This session, organized by Greg Allenby of Ohio State University, brings together academics pushing the frontiers of conjoint analysis theory and practice. The academic session will run concurrent with the main session of the A&I Summit on Thursday 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 Now
We examine the effect of a U.S. Supreme Court decision regarding abortion laws on Americans’ preferences for political candidates. The decision was leaked in advance of the official announcement, and we track the evolution of political preferences from before the leak to after the leak, and eventually after the formal announcement. The abortion issue was already very important to voters before the leak, but the Court’s decision did not simply make it more important for everyone. We find that the decision decreased the importance weight of abortion for Republicans while increasing it for independents/non-voters. Further, the decision increased Republican support for candidates who want to ban abortions, although this effect is diminished for candidates that oppose exceptions for rape, incest, or the mother’s health. Non-affiliated voters move sharply away from candidates who want to ban abortions without exceptions. The decision also resulted in a lasting polarization along gender lines whereby men became more likely to vote for a candidate that supports a ban on abortion, while women are less likely to support candidates that ban abortions.
Under what political conditions will citizens punish politicians who violate democratic norms? Although Americans routinely express support for democracy and a host of related principles, including free and fair elections, checks and balances, and the rule of law, when it comes to choosing candidates partisanship often trumps citizenship (Carey et al 2022; Svolik and Graham 2019). As well, the literature has argued that Americans frequently engage in so-called “democratic hypocrisy,” whereby voters’ willingness to discipline politicians who threaten democracy is also conditioned by whether their party is in or out of office (Simonovits, McCoy, and Littvay 2022). Building on this growing line of research, we explore the impact of beliefs about whether government will be divided on the choices citizens make about a slate of candidates in a general election. To study this problem, we experimentally assign participants to treatments with different hypothetical data about the likelihood of the general election outcomes for the House, Senate, and President, and leverage a choice-based conjoint analysis (Ben-Akiva, McFadden, Train 2019) that allows candidates to choose whether and who to vote for among hypothetical House, Senate, and Presidential candidates. These hypothetical candidates vary in party and willingness to break democratic norms. The analysis proceeds via Bayesian hierarchical analysis and conditions the preferences of individuals on their partisanship and previous voting behaviors.
Consumers often make choices using incomplete information. When an information set presented in a choice setting is incomplete, consumers may make inference about the existence of missing attributes. Such inferences create what we call `phantom attributes,' which influence choice but are latent artifacts of the decision process. Phantom attributes increase the original set of attributes and, as such, are distinct from a mediation processes which consolidate attributes. In three studies, we demonstrate that phantom attributes can meaningfully impact decisions as they interact with price, brand, and other complex or perceptual attributes. We provide a practical approach to measuring phantom attributes and determining their impact on choice. This approach has practical implications for product positioning and pricing, especially in contexts with limited information like digital advertising, package design, and item listings in digital marketplaces. Phantom attributes generalize price-quality effects and can serve as a mechanism for brand-based contrast effects.
The main aim of this paper is to showcase the development of adaptive choice-based conjoint analysis (ACBC) experiment in the unconventional setting of behavioural economics. I start by highlighting the problems and pitfalls from the initial stages of designing the experiment, i.e. defining the attributes; through scale development; to end with the analysis and graphical presentation of part-worth functions, while giving practical solutions at each stage. The presentation is based on the results of the empirical research on the career choice problem. The emphasis is put on the problems pertaining to usage of more abstract and less tangible attributes and level descriptions that require careful treatment to guarantee high quality data output and subsequent results. Finally, I contrast the results of the study with previous research based on traditional conjoint experiments in order to show the importance of proper and conscious scale development, as it might affect the results and conclusions of the experiment. The whole presentation is based on research which might be useful for practitioners, especially HR departments, interested in enhancing their recruitment procedures. Nevertheless, presented ideas and solutions are considered useful and applicable for anyone designing any ACBC study.
Join us for a comprehensive introduction to Sawtooth Software's new Discover tool. With Discover you can easily build surveys, collect data, and analyze the results. Discover makes it is super simple to do conjoint analysis, MaxDiff, and everything in between. In this session we will cover MaxDiff, conjoint analysis, all question types, skip logic, quota control, passing data into/out of surveys, collaboration, multi-language, scripting, analysis, and more. You will even have time to try Discover yourself with some in class exercises. At the end of the session we will discuss our future plans and listen to your ideas for improvement. We can't wait to show you Discover!
Choice-based conjoint (CBC) studies started to increasingly rely on concepts from industrial organization such as equilibrium pricing. A driver of this development is the recently prominent role of inference from CBC in lawsuits. While large stake decisions depend on CBC based inference, the economic assumptions implied by standard models may not apply to high-ticket purchases. Despite the prevalence in economic theory, standard CBC models often ignore or insufficiently approximate consumers' budget constraints. We offer a theoretically motivated improvement to the CBC model, especially for high-ticket durable goods, and develop a Bayesian method for the inference of unobserved budget constraints. The proposed MCMC method leverages respondents' stated budget constraints that suffer from measurement error and respondents' financial demographic variables as additional information to reduce the dependency on arbitrary functional form assumptions in the estimation. We show that - by disentangling price-sensitivity within a budget from the budget constraint itself - accounting for unobserved budgets substantially increases model fit and the accuracy of competitive prices in an industry-grade discrete-choice experiment on consumer preferences for high-end laptops.
Collecting and analyzing ranking, rating, paired comparisons, or choices for optimal attribute-level combinations has a long tradition in marketing research (see e.g., Luce and Tukey (1964), Green and Rao (1969) Johnson (1974) and Louviere and Woodworth (1983)). A typical problem can be stated as follows: A focal firm wants to introduce new products (e.g., goods, services, or hybrid offers) into a market where own and/or foreign products already exists. However, the number of possible attribute-level combinations often is high and a complete enumeration by evaluating all possible product-lines is not feasible. In this paper, we apply the recent winners of comparisons with respect to accuracy and computation time – Ant Colony Optimization, Genetic Algorithms, Particle Swarm Optimization, Simulated Annealing, Tabu Search – and two new solution methods – Cluster-based Genetic Algorithm and Max-Min Ant Systems – to 538 small- to large-size product-line design problems with up to 7.3310200 possible solutions. In this talk we discuss typical product-line problems formally. Then, an overview on proposed and applied solution methods follows. The talk closes with conclusions and an outlook.
Maximum difference scaling (MaxDiff) has experienced a surge in popularity to measure preferences of items in a list on a common scale. This scaling approach is similar to conjoint analysis, where respondents are presented with multiple alternatives described by features of interest. However, unlike a conjoint analysis where price is included as a product feature and can be used to convert part-worth utility to a monetary scale, MaxDiff measures preferences on an ordinal scale. Ordinally scaled preferences give rise to the need to employ an arbitrary origin in an analysis for estimation and inference. We propose an integrative model that combines MaxDiff and conjoint data by assuming that the coefficients come from the same generative process but provide different information about product features. We show how our integrative model can be used to estimate the null level of product attributes useful for assessing the drivers of demand and menu pricing.