Car-Rider Segmentation According to Riding Status and Investment in Car Mobility

Transcription

1 Car-Rider Segmentation According to Riding Status and Investment in Car Mobility Alon Elgar and Shlomo Bekhor Population segmentations for mode choice models are investigated. Several researchers have shown that car and bus users differ substantially in their value of time, in both revealed-preference and statedpreference surveys. This line of research is followed, and a new methodology to segment the population is presented. The hypothesis that two further segmentation dimensions will produce meaningful results is verified. In the first segmentation, car drivers are distinguished from car passengers. In the second segmentation, the population is divided according to their investment in mobility by car. With data from a revealed-preference survey conducted in the Tel Aviv, Israel, metropolitan area, several multinomial logit models were estimated with respect to each population segment. The results show clearly that households with a high investment in car mobility have a much higher value of time than do those with a low investment in mobility. In addition, car drivers perceive higher transit penalties than do car passengers. An interesting result, which exceeded the expectations, was the strong increase of the waiting time penalty at high investment in car mobility. The cumulative effect of the value of time, bus transit penalties, and additional in-vehicle travel time by transit modes prevents most car drivers from potentially changing their travel mode to transit. The proposed procedure, if adopted in the estimation of transportation models, would reduce much of the evident gap between the expected and actual demand generated by changes in transit services. Since the late 1980s there has been a growing awareness of the fact that forecasting procedures used by planners for the evaluation of new transit (mostly rail) construction projects are systematically overestimating transit ridership. Pickrell (1) and Flyvbjerg et al. (2) have strongly confronted this issue. They showed that there was a systematic bias in the forecasting procedures that tended to exaggerate by tens (and even hundreds) of percentage points, in comparison with actual figures, the foreseeable number of transit riders as a result of introduction of new urban rail systems. Their criticism was concentrated on the interested parties (local authorities, planners, and entrepreneurs) and hence the structure of the grant programs and funding sources as a disincentive to local officials and decision makers to seek accurate information in evaluating alternatives. Some major methodological aspects that enabled the erroneous conclusions by the planners, and hence by the decision makers, are A. Elgar, Mevo-Hazait 9, Har Adar 90836, Israel. S. Bekhor, Faculty of Civil and Environmental Engineering, Technion Israel Institute of Technology, Haifa 32000, Israel. Transportation Research Record: Journal of the Transportation Research Board, No. 1894, TRB, National Research Council, Washington, D.C., 2004, pp investigated here. It appears that the arsenal of planning models available to the planner nowadays covers most of what is needed for proper planning and forecasting. What appears to be lacking is a well-defined, tested, and proven methodology for use of the available tools (that is, the models) in order to avoid bias in forecasting results. Most of the current transportation models in practice were estimated using a uniform set of parameters representing some sort of average rider in transportation, which is to be blamed for a substantial part of the error. Therefore, this research is confined to one part of the estimation methodology the segmentation of the car-rider population in order to properly identify the preferences of different car-rider segments. In the authors opinion, a first step toward improvement of the estimation procedures is the segmentation between two populations: transit riders and car riders. Wardman (3) showed that there is a substantial difference in the preference systems between those two populations, as reflected by stated-preference (SP) studies. As stated by Wardman (3), The results would seem to indicate that bus users have the lowest values and car users the highest values in the urban context, and the differences are quite substantial. This statement is supported by data from the Netherlands and Norway for the in-vehicle value of time (VOT) and from Finland for total travel time. In the study conducted by Liu et al. (4) for New York and New Jersey commuting corridors with revealed-preference (RP) and SP data for car and transit riders, an intermodal transfer penalty of 15 min was found. In a report by the Federal Transit Administration (5) for work trips in Boston, the transfer penalty ranges from 12 to 15 min of in-vehicle time for urban mode choice modeling. It may be noted that in Israel, planning agencies are also currently using 12 to 15 min of in-vehicle time in transit mode choice and assignment model implementations. The authors believe that although these penalties appear to be higher than those traditionally used in travel demand models, as stated by Liu et al. (4), they may be still too low and misleading for estimating transit ridership, especially the mode shift from car to transit. This opinion may be explained by the fact that the penalties were calculated as an average between the preferences of transit and car riders. The aforementioned average is destructive to the modeling effort because when an investment in a new mode of transit is evaluated, it is crucial to accurately estimate how many car riders would change their behavior and start using the new transit system. For that purpose, any artificial reduction of the transfer penalty, as well as other transit penalties, exaggerates the number of car riders shifting to the transit mode. The foregoing line of research is followed here, and further segmentation of car riders is proposed. The assumption is that further 109

