Paper submitted to the Transportation Research Board for presentation at the 75th Annual Meeting, January 1996.
| A final version of this paper appears in the Transportation Research Record #1556 (1996) pp. 37-45. |
Abstract
Introduction
Accessibility in Trip Generation Models: A Review of the Literature
Estimation of Disaggregate Trip Generation Models
Disaggregate Validation of Trip Generation Models
Sensitivity Analysis
Application in a Travel Demand Model Forecasting System
Conclusions
Acknowledgments
References
|
ABSTRACT
This paper reports on efforts to include disaggregate work trip accessibility in models of non-work trip generation. Reported household-level one-way average home-based work trip duration is used in home-based shop/other and home-based social/recreation models for the San Francisco Bay Area. The survey data and models show an inverse relationship between work trip duration and home-based non-work trip frequency: as work trip duration increases, non-work trip frequency decreases. Hybrid trip generation models using multiple regression techniques, cross-classified by workers in household level and vehicles in household level, are estimated using data from the 1981 and 1990 household travel surveys. Work trip duration is excluded in models estimated for non-working households and is included in models estimated for single worker and multi-worker households. Elasticity analyses show that a 10 percent decrease in the regional work trip duration yields a 1.2 percent increase in regional home-based shop/other trips and a 0.9 percent increase in regional home-based social/ recreation trips. This research helps to identify practical means to incorporate workplace accessibility in regional travel demand model forecasting systems, to better analyze the issue of induced trip-making, and to provide a better understanding of the linkage between congestion and trip frequency choice behavior. |
This paper reports on work conducted at the Metropolitan Transportation Commission (MTC) in the San Francisco Bay Area to estimate new sets of trip generation models that address the issue of induced trip-making. This is done by including a disaggregate, household-level accessibility variable in the multiple regression models estimated for home-based shop/other (HBSH) and home-based social/recreation (HBSR) trip purposes. The work at MTC is a practitioners' extension to the research conducted by Golob, Kitamura and Lula (1). One of their research hypotheses was to explore the inverse relationship between work trip duration (and work activity duration) and non-work trip frequency. Their success in showing statistically significant inverse relationships between work trip duration and non-work frequency choice helped in the redesign of MTC model development strategies.
Purvis' analysis of aggregate trip frequency, trip duration and travel time expenditure trends based on San Francisco survey data from 1965, 1981 and 1990 also suggested some sort of inverse relationship between trip frequency and trip duration (2). Both Purvis and the paper discussant reported on the need to go beyond the aggregate analysis presented in the paper to a disaggregate analysis of trip frequency / trip duration choice behavior.
Traditional trip generation models estimated for or by metropolitan planning organizations in the United States have almost universally excluded the impact of travel time changes on overall trip frequency choice. Simpler forms of trip generation models, say, cross-classification models incorporating two to three independent variables, are typical of those found in many metropolitan areas. These simpler models tend to exclude continuous variables such as density or accessibility, or treat a continuous variable (e.g., household income) as a categorical variable (e.g., household trip rates stratified by income tertile or quartile).
Previous and current travel demand models used in the San Francisco region have indirectly incorporated accessibility in trip generation models by basing non-work trip frequency choice on auto ownership levels which are in turn influenced by the relative accessibility of transit to auto. Neighborhoods with higher transit accessibility relative to auto will tend to have lower auto ownership levels, and thus lower non-work trip frequency levels. This accessibility linkage is only true for the non-work trip generation models, as the current MTC home-based work trip generation models are based strictly on work trips per worker as a function of household size, household income, and neighborhood density characteristics (3-5).
This paper provides a brief review of the literature related to accessibility in trip generation models, model estimation results using data from the San Francisco Bay Area 1981 and 1990 household travel surveys, disaggregate validation analyses, sensitivity analyses, issues related to incorporating these models into an aggregate travel demand forecasting system and conclusions and recommendations for other metropolitan area practitioners.
