This session, organized by Greg Allenby of Ohio State University, brings together academics pushing the frontiers of research in marketing science. The academic session will run concurrent with the main session of the A&I Summit on Wednesday so that conference attendees can switch between sessions. Our hope is that the academic sessions will stimulate new practice, and that the academic attendees will learn about new research opportunities from main session talks.
Register nowThis paper explores the uses and benefits of LLMs for researchers and practitioners who aim to understand consumer preferences. We show that GPT-3.5, a widely used LLM, responds to sets of survey questions in ways that are consistent with economic theory and well-documented patterns of consumer behavior, including downward-sloping demand curves and state dependence. We also show that estimates of willingness-to-pay for products and features generated by GPT-3.5 are of realistic magnitudes.
This paper explores the uses and benefits of LLMs for researchers and practitioners who aim to understand consumer preferences. We show that GPT-3.5, a widely used LLM, responds to sets of survey questions in ways that are consistent with economic theory and well-documented patterns of consumer behavior, including downward-sloping demand curves and state dependence. We also show that estimates of willingness-to-pay for products and features generated by GPT-3.5 are of realistic magnitudes.
We show different applications of LLMs for predicting consumer preferences, drawing from real choice-based conjoint studies. We demonstrate that fine-tuned LLMs can accurately extrapolate average utilities from databases of tested concepts, offering insights into potential market reception and using self-explanatory capabilities of LLMs to identify key drivers. Further, by leveraging the synthetic respondent idea, we can capture heterogeneity in preferences. This allows us not only to extrapolate new concepts, but also enables counterfactual analyses based on target customer characteristics. We discuss the challenges of taking this approach to complex choice situations.
We show different applications of LLMs for predicting consumer preferences, drawing from real choice-based conjoint studies. We demonstrate that fine-tuned LLMs can accurately extrapolate average utilities from databases of tested concepts, offering insights into potential market reception and using self-explanatory capabilities of LLMs to identify key drivers. Further, by leveraging the synthetic respondent idea, we can capture heterogeneity in preferences. This allows us not only to extrapolate new concepts, but also enables counterfactual analyses based on target customer characteristics. We discuss the challenges of taking this approach to complex choice situations.
The rise of generative AI provides researchers with the ability to rapidly create novel stimuli that can be used to measure a variety of constructs. This paper adopts the perspective of psychometric scale development and offers guidance on how to best implement this process to derive estimates of these quantities of interest that are both valid and reliable. Specific attention is paid to applications relevant for discrete choice experimentation.
The rise of generative AI provides researchers with the ability to rapidly create novel stimuli that can be used to measure a variety of constructs. This paper adopts the perspective of psychometric scale development and offers guidance on how to best implement this process to derive estimates of these quantities of interest that are both valid and reliable. Specific attention is paid to applications relevant for discrete choice experimentation.
Drunk driving is a huge public health issue. Ignition interlock devices can reduce DUI recidivism after a DUI conviction. But there is little evidence on how interlock devices might deter first-time DUI offenders. We use a conjoint analysis to measure the deterrence potential of an ignition interlock device as DUI penalty, and find the potential is large: on par with a dramatic increase in police enforcement activity, or a $2,000 increase in DUI fines.
Drunk driving is a huge public health issue. Ignition interlock devices can reduce DUI recidivism after a DUI conviction. But there is little evidence on how interlock devices might deter first-time DUI offenders. We use a conjoint analysis to measure the deterrence potential of an ignition interlock device as DUI penalty, and find the potential is large: on par with a dramatic increase in police enforcement activity, or a $2,000 increase in DUI fines.
Our work provides a critical input into crafting effective firearms policy: an understanding of consumer demand for guns. We estimate individual-level price sensitivity and substitution patterns across gun types using stated choice conjoint experiments. We find that potential firearm buyers are price insensitive overall, but that first-time handgun buyers are the most price sensitive. We also estimate considerable substitution from semi-automatic long guns to handguns. This finding suggests that firearm restrictions specifically targeting semi-automatic long guns would have minimal impact on gun ownership.
