Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
We estimated new MNL models for the six trip pur- pose and direction combinations, using 32 alternatives. Compared with the remainder of the modeling system, the a.m. and p.m. peak periods are both expanded to include wider shoulder periods. The a.m. peak, midday, and p.m. peak periods are set to be 5 h long and contain ten half-hour subperiods. The evening and night periods remain as single periods, spanning 3 and 6 h, respectively. We use period-specific variables and shift variables to move trips earlier or later within each of the three larger periods. The shift variables are nonlinear; for exam ple, it may take more than twice as much a.m. peak congestion to get someone to shift their departure time 30 min ear- lier than it does to get the same person to shift 60 min earlier. The following variables were tested in the model esti- mation: sociodemographic (i.e., income and household size); land use and accessibility (i.e., total employment accessible by auto within 6 min and retail employment accessible by auto within 15 min); and originâdestination and level of service (i.e., auto in-vehicle generalized cost in minutes during each of the five periods, bridge dummy variable, and shared ride dummy variable). All the variables that were tested and either retained or excluded in the utility equation for the logit choice mod- els were based on their significance, whether the sign of the coefficient was logical, and whether the data can be forecasted by the metropolitan planning organization or the DOT. Variables that were tested but not included in the final models are the employment accessibility vari- ables. A description of the variables retained in the final models and the impact of these variables on the temporal choice behavior of travelers are as follows: ⢠Household incomeâThe dummy variable that indicates high-income group (>$75K) has a significant coefficient specific to p.m. peak period in the WH model whereas in the HW model, the coefficient is significant in the a.m. peak period. This indi cates that commuters from higher-income households are more likely to travel to work during the a.m. peak period and less likely to travel during the p.m. peak period. This is further cor- roborated in the WH model where the income coeffi- cient for the a.m. period was insignificant (and not included), suggesting that commute trips from higher- income households are not as likely to be destined to home during the morning peak period. The income coef- ficients for the p.m. time period are greater than for other time periods in the WH model, indicating that higher-income commuters are more likely to return home during the p.m. peak period. The lower-income variable (<$45K) has a negative coefficient in the a.m. peak period of the HW model, probably because lower- income jobs have more irregular hours than high- income jobs and are more likely to occur in off-peak periods. ⢠Household sizeâLarger households are less likely to travel to work in the a.m. peak than smaller house- holds, as indicated by the negative and significant house- hold size coefficient in the HW model. It may be that larger household sizes indicate the presence of children or more complicated household structures, which, com- bined with multiple workers in the household, lead to flexible or extended work schedules resulting in more reverse direction work trips. By contrast, smaller house- holds are more likely to return home from work in the p.m. peak period, as indicated by the negative and sig- nificant coef ficient in the WH model. It is possible that smaller households have fewer out side constraints on work hours and schedules, and work trips can occur in more traditional work hours. ⢠Carpool dummyâIf a WH trip is made using the carpool mode of travel, then this variable is equal to 1; otherwise, it equals 0. The coefficient of this variable is very significant and positive in the HW model, and very significant and negative in the WH model. This coeffi- cient is negative in the WH model, indi cating that car- pool trips from work to home are less likely to occur, and it is hypothesized that there are fewer opportunities for casual carpooling from work to home than vice versa. For nonwork trips, this is significant and positive in both directions of the trip except for trips returning home dur- ing the a.m. peak period where it is negative because car- pooling is usually not an option from a nonhomeâ nonwork location. ⢠Bridge dummyâIf a trip is made using one of three bridges in the Puget Sound region (namely Tacoma Nar- rows, I-90, and SR 520), then this variable is equal to 1; and, if not, it is 0. In the HW model, this coefficient was significant and positive in the a.m. peak period, indicat- ing that it is more likely that trips across the bridge will be made during morning peak hours solely for work- related purposes. These coefficients were more signifi- cant in the midday and p.m. peak periods of the WH model, indicating a higher likelihood of trips across bridges in the reverse work commute direc tion. This was found to be significant and positive in the nonwork mod- els during the midday period, indicating the propensity of nonwork travelers to opt for uncongested periods to perform nonwork activities. ⢠Congestion levelâThe level of congestion or delay is measured by the difference in gener alized cost (in min- utes) for a.m., midday, p.m., and evening time periods and the general ized cost for nighttime period. This vari- able is found to be negative and significant in all models, indicating that delay affects travel decisions by time-of- day choice significantly. The size of the coefficient in the HW model is less negative than in the WH model, indi- 145INNOVATIVE METHODS FOR PRICING STUDIES