubes 6
3. Miscellaneous maintenance 4-5-6
4. Wages (drivers and helpers) 8
5. Fuel and oil 9+10+11
6. Miscellaneous transportation 7+12-8-9-10-11
7. Traffic and advertising 13
8. Insurance and safety 14
9. Operating taxes and licenses 19
10. Administration and general 16
The output variables:
1. Bus miles 36
2. Bus hours 39
3. Buses active 33
243
-------
Table 67. COST ESTIMATION EQUATIONS FOR 1973
Operating costs
1. Repairs to revenue
equipment
2. Ti res and Tubes
3. Mi seel laneous
maintenance
4. Wages (drivers
and helpers)
5. Fuel and oi 1
6. Mi seel laneous
transportation
7. Traffic and
advert! si ng
8. 1 nsurance and
safety
9. Operating taxes
and 1 [censes
0. Administration
and general
Variables and
parameters
2. 08306 (HR)
(20.390)
0. 00957 19 (MI)
(16.683)
0. 0820747 (MI)
(15.402)
4.??097(HR) + 0.00000108836(HP2)
(9.636) (2.904)
0. 464492 (HP)
(13.031)
0. 829355 (HR)
(15.569)
0. 0122232 (MI)
(8.166)
10 67. 7 8 (NBA)
( 1 1 .817)
16 71. 24 (NBA)
(1 1 .585)
0. 158845 (EX)
(14.726)
R2
0.8905
0.8190
0.8402
0.9710
0.7839
0.837!
0.6183
0.7126
0.7379
0.8181
NBA = number of active buses
MI = bus miles
HR = bus hours
EX = total operating expenses exclusive of administrative and
general costs
t-statistics are in parentheses
244
-------
a linear regression yielded a constant term that was not
statistically significant. Therefore, the form chosen was
the proportional relationship between costs and the number
of buses shown in Table 67.
These costs are aggregated into the total operating cost
function in 1973 dollars shown in Equation 59. The cate-
gories one through nine in Table 67 are summed and the coef-
ficients of each output variable are multiplied by one plus
the coefficient of total operating expenses used to estimate
category 10.
Express and Conventional Bus — The primary difference between
the operation of conventional and express buses is that
express buses operate at higher speeds. Therefore, if the
functional forms used to estimate costs for conventional
buses are not affected by changes in the speed at which buses
are operated, then the conventional bus cost model can be
used to estimate the costs of express buses.
In particular, we assume that categories 1, 4 and 5 —
repairs to revenue equipment, wages, and miscellaneous trans-
portation -- are a function of the bus hours of service
independent of the speed. Similarly, we assume that cate-
gories 8 and 9 (insurance and safety, and operating taxes
and licenses) are a function of the number of buses used.
Therefore, for express bus service the average cost per mile
in these categories will be slightly less than for conven-
tional bus service. These assumptions are consistent with
*
the approach and costing methodology used in the IDA study.
*See Institute for Defense Analyses. Evaluation of Rail 'Rapid
Transit and Express Bus Service in the Urban Commuter Market. p. A-7.
The average cost per mile for conventional bus service is
allocated to express bus on the basis of hours of bus ser-
vice per mile consistent with the treatment of these three
categories here. Also, a comparison of the category which
includes repairs to revenue equipment for conventional and
intercity bus service supports the decrease per mile in the
category.
245
-------
In addition, we assume categories 2, 3 and 7 (tires and
tubes, miscellaneous maintenance, and traffic and advertising)
all vary with bus miles independent of speed. That is, the
cost per mile in these categories would be the same for
express and conventional bus.
However, category 5, fuel and oil, calibrated using data
which reflect average speed of conventional bus operations,
will overestimate the fuel and oil expenses of express buses.
Therefore we have calculated factors to adjust the estimates
of fuel and oil costs for conventional bus to estimates for
express bus. In particular, the fuel and oil costs per mile
for express bus are obtained by multiplying the fuel and oil
costs per mile for conventional buses by the following
factors:
• .61 for express bus operations averaging 30 miles/hour;
« .51 for express bus operations averaging 40 miles/hour.
These factors were calculated by dividing average bus
fuel consumption at speeds of 30 (or 40) miles per hour by
*
average bus fuel consumption at 10 miles per hour. The
average bus speed in the sample of bus firms used to cali-
brate the conventional cost model was 11 miles per hour.
(Note that this includes bus wait time.) Therefore, the
operating costs for service including express bus are
estimated as:
TOC77 = 1.15845(6. 68338H + .464492(H^ + f PH „)
/ 6 ^ ^ t &
+ .00000109H2 + .091647M + 2739.02(j) (60)
*The data on fuel consumption were taken from Table 23 in;
.">:.z?2.':~:-:.<:~'l2S rf '-:>b^*: ~:ra>\sf:r>tati'*: S^s's^s. U.S. Department
of Transportation. Washington, D.C. May 1974.
246
-------
where H, M, U = same as defined for the conventional bus
model;
Hr =. the annual bus hours operated in conventional
0
bus service;
H = the annual bus hours operated in express bus
service;
/„ E the factor: .61 for buses averaging JO mph
L
.51 for buses averaging 40 mph.
Minibus and Conventional Bus — Recent experience in opera-
ting minibuses indicates that the vehicle operating and main-
tenance costs are approximately the same as those of conven-
*
tional buses. To our knowledge no model of minibus costs
as a function of output variables has been calibrated. How-
ever, initial OCTD (Orange County Transit District) operating
costs in the first year of operation (February 1973 to Janu-
ary 1973) were approximately $1.00 per mile. OCTD indicates
that the wage rate for dial~a-ride was no greater than three-
fifths of the wage rate for their fixed route system. The
actual average operating cost per bus mile is calculated at
$1.27 per mile using data comparable to that in the 1973 ATA
Transit Operating Report for SCRTD. When the labor costs in
ATA category 8 are adjusted downward by three-fifths, the
average cost per mile is reduced to $1.04, approximately the
cost experienced by OCTD.
Therefore it can be assumed that the average operating
cost per mile for minibuses is the same as the average cost
per mile for conventional bus when the service is operated by
the same management. If the service is operated as a taxi
service with lower labor rates, then the labor costs are
adjusted to represent three-fifths of the labor costs.
*Conversation with N. Wilson at MIT and P. Conway at SCRTD
(Southern California Rapid Transit District). At present,
maintenance costs are slightly higher than conventional bus
and offset any savings in other operating costs.