ACCESSIBILITY IN TRIP GENERATION MODELS: A REVIEW OF THE
LITERATURE
Incorporating accessibility in trip generation models to predict if adding transportation capacity induces trips has broiled up as a major topic for discussion and debate over the past several years. An air quality lawsuit in the Bay Area, EPA regulations, and travel model guidance documents have all been quite vocal about this issue, though none have offered a methodology to accomplish it.
Kitamura (6-7) provides a good coverage and critique of the literature on induced travel. Also worth noting is his critique of the 1972 Nakkash and Grecco study (8) which tested aggregate, zone-level accessibility measures rather than disaggregate, household-level accessibility measures. Kitamura expresses concern that this method would be problematic due to too little variation between zones (and no variation within zones) and that zone-based accessibility measures are too insensitive to detect the effect of accessibility on trip frequency.
The Nakkash and Grecco study of Indianapolis is typical of other research in this field in that it incorporates aggregate accessibility measures in trip generation. Similar studies in this genre are Vickerman's (9) analysis using data from the Oxford Transportation Study; Golding and Olsen's (10) analysis of data from two Australian travel surveys conducted in the 1960s; and Leake and Huzzayin's (11) 1979 review of the literature. Also of note is Southworth's (12) research on estimating combined trip frequency-trip destination choice models using data from the 1966 Leeds survey, albeit for work trips only.
Developers of disaggregate, activity-based models have also contributed research applicable to analyzing induced travel. Landau, Prashker and Hirsh (13) estimated models using single-worker household data from the 1972 Tel Aviv survey. In this study the authors adopted a three-class typology of activities: subsistence activities (work, school); maintenance activities (shopping, personal business); and leisure activities (visiting, recreation). The authors assumed that the schedules of subsistence activities are given and fixed in the short term. The demand for maintenance and leisure activities are elastic. Landau analyzed both the time-at-work and duration of the trip to work in their models. The authors concluded that "as expected, as the work time and/or travel time to work of the household head increases, his probability of executing a maintenance trip in [the middle of day time period] decreases."
Two related papers on trip chaining by Goulias, Pendyala and Kitamura (14) and Goulias and Kitamura (15) use a two-class typology of activities: mandatory activities (such as work and school) and discretionary activities (such as shopping, personal business and social). The authors hypothesize that the "presence and number of compulsory trips influence the presence and number of discretionary trips." The authors found a significant inverse relationship between work trip frequency and shopping trip and social trip frequency. The authors, however, did not test work trip duration, or work activity duration, in their non-work trip generation models.
Time use analyses by Kitamura, et. al. (16) provide further evidence using data from the Netherlands and California surveys on the apparent tradeoffs between work duration on one hand and non-work activities, non-work travel, in-home activities and out-of-home activities, on the other.
And, lastly, a recent paper by Golob and McNally (17) revisits and modifies the three-class activity typology - work activities, maintenance activities, and discretionary activities - in their analysis of the 1994 Portland, Oregon household travel survey. Golob and McNally explore the impact of reduced commute times for male and female heads of household, noting that females increase both discretionary travel times and discretionary activities whereas males only increase their discretionary travel times. The aforementioned study by Golob, Kitamura and Lula (1) used similar model development techniques using data from the 1985 National Time Use Survey in the Netherlands.
It is apparent from the recent work in activity-based models that researchers are finding a clear inverse relationship between work activity and commute duration on one hand and non-work activities and non-work travel on the other hand. In contrast to research efforts in the 1970s which used aggregate, zone-level estimates of accessibility as factors in trip generation models, the newer breed of activity-based models are beginning to take full advantage of disaggregate, household-level accessibility data. The new trip generation models discussed in this paper are not, however, activity-based travel models. They merely represent some simple and tractable enhancements to traditional, trip-based trip generation models.
ESTIMATION OF DISAGGREGATE TRIP GENERATION MODELS
The data sets used in the estimation of the disaggregate home-based shop/other and home-based social/recreation trip generation models are the 1981 and 1990 San Francisco Bay Area household travel surveys (18-20). Both surveys are traditional, trip-based telephone surveys using mail-out "memory jogger" trip diary cards and phone retrieval of all household, person and trip information. Unfortunately, no information on in-home activity data was collected in either survey. Thus, we are unable to detect if a worker not traveling to work is working at home, ill, on vacation, or on a scheduled day off.