Our work provides a critical input into crafting effective firearms policy: an understanding of consumer demand for guns. We estimate individual-level price sensitivity and substitution patterns across gun types using stated choice conjoint experiments. We find that potential firearm buyers are price insensitive overall, but that first-time handgun buyers are the most price sensitive. We also estimate considerable substitution from semi-automatic long guns to handguns. This finding suggests that firearm restrictions specifically targeting semi-automatic long guns would have minimal impact on gun ownership.
We examine how preferences for abortion and healthcare policies and outcomes change with an election results, which included better than expected results for Democrats. This result was attributed to being because of support for abortion rights. In response, we see an increase in preferences for abortion rights and for lower maternal deaths, while we see no changes in preferences towards healthcare policies or outcomes.
We examine how preferences for abortion and healthcare policies and outcomes change with an election results, which included better than expected results for Democrats. This result was attributed to being because of support for abortion rights. In response, we see an increase in preferences for abortion rights and for lower maternal deaths, while we see no changes in preferences towards healthcare policies or outcomes.
Prior research on nonprofit fundraising has primarily examined how financial transparency and organizational accountability influence individual donor behavior. Over the past two decades, however, an increasing number of host governments have cracked down on non-governmental organizations (NGOs). We use a conjoint survey experiment to explore how organizational features and the domestic environments of NGO host countries influence preferences of private donors.
Prior research on nonprofit fundraising has primarily examined how financial transparency and organizational accountability influence individual donor behavior. Over the past two decades, however, an increasing number of host governments have cracked down on non-governmental organizations (NGOs). We use a conjoint survey experiment to explore how organizational features and the domestic environments of NGO host countries influence preferences of private donors.
Understanding and predicting consumer decision-making via discrete choice experiments has long relied on random utility models (RUMs). However, these models often struggle to account for various contextual influences on decision-making, leading to non-unified adjustments. In response, our article proposes a new model rooted in rational inattention theory. It considers how individuals process information and make choices based on limited cognitive resources, shedding light on context-dependent behaviors that challenge established RUMs.
Understanding and predicting consumer decision-making via discrete choice experiments has long relied on random utility models (RUMs). However, these models often struggle to account for various contextual influences on decision-making, leading to non-unified adjustments. In response, our article proposes a new model rooted in rational inattention theory. It considers how individuals process information and make choices based on limited cognitive resources, shedding light on context-dependent behaviors that challenge established RUMs.
Using conjoint analysis and varying the set of races in the election, we find voters’ willingness to support democractic norms is higher than previously reported (Carey et al 2022; Svolik and Graham 2019). With multiple races voters more often split their ticket to support democracy. We also find that a specific mechanism of moral licensing or balancing (Merrit, Effron, and Monin 2010) appears to be relevant to this willingness to vote for the opposing party.
Using conjoint analysis and varying the set of races in the election, we find voters’ willingness to support democractic norms is higher than previously reported (Carey et al 2022; Svolik and Graham 2019). With multiple races voters more often split their ticket to support democracy. We also find that a specific mechanism of moral licensing or balancing (Merrit, Effron, and Monin 2010) appears to be relevant to this willingness to vote for the opposing party.
The representation of consumer heterogeneity has a long history in marketing, beginning with the use of finite mixture models (Kamakura and Russell, 1989) that approximate the distribution of heterogeneity as a fixed set of mass points, to hierarchical Bayes models (Rossi, Allenby and McCulloch, 2005) that employ continuous distributions with observable covariates (Allenby and Ginter, 1995) to shift the mean of the heterogeneity distribution. A limitation of these approaches to modeling heterogeneity is that they assume a common pooling mechanism for all respondents. This is particularly problematic because of the shallow nature of marketing data, making it difficult to identify alternative mechanisms. In this presentation we develop a hierarchical Dirichlet process (HDP) model for heterogeneity in the context of a conjoint model for charitable giving. The heterogeneity distribution employs a Grade of Membership model for scaled response questions that shifts the location of the heterogeneity distribution. We show that the HDP model avoids imposing unnecessary restrictions in the heterogeneity distribution, resulting in a cleaner classification of respondents useful for customer segmentation and understanding alternative pooling mechanism. We apply our model to a conjoint study of charitable giving.
