Factor-Analytic Specifications of Mixed Logit

Abstract

Factor-analytic specifications of mixed logit are especially useful in representing patterns of preference heterogeneity. The latent variables of factor-analytic specifications represent unobserved sources of heterogeneity in the effects of the explanators of the mixed logit model. One such application is market structure analysis – an important topic within the marketing literature on brand choice. This paper reports an analysis of brand preferences measured using best-worst Type I (the object case) and paired-comparison methods. The analysis leverages the within-subjects design – the same respondents complete both methods. The paper makes an empirical contribution based on the specification and testing of the theoretical model of interest – a multitrait-multimethod (MT-MM) specification of mixed logit (a type of factor-analytic specification). The MT-MM model partitions the heterogeneity into structural, trait, and method components and is compared to three rival specifications. The MT-MM model is more parsimonious, offers better fit (based on measures of absolute and comparative fit), and has better interpretations than the competing models. We suggest future applications of the MT-MM model that should be of interest to choice modellers in a range of fields. The factor-analytic approach opens a new pathway for developing and testing theory of preference heterogeneity in a systematic way.