Documentation of the trip generation models discussed in this paper is included in two compilations (21, 22). The documentation describes in detail the "chosen" model and all of the "rejected" models.
The dependent variables in these models are home-based shop/other and home-based social/ recreation total trips per household, including bicycle and walk trips. Previous generations of MTC trip generation models focused solely on motorized trips (auto person and transit person trips).
The estimated home-based shop/other trip generation models (HBSHG) are "hybrid" linear regression models cross-classified by workers in household level (WHH). This means that there are separate regression equations applied to zero-worker, single worker, and multi-worker households. Linear regression models were also estimated for nine workers/household and autos/household (AO) market segments, but, based on disaggregate validation analyses, these models were inferior to the simpler, three segment model (22). The final results of the HBSHG models are shown in Table 1.
TABLE 1 Home-Based Shop/Other Trip Generation Models
Linear Regression Models Cross-Classified by Workers in HH
Based on 1990 Bay Area Household Travel Survey
Workers in Household
Group
Parameter WHH=0 WHH=1 WHH=2+
Intercept 0.3141 0.03746 0.3763
HHSize 0.4709 0.7453 0.5984
Vehicle 0.4034 0.2661 0.1054
Income 0.02052
Income2 -0.000131
HBWTDUR -0.02178 -0.03065
HBWTDUR2 0.000158 0.000235
Sample Size (N) 1,627 3,622 4,110
R2 .1179 .2233 .0990
Where:
HHSize = Household Size
Income = Household Income, in 1000s of 1989 dollars.
Income2 = Household Income, in 1000s of 1989 dollars, squared
HBWTDUR = Average one-way HBW Trip Duration, in minutes
HBWTDUR2 = Average one-way HBW Trip Duration, in minutes, squared
The home-based shop/other trip generation model uses second-degree polynomials for household income and average one-way home-based work (HBW) trip duration. This is done to better represent the nonlinear relationships between the independent variables and the dependent variable (i.e., home-based shop trips per household). The home-based work trip duration variable is excluded from the models estimated for non-working households. This is because non-working households do not have workers commuting to work.
The t-statistics for all variables are greater than 2.0, signifying that the coefficients are significantly different from zero. The t-statistics for HBW trip duration in the single worker household model are -5.4 for HBWTDUR and 3.2 for HBWTDUR2. For the multi-worker model, the t-statistics are -5.6 for HBWTDUR and 3.6 for HBWTDUR2 which indicates a strong inverse relationship between home-based work trip duration and home-based shop/other trip frequency.
The estimated home-based social/recreation trip generation models (HBSRG) are also "hybrid" linear models market segmented by workers in household level (0, 1, 2+) and vehicles per household level (0, 1, 2+). Therefore we have nine linear regression models which correspond to nine different market segments: three workers in household groups and three vehicles in household groups. Final results for the home-based social/recreation trip models are summarized in Table 2.
For the home-based social/recreation trip generation model, the regression intercept is suppressed in seven out of nine models. This is because the intercepts are either slightly negative or close enough to zero, so that suppressing the intercept makes good sense.
