Standard Discrete Choice Research Articles

Formative versus reflective attitude measures: Extending the hybrid choice model

The inclusion of attitudinal indicator variables within discrete choice models is now largely common practice. Typically, this involves the estimation of multiple indicator multiple cause (MIMIC) type models which are used to construct latent attitudinal variables that are then employed as independent variables within standard discrete choice models. Such models, collectively termed hybrid choice models (HCM) assume a particular causal relationship between the indicator variables, latent construct, and choice. In effect, the underlying assumption of such a model system is that latent variables of interest exist independent of the indicator variables used to measure them, and that the survey items used are reflective in nature insofar as responses to such questions reflect the underlying constructs. In this paper, we describe an alternative form of attitude measure, known as formative measures, where the items themselves are used to create the latent variable rather than the other way around. In addition to making a distinction between formative and reflective attitudinal measures, the paper seeks to describe how the HCM can be adapted to model different types of attitude question formats. Further the paper seeks to act as a catalyst for choice modellers to think more about the quality and validity of attitudinal items capture in survey questionnaires, by placing more emphasis on proper scale development techniques.

Estimating a model of forward-looking behavior with discrete choice experiments: The case of lifetime hunting license demand

Sales of deer licenses, one of the most important revenue sources for wildlife management at the Indiana Department of Natural Resources (IDNR), have been declining for a decade. To increase its revenue, the IDNR is considering introducing a new lifetime deer license for sale. This license would allow hunters to harvest deer (and possibly other species) each year for the rest of their lives in exchange for a relatively large up-front fee. The forward-looking nature of the decision to buy a lifetime license means hunters' choice behavior is necessarily dynamic. Prior work estimates preferences for long-lived, durable goods using standard discrete choice experiments underpinned by static models. We derive a dynamic discrete choice model of lifetime license purchases. Our model informs the design of a novel, dynamic discrete choice experiment, generating data that allows us to consistently estimate individuals’ forward-looking preferences for lifetime hunting licenses. We use our model to estimate the price of lifetime licenses that maximizes IDNR revenues.

Demand estimation with infrequent purchases and small market sizes

We propose a demand estimation method that allows for a large number of zero‐ sale observations, rich unobserved heterogeneity, and endogenous prices. We do so by modeling small market sizes through Poisson arrivals. Each of these arriving consumers solves a standard discrete choice problem. We present a Bayesian IV estimation approach that addresses sampling error in product shares and scales well to rich data environments. The data requirements are traditional market‐level data as well as a measure of market sizes or consumer arrivals. After presenting simulation studies, we demonstrate the method in an empirical application of air travel demand.

Robust discrete choice models with t-distributed kernel errors

Outliers in discrete choice response data may result from misclassification and misreporting of the response variable and from choice behaviour that is inconsistent with modelling assumptions (e.g. random utility maximisation). In the presence of outliers, standard discrete choice models produce biased estimates and suffer from compromised predictive accuracy. Robust statistical models are less sensitive to outliers than standard non-robust models. This paper analyses two robust alternatives to the multinomial probit (MNP) model. The two models are robit models whose kernel error distributions are heavy-tailed t-distributions to moderate the influence of outliers. The first model is the multinomial robit (MNR) model, in which a generic degrees of freedom parameter controls the heavy-tailedness of the kernel error distribution. The second model, the generalised multinomial robit (Gen-MNR) model, is more flexible than MNR, as it allows for distinct heavy-tailedness in each dimension of the kernel error distribution. For both models, we derive Gibbs samplers for posterior inference. In a simulation study, we illustrate the excellent finite sample properties of the proposed Bayes estimators and show that MNR and Gen-MNR produce more accurate estimates if the choice data contain outliers through the lens of the non-robust MNP model. In a case study on transport mode choice behaviour, MNR and Gen-MNR outperform MNP by substantial margins in terms of in-sample fit and out-of-sample predictive accuracy. The case study also highlights differences in elasticity estimates across models.