247
-------
In 1973, SCRTD's average cost per mile was about $1.27
based on published figures. Using the operating cost model
in this study for the same volume, we estimated $1,36 per
mile. Correspondingly, when the labor rate is adjusted down-
ward, we obtain an average cost of $1.04 per mile using pub-
lished figures and a cost estimate of $0.98 per mile using
the model. (Note that the nonlinear form of category 4 is
probably reflecting the difference in labor rate (unionized
vs. nonunionized) between large and small firms.)
Capital Costs
Several types of capital costs can be affected by
changes in the service offered. These are equipment costs,
related costs in shop and yard, costs of CBD terminals and
stations, and subsidiary costs. For the service changes
considered in this study, the only capital costs affected
will be equipment costs and costs in shop and yard.
Equipment Costs — Annual capital costs for bus in 1972 and
1974 dollars are presented in Table 68. These costs are
developed from two sources. One is average purchase costs
in 1972, published in the IDA study:1
Bus Type Passengers Purchase Cost
Conventional 50 $43,000
Express 50 48,000
Mini 19 14,000
The other is bus costs of the type used by SCRTD at the
beginning of 1975. These are as follows:
Bus Type Passengers Purchase Cost
Conventional 50 $62,000
Mini 19 32,000
248
-------
Table 68. APPROXIMATE PURCHASE AND ANNUAL CAPITAL
COSTS IN 1973 DOLLARS
Purchase Costs
Bus type
Conventional
Express
Mini
IDA source average
$43,331
48,370
14,108
SCRTD source
$58,050
48,370
29,961
Selected
$58,000
48,370
30,000
Annual Capital Costs in 1973 Dollars
Bus type
Conventiona 1
Mini
IDA source average
$5,689
2,644
SCRTD source
$7,622
5,615
Selected
$7,622
5,615
249
-------
0.1313
0.1468
0.1874
10
10
10
0
0
0
15
12
8
SCRTD notes that while there are less expensive minibuses
on the market, the more expensive models are cost effective
when maintenance costs are considered.
To calculate the annual cost the following capital
recovery factors were used:
Bus Type CRT Interest Rate Residual Rate Life
Conventional
Express
Mini
The figures for 1973 were obtained by using the following
Wholesale Price Index — Motor Buses2 applied to the capital
costs:
Average Annual
Year Wholesale Price Index
1971 115.0
1972 116.8
1973 117.7
1974 125.7*
Shop and Yard — For the estimate of investment in shops
and yards required to support each bus owned, we use the
same procedure as that cited in the IDA study (that is, the
1964 estimate of investment of $4,500 per bus used in the
Meyer, Kain and Wohl study, "The Urban Transportation Prob-
lem"3). The IDA study adjusted these costs to 1972 dollars
giving an estimate of $6,100."
These costs were then put on an annual basis assuming
a life of 50 years, no residual value, and an interest rate
of 10 percent. The corresponding capital recovery factor
is 0.1009.
*Based on the first eight months of 1974.
250
-------
The consumer price index for 1972 was 125.3; for 1973,
133.1. Therefore, the estimate of investment in shop and
yard used for 1973 is $6480, and the estimate of 1973
annualized cost is $654.
CHANGES IN BUS SERVICE OUTPUT VARIABLES RESULTING FROM
CHANGES IN SERVICE CHARACTERISTICS
General Approach for Conventional Bus Service
The changes in transport services considered in this
study include changes in departure frequency and routes.
Ideally, the service change would be specified so that the
total annual change in the variables of the cost model
(annual bus hours, annual bus miles, and number of vehicles
available for service) due to introducing the alternative
can be calculated. The cost model can then be applied to
calculate the incremental change.
In particular, the following variables should be
specified for each bus route or mode introduced or removed:
v = vehicle type — conventional, express, or
minibus;
i7ji0...i = labels of the bus stops on the route;
J. Ci ft
m. . = the distance (in miles) between bus stop i .
^ i ^ n-, 7 3
<* d and i. for all j;
v
ss . = the average speed (in miles/hour) in the peak
J J hours between stops i . and i. ,, including the
3 J + 1
effect of acceleration and deceleration. For
off-peak hours, the average speed is repre-
sented as Si .i- . ;
P C 3 "*"•*
u\ = the average stop time (in hours) in peak hours
3 at stop i. for loading and unloading. For
3
off-peak hours, the average stop time is repre-
sented as w°. ;
251
-------
Irr = the average headway or time between buses
(in hours) in peak hours on route r. For
off-peak hours, the average headway is repre-
sented as h ',
dt . = the average time at origin i between trips.
o
The changes in annual bus miles, bus hours, and active
buses expected to result from a proposed alternative are
calculated from this data as described below. As the first
step, the bus miles per trip, bus hours per trip and number
of scheduled buses per hour are calculated for peak and off-
peak service for each bus route or mode introduced or removed,
The variables for each trip are calculated as:
mt Bus miles per trip on route r> where:
r>
mtv = I
m .
jcr l'j j + 1
Bus hours per trip on route r in peak hours:
ht^ = y —"•—"• + y u . + dt .
r . v . i . i
J£" y3 Q . TPyT /o
t. / fa . • JC./ J U
ht° Bus hours per trip on route r, off-peak hours:
m .
ht° = y —L.A+L + y w° + dt.
K* Q It If
jer a. • jer j o
V * (* • M
3 J + 1
252
-------
•IP
provide service on route r:
Buses required per hour in peak hours to
U Buses required per hour in off-peak hours to
provide service on route r:
Annual variables are then calculated for route r as
follows:
AM Annual bus miles on route r is defined as:
mt kp mt kc
Y>
AM = — *- - +
h°
where k^ is peak hours per year;
k° is off-peak hours per year;
other parameters are defined above.
AH Annual bus hours on route r is defined as
AH = kPUP + k°U°
where parameters are as defined above.
253
-------
NBA Number of active buses servicing route r:
where N is the ratio of active buses to
buses used in peak hours.
If the transport alternative consists of the addition
A
of several routes, say R , and the deletion of several
p
routes, say P , then the incremental annual bus hours,
miles, and buses are defined as shown below.
The change in annual bus miles is defined as:
AM - AM
- -
The change in annual bus hours is defined as:
HAH = I AH - I AH
DA ^
rcR re/?
t\NBA The change in the number of active buses is:
LNBA = I NBA ~ £ NBA
rtPA r —D*
Ratio of Active to Peak Buses — The ratio /V of active to
peak buses used in this analysis is 1.25. The average of
the ratio of active to peak buses in the sample used to
estimate bus operating costs is 1.28. The average of the
ratio of the largest ten companies in the sample was 1.22.
