Exploring the nonlinearity of commute-time utility is useful in predicting urban travel demand. However, existing studies assume that commute utility decreases linearly with commute time. This ignores the influence of commuters’ ideal preferences and tolerance thresholds on commute utility. To reveal the nonlinear variation of commute utility with commute time and its influence on the choice of commute mode, the ideal commute time (ICT) and the tolerance threshold for commute time (TTCT) were introduced. Three-piecewise linear utility models (Models 2, 3, 4) were constructed and compared with the linear utility multinomial logit model (Model 1). The results of empirical study showed that: (a) the goodness of fit of these three modified models is higher than that of Model 1, indicating that the fitting effect of the commute-time utility model can be improved with either or both the ICT and TTCT; (b) there is a nonlinear relationship between the commute utility and the commute time. The commute-time utility decreases slowly before ICT, declines steadily between the ICT and the TTCT, and falls significantly after TTCT; and (c) when the commute time exceeds the TTCT, the perceived utility of commuters traveling by walking or cycling decreases significantly, and there are few changes in the perceived utility of commuters by car, which increases the probability that commuters who use active modes will transfer to commuting by car. The research results have implications for improving the prediction capacity of commute mode choice model and could guide commuters to switch to more sustainable commute modes.

No abstract

The tolerance threshold of commute time (TTCT) reflects the longest commute time that commuters can tolerate from home to the workplace. When the commute time exceeds the TTCT, the commuting utility significantly reduces, which has a nonlinear influence on commuting mode choice. To reveal the nonlinear relationship between the commuting utility and commute time, the TTCT is introduced to constrained multinomial logit (CMNL) model based on the semicompensatory decision-making mechanism. In addition, an empirical study is carried out on 405 commuters in Kunming, China. The results show that the CMNL model has a higher fitting accuracy than the MNL model, which indicates that the TTCT is a significant explanatory variable for the commuting mode choice. Moreover, the commuting utility does not decrease linearly with the commute time. An appropriate commute time range (about 5–25 min) could bring positive commute utility to the commuters, but the commute utility is negatively impacted when the commute time is larger than the TTCT. Therefore, it is necessary importing the TTCT in the utility function to improve the predictive power of the commuting mode choice model.