Research

Mediation analysis empirically investigates the process underlying the effect of an experimental manipulation on a dependent variable of interest. In the simplest mediation setting, the experimental treatment can affect the dependent variable through the mediator (indirect effect) and/or directly (direct effect). Recent methodological advances made in the field of mediation analysis aim at developing statistically reliable estimates of the indirect effect of the treatment on the outcome. However, what appears to be an indirect effect through the mediator may reflect a data generating process without mediation, regardless of the statistical properties of the estimate. To overcome this indeterminacy where possible, we develop the insight that a statistically reliable indirect effect combined with strong evidence for conditional independence of treatment and outcome given the mediator is unequivocal evidence for mediation (as the underlying causal model generating the data) into an operational procedure. Our procedure combines Bayes factors as principled measures of the degree of support for conditional independence, i.e., the degree of support for a Null hypothesis, with latent variable modeling to account for measurement error and discretization in a fully Bayesian framework. We illustrate how our approach facilitates stronger conclusions by re-analzing a set of published mediation studies.

Abstract:

Models derived from random utility theory represent the workhorse methods to learn about consumer preferences from discrete choice data. However, a large body of literature documents various behavioral patterns that cannot be captured by basic random utility models and require different non-unified adjustments to accommodate these patterns. In this article, we discuss strategies how to apply rational inattention theory—which explains a large variety of such departures—to the analysis of discrete choice among multiple alternatives described along multiple attributes. We first review existing applications that make restrictive belief assumptions to obtain choice probabilities in closed multinomial logit form. We then propose a model that allows for general consumer beliefs and demonstrate its empirical identification. Further, we illustrate how this model naturally motivates stylized empirical results that are hard to reconcile from a random utility perspective.

Abstract:

The hypothetical nature of choices collected in typical discrete choice experiments (DCEs) for market research has been a source of concern for both researchers in academia and industry. Recent studies in marketing indeed demonstrate that choices change qualitatively when respondents are properly incentivized. However, studies that try to model these changes in the framework of random utility models show that incentives make choices less consistent reflected in a decrease of the scale of preference estimates, which is widely considered as a measure of choice quality in the literature. This result is concerning for market researchers that consider incentive-alignment and seemingly contradicts the large body of behavioral literature that suggests incentives increase decision effort and performance. In this paper, we propose a model that sheds light on the mechanism underlying the differences between hypothetical (HYP) and incentivized (ICA) multi-attribute choice. The model is in the class of sequential sampling models of choice and measures the differential cognitive processing between respondents in HYP and ICA choice settings. The empirical application of the model shows that respondents in the ICA group attend to larger sets of attributes on average and therefore resolve more trade-offs than those in the HYP group. Consequently, ICA respondents process more information and across more diverse sources of utility when making inside choices and make less consistent choices overall that nevertheless are of higher quality than HYP choices in that they are more aligned with respondents’ deep preferences. Insights from this paper can help market researchers when deciding to invest in incentive-alignment mechanisms to collect better data.

Abstract:

Durables such as cars can often be configured, and equally often, a choice between the financed and non-financed (cash) purchase is possible. Modeling demand in these settings is challenging for several reasons: i) the number of possible combinations, i.e., the size of the implied choice set, is large, ii) the typical configurator combines multinomial choices (e.g., the choice of a car model or engine) with menu choices (optional features that can be freely combined), iii) possible configurations may span enormous price ranges, and iv) configurators for high-ticket items often include financing options. The resulting choice surface is both extremely high-dimensional and highly structured. We extend the model proposed in Kosyakova et al. (2020) to meet these challenges. We construct an exchange-algorithm that handles high-dimensional combinations of multinomial and menu choices that determine the final configuration to infer individual preferences for items and (potentially highly structured) demand interactions. Further, we incorporate liquidity constraints into the model that we again infer without computing the likelihood function, which is prohibitive in all but the smallest configurators. The underlying economic model of choice allows for two liquidity constraints - the maximum cash purchase price and the maximum monthly payment for a financed purchase - possibly constraining the feasible set of choices. A cash purchase requires a higher amount of money upfront explaining choices of financing options. We illustrate our model in a case study that uses configurators in a discrete choice experiment in the car category. We demonstrate implications from ignoring budgetary restrictions and perform counterfactual demand analysis for cash and financing options.

Abstract:

Recent developments in cognitive models in psychology have motivated economics and marketing scholars to adopt these models for economic choice. However, application of cognitive models to economic choice is typically limited to binary choices because of computational difficulties, and generalizing them to larger choice sets is the prerequisite to fully exploit the advantages of these models in empirically interesting discrete choice settings. An example is the dependent Poisson race model (DPRM), a multi-attribute cognitive model with an intractable likelihood for choices

larger than two. In this paper we show how to extend the DPRM to larger choice sets by applying an MCMC based on likelihood approximation. We then propose a massively parallel algorithm that facilitates feasible implemetation of the estimation procedure. The implementation utilizes GPU resources in the Cuda C environment to parallelize likelihood approximation, uses C++ for an efficient execution of loops, and the R environemnt for a user-friendly interface. We discuss the viability of the estimation algorithm and give practical guidelines for an efficient execution of it. Finally, we apply the extended DPRM to a laptop discrete choice experiment data with 4-alternative choice sets, and show that it outperforms the HB logit in out-of-sample predicitions.