Therefore, as an approximate value we selected 1.25.
254
-------
Calculation Method of Bus Outputs for Conventional Bus
In practice, the only improvements considered were
direct service for certain segments of routes. Possible
interconnections of routes, which can have some effect on
costs, were ignored. Two sets of output variables were cal-
culated. One represents the bus miles, bus hours, and buses
required for the improved service for one-way trips. The
other represents the output when the bus must make the
reverse trip as well. For each improved segment the
following were specified:
m. . segment distance;
^l^2
tt . . trip time for direct (improved) service
^ l^ 2
from pickup to delivery point;
h • • headway in peak hours for the improved
12
service on this route;
h . . headway in off-peak hours for the
t ^ 2
improved service on this route.
The following output variables are calculated for one-
way trips on the segment:
mt . . one-way bus miles per segment trip are
1 11 2
equal to m . . ;
ht . . one-way bus hours per segment trip are
tl12
equal to tt . . ;
255
-------
rt U Buses required per segment are equal to:
ht . .
per peak hour;
ht . .
I ^ c,
U° = per off-peak hour.
°
Then annual output variables per segment due to the
new service are calculated as:
AM Annual bus miles on route r:
r
A \A —
fifl —
r
mt . . mt . .
11 11
7? 7 ?
f "* n ) i f i n \
p Q
•7* -7* *j
-------
The change in annual output variables for all improve-
ments are then given by:
= IAH
L
ESTIMATES OF INCREMENTAL COSTS OF CHANGES IN BUS SERVICE
General Method for Conventional Bus
In theory, the incremental cost of changes in bus ser-
vice is calculated by taking the difference of the cost under
the previous and changed service. Using the cost model in
the previous section, the change is defined as:
&TOC?2 = 1. 15845(6. 68338(H*-H) + . 464492 [( H £-H c)
+ fE(HE-HEn + .00000109((H2)*-H2]
+ .091647(M*-M) + 2739.02(U*-U)}
where H* , H*, H*, M* , U* are the annual output of bus hours,
L Ci
bus miles, and active levels after the service change; H,
H~, H.,, M, U are the annual output of these variables before
C D
the service change; and kTOC _ is the change in operating
cost. This can be written in a simpler form in terms of the
change in each of the output variables, A#, A#6,, A#£, and
Ai/ as follows :
= 1. 1-5845(6. 68338(bH) + .464492(&H + /^.AtfgJ (61)
00000209[&H(2H + kH)\ + .09164?(kM) + 2729.02(kV)}
257
-------
The change in annual capital costs for conventional
(and express) bus resulting from the type of changes intro-
duced in this study is due to increased purchase of equipment
and expansion of shop and yard. Therefore the changes in
annual capital costs for conventional and express service
(using the data in Equation 61) is described by the following
equation:
&TOC?2 = ($7615 + $654}kU (62)
where A£/ is the change in the number of active buses.
Calculation of Incremental Costs^ of Bus Service Changes
The changes in bus service considered in this study
require additions to service. Therefore, the changes in
output variables (A#, A// and Atf) are equal to the annual
hours, routes and buses required to provide additional
frequency on new routes. Thus these can be calculated by
the method described in the previous section.
An adjustment in the model is necessary to make it
equivalent to 1974 experience — the base case year for
travel demand forecasts in Chapter 3. A 10 percent infla-
tionary increase in costs is assumed between 1973 and 1974;
this is accounted for by increasing the multiplicative
constant in Equation 59 by 10 percent.
258
-------
List of References, Appendix C
JU.S. Department of Transportation, Institute for Defense
Analyses. Evaluation of Rapid Transit and Express Bus Service.
Washington, D.C., Government Printing Office, October 1973.
p. 27.
2U.S. Department of Labor, Bureau of Labor Statistics.
Wholesale Prices and Price Index. Washington, D.C. , Government
Printing Office.
3Meyer, j. R., j. p. Kain, and M. Wohl. The Urban Transportation
Problem. Cambridge, Harvard University Press, 1966.
11 U.S. Department of Transportation, Institute for Defense
Analyses, op. cit. p-. A-32.
259
-------
Appendix D. MODELS AND DATA
TRAVEL DATA
The only comprehensive primary data describing travel
in the Los Angeles region is the Los Angeles Regional Trans-
portation Study (LARTS) Origin-Destination Survey conducted
in 1967. LARTS, organized within the California Division of
Highways and under the surveillance of the Southern Cali-
fornia Association of Governments (SCAG), was charged with
developing a comprehensive transportation plan for Los
Angeles, Orange, Ventura and parts of Riverside and San
Bernardino Counties. Approximately 9,000 square miles fall
within the LARTS study area. Because all LARTS travel data
refer to the year 1967, this year was taken as the base year
for development of the demand model.
Two surveys were conducted by LARTS during the 1966-67
period. The major survey was the Home Interview Origin-
Destination Survey of a 1 in 100 sample of households within
the study area. Data from this survey consist of 33,030
records of households interviewed and 221,895 records of
individual one-way trips made by these households. The
particular information items from the household interviews
which we have incorporated into our research are discussed
in detail in this appendix.
The Home Interview Survey forms appear in Figure 4.
For each household, one housing unit form was completed
together with as many trip interview records as were neces-
sary to describe all of the trips made the day before
260
-------
Figure 4. Excerpt from LARTS Travel Interview Form
to
Figure continued on following page.
-------
Figure 4 (continued). Excerpt from
LARTS Travel Interview Form
I •' I
u •
c-
N = I
»;»< '.••[
Y N
N I
N = I
.-'-.A
-E
262
-------
by all occupants of the housing unit. These forms are
referred to in subsequent discussion of the specific data
items.
LARTS also conducted a roadside interview survey at
various points along a cordon encircling the study area in
order to obtain data concerning travelers who do not reside
in the study area as well as to check the accuracy of some
of the home interview data. All in all, there were 47,130
roadside interviews. In addition, LARTS took actual counts
of vehicular traffic crossing two selected screenlines and
passing through each of seven corridors. Because LARTS has
not released the actual records and/or counts from these
roadside interviews and ground counts, particular items
from these interviews are not included in our data analyses.
However, LARTS has compared the results of the Home Interview
Survey counts and external counts and has calculated and
released adjustment factors to be used when dealing with
data from the Home Interview Survey; we have used these
factors to adjust the trip counts estimated by the Home
Interview Survey.
Zonal System
In our analysis we divided the LARTS study area into
zones and organized all data with respect to these zones.