2 110 Transportation Research Record 1894 segmentation will improve the quality and the reliability of the model estimation process. This improvement would be related to the expected demand for alternative modes as a result of foreseeable changes in the level of service provided by the different modes of travel. The methodology to segment the car-rider population and prepare the data from an RP survey is presented first, followed by the results obtained from the RP survey and the conclusions from the data analysis. METHODOLOGY Population Segmentation Wardman s (3) line of research is used as a starting point for further segmentation of the car-rider population in two dimensions: 1. Car drivers, car passengers, and bus riders distinguished according to their driving status in the household and 2. The level of investment in car mobility in the household. For both segmentations, a significant difference in the VOT and selected level-of-service parameters was expected, motivated by the fact that in the same household, it may be acknowledged that a higher necessity for taking the car should be expressed in a stronger rejection of transit attributes, such as walking, waiting, and transfer time, coupled with a higher VOT. At the same time, it was expected that members of the household with a higher investment in car mobility would express a higher VOT and would perceive higher transit penalties associated with transit usage. At this point, it is appropriate to comment on two other common sources for segmentation: the number of cars in household and household income. The investment in car mobility, defined as the total market value of the cars available to household members, may be correlated with both sources. The number of cars in the household is included as a variable in the models presented here, and therefore this factor is already taken into account. It is believed that the investment in car mobility represents a revealed preference of the household that reflects its wealth, size, and needs. Thus, this variable may represent household lifestyle better than income. Lifestyle covers not only the income factor but also household priorities, such as the way in which income is allocated for different uses, either for production or for consumption, as far as mobility is concerned. The investment in mobility by car is not presented here as a proxy for income. Rather, it is used as a presentation of lifestyle. There is no question that lifestyle and investment in cars are heavily influenced by household wealth. However, there are other considerations that derive from social group identification, personal psychology, and functional needs, which are not captured using the income variable alone. There is a practical problem with respect to the income variable. In Israel, the information collected in RP surveys about income level for each household is not reliable. Experience in various surveys has shown that respondents are reluctant to provide such information, and very often the information collected is not accurate enough to be included as a variable in the models. The investment in car mobility is a much more reliable variable in this respect. Total investment was chosen rather than average investment per household. For example, a family with two adults and two children with a total car investment of $20,000 would be defined here as a high-investment family compared with a single person living alone with an investment of $5,000, which would be defined as a low investment in car mobility. If the average measure was used instead of the total amount, both households would be considered equally as having a low investment in car mobility. In this example, the first household chose to invest more in car mobility than the second household; the hypothesis here is that the first household is significantly different in its set of preferences related to travel mode as compared with the second household. In another example, a household composed of two adults with $20,000 investment in car mobility, the average would be higher compared with the household with four persons. In this example, it would be expected that the two households would exhibit a rather similar set of preferences relating to travel mode. Modeling Procedure This research focuses on the methodological aspects of population segmentation rather than on model structure and calibration. For this reason, the multinomial logit model was used with the same utility function for all models tested. In this way, the modeling estimation procedure was kept constant throughout, and the focus was concentrated on different segments of the population. The database used for model estimation is a subsample from the National Travel Habits Survey (6) carried out by the Israeli Central Bureau of Statistics in on behalf of the Ministry of Transport. The study reported here was confined to the Tel Aviv metropolitan area (about 1.7 million inhabitants in 1996), since reliable level-of-service data were available only for this region. The survey is a typical RP study. About 1% of the households were surveyed, and for each person above age 14, a 3-day diary was filled. A total of 29,506 observations, corresponding to trips departing from home, were selected for the analysis. Chained trips were purposely avoided, since for more than 95% of these cases, the chosen mode was identical to the mode used in the trip departing from home. The travel times for car and transit modes were imported from Emme/2 networks used for modeling a light-rail transit project planned in the Tel Aviv metropolitan area (7). The survey collected additional information about the cars in the household. According to the information on the questionnaire about the year of production and engine size, average market values were calculated for each car in the household. Table 1 presents average car values (December 1996 prices) according to the price booklet used for car insurance companies (8). TABLE 1 Average Car Prices for Given Engine Size and Production Year Engine Size (cm 3 ) Production Year Up to and more Up to December 1996 prices in thousands of NIS; 1 U.S. dollar = NIS.

