Modeling Destination Choice for Shopping Using a GIS-based Spatiotemporal Framework: An Investigation of Choice Set Generation and Scale Effects

Abstract

<p>Incorrect specification of destination choice sets and changes in scale or unit definition can lead to biased parameter estimates and predictions as well as influence findings in statistical tests. Through an empirical study, this thesis evaluates bias as a result of unconstrained destination choice sets and scale effects. Given the deficiency of most current destination choice models, which is the lack of integration of spatio-temporal constraints in generating destination choice sets, the activity-based approach is proposed as a solution by taking into consideration both spatial and temporal constraints in the generation process. Standardized industrial classification (SIC) codes adjoined to shopping opportunities are used to facilitate the discrimination of different shopping types and the classification of shopping stores in order to better understand shopping behaviour.</p> <p>Analytical results obtained by techniques sensitive to scale effects or zomng effects are likely to change as the aggregation level or area boundary varies. Traffic analysis zones (TAZ) as predefined analysis unit in transportation-related research may not be an optimal choice in the context of destination choice behaviour. Documenting the results on model estimations at different scales and zoning levels is important to investigate the modifiable areal unit problem (MAUP) effects and critically assess the reliability of the estimates. Sensitivity of parameter estimations and model goodness-of-fit between the TAZ system and 10 randomly generated grid systems show remarkable differences. Under a series of criteria, the best zoning system is recommended with certain conditions applied. Our results support the suspicion on the suitability of predefined analysis units like TAZ and suggest grid systems could be a potential substitution.</p> <p>The study area in this research spans seven counties of the Louisville MSA. Three primary data sources are used in our analysis: (1) a travel diary survey conducted in 2000 for seven counties of the Louisville MSA; (2) a 2002 Dynamap/Transportation 4.0 network produced by Geographic Data Technology Inc. (GDT); and (3) an urban opportunities data set for the Louisville MSA as geocoded from a database obtained from ReferenceUSA.</p> <p>A time geography perspective, constrained destination choice set, discrimination of grocery and nongrocery shopping and MAUP effects on destination choice model characterize the contributions of this research to the literature.</p>