We chose our zonal system from previously existing LARTS
breakdowns. The records from the LARTS survey were origin-
ally coded and tabulated using a system consisting of 1,246
traffic analysis zones (AZ's). These AZ's were primarily
based on aggregations of 1960 census tracts. In 1970 LARTS
revised its zonal system into a new arrangement of 1,285
AZ's based on 1970 census tracts. These AZ's were further
aggregated to 107 regional analysis zones (RAZ's) for
sketch planning. Because of census tract redefinitions, the
1967 and 1970 analysis zones are considerably different.
263
-------
Since the Los Angeles study area is so large, the size
of an average 1970 RAZ is on the order of 25 square miles,
with some in less densely populated areas being considerably
larger. Applying transportation system data to these zones
must be done with great care since much information is lost
in aggregations of this size. However, using a larger number
of smaller zones exponentially increases the number of zonal
pairs and makes the data base too cumbersome for in-depth
analysis. Consequently, the 1970 RAZ's were chosen as the
zonal system to be used in this study.
In order to convert data referring to the 1,246 original
1967 AZ's to the 107 larger 1970 RAZ's, it was necessary to
prepare a zone dictionary. This dictionary was prepared
from maps of the respective zone systems provided by LARTS.
A considerable number of the 1967 AZ's were split when the
zone system was revised in 1970. The zone dictionary
accounted for this by making fractional allocations (to the
nearest tenth) of split AZ's based on land area.
Seven of the original 107 zones have been excluded from
our analysis; these lie to the extreme north and west of
the major metropolitan areas which are normally thought to
comprise Los Angeles; they are also separated by natural
boundaries and long driving distances.
Trip Counts
The number of interzonal round trips, the unit measure
of the dependent variable, was obtained from the 1967 LARTS
household survey. Our total data base consists of 100 zones,
or 10,000 zonal pair observations.
Trip Purpose and Coding of Round Trips — LARTS recorded
each one-way leg of a trip separately, associating it with
a purpose defined by land use at the origin and destination
of the trip. The different land uses that were specified
were as follows (see questionnaire): work place, related
264
-------
business, home, social or entertainment, recreational, shop-
ping, educational, and other. This allows for 64 actual trip
purpose categories, i.e., any combination of land uses at
origin and destination. The number of categories was reduced
in two ways. First, since the relevant unit measure of
trips was the round trip, a method was devised to identify
each one-way trip leg as the initial or return portion of
the round trip, using assumptions listed in the following
pages. Trips in either direction between any two land uses
were then grouped into one trip purpose category, e.g.,
home-shopping. This automatically halves the number of
trip purposes.
In order to make certain that this approach to classi-
fying one-way legs as round trips was valid, we took a 1 in
500 sample of all trip records and checked to see whether
the number of trips between each combination of the eight
LARTS land use purposes was approximately the same in each
direction. The results of this exercise assured us of the
appropriateness of our approach. For example, there were
51 home-to-work one-way legs captured by the sample and 50
work-to-home one-way legs; 26 home-to-shopping one-way legs
and 30 shopping-to-home one-way legs; and 40 home-to-other
one-way legs and 39 other-to-home one-way legs.
The number of trip purpose categories was further
reduced for the purpose of modeling by aggregating land use
purposes when the number of trips in a particular land use
was small and when that land use could conceptually be
grouped with others.
The final set of trip purposes as used in our analysis
is as follows:
1. home-work (includes home-related business)
2. home-shopping
3. home-social, entertainment, and recreation
4. home-education
265
-------
5. home-other
6. home-home
7. work (or related business)-work (or related business)
8. other-work ("other" here refers to any non-home and
non-work land use)
9. other-other (i.e., non-home, non-work-non-home, non-
work) •
After specifying the trip purpose breakdown, each one-
way leg was then assigned to one of the 90,000 possible com-
binations of trip purpose (9 possibilities) and zonal pair
(100 x 100 possibilities). As previously explained, the end
nodes of each one-way leg were identified as origin or desti-
nation not according to the direction of travel specified
in the LARTS survey records, but rather according to the
assumed origin and destination of the round trip of which
this one-way leg was considered a part. The assumptions that
were made to associate one-way legs with a round trip purpose
and to identify the end-points as origin or destination of
a round trip are as follows. For all trips which had home
at either end of the trip, home was considered to be the
origin of the assumed round trip. The other end of each of
these one-way legs was assumed to be the destination of the
round trip. These trips account for the trip purposes:
home-work, home-shopping, home-social/entertainment/recre-
ation, home-education, home-other, home-home. All one-way
legs which did not record home at either their beginning or
end, but had work at one end (and one end only), were
assumed to be journeys to/from work in which the tripmaker
had made an intermediate stopover along the way, such as
a shopping errand on the way home from the office. In
keeping with this logic, the residence of the person making
these trips was recorded as the origin of the total journey-
to work round trip, the workplace was recorded as the
266
-------
destination, and the secondary stopover was ignored. These
trips were stored separately in our data base as other-work
trips (purpose 8) but were included in the home-work model.
For one-way legs not included in either of these two cate-
gories, neither end node could automatically be identified
as an origin or destination of a round trip. These trips
were therefore aggregated by their recorded origin and desti-
nation.
After one-way legs were identified with a trip purpose,
the legs associated with each purpose-zonal pair combination
were counted and then divided by two to obtain the number
of round trips.
An example of a hypothetical tripmaker's trips during
the survey will illustrate how trips were aggregated by pur-
pose and associated with a zonal pair. If the person left
his home in the morning, went to work, went shopping,
returned home and in the evening went out to a movie and
again returned home, his day's trip record would be recorded
on our data base as:
1 round trip, home-work, origin of round trip = home
zone; destination of round trip = zone of work place
h round trip, work-other, origin of round trip = zone
of work place; destination of round trip = zone of
shopping place
1 round trip, home-social, entertainment and recreation,
origin of round trip = home zone; destination of round
trip = zone of theater-
The assignment of each of the 64 possible combinations
of LARTS trip purposes to the 9 round-trip purposes is sum-
marized by the matrix in Table 69.
