The usefulness of socio-demographic variables in predicting purchase decisions: Evidence from machine learning procedures

Research has long debated the effectiveness of socio-demographics in understanding purchase behavior, with mixed conclusions. The appeal of socio-demographic data for customer relationship marketing is based on its low acquisition cost and the growing array of variables on which marketers can condition messages and offers. We reinvestigate the value of socio-demographic variables, focusing on the potential of machine learning procedures (MLPs) to extract a stronger and reliable signal than the standard linear-in-parameters (logistic) regression models. We explore how predictive power can be increased through the nonlinearities and interactions identified with MLPs; our experimental set ranges from well-established procedures to newer entrants in this space. We also examine causality vis-a-vis predictability using a propensity scoring approach. Empirics are based on six grocery product categories and more than 7,000 panelists. We find that, relative to logistic regression models, MLPs using demographic variables yield a 20% to 33% improvement in out-of-sample predictive accuracy.

Automated summary

This study reinvestigates the value of socio-demographic variables in understanding purchase behavior and demonstrates that machine learning procedures can improve predictive accuracy.