A Statistical Framework for Dealing with Endogeneity

Abstract

We propose a general framework for dealing with endogeneity in models in marketing and economics. It consists of a multivariate, hierarchical, mixed discrete/continuous representation of behavioral response variables. Importantly, it includes a non-parametric approximation to unobserved sources of exogenous information. It complements the instrumental variables (IV) approach in that it may but does not need to include, observable instruments. After presenting the theoretical basis of the method, a simulation study reveals that parameters can be estimated consistently even if instruments are not observed. The proposed approach is applied in three case studies in business and economics. They include a case where a standard IV is inadequate in correcting for endogeneity bias, and two cases where IVs are not available. In the examples, the proposed framework corrects for endogeneity bias without recourse to IVs. Resulting policy actions are shown to be different from equivalent models that ignore endogeneity. We conclude that the approach has applications in marketing and economics as a framework for testing for conjectured endogeneity. The development of theoretical arguments motivating the investigation of endogeneity remains crucial, but even after such a rigorous theoretical analysis there will remain instances in which instruments are not available, cannot be found, or where empirically their quality is insufficient, in which case the proposed framework provides a useful alternative.