267
-------
Table 69. ALLOCATION OF ONE-WAY TRIPS BETWEEN
LAND USES (LARTS CATEGORIES)
TO ROUND TRIP PURPOSE CATEGORIES
x. Desti-
\nation
Origin \^
Work place
Related
busi ness
Home
Social or
enterta i nment
Recreat iona 1
Shopp i ng
Educat iona 1
Other
Work place
work-
work
7 work-
work
home-
work
o work-
other
o work-
other
o work-
other
o work-
other
_ work-
other
Related
business
-. work-
work
7 work-
work
. home-
work
o other-
other
g other-
other
Q other-
other
other-
other
q other-
other
Home
. home-
work
home-
work
fi home-
home
, hcme-soc/
eat/rec
, home-soc/
ent/rec
- home-
shopp i ng
. home-
education
_ home-
other
Social or
entertainment
_ work-
other
p other-
other
, home-soc/
ent/rec
p other-
other
g other-
other
other-
other
q other-
other
g other-
other
Table continued on following page.
268
-------
Table 69 (continued). ALLOCATION OF ONE-WAY TRIPS
BETWEEN LAND USES (LARTS CATEGORIES)
TO ROUND TRIP PURPOSE CATEGORIES
>< Desti-
^\nation
Origin \^
Work p 1 ace
Related
bus! ness
Home
Social or
enterta i nment
Recreational
Shopp i ng
Educational
Other
Recreational
R work-
other
q other-
other
home-soc/
ent/rec
q other-
other
q other-
other
other-
other
other-
other
g other-
other
Shopping
work-
0 other
g other-
other
„ home-
shoppi ng
g other-
other
g other-
other
q other-
other
q other-
other
g other-
other
Educational
g work-
other
other-
other
. home-
education
g other-
other
g other-
other
other-
other
g other-
other
other-
other
Other
o work-
other
q other-
other
,- home-
other
g other-
other
g other-
other
Q other-
other
g other-
other
g other-
other
269
-------
Mode Categories — In the LARTS data, the mode of travel was
specified for each trip record (see questionnaire, column 5
of trip record). These modes were aggregated as follows:
1. auto drivers, including pick-up truck drivers;
2. auto passengers, including pick-up passengers;
3. transit trips.
Several of the modes listed in the questionnaire were not
incorporated in our model because they comprise less than
a quarter of one percent of the total trips captured by the
survey and had little relevance to the issues being consi-
dered. These include ferry, taxi, school bus, and motorcycle
trips.
A fourth mode considered in the demand model was auto
serve passenger trips; however, the LARTS definition was
different from ours. These trips were indicated in answer
to a question concerning the purpose of the trips (see ques-
tionnaire, column 7). As the questionnaire shows, in addi-
tion to defining trip purpose by land use at the origin and
destination, it asks whether the primary purpose of making
a particular stop was to assist a passenger arriving at his
destination. This question was asked separately for the
origin of the trip leg and the destination of the leg.
(Note: if, in a continuous trip linking several legs, the
destination of one leg was "serve passenger," then the
origin of the next leg would also be recorded as "serve
passenger.") Drivers who indicated that their purpose was
to serve a passenger were then asked to indicate the "'ulti-
mate purpose1 that the driver has in reaching his desired
destination."1 If the entire excursion was made by the
270
-------
driver expressly to serve the passenger, and the driver
began and ended his trip in the same place, no land-use
related purpose was recorded at the end nodes of legs identi-
fied as serve-passenger stops.
In our aggregations of the LARTS records, we linked
together all continuous trips which had serve passenger at
either end mode, and considered this to be a single one-way
trip. (Trip legs were recorded in consecutive order on the
LARTS files, and therefore linking was possible.) This trip
was then classified as serve passenger mode, and the purpose
of the trip was inferred from the land uses at the origin
of the first leg and the destination of the final leg.
It would have been preferable to be able to record the
serve passenger trips with respect to the purpose of the
passenger being served; however, this information was not
available from the LARTS records. For instance, if a wife
left her home with her husband, dropped him off at work, and
went shopping, LARTS records would show two legs:
*
1. non-serve passenger, home to serve passenger,
shopping
2. serve passenger, shopping to non-serve passenger,
shopping.
Rather than considering this half of a home-shopping round
trip and half of a shopping-shopping round trip, we
*Since the driver's purpose in being at the origin of the
entire trip, home, was not to serve a passenger but rather
for personal reasons, the origin point is classified as non-
serve passenger even though, in this exanple, the passenger
began his trip at this point.
-HARTS interviewers were instructed to consider the purpose
at the origin of a trip leg to be the purpose at the desti-
nation of the previous trip leg if it was part of the same
excursion. While this was functional in non-serve passenger
trips when considering trip purpose as defined fay actual
land use at that location, it is meaningless when considering
trip purpose as defined by the ultimate purpose of the
driver.
271
-------
considered this to be a single one-way leg, whose mode of
travel was "auto, serve passenger" and whose purpose was
"home-shopping."
It will be noted that in Appendix A we aggregated serve
passenger trips into auto driver trips, associating them
with the land use of the driver's origin-destination. There
are some VMT's that are not counted within any trip purpose,
owing to the added circuity of serving a passenger. These
are, however, estimated in the demand model. Note that the
passenger's trip was recorded in a separate trip record,
and therefore all passenger trips were captured and the
proper purpose assigned to them.
Automobile System Performance Variables
The variables listed in this section were used to
characterize Los Angeles automobile travel. Separate data
for each trip purpose were compiled for all auto mode vari-
ables.
Auto Line-Haul Time — Interzonal auto line-haul time was
averaged from the times recorded on the LARTS survey for
each purpose and mode specific interzonal one-way trip.
The times appear on a questionnaire as actual clock times
of trip beginning and trip end, from which time in minutes
was calculated. The trip times were then converted into
line-haul times by subtracting out the time spent walking
at both trip beginning and trip end, as recorded by the
LARTS survey, and then doubled to represent round trips.
Auto Trip Distance — In addition to its household survey,
LARTS coded and skimmed networks to compile centroid-to-
centroid automobile distances between each of the 1,246
1967 analysis zones. These distances represent the
shortest route from one zone to another using any combi-
nation of city streets and expressways. The distances
between the 100 larger regional analysis zones were obtained
272
-------
from these inter-analysis zone distances by weighting the
inter-AZ distances pertaining to each RAZ pair by the propor-
tional frequency of inter-AZ trips. These trip frequency
distribution figures were made available from a 1970 LARTS
estimated auto driver trip table.
Auto Costs -- 1967 automobile operating costs were estimated
as $0.03 per mile and multiplied by the interzonal (RAZ)
distances (explained above) to arrive at the average cost
for trips made between each RAZ pair.
