Performance of the integrated random coefficients covariance probit model: Implications for brand choice

Abstract

In probit brand choice models, researchers increasingly focus on random variation of key parameters of interest. An important source of such variation may be due to differing consumer response to marketing efforts across purchase occasions. Also, variables common across alternatives influence purchase behavior, but may not be measured. Failure to adequately control for both these factors may cause biased estimates. More importantly, the incorrect estimates may lead to erroneous managerial conclusions. While either problem can be directly addressed in a (non-independent) probit model, we focus on whether a joint specification of both factors is necessary. We highlight the potential for misleading managerial inferences by contrasting the model with both sources of variation against nested models, using empirical data on ketchup product category. Further, in a simulation study we establish the extent of bias in recovery of parameters due to different specifications of both factors.