3 Elgar and Bekhor 111 A hypothesis is posed that there is a significant correlation between the level of investment in mobility with private vehicles, as reflected by the market value of the cars possessed by household members, and typical perceived preferences derived from mode choice model estimations as follows: VOT, Transfer penalty, Waiting time penalty, and Walk time penalty. The investment in car mobility is defined as the sum of the estimated cost of all cars possessed by the household members. A positive relationship was expected between the level of investment and the penalties perceived by car passengers concerning travel by transit. For simplicity, two classes were concentrated on: households with high investment in mobility by car and households with low investment in mobility by car relative to a reference value of 51,000 New Israeli Shekels (NIS) (U.S.$15,721). This value divides the observations close to their median value, which is the reason for its choice. Most households do not purchase a new car every year. However, they tend to keep up with a standard, which means buying a newer car every 2 to 4 years of use. The result is an overlap between those defined as high investors in mobility at the lower margin of 51,000 NIS and those defined as low investors at the upper margin of 51,000 NIS. This problem is inherent in any continuous variable cut into two categories, and the hypothesis is that this phenomenon would not basically alter the results. Figure 1 illustrates the segmentation procedure. RESULTS Model Estimation The segmentation procedure refers to all trips departing from home. Seven modeling runs were performed, according to the seg- mentation diagram presented in Figure 1. The first three runs were as follows: Model A: the average preference system of all trips departing from home (all hours) with three choice alternatives between car (either driver or passenger) and bus (29,506 observations). Model B: all observations from households with at least one car driver in the household (does not include households without car drivers). The choice alternatives in this case are also three modes (20,133 observations). Model C: the complementary population with respect to Model B, that is, households without car drivers. The choice alternatives in this case are only two modes: car passengers and bus riders (9,373 observations). Table 2 shows the results for Models A, B, and C. The next four runs reflect two dimensions of the segmentation: the first concerns driving passenger status and the second, level of household investment in mobility by car. Models D and E respectively refer to car drivers with an investment under 51,000 NIS (about U.S.$15,720 in 1996 prices) and those with an investment higher than U.S.$15,720. This value represents the average household investment for the sample. Table 3 shows the results for Models B (presented for comparison purposes), D, and E. Models F and G respectively refer to car passengers in households with an investment lower and higher than 51,000 NIS. Table 4 shows the results for Models C (presented for comparison purposes), F, and G. Analysis of Findings VOT Comparison The VOT is calculated as follows: VOT m = β β in-vehicle time, m cost, m 60 where m represents the travel mode. All users (Model A) 29,506 observations car drivers (Model B) 20,133 observations out car drivers (Model C) 9,373 observations low investment in car mobility (Model D) 11,681 obs. high investment in car mobility (Model E) 8,452 obs. low investment in car mobility (Model F) 7,648 obs. high investment in car mobility (Model G) 1,725 obs. FIGURE 1 Segmentation diagram.

4 112 Transportation Research Record 1894 TABLE 2 Results for Models A, B, and C Model B Model C Variable Description Model A (entire sample) (households with drivers) (households without drivers) coefficient t-stat coefficient t-stat coefficient t-stat Cd- constant Cp- constant Bus- number of transfers Bus- walk time (min) Bus- in-vehicle time (min) Bus- wait time (min) Bus- fare (NIS) Cd- in-vehicle time (min) Cd- cost (NIS) Cd- park cost (NIS) Cd- park search time (min) Cp- in-vehicle time (min) Cp- cost (NIS) Cp- park cost (NIS) Cp- park search time (min) Cd- dummy for 2+ cars in hh Cp- dummy for 2+ cars in hh Total number of observations Car drivers Car passengers Bus riders Likelihood (Zero) Likelihood (Constants) Likelihood (Final) "Rho-Squared" w.r.t. Zero "Rho-Squared" w.r.t. Const Cd- VOT (NIS/hr) Cp- VOT (NIS/hr) Bus- VOT (NIS/hr) Bus- wait time penalty Bus- walk time penalty Bus- transfer penalty $1U.S. = NIS. Cd = car driver; Cp = car passenger; hh = household; w.r.t. = with respect to. Model A was estimated with all available observations. The VOT results represent the average VOT of car drivers equal to 52.8 NIS/h (about U.S.$16.30/h in December 1996 prices) and the VOT of car passengers equal to 25.6 NIS/h (U.S.$7.90/h). Bus riders had a much lower VOT: 5.0 NIS/h or U.S.