Transit System Performance Variables
The LARTS survey included data on the performance of
the bus system, the only form of transit in the Los Angeles
region. However, due to the low level of bus usage in Los
Angeles, the number of bus trip records available from the
LARTS survey is small. A random 1 in 500 survey taken of
the LARTS records indicates that only 2.3 percent of trips
were made by bus; when expanded to the size of the actual
LARTS survey, this means that of the 220,000 trips described,
only about 5,000 were transit trips. A 5,000 trip sample
is clearly inadequate to describe the performance of the
transit system at the zonal pair level of accuracy, since
there are 10,000 possible zonal pair combinations (100 x
100) .
Instead, transit network data had to be extracted
directly from Los Angeles area bus maps and schedules. This
material was obtained for the 1967 bus systems run by the
Southern California Rapid Transit District (SCRTD), the Long
Beach Public Transit Company, the Santa Ana Transit Company
and the South Coast Transit Company. These companies pro-
vided transit service for most of Los Angeles and Orange
Counties in 1967.
273
-------
To reduce the number of zonal pairs for which transit
system data were developed, a random sample of 400 zonal
pairs was taken from the actual 10,000. Of these 400 zonal
pairs, only 172 lay within the service area covered by the
transit companies listed above. Detailed transit data were
derived for these 172 zonal pairs only. It was assumed that
for any interzonal round trip between these 172 zones, the
transit system performance characteristics of the return
trip were exactly the same as the transit data computed for
the original trip.
Description of the bus service for each leg of the
interzonal traffic was a three-step process:
• measuring accessibility to bus lines in each zone;
• identifying the bus line(s) or sequence of bus lines
that connect the zonal pair; and
• quantifying the characteristics associated with inter-
zonal transit trips by use of the schedules of the bus
lines identified.
These three steps are described in greater detail below.
Access Time — Average access time, or average walk time to
transit within a zone, was calculated separately from the
other transit system variables. Average access time in any
zone was defined as the average time that it takes to walk
to the nearest transit line. (The method used to calculate
these times is explained below.) Given this definition,
the same average access time figure could be used for all
interzonal transit trips originating or ending in a par-
ticular zone. The task of calculating access times was
involved and time-consuming, even given this simplifying
definition. Because most zones have many different bus
routes traversing them, calculating access times separately
for each zonal pair would have meant calculating them for
each possible interzonal route for all the 172 pairs, and so
would have greatly increased the task of measuring access
274
-------
times. Moreover, doing this most likely would not have
yielded significantly different access times — since walk
time is usually weighted about three times as heavily as
line-haul time in tripmakers1 perceptions, transit users are
*
most likely to use the bus line nearest them. In addition,
the Los Angeles bus system is such that usually either the
second closest bus line to any point is so far away that the
tripmaker would always take the nearest bus, or else the
second closest bus line is approximately as far as the first
closest bus line.
Given this definition of access time, average access
times were calculated by first superimposing the zonal
maps on bus route maps. Using these maps, the populated
area of each zone considered was then divided into market
areas such that each market area was served by one and only
one bus line. The market areas were defined by the layout
of the bus lines and, due to intersecting bus lines, were
each usually associated with only a small segment of that
bus line's entire route through the zone. The market areas
were often quite small, on the scale of 1 to 10 square blocks
For each market area, the average distance to the
associated bus line was calculated by hand. Ideally, these
distances would then have been weighted by the population
within the market area. The amount of detailed calculation
that this would entail, necessitating use of census block
statistics, rendered the task impossible within time and
budget limits. Instead, the distances were weighted by a
rough estimator of the population living within each market
area from the number of census tracts (usually a fraction)
within the market area. (The Bureau of the Census defines
census tracts so that they are all on the same scale, with
approximately 4,000 inhabitants.)
*Domencich, Thomas A. and Gerald Kraft. Free Transit.
A Charles River Associates Incorporated Report. Cambridge,
Lexington Books, 1970. This relationship also obtains for
the models used in this study.
275
-------
The as-the-crow-flies average distances derived in
this manner were converted to walking distances along a
street grid by a 1.28 expansion factor. A walking speed of
3.16 miles per hour was then applied to arrive at final
figures for access times.
Identifying Interzonal Routes -- Our objective in identifying
*
interzonal transit routes was to choose the route that a
transit user would be most likely to take in traveling
between any particular zonal pair. This task involved a
great deal of discretion, especially in making the trade-
offs between the different positive and negative aspects of
the various routes. As a general rule-of-thumb, it was
assumed that tripmakers perceive walk time to a bus as being
approximately three times more onerous than the time actu-
ally spent on the bus.2
Because of the large size of the zones, the most likely
interzonal route via transit varied depending on the exact
location within the zones of the tripmaker's origin and
destination. Therefore, if there were five or fewer likely
interzonal routes via transit, each route was identified
separately. If there were more than five likely routes,
interzonal line-haul times, the number of transfers and bus
headways were found for all possible routes, and then up to
five routes were chosen which together represented all the
possible interzonal routes embodying these three charac-
teristics .
The different typical routes thus identified were then
weighted by the percentage of the populated area within the
origin and destination zones respectively for which either
the typical route was the most likely one to be taken in an
*In the following discussion, "route" will refer to the total
sequence of buses necessary to complete a specified inter-
zonal trip rather than to individual bus lines.
276
-------
interzonal trip, or for which the typical route was similar
to the most likely route to be taken in an interzonal trip
(in terms of such factors as line-haul time). For instance,
in a very simple case, there may be only one bus line
passing through the origin zone and it may also be more
convenient than any nearby bus line to all points in that
zone. This same bus line runs through the destination zone;
however, a feeder line runs perpendicular to the interzonal
*
line, and people living in a quarter of the populated area
of the destination zone use the feeder line to reach the
main interzonal bus. In this case the two likely routes
would be 1) interzonal line only and 2) interzonal line,
feeder line. All variables describing these two routes
(e.g., average line-haul time, wait time, etc.) would have
been weighted by .75 and .25 respectively.
Quantification of Characteristics of Interzonal Bus Travel —
After identifying the interzonal routes and determining the
appropriate weights of these routes, the following vari-
ables were collected to describe interzonal bus transit
travel along these routes.
Line-Haul Time — Line-haul times were measured from the
mid-point in each zone of the identified typical interzonal
routes. Two different line-haul times were collected, one
for peak hours and another for offpeak hours. Peak hours
were defined as approximately 6:30 to 9:00 a.m. and 3:00
to 5:30 p.m., varying slightly with distance from the CBD.
Wait Time -- The headways for the different bus lines in
each interzonal route were used to compute the average wait
time which passengers would spend at bus stops when taking
each route. The average headways were calculated directly
*The exact proportion was derived from maps, using rules-of-
thumb described earlier to make trade-offs between walk time
to the interzonal bus and time spent waiting for and tra-
veling on the feeder line.