TABLE 2 Home-Based Social Recreation Trip Generation Models
Linear Regression Models Cross-Classified by Workers in HH and Autos/HH
Based on 1990 Bay Area Household Travel Survey
WHH=0 WHH=0 WHH=0 WHH=1 WHH=1 WHH=1 WHH=2+ WHH=2+ WHH=2+
Parameter AO=0 AO=1 AO=2+ AO=0 AO=1 AO=2+ AO=0 AO=1 AO=2+
Intercept 2.241 1.014
HHSize 0.548 -2.8455 0.873 0.497 0.834 -0.0625
LogHHSize 0.3638 0.709 1.018
HHSize2 -0.1889 1.32 -0.1389 -0.083 -0.283
HHSize3 0.01637 -0.1492 0.03
LogHHInc 0.15 0.23 0.183
AreaType -0.651 -0.056 -0.584
AreaType2 0.252 0.096
AreaType3 -0.028
4
HBWTDUR -0.0067 -0.00427 -0.002533
8
LogHBWTDUR -0.0666 -0.086 -0.074
8
Sample Size 302 757 568 238 1,720 1,664 73 500 3,537
(N)
R2 .1909 .0838 .2901 .1899 .0458 .2896 .2039 .2021 .2854
Where:
HHSize = Household Size
LogHHSize = Natural Log of Household Size
HHSize2 = Household Size, Squared
HHSize3 = Household Size, Cubed
LogHHInc = Natural Logarithm of Household Income in 1000s of $ (MAX(0,LN(Income))
AreaType = Density Based Area Type
AreaType2 = Density Based Area Type, Squared
AreaType3 = Density Based Area Type, Cubed
HBWTDUR = Average One-Way HBW Trip Duration, in minutes
LogHBWTDUR = Natural Log of Average One-Way HBW Trip Duration (MAX(0,LN(HBWTDUR))
Area Density = (Total Population + 2.5 * Total Employment / Developed Acres)
AREATYPE Area Density
0 - Regional Core > 300.0
1 - CBD 100.0 - 300.0
2 - Urban Business 55.0 - 100.0
3 - Urban 30.0 - 55.0
4 - Suburban 6.0 - 30.0
5 - Rural < 6.0
Temporal Stability of Trip Generation Models
To investigate the temporal stability of trip generation model parameters, similar model specifications were tested using data from the 1981 Bay Area household travel survey. Two styles of comparisons were used: an informal comparison of model coefficients, focusing on the magnitude and sign of the coefficients; and a formal comparison of the equality of the two sets of regression coefficients. This formal test, Chow's test (23), evaluates the null hypothesis of equality of the sets of regression coefficient (H0: b1 = b2). Chow's test was used by Smith and Cleveland (24) in their investigation of the temporal stability of trip generation models for the Detroit region using data from the 1953 and 1965 household surveys. The null hypothesis is rejected at a (1 - a) percent confidence level if the Fisher test statistic (F) is greater than F1 - a with k and (m + n - 2k) degrees of freedom. The Fisher test statistic for Chow's test is computed as:
F = [(Q1 - Q2) / k] / [Q2 / (m + n - 2k)]
where:
Q1 = sum of squared errors (SSE) from pooling the observations;
Q2 = sum of squared errors (SSE) from separate regression for 1981 and 1990;
m = numbers of observations in year 1 (1981);
n = number of observations in year 2 (1990); and
k = number of independent variables plus 1.
Application of Chow's test required that we pool the household samples from the 1981 and 1990 survey to estimate a combined-survey model. Comparison of the home-based shop/other trip generation models using the 1981 and 1990 surveys is shown in Table 3.
TABLE 3 Home-Based Shop/Other Trip Generation Models
Linear Regression Coefficients based on 1981 and 1990 Bay Area Household Travel Surveys
Non-Working Household Single-Worker Multi-Worker Household
Household
Parameter 1981 1990 1981 1990 1981 1990
Intercept .206 .314 -.583 .0375 .241 .376
HHSize .443 .471 .957 .745 .771 .598
Vehicle .472 .403 .303 .266 .036 .105
Income ($1989) .0412 .0205
Income2 -.000269 -.000131
($1989)
HBWTDUR -.00114 -.0218 -.019 -.0306
HBWTDUR2 1.025E-6 .000158 4.73E-5 .000235
Sample Size 859 1,627 1,952 3,622 2,158 4,110
(N)
R2 .1409 .1179 .3179 .2233 .1380 .0990
Chow's Test 8.2 10.9 8.4
(F)
The coefficients that relate to demographic characteristics - household size, household income, and vehicles per household - are relatively consistent in terms of sign and magnitude. In contrast, the work trip duration coefficients are significantly larger in 1990 than the 1981 survey-based models, suggesting that the impact of work trip duration on shop trip generation has increased over time.
The t-statistics for home-based work trip duration are very weak for the 1981 single worker household model (-0.4 for HBWTDUR and 0.2 for HBWTDUR2) and quite strong for the 1981 multi-worker household model (-3.9 for HBWTDUR and 2.1 for HBWTDUR2).