The representation of consumer heterogeneity has a long history in marketing, beginning with the use of finite mixture models (Kamakura and Russell, 1989) that approximate the distribution of heterogeneity as a fixed set of mass points, to hierarchical Bayes models (Rossi, Allenby and McCulloch, 2005) that employ continuous distributions with observable covariates (Allenby and Ginter, 1995) to shift the mean of the heterogeneity distribution. A limitation of these approaches to modeling heterogeneity is that they assume a common pooling mechanism for all respondents. This is particularly problematic because of the shallow nature of marketing data, making it difficult to identify alternative mechanisms. In this presentation we develop a hierarchical Dirichlet process (HDP) model for heterogeneity in the context of a conjoint model for charitable giving. The heterogeneity distribution employs a Grade of Membership model for scaled response questions that shifts the location of the heterogeneity distribution. We show that the HDP model avoids imposing unnecessary restrictions in the heterogeneity distribution, resulting in a cleaner classification of respondents useful for customer segmentation and understanding alternative pooling mechanism. We apply our model to a conjoint study of charitable giving.
In this presentation, we examine the performance of the hierarchical Dirichlet process (HDP) model in a high-dimension conjoint exercise for an adhesive product used to join materials together in home construction projects. We relate preferences for attribute-levels to the types of projects undertaken and alternative joining materials used (e.g., nails, screws and fasteners). We examine the implications of alternative models of heterogeneity for optimal product line design, and show that our proposed model can better identify the optimal number of products to offer and provides better consumer insights for market opportunities.
In this presentation, we examine the performance of the hierarchical Dirichlet process (HDP) model in a high-dimension conjoint exercise for an adhesive product used to join materials together in home construction projects. We relate preferences for attribute-levels to the types of projects undertaken and alternative joining materials used (e.g., nails, screws and fasteners). We examine the implications of alternative models of heterogeneity for optimal product line design, and show that our proposed model can better identify the optimal number of products to offer and provides better consumer insights for market opportunities.
Compound decisions are characterized by the presence of sub-decisions, or components, that affect overall consumer choice. In this presentation we develop conjoint analysis for compound decisions where channel members play a critical role in consumer preferences in addition to the core product. We develop our model in the context of charitable fundraising that involves the simultaneous consideration of three actors – the donor, the charitable organization raising the funds, and the beneficiary – using data from a national survey in Australia. We find evidence of the need to consider interactions among the actors and show that an analysis that focuses exclusively on the consumer (i.e., the donor) does not predict actual giving amounts as well as analysis that incorporates the channel members (i.e., charities) and the core product offering (i.e., the beneficiaries). We propose integrated, conditional and competitive analyses useful for compound decisions.
Compound decisions are characterized by the presence of sub-decisions, or components, that affect overall consumer choice. In this presentation we develop conjoint analysis for compound decisions where channel members play a critical role in consumer preferences in addition to the core product. We develop our model in the context of charitable fundraising that involves the simultaneous consideration of three actors – the donor, the charitable organization raising the funds, and the beneficiary – using data from a national survey in Australia. We find evidence of the need to consider interactions among the actors and show that an analysis that focuses exclusively on the consumer (i.e., the donor) does not predict actual giving amounts as well as analysis that incorporates the channel members (i.e., charities) and the core product offering (i.e., the beneficiaries). We propose integrated, conditional and competitive analyses useful for compound decisions.