277
-------
from actual schedules of individual bus lines; peak and off-
peak headways were calculated separately. All calculations
described below were carried out twice for each interzonal
route, once for peak hour schedules and once for offpeak
schedules.
In order to derive average wait times from these head-
ways, it was necessary to make several behavioral assumptions
about the rationality of tripmakers' decisions. Specific-
cally, it was assumed that passengers were familiar with the
schedules of the bus lines, planned their trips to minimize
wait time, all other things being equal, and perceived time
spent waiting at a bus stop as more onerous than time spent
waiting at the origin point before starting their trip.3
The method by which average wait times were derived is
best explained by describing a typical trip. If more than
one bus must be taken to travel from a specified origin to
a specified destination, a tripmaker will first examine the
bus schedules before starting on his trip. If he finds that
the last bus he must take runs more infrequently than the
previous buses along his interzonal route (or, more gener-
ally, if any bus along the route is more infrequent than
previous ones), the tripmaker will plan his trip around the
more infrequent bus. He would first calculate which specific
bus along the more infrequent bus line he could catch if he
left his origin immediately. He would then take the last
possible combination of previous buses that would bring him
to the bus stop of the more infrequent bus line in time to
catch that specific bus. For instance, if a particular
interzonal trip included taking three different buses whose
headways were 5, 10 and 30 minutes respectively, the trip-
maker would first consult his schedule to find the first #3
bus he could catch, including the wait times and line-haul
278
-------
times of the first two buses in his calculations. Should he
find that, leaving immediately, he would arrive at the #3
bus stop just after a bus has left, he would wait at his
origin to catch the last combination of previous buses that
would bring him to bus stop #3 in time for the next bus to
leave. The longest time he would ever wait at the #3 bus
stop is 10 minutes, since he would be able to take a later
series of buses if he had to wait more than 10 minutes.
This examples can be generalized to the overall rule
that the maximum waiting time for each bus in an interzonal
route is the headway of that bus itself, or the largest head-
way of the previous buses along that route, whichever is
less. Once maximum wait times were calculated, average wait
time was obtained by taking the average of the maximum and
the minimum, or in other words, by halving the maximum
figures, since the minimum wait time is zero.
This rule was used to calculate maximum wait times for
all buses along a route except the first. For the first
bus, tripmakers can time their departure in order to mini-
mize wait time at the bus stop. Because people typically
will time themselves to arrive at the first bus stop slightly
before the bus is scheduled to leave, the average wait time
for the first bus was taken to be 5 minutes, or one-half the
headway of that bus line, whichever was less.
Using these rules, the average wait times for each bus
along an interzonal route were computed and then added
together. After this process was completed for each route
between any designated zonal pair, a weighted average was
calculated for each zonal pair, using the system of weighting
the routes explained above.
For the return leg of the round trip, buses are used
in the opposite order of the original trip. For this reason,
the wait times would be somewhat dissimilar. However, as a
result of the method by which wait times were calculated
279
-------
from headways, the maximum difference that can occur is five
minutes, and in most cases there would be no difference at
all. To illustrate this point, assume that the interzonal
bus route is composed of three bus lines with different head-
ways, with the smallest headway being larger than 10 minutes.
Recalling that the maximum wait time for any bus along an
interzonal route is the headway of the bus itself, or the
largest headway of the previous buses along the route,
whichever is less, then no matter how the three buses are
ordered along the route, the following will hold:
a) The largest headway will never be counted in wait
time calculations — if it is the first bus along the route,
five minutes will be the average wait for that bus itself;
if it is the second or third bus along the route, either the
medium or smallest headway will be counted as the maximum
wait for the infrequent bus.
b) The bus with the second largest headway will always
be counted once in wait time calculations: if it is the
first bus along the interzonal route, five minutes will be
the average wait time for that bus itself, and the medium-
sized headway will be the maximum wait for the bus with the
largest headway; if the bus with the second largest headway
is the second or third bus along the route and the bus with
the largest headway comes somewhere before it along the
interzonal route, the maximum wait at the medium1 frequency
bus will be the headway of that bus itself; if it is the
second bus along the route and the bus with the largest
headway comes after it, the maximum wait for the medium
frequency bus will be the smallest headway, and the middle-
sized headway will be the maximum wait for the bus with the
largest headway.
c) The smallest headway will always be counted once
in wait time calculations: if it is the first bus along
the interzonal route, five minutes will be the average wait
280
-------
for that bus itself, and the smallest headway will be the
maximum wait for whichever bus follows it along the route;
if it is the second or third bus along the route, the maximum
wait for the bus will be its own headway.
The only time that the direction of traveJ (i.e., the
ordering of the buses along the route) will make a difference
in wait time calculations is when the smallest headway is
less than 10 minutes, and when it is the first bus along the
route in either direction. In this case, in one direction
the average wait time for the first bus will be half the
headway itself instead of five minutes. The total average
wait time would be half (smallest headway plus medium headway
plus smallest headway). In the other direction, the average
wait time would be five plus half (medium headway plus
smallest headway). The difference is five minutes minus
one-half of the smallest headway. This difference can
clearly never be more than five minutes.
Because five minutes is insignificant compared to the
total trip time, the wait times for return trips were not
calculated, but were assumed to be identical to the original
figure.
Schedule Delay — In any trip the passenger will choose the
last series of buses that will get him to his destination
before his preferred arrival time. The difference between
the time that this chosen series of .buses will bring him to
his destination and his preferred arrival time, or the
schedule delay, is a measure of the inconvenience cost to
the tripmaker as a result of the bus scheduling. The maximum
possible schedule delay is the sum of the headways of the
buses along any particular route; the average is one-half of
this sum. This calculation assumes a symmetric distribution
of bus arrival times around preferred arrival time.
281
-------
Total Line-Haul Time — One variable was created from the
previous three variables to represent the total round trip
transit time excluding access time for each zonal pair in
the sample. This variable was named line-haul time to dis-
tinguish it from access time; however, it includes not only
time spent on the buses along the interzonal route but also
time spent waiting for the buses as well as the schedule
delay at one end of the round trip. The primary reason for
including schedule delay at one end of the trip only is that
in a typical trip, a person will be concerned about arriving
on time at only one end of the trip. For instance, in home-
work round trips, the tripmaker must arrive at work by a
certain hour, whereas his return timing is more flexible.
This treatment of schedule delay also reflects the fact that
people consider time spent at origin and destination to be
less onerous than time spent en route. Total line-haul time
was calculated separately for peak and offpeak hours.
