How does the length of residence in a neighborhood vary the effects of neighborhood land use on commuting trip time and mode choice?

Abstract

Recent studies that estimated the land use–travel relationship while controlling for residential self-selection mostly acknowledged the potential for the self-selection in the reverse direction according to learning and self-justification over time. Thus, this study empirically tests how the relationship varies by the length of time after residential move in Seoul, South Korea. Analytical results show that population density encourages the choice of automobile alternatives in extreme cases, that is, for newly moved first-year residents who decided to tolerate congestion and for longtime residents of 4 years or more who had a due chance to learn and appreciate neighborhood travel options and also, it extends commuting time for these short and longtime residents. Employment density reduces commuting time for all but longtime residents because of better jobs–housing balance while it is not necessarily accompanied by mode choice variations. Street intersection density is associated with commuting time variations in all cases. Land use balance entropy is unrelated to the time variations, but consistently related to the alternative mode choice, and possibly beneficial to trip-chaining. While public transit densities are classified into subway and bus densities, the former turns out to be significant regarding commuting time and mode choice for most residents, but the latter only for first-year residents. Particularly on the mode choice, the gradual learning effect is associated with subway density (significant after 1 year), whereas the self-justification effect by bus density is rather quick, but not lasting (significant only for 1 year).

Appendix

As shown in “Literature Review,” this study differentiated the residence of four or more years from the shorter continuous residence (i.e., 1, 2, and 3 years). As a sensitivity analysis, this study conducted linear and logistic regression using the cut-off points of the five-plus years (n = 8904) and six-plus years (n = 7200) instead of the original four-plus years (n = 10,177). As shown in Tables

Table 6 Sensitivity analysis of linear regression: trip time

6 and

Table 7 Sensitivity analysis of logistic regression: mode choice (alternatives vs. automobile)

7, regardless of the cut-off point, results in the alternative models were generally consistent with those from the original four-plus-year model in terms of the coefficient significance and direction (+/−). Among differences in the linear regression models, C1_LvEnvRsk was insignificant in the six-plus-year model (while significant in the five-plus-year and original four-plus-year models). Among the logistic regression models, Emp_Den became significant in the six-plus-year model instead of Sub1000_Den and also, C2_LvEnvStf turned to be insignificant. Except for these differences, compared to the models of the 1, 2, and 3 years, the results of the above alternative models were substantially similar and consistent.