$1.50/h, which indicates that it is related to the population that is less economically productive. In Model B, the first segmentation step is taken in reference to households with at least one car driver. Here the VOT for all users increased with respect to Model A: the car-driver VOT increased to 57.3 NIS/h (U.S.$17.70/h), the car-passenger VOT increased to 28.6 NIS/h (U.S.$8.80/h), and the VOT for bus riders increased to about 8.7 NIS/h (U.S.$2.70/h). Model C refers to car passengers and bus riders only. The carpassenger VOT decreased compared with Model A 21.7 NIS/h (U.S.$6.70/h) and the VOT for bus riders increased to 10.9 NIS/h (U.S.$3.40/h). In this segment, in which there are no car drivers, every person in the household is confined mainly to using the bus and riding as a passenger in cars from other households. This population also includes productive members in the households, whereas bus riders in Model B are mostly the weak and less productive persons in the household, expressed by the low VOT. Models D and E are segmented with respect to the population in Model B according to the level of investment in mobility by car. Model D shows that both car drivers and car passengers in households with a low investment in car mobility have lower VOTs (38.0 NIS/h and 17.8 NIS/h, respectively), whereas bus riders have a higher VOT (12.5 NIS/h). Model E shows that for households with a high investment in car mobility, the VOT results exhibit the opposite pattern from that for Model D: both car drivers and passengers increased their VOT, and bus riders decreased their VOT. Models F and G respectively refer to car passengers in households with less than a 51,000 NIS investment in mobility by car (Model F) and with higher investment (Model G). The comparison between Models F and G illuminates the results of Model C in a new way and emphasizes the benefit of relevant population segmentation for modeling preferences and behavior. The results in Models F and G show a similar pattern to that in Models D and E. In households with a low investment in car mobil-

5 Elgar and Bekhor 113 TABLE 3 Results for Models B, D, and E Variable Description Model B Model D Model E (households with drivers) (low investment in car mobility) (high investment in car mobility) coefficient t-stat coefficient t-stat coefficient t-stat Cd- constant Cp- constant Bus- number of transfers Bus- walk time (min) Bus- in-vehicle time (min) Bus- wait time (min) Bus- fare (NIS) Cd- in-vehicle time (min) Cd- cost (NIS) Cd- park cost (NIS) Cd- park search time (min) Cp- in-vehicle time (min) Cp- cost (NIS) Cp- park cost (NIS) Cp- park search time (min) Cd- dummy for 2+ cars in hh Cp- dummy for 2+ cars in hh Total number of observations Car drivers Car passengers Bus riders Likelihood (Zero) Likelihood (Constants) Likelihood (Final) "Rho-Squared" w.r.t. Zero "Rho-Squared" w.r.t. Const Cd- VOT (NIS/hr) Cp- VOT (NIS/hr) Bus- VOT (NIS/hr) Bus- wait time penalty Bus- walk time penalty Bus- transfer penalty $1U.S. = NIS. Cd = car driver; Cp = car passenger; hh = household; w.r.t. = with respect to. ity, the car-passenger VOT (16.9 NIS/h) decreases and the bus-rider VOT (10.9 NIS/h) remains the same. In households with a high investment in car mobility, car-passenger VOT (28.8 NIS/h) increases and bus-rider VOT (4.6 NIS/h) decreases. In summary, the VOT results indicate that the segmentation according to investment in car mobility produces very different models, which should be accounted for in estimation procedures. Transit Penalties Tables 2, 3, and 4 present results for three transit penalties: walk time, wait time, and number of transfers. All penalties were computed by dividing the respective coefficient by the bus in-vehicle time coefficient. As with the VOT comparison, Model A was used as the basis for comparison with all other models. It should be remembered that this model was estimated by using all available observations, that is, for persons living in households with at least one car. Model B is used as the basis for comparison with Models D and E, and Model C is compared with Models F and G. Bus Wait Time Penalty In Model B, the penalty is 0.9 min only compared with the average value of 1.3 min (Model A). This finding may be explained by the fact that in this segmentation there is a healthier population with respect to those in other segments, who are less sensitive to walking time and distance. It should be noted that in Model C the wait time penalty increased to 1.6 min. When the population is segmented according to investment in car mobility, a very similar pattern to that for the VOT is obtained: lower sensitivity to wait time for Model D (0.8 min), and higher sensitivity to wait time for Model E (1.0 min). The same pattern is observed when Models F and G are compared with Model C. Again, using solely models like A, B, and C as a basis for planning, biased estimates may be obtained. Bus Walk Time Penalty The walk time penalties exhibit a similar pattern to the wait time penalties except for the relation between Models B and C. In this case, the walk time penalty in Model B (2.1 min) is higher than the corresponding value in Model C (2.0 min), which is the opposite of the relation found in the wait time penalties.

