Hybrid BYO Conjoint with Boosting for Data Fusion

Abstract

This research advances build-your-own (BYO) conjoint analysis methods that elicit subjects’ preferences by tasking construction of ideal products from a list of items. We propose a novel BYO with data fusion model that enables researchers to integrate BYO conjoint with follow-up tasks without overly taxing subjects, and show this model efficiently improves estimation and prediction accuracy over prior BYO methods. We then introduce a BYO with boosting (BYOB) estimation algorithm to reduce high computational costs of calculating nonlinear utilities in the BYO conjoint. We demonstrate the BYOB algorithm, with and without data fusion, outperforms competing BYO methods in a simulated test. Further, we show the BYOB algorithm is competitive with hierarchical Bayes choice-based conjoint (CBC) while also enabling researchers to elicit preferences for decisions with a large number of attributes that is difficult for CBC. Our empirical application tasks 563 marketing managers to construct dashboards from a list of metrics for different marketing budget decisions and rank selected metrics for internal and external decisions. The BYOB with data fusion model reveals novel insights not available with BYO or CBC methods alone, such as infrequently selected metrics can be highly important and frequently selected metrics are not always the most useful.

Keywords: menu-based conjoint; data fusion; Bayesian inference; metrics; dashboards; boosting

JEL Classification: C00, C11, C15, C52, C53, C81, C83, M31