Abstract
Activity and travel demand analysis can be greatly improved by a more realistic description of the relationship between activity engagement and trip-making behavior when person and household variation are accounted for jointly. Four activity pattern groups and four travel pattern groups, obtained in the past using cluster analysis and applied to behavioral indicators (e.g., activity frequency and duration), are analyzed using multinomial logit and multilevel multinomial logit (multi-MNL) models. All models include temporal, spatial (residence and workplace), household, and person effects. The multilevel models, designed for hierarchical data structures, also include correlated random components among the pattern options at the household and person levels. First, past research results are confirmed that contextual variables (e.g., household, neighborhood, time) are important explanatory variables for these activity and travel patterns and that temporal effects are much stronger for activity pattern groups than for travel pattern groups. Then, using the multi-MNL models, a complex structure in unobserved heterogeneity is identified in which choices are correlated among some patterns at the household level and among other patterns at the person level. Both travel and activity pattern analyses, however, indicate that unobserved heterogeneity should be accounted for at both levels when regression models are estimated.