6 114 Transportation Research Record 1894 TABLE 4 Results for Models C, F, and G Model C Model F Model G Variable Description (households without drivers) (low investment in car mobility) (high investment in car mobility) coefficient t-stat coefficient t-stat coefficient t-stat Cd- constant Cp- constant Bus- number of transfers Bus- walk time (min) Bus- in-vehicle time (min) Bus- wait time (min) Bus- fare (NIS) Cd- in-vehicle time (min) Cd- cost (NIS) Cd- park cost (NIS) Cd- park search time (min) Cp- in-vehicle time (min) Cp- cost (NIS) Cp- park cost (NIS) Cp- park search time (min) Cd- dummy for 2+ cars in hh Cp- dummy for 2+ cars in hh Total number of observations Car drivers Car passengers Bus riders Likelihood (Zero) Likelihood (Constants) Likelihood (Final) "Rho-Squared" w.r.t. Zero "Rho-Squared" w.r.t. Const Cd- VOT (NIS/hr) Cp- VOT (NIS/hr) Bus- VOT (NIS/hr) Bus- wait time penalty Bus- walk time penalty Bus- transfer penalty $1U.S. = NIS. Cd = car driver; Cp = car passenger; hh = household; w.r.t. = with respect to. It is worth noting that walk time penalties are constantly higher than wait time penalties. In Israel, many bus stops have seats and roofs, reducing the inconvenience of waiting for a bus. The walk time penalty reflects the additional effort and inconvenience imposed on travelers. Bus Transfer Penalty In Model A, the transfer penalty is equivalent to 21.2 min of in-vehicle bus travel time. This value is already high compared with other transfer penalties derived for bus riders in other RP surveys in the Tel Aviv metropolitan area. Local agencies have traditionally used 12- to 15-min transfer penalties in transit assignment models. The first segmentation dimension (Models B and C) shows that the transfer penalties change to about 18.5 min and 16 min, respectively. Further segmentation in relation to both driving status and household investment in car mobility produces in Model D a transfer penalty of 17.9 min for low investment and a penalty of 19.4 min for households with a high level of investment for car drivers (Model E). The magnitude of these transfer penalties is still substantial, and it becomes more pronounced when combined with the other transit penalties, as will be shown. Car and Transit In-Vehicle Travel Times The average car travel time in the survey for the Tel Aviv metropolitan area is 19.1 min, and the respective in-vehicle travel time for transit is 23.8 min. Thus, the average difference is 4.7 min, and this value should be added to the total cost of moving a person from car to transit. Cumulative Effect of Bus Penalties To get a better perception of the bus penalties, their cumulative effect was analyzed. Table 5 presents the combined travel time results. The combined wait and walk time was calculated according to the average transit trip in the data, which involves 7 min waiting time and 5 min walking time. The total time was calculated by adding 4.7 min (the difference between in-vehicle travel times) to the combined times calculated from the transit penalties. It should be noted that in Model A, the total walk and wait time is very similar to the transfer time. This finding is explained by the fact that the average number of transfers in the Tel Aviv metropolitan area is quite low (15% in the sample).