Number of Transfers — The average number of transfers
necessary to complete each interzonal bus trip was recorded
along with the other variables describing interzonal transit
trips.
Transit Fares — During 1967 fares charged by the SCRTD
varied by distance at a formula of $0.30 initial fare for
travel within and between up to two transit fare zones,
$0.08 per zone up to a $1.18 fare, and $0.07 per zone there-
after. There were 320 transit zones in the SCRTD system;
these zones were determined by historical and political
factors as well as by distance. As a result fares did not
vary evenly with distance traveled. However, because zones
usually ranged from approximately 2 to 4 miles, a 3 mile
figure was taken as an approximation of typical zone breadth.
Transfers were an additional $0.05, and were added to our
fare calculations for all routed trips which involved more
than one bus.
282
-------
SOCIOECONOMIC AND DESCRIPTIVE VARIABLES
The following data on zones were also used in applica-
tion of the travel demand model: area, number of households,
vehicle availability, income, and retail employment.
Area
Data on areas of 1970 census tracts were aggregated to
the RAZ level by National Planning Data Associates.
Number of Households
Data on number of households in 1970 census tracts were
similarly aggregated to the RAZ level.
Vehicle Availability
Vehicle availability was calculated from 1970 U.S. Census
of Population data, which record the number of households in
each zone having zero, one, two and three or more cars.
Vehicle availability for different trip purposes was
calculated from these figures. For home-work trips, it was
assumed that the number of vehicles available for work trips
in a zone was proportional to the number of families with
one or more cars. The vehicle availability per household
for work trips was therefore estimated as the sum of the
number of households with one, two, and three or more avail-
able vehicles divided by the total number of households.
The number of cars per household available for shopping
trips was computed to be the sum of the number of households
with two vehicles (one of which was already used for the
work trip) plus 2.1 times the number of households with
three or more vehicles (i.e., 3.1 cars minus one car used
for the work trip), divided by the total number of house-
holds.
283
-------
Income
Median income of households in each zone was obtained
from 1970 U.S. Census of Population tapes. The actual figure
available was the 1969 median income reported by sample
households as part of the 1970 census.
Retail Employment
Employment figures by place of work for retail trade
were used as a proxy for shopping-oriented land uses. The
initial source of these figures is the Bureau of Census 1970
Urban Transportation Planning Package. The Bureau of the Census
obtained this information from the locations of work places
given by census respondents interviewed at their place of
residence, rather than from a direct census taken at the
work locations themselves. There were severe underestimates
caused by wrong addresses, etc. For its own purposes, LARTS
adjusted the Urban Transportation Planning Package (UTPP) figures
to compensate for these irregularities, using control totals
from other sources. The LARTS data were then aggregated to
correspond to the regional analysis zones used in this study.
284
-------
List of References, Appendix D
1LARTS. Manual of Instructions to Home Interviewer's. January 1967
p. 46.
2Domencich, Thomas A. and Gerald Kraft. Free Transit.
A Charles River Associates Incorporated Report. Cambridge,
Lexington Books, 1970.
3Domencich and Kraft. Free Transit, op. ait.
285
-------
TECHNICAL REPORT DATA
(Please read InUsiicticus on the reverse be/ore coin/'lctinxl
I. REPORT NO.
EPA-600/5-77-014
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
Regional Management of Automotive Emissions: The
Effectiveness of Alternative Policies for Los Angeles
5. REPORT DATE
December 1977
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Charles River Associates Inc.
1050 Massachusetts Avenue
Cambridge, Massachusetts 02138
10. PROGRAM ELEMENT NO.
1HC619
11. CONTRACT/GRANT NO.
68-01-2235
12. SPONSORING AGENCY NAME AND ADDRESS
Office of Air Land and Water Use
Office of Research and Development
U.S. Environmental Protection Agemcy
Washington, D.C. 20460
13. TYPE OF REPORT AND PERIOD COVERED
14. SPONSORING AGENCY CODE
EPA/600/16
15. SUPPLEMENTARY NOTES
16. ABSTRACT This study has two objectives: first, to develop procedures to evaluate
policies for controlling automobile emissions; and second, to use these proce-
dures to evaluate specific pollution control strategies for Los Angeles.
The first objective is achieved by developing, a relatively quick .ind re-
liable method for estimating the cost effectiveness of travel related policies.
The methods used include application of a behavioral demand model for automo-
bile travel by mode, purpose and destination, and a model which predicts the
size of the auto stock and its age distribution. These models are used to
compute the costs to society and individual travelers of various policies, and
to compute the emission reduction effects of various policies.
In applying these procedures to Los Angeles, the following specific
strategies were evaluated:
O increased gas taxes;
O taxes on vehicle emissions per mile based on odoneter
readings and emissions tests;
O nonresidential parking surcharges;
O extensions of route miles by conventional bus;
O annual taxes based on vehicle model, make and year.
The report's findings indicate that implement at ion of these policies could
significantly decrease pollution. Emission taxes and gasoline taxes arc
particularly effective strategics; park inf. taxes arc a less effective but
still viable policy. Tax-induced decreases in pollution nro reinforced by
improvements in conventional bus service.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
Environment, Regional Planning, Travel
Demand, Bus Cost Model, Emissions Control
Mobile Source Air Pollution, Cost
Benefit Analysis
). IDENTIFIERS/OPEN ENDED TERMS
Air Quality Standards
Transportation Mgmt.
Cost Effectiveness
Air Pollution
COSATI FicKI/Group
05A
Behavioral and
Social Sciences/
Administration
and Mgmt.
18. DISTRIBUTION STATEMENT
UNLIMITED
19. SECURITY CLASS (This Report)
JlKCLASSIIiEDL
21. NO. OF PAGES
286
20. SECURITY CLASS (Tillspage)
UNCLASSIFIED
22. PRICE
Ef A Form 2220-1 (9-73)
«U.S. GOVERNMENT PRINTING OFFICE: l!>78 2CO-S80/35
286
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Research and Development
Technical Information Staff
Cincinnati, Ohio 45268
OFFICIAL BUSINESS
PENALTY FOR PRIVATE USE, S30O
AN EQUAL OPPORTUNITY EMPLOYER
FOURTH-CLASS MAIL
Postage & Fees Paid
EPA
Permit No. G-35
I
56
\
// your address is incorrect, please change on the above label;
tear off; and return to the above address.
If you do not desire to continue receiving this technical report
series, CHECK HERE (~|; tear off label, and return it to the
- above address.
------- |