Hybrid BYO Conjoint with Boosting for Data Fusion: Estimating Context Effects for Marketing Dashboards
57 Pages Posted: 1 Jan 2020 Last revised: 2 Sep 2021
Date Written: September 2021
This research combines prior approaches to build-your-own (BYO) conjoint with a data fusion methodology that includes ranking of the selected attributes for different use contexts after constructing the profile. Our joint model uses boosting methods to produce randomly generated latent choice sets for each iteration of the estimation algorithm that integrate the posterior distribution of the parameters over the latent choice sets. The model also accounts for attribute interaction effects and overcomes the curse of large dimensions that often arise in multiple marketing contexts and complicates BYO computation. The empirical application focuses on 563 marketing managers who constructed metrics dashboards for idealized marketing budgetary decisions and ranked the selected metrics for different use contexts. The data fusion allows a more nuanced analysis than just BYO. For instance, the “Tobin’s q” metric is infrequently selected, but highly ranked among the managers who do select it, while the “total customers” metric is frequently selected, but not highly ranked. Thus, the proposed methodology extends BYO conjoint applications to more generalized settings in which marketing researchers commonly elicit individual preferences.
Keywords: menu-based conjoint; data fusion; Bayesian inference; metrics; dashboards; boosting
Suggested Citation: Suggested Citation