7 Elgar and Bekhor 115 TABLE 5 Cumulative Effect of Bus Penalties Model Wait Penalty Walk Penalty Transfer Penalty Wait + Walk Time Total Time (min) (min) (min) (min) (min) A- entire sample B- with drivers C- without drivers D- low investment E- high investment F- low investment G- high investment A quick glance at the last column in Table 5 can explain why there is practically no chance to divert car users from using vehicles when a transfer between different travel modes is needed. The lowest value of the combined total time caused by the transit penalties is as high as 38.4 min. Table 6 presents the monetary values of the total times to illustrate the perceived cost of each segmented population for a shift from car to transit. The results presented in Table 6 may explain why in most cases of transit improvements, the observed reduction in congestion is small. It should be recalled that the results presented in Table 6 were computed by using average travel time values. Therefore, it is expected that most transit passengers would switch from being car passengers, and relatively few of them would switch from being car drivers. This switch in turn would have a minimal impact on car trips and hence congestion reduction would be limited. SUMMARY AND CONCLUSIONS The analysis of the findings shows that proper segmentation procedure yields a much deeper understanding of mode choice behavior. The modeling procedure used here illuminated the fact that the average VOT for car riders is quite high for all trips departing from home (about U.S.$16.30/h), which is much higher than the figure acknowledged by the Israeli Ministry of Finance, 19 NIS, or U.S.$5.90 (9). Further segmentation to households with and without car drivers showed that the foregoing VOT differs significantly for car drivers, car passengers, and bus riders. The second segmentation dimension (investment in mobility by car) significantly sharpened the differences in preferences among car-rider segments. The estimated models showed that under the prevailing level of service of the bus system in the metropolitan area of Tel Aviv, only persons with a very low VOT choose bus if their alternative is car driving. A very similar pattern emerged in the three bus penalty parameters (walk time, wait time, and transfer). A higher level of investment in car mobility is correlated with a higher time penalty. The estimation results indicate that even with massive improvements in transit, the diverted users to transit would come mainly from car passengers and only marginally from car drivers. Thus, the impact of improved transit on road congestion would be quite limited. The combination of high VOT for most car riders (car drivers and part of the car passengers) with a substantially higher transfer penalty for car drivers, in comparison with current values used in demand modeling, may explain a considerable part of the perceived errors. A proper segmentation procedure can not only narrow the error in estimation but can also help to focus the effort of the transportation planner on those segments of the population that can be influenced to change their behavior toward transit by directing the limited resources to improve the elements of transit service that are most sensitive to those segments of the population, mostly car passengers. FURTHER RESEARCH The specific results are valid for the Tel Aviv metropolitan area subject to the RP survey procedure of home interviews. The study reflects the attitudes toward the transportation system and the conditions prevailing at that time (1996) and of course did not consider possible new modes of travel. The bus is still the main transit mode in the Tel Aviv metropolitan area. A light-rail transit (LRT) system is currently being planned (7 ). It would be interesting to compare the results with an SP study conducted to evaluate the potential market for the LRT system. However, the SP survey did not collect information about car characteristics, so only in a forthcoming SP survey will it be possible to compare the results. Some parameters like parking cost and parking search time were found not statistically significant in some modes because in 1996 there were parking limitations in only a few traffic zones. In addition, the level-of-service data were not produced from parking surveys. Instead, they were inferred from qualitative land use data. For these reasons, an investigation of the influence of parking variables was avoided in the segmentation procedure. A parking survey was conducted recently, and the results may be used in subsequent model estimations. The consistency of the results was encouraging for the validity of the procedure proposed here, and its adoption will improve modeling TABLE 6 Model VOT Total Time Total Cost of Bus Penalties ($/hour) (min) ($) Car Car Car Car Driver Passenger Bus Driver Passenger Bus A- entire sample B-with drivers C- without drivers D- low invest E- high invest F- low invest G- high invest $1U.S. = NIS. Monetary Values of Cumulative Effect of Bus Penalties

8 116 Transportation Research Record 1894 quality when changes in modal transportation services are considered. It should be interesting to compare forecast results using the proposed segmentation with current forecasts. This study is also left for further research. The authors are aware of the fact that the results of this study are applicable mainly to the Tel Aviv metropolitan area, and of course it is advisable that each urban area carry on its own survey and develop its own set of models and parameters. However, the general applicability of the methodology discussed here is believed to render better insight into the searched segments and will demonstrate the significant differences among them. REFERENCES 1. Pickrell, D. H. A Desire Named Streetcar: Fantasy and Fact in Rail Transit Planning. Journal of the American Planning Association, Vol. 58, No. 2, 1992, pp Flyvbjerg, B., N. Bruzelius, and W. Rothengatter. Megaprojects and Risk. Cambridge University Press, Cambridge, United Kingdom, Wardman, M. Public Transport Values of Time. ITS Working Paper 564. University of Leeds, United Kingdom, Liu, R., R. M. Pendyala, and S. Polzin. Assessment of Intermodal Transfer Penalties Using Stated Preference Data. In Transportation Research Record 1607, TRB, National Research Council, Washington, D.C., 1997, pp Central Transportation Planning Staff. Transfer Penalties in Urban Mode Choice Modeling. FTA, U.S. Department of Transportation, Central Bureau of Statistics. National Travel Habits Survey (in Hebrew). Israeli Ministry of Transport, Perlstein Galit Ltd. Travel Demand Forecasting Model. NTA (mass transit system for Tel Aviv metropolitan area), Tel Aviv, Israel, Car Price Tables (in Hebrew). Levi-Itzhak, Ltd., Israel. Dec Hague Consulting Group. Recommendations for a Revised Nohal Prat. Israeli Ministry of Infrastructure, Publication of this paper sponsored by Traveler Behavior and Values Committee.