Suggested Citation:
Mokhtarian, Patricia L. and Wei Tang (2011) Trivariate Probit Models of Pre-purchase/Purchase Shopping Channel Choice: Clothing Purchases in Northern California. Institute of Transportation Studies, University of California, Davis, Research Report UCD-ITS-RR-11-15

This study analyzes the joint choice of pre-purchase and purchase shopping channels for clothing purchases, using data collected from an internet-based survey of two university towns in Northern California (final Ns=390 and 452). Descriptive analysis clearly shows dependence across these three choices: in particular, the “sticky” combinations of {only-store pre-purchase + store purchase} and {only internet pre-purchase + internet purchase} occur substantially more often than independent choices would predict. We develop two trivariate probit (TVP) models, consisting of two binary choice equations for the pre-purchase channel (respectively measuring the use of store or not, and the use of internet or not) and one binary choice equation for the purchase channel (store or internet). One model allows prior channel purchase experience variables to enter while the other model excludes them. The results further confirm the dependency among pre-purchase and purchase channel choices, with all three error term correlations strongly significant. In addition to breadth and depth of experience variables and channel-specific perceptions (post-purchase satisfaction, cost savings, enjoyment, and convenience), significant explanatory variables include general shopping-related attitudes (pro-exercise, shop enjoyment, and store enjoyment), context variables, and sociodemographic traits (age and income). Prediction of joint choice probabilities was considerably better for the TVP model than for independent binary choice models, confirming the value of simultaneously modeling pre-purchase and purchase channel choice bundles.

Keywords: internet shopping, online shopping, store shopping, pre-purchase channel, purchase channel, multichannel shopping, multivariate probit model, trivariate probit model