With six degrees of freedom (k) and at a 1 percent confidence level (F.01), the calculated Chow F statistics are larger than the critical F value (6.88). Therefore, we reject the null hypothesis and conclude that the 1990 set of model coefficients are statistically significantly different than the 1981 model coefficient set.
Due to the complexity of the 1990 home-based social/recreation trip generation models, with its nine linear regression models, we chose the 1990 HBSHG model specification for our 1981 versus 1990 model comparison since all of the variables in the HBSHG model are also in the HBSRG model. This comparison is shown in Table 4.
TABLE 4 Home-Based Social/Recreation Trip Generation Models
Linear Regression Coefficients based on 1981 and 1990 Bay Area Household Travel Surveys
Non-Working Household Single-Worker Multi-Worker Household
Household
Parameter 1981 1990 1981 1990 1981 1990
Intercept .0253 .193 .207 .155 .037 -.068
HHSize .173 .132 .366 .272 .378 .269
Vehicle .171 .213 .169 .123 .155 .150
Income ($1989) .0275 .00318
Income2 -.000151 -7.12E-6
($1989)
HBWTDUR -.00668 -.0100 -.00622 -.00838
HBWTDUR2 8.24E-6 5.52E-5 1.74E-5 6.61E-5
Sample Size 859 1,627 1,952 3,622 2,158 4,110
(N)
R2 .0635 .0402 .0872 .0759 .1099 .0414
Chow's Test 8.0 17.6 24.5
(F)
As with the home-based shop generation model, the coefficients for the demographic variables in the home-based social/recreation model are more stable and consistent than the work trip duration variables when we compare 1981 models to the 1990 models. The t-statistics for the work trip duration variables are higher in the 1981 single worker model (-2.7 for HBWTDUR and 1.5 for HBWTDUR2) than the multi-worker model (-1.7 for HBWTDUR and 1.0 for HBWTDUR2). The t-statistics for the 1990 work trip duration variables are also higher for the single worker model (-3.8 for HBWTDUR and 1.7 for HBWTDUR2) than the multi-worker model (-2.4 for HBWTDUR and 1.6 for HBWTDUR2).
Based on the Chow's test results for the home-based social/recreation trip generation models, we reject the null hypothesis that the sets of model coefficients are the same.
DISAGGREGATE VALIDATION OF TRIP GENERATION MODELS
Disaggregate validation is the stage in the model development process where estimated models are applied to disaggregate records to analyze the observed versus predicted choice behavior, typically by market segment. For the MTC home-based shop/other and home-based social/ recreation trip generation models, we conducted numerous disaggregate validation tests by several market segments, including household size, area type, vehicle ownership level, county of residence, income quartiles and detailed income groups, and home-based work trip duration deciles. For this paper we report on just the disaggregate validation by home-based work trip duration decile. The deciles are defined so as to have approximately 10 percent of all working households included in each group.
Observed and model predicted trips per working household by work trip duration is shown in Table 5. A special category of households, working households with no one taking a HBW trip, is reported separately. This group may include working households with workers who did not commute to work on the assigned travel day, or workers who chained trips to other non-work activities, say, conducting personal business on the way to and from work. This group of non-HBW trip-taking working households tend to have the highest trip rates for home-based shop and home-based social/recreation trip purposes. This makes intuitive sense: workers not commuting to work may partake in more out-of-home non-work activities than workers commuting long distances from home-to-work and work-to-home.
One of the limitations of our analysis is that each trip in a trip chain is independent of prior and subsequent trips. Therefore, an alternative explanation could be that workers with no reported HBW trips are, in fact, chaining their trips to work with other non-work activities at both ends of their commute. Incorporation of this trip chaining phenomenon is the subject of future research at MTC.
TABLE 5 1990 Survey and Predicted Household Trip Rates by HBW Trip Duration
HBW Trip Home-Based Shop Trips per HH Home-Based Soc./Rec. Trips per HH
Duration Survey Model Survey Model
No HBW Trips 2.29 2.15 .89 .92
< 12 min. 1.91 1.95 .99 .84
12 - 15 min. 1.91 1.95 .77 .85
16 - 20 min. 1.92 1.91 .89 .86
21 - 25 min. 1.74 1.86 .95 .88
26 - 30 min. 1.65 1.80 .71 .83
31 - 39 min. 1.77 1.73 .81 .85
40 - 52 min. 1.66 1.63 89 .82
> 52 min. 1.60 1.48 .64 .66
Total, WHH 1.87 1.85 .82 .82
The 1990 survey (observed) versus predicted trip rates by work trip duration level is depicted in Figure 1 for home-based shop trips and Figure 2 for home-based social/recreation trips.
Models that exclude home-based work trip duration underpredict HBSHG and HBSRG trips for households with low average work trip duration and overpredict trips for households with longer work trip duration. This is illustrated in Figure 3 which compares survey home-based shop trips per household to a model that excludes HBW trip duration.
As an extension to the disaggregate validation analysis, sensitivity analyses were conducted by adjusting the reported HBW trip duration by plus-or-minus ten percent, reapplying the models to the sample households, and reporting the aggregate changes in trip making by workers in household level. Results of this sensitivity analysis are shown in Table 6. This analysis shows demand elasticities ranging from -0.1 to 0.176 for home-based shop trips and from -0.069 to 0.152 for home-based social/recreation trips. For the combined non-work trip purposes, a ten percent decrease in the average home-based work trip duration would yield a 1.11 percent increase in the total number of home-based shop and home-based social/recreation trips generated in the Bay Area. Given that these two trip purposes combined comprise 36 percent of all Bay Area trips in 1990, the overall impact of a ten percent decrease in work trip duration would be a 0.40 percent increase in total trips generated by Bay Area residents.
TABLE 6 Sensitivity Analysis for HBSH and HBSR Trip Generation Models
Percent Change in Predicted Trips due to 10% Change in Work Trip Duration
10 Percent Decrease, 10 Percent Increase,
Avg. Work Trip Duration Avg. Work Trip Duration
Home-Based Shop/Other
Trips, Percent Change
Workers in HH = 0 0.00% 0.00%
Workers in HH = 1 1.19% -1.00%
Workers in HH = 2+ 1.76% -1.50%
Total Households 1.23% -1.04%
Home-Based
Social/Recreation Trips,
Percent Change
Workers in HH = 0 0.00% 0.00%
Workers in HH = 1 1.52% -1.47%
Workers in HH = 2+ 0.69% -0.69%
Total Households 0.85% -0.83%
Combined Non-Work Trip
Purposes
Total Households 1.11% -0.97%
The sensitivity analysis further shows that multi-worker households are more sensitive to work trip duration for home-based shop trips than are single worker households. For home-based social/recreation trips the opposite was found: single workers households are more sensitive to HBW trip duration than are multi-worker households. Non-working households are insensitive to work trip duration because residents in non-working households don't commute.
APPLICATION IN A TRAVEL DEMAND MODEL FORECASTING
SYSTEM
One of the last remaining challenges is how to implement these disaggregate trip generation models in a zone-level, aggregate travel demand model forecasting system. The critical concern here is the need to convert from disaggregate, household-level average work trip duration to zone-level average work trip duration. In aggregate models, the between-household variability in work trip duration is sacrificed; only between-zone variability can be captured in the model application.
The current MTC model system, illustrated in Figure 4, shows the typical sequential structure of travel demand models starting with demographic and auto ownership forecasts and ending with trip assignment. The dotted arrows pointing upwards in the sequence represent logsum linkages between the lower nest mode choice models and the upper nest destination choice and auto ownership choice models. Optional feedback loops (the solid lines leading from lower nest to upper nest models) are also used to ensure consistency between travel times input to and output from the work trip destination and mode choice models. The current MTC model system also supplies the household auto ownership forecast data to the non-work trip generation models. For a discussion of the adaptation of disaggregate choice models into a regional travel forecasting system, the reader may refer to the paper by Kollo and Purvis (25).
Figure 4 shows the added model linkage between trip assignment and trip generation. The average work trip duration by travel analysis zone-of-production is used as input to the non-work trip generation modules. This data is calculated by weighting total door-to-door travel times by predicted HBW trips by means of transportation (k) (auto, transit, non-motorized) then reducing this matrix of weighted zonei-to-zonej travel times to zone-of-production (i). This is represented in the following equation:
Weighted Timei = Summationi (Timeijk * Tripsijk) / Summationi Tripsijk
An additional complexity is the need to incorporate "background" non-work vehicle trips in peak period traffic assignment. For the first iteration in any future year travel forecast, a "seed" non-work vehicle trip table, say, from previous forecasts, is needed as input into the peak period traffic assignment. Feedback loops are needed to iterate on these "seed" non-work trip tables and "final" non-work trip tables, in a similar fashion to feedback loops used in forecasting work trip distribution and mode choice.
New non-work trip generation models for the San Francisco Bay Area have been estimated which incorporate the effect of work trip duration on non-work trip frequency choice. By incorporating work trip duration as a variable in home-based shop/other and home-based social/ recreation trip generation models, we are able to address, at least in part, the issue of induced travel due to changes in congestion levels as they affect work trip duration.
This research is a practical extension of research by activity-based modelers and time use researchers in terms of defining linkages between mandatory activities (e.g., work, school) and discretionary trip-making (e.g., shop, personal business, recreation). The primary emphasis of this research is to develop practical, yet sophisticated models for use in traditional regional travel demand model forecasting systems. The secondary role of this research is for improved understanding of travel behavior.
Disaggregate validation and sensitivity analyses suggest the effect of including or omitting work trip duration as a variable in non-work trip generation models. Sensitivity analyses indicate that a 10 percent decrease in average work trip duration would yield a 1.2 percent increase in home-based shop trips and a 0.9 percent increase in home-based social/ recreation trips, by all means of transportation.
This research may serve to assist other metropolitan area practitioners in their efforts to address the issue of induced travel, and to create better tools for evaluation of transportation investment scenarios.
Miguel Iglesias performed all of the estimation and validation work on the home-based social/recreation models. Victoria Eisen performed all of the estimation and validation work on the home-based shop models. Chuck Purvis was responsible for overall model development activities, review of literature, and model analysis using the 1981 survey database. The authors would like to acknowledge our fellow workers on the MTC model development project: Ron West, Rupinder Singh, Lisa Klein and Pat Hackett. The authors would also like to acknowledge the inspiring work of Tom Golob and Ryuichi Kitamura. Their research results have had an immediate impact on our model development activities.
1. Golob, Thomas F., Ryuichi Kitamura and Clarisse Lula. Modeling the Effects of Commuting Time on Activity Duration and Non-Work Travel. Presentation to the 73rd Annual Meeting of the Transportation Research Board, Washington, D.C., January 1994.
2. Purvis, Charles L. Changes in Regional Travel Characteristics and Travel Time Expenditures in the San Francisco Bay Area: 1960-1990. In Transportation Research Record 1466, Transportation Research Board, NRC, Washington, D.C., 1995, pp. 99-109.
3. Cambridge Systematics, Inc. Travel Model Development Project: Phase 2 Final Report: Volume 1: Summary Report. Metropolitan Transportation Commission, Berkeley, CA, June 1980.
4. Kollo, Hanna P.H. Home Based Work Trip Models: Final Disaggregate Version: Travel Model Development with 1980/81 Data Base: Working Paper #2. Metropolitan Transportation Commission, Oakland, CA, Revised December 1987.
5. Kollo, Hanna P.H. and Charles L. Purvis. Non-Work Trip Models: Final Version: Technical Summary: Travel Model Update with 1980/81 Data Base. Metropolitan Transportation Commission, Oakland, CA, August 1987.
6. The Effects of Added Transportation Capacity: Conference Proceedings. Edited by Gordon A. Shunk. U.S. Department of Transportation, Washington, D.C., DOT-T-94-12, December 1991.
7. Kitamura, Ryuichi. The Effects of Added Transportation Capacity on Travel: A Review of Theoretical and Empirical Results. In The Effects of Added Transportation Capacity: Conference Proceedings. Edited by Gordon A. Shunk. U.S. Department of Transportation, Washington, D.C., DOT-T-94-12, December 1991, pp. 79-95.
8. Nakkash, T.Z. and W.L. Grecco. Activity-Accessibility Models of Trip Generation. In Transportation Research Record 392, Transportation Research Board, NRC, Washington, D.C., 1972, pp. 98-110.
9. Vickerman, R.W. Accessibility, Attraction, and Potential: A Review of Some Concepts and their Use in Determining Mobility. In Environment and Planning A:6, 1974, pp. 675-691.
10. Golding, S. and J.L. Olsen. Analysis of Household Trip Production Rates. In Australian Road Research Board Conference Proceedings, vol. 8, 1976, pp. 26-40.
11. Leake, G.R. and A.S. Huzayyin. Accessibility Measures and their Suitability for Use in Trip Generation Models. In Traffic Engineering and Control 20:12, 1979, pp. 566-572.
12. Southworth, Frank. A Disaggregated Trip Distribution Model with Elastic Frequencies and Expenditures. In Journal of Transport Economics and Policy 13:2, 1979, pp. 209-224.
13. Landau, U., J.N. Prashker and M. Hirsh. The Effect of Temporal Constraints on Household Travel Behavior. In Environment and Planning A, 1981, pp. 435-448.
14. Goulias, Konstadinos G., Ram M. Pendyala and Ryuichi Kitamura. Practical Method for the Estimation of Trip Generation and Trip Chaining. In Transportation Research Record 1285, Transportation Research Board, NRC, Washington, D.C., 1990, pp. 47-56.
15. Goulias, Konstadinos G. and Ryuichi Kitamura. Recursive Model System for Trip Generation and Trip Chaining. In Transportation Research Record 1236, Transportation Research Board, NRC, Washington, D.C., 1989, pp. 59-66.
16. Kitamura, Ryuichi, John P. Robinson, Thomas F. Golob, Mark A. Bradley, John Leonard and Toon van der Hoorn. A Comparative Analysis of Time Use Data in the Netherlands and California. Research Report, Institute of Transportation Studies, University of California, Davis, UCD-ITS-RR-92-9, June 1992.
17. Golob, Thomas F. and Michael G. McNally. A Model of Household Interactions in Activity Participation and the Derived Demand for Travel. Paper presented at the Conference on Activity-Based Approaches: Activity Scheduling and the Analysis of Activity Patterns, Eindhoven, The Netherlands, May 1995.
18. Crain and Associates. 1981 Bay Area Travel Survey. Metropolitan Transportation Commission, Oakland, CA, August 1981.
19. Reynolds, Marilyn M., Sydwell M. Flynn and David B. Reinke. 1981 San Francisco Bay Area Travel Survey. In Transportation Research Record 877, TRB, National Research Council, Washington, D.C., 1982, pp. 51-58.
20. Charns, Harold, Bonnie Nelson and Nancy Pitta. 1990 Bay Area Travel Surveys: Final Report. Metropolitan Transportation Commission, Oakland, CA, March 1991.
21. San Francisco Bay Area 1990 Travel Model Development Project: Compilation of Technical Memoranda: Volume I. Metropolitan Transportation Commission, Oakland, CA, March 1995.
22. San Francisco Bay Area 1990 Travel Model Development Project: Compilation of Technical Memoranda: Volume II. Metropolitan Transportation Commission, Oakland, CA, June 1995.
23. Chow, Gregory C. Tests of Equality Between Sets of Coefficients in Two Linear Regressions. In Econometrica 28:3, July 1960, pp. 591-605.
24. Smith, Robert L., Jr., and Donald Cleveland. Time-Stability Analysis of Trip-Generation and Predistribution Modal-Choice Models. In Transportation Research Record 569, TRB, National Research Council, Washington, D.C., 1976, pp. 76-86.
25. Kollo, Hanna P.H. and Charles L. Purvis. Regional Travel Forecasting Model System for the San Francisco Bay Area. In Transportation Research Record 1220, TRB, National Research Council, Washington, D.C., 1989, pp. 58-65.