Consumer Search and Prices in the Automobile Market

Abstract This article develops a discrete choice model of demand with optimal sequential consumer search. Consumers first choose a product to search; then, once they learn the utility they get from the searched product, they choose whether to buy it or to keep searching. We characterize the search problem as a standard discrete choice problem and propose a parametric search cost distribution that generates closed-form expressions for the probability of purchasing a product. We propose a method to estimate the model that supplements aggregate product data with individual-specific data which allows for the separate identification of search costs and preferences. We estimate the model using data from the automobile industry and find that search costs have non-trivial implications for elasticities and markups. We study the effects of exclusive dealing regulation and find that firms benefit at the expense of consumers, who face higher search costs and higher prices than would be the case if multi-brand dealerships were used.

Market counterfactuals and the specification of multiproduct demand: A nonparametric approach

Demand estimates are essential for addressing a wide range of positive and normative questions in economics that are known to depend on the shape—and notably the curvature—of the true demand functions. The existing frontier approaches, while allowing flexible substitution patterns, typically require the researcher to commit to a parametric specification. An open question is whether these a priori restrictions are likely to significantly affect the results. To address this, I develop a nonparametric approach to estimation of demand for differentiated products, which I then apply to California supermarket data. While the approach subsumes workhorse models such as mixed logit, it allows consumer behaviors and preferences beyond standard discrete choice, including continuous choices, complementarities across goods, and consumer inattention. When considering a tax on one good, the nonparametric approach predicts a much lower pass‐through than a standard mixed logit model. However, when assessing the market power of a multiproduct firm relative to that of a single‐product firm, the models give similar results. I also illustrate how the nonparametric approach may be used to guide the choice among parametric specifications.

Risky business, healthy lives: how risk perception, risk preferences and information influence consumer's risky health choices.

Consumer's understanding of health risks is a key determinant of effective risk communication about health care options. Individuals make decisions about their health under uncertainty, yet the influence of risk perception, risk preferences and information processing has not been well researched in the health economics literature. Here, a novel extension of the standard discrete choice experiment framework is developed to investigate the effect of a "risk shock" (the difference between an individual's perceived risk and updated information) on consumption and health care decisions. In the experiment, consumers were asked to evaluate a basket of health service and lifestyle choices to lower cardiovascular risk. While most individuals, on average, aim to lower their CVD risk to a level of 1.37% points below the updated information provided, the response was asymmetrical, whereby the response to a "negative shock" was less than a "positive shock". Furthermore, we find evidence that risk perception, rather than risk preferences, are more likely to affect consumer health and behaviour changes. Given the global pandemic of CVD, these findings suggest there are public health gains to be made from personalised risk communication if it is better tailored to account for individuals' preferences and risk perception.

Estimating Complementarity With Large Choice Sets: An Application to Mergers

Standard discrete choice demand models assume that products are substitutes. Merger analyses based on these models may overstate consumer harm when producers of complementary products merge. Allowing for demand complementarity greatly complicates demand estimation, particularly when the number of choices is large. We introduce a straightforward Generalized Method of Moments estimator that identifies preferences allowing for (1) potential consumption complementarity, (2) price endogeneity and (3) large choice sets. Our estimator parsimoniously leverages information on consumer level bundle specific purchases and aggregate sales data. We apply this estimator to the chips and soda market and find a high degree of complementarity between these product groups. We show that a merger between PepsiCo/Frito-Lay and Dr. Pepper would increase soda prices by 30% less than suggested by a model that does not account for complementarity. Post-merger chip prices decrease. Overall, accounting for complementarity leads to positive welfare gains in some markets with large numbers of Frito-Lay varieties.

Identification of Intertemporal Preferences in History-Dependent Dynamic Discrete Choice Models

We study the identification of intertemporal preferences in a stationary dynamic discrete decision model. We propose a new approach which focuses on problems which are intrinsically dynamic: either there is endogenous variation in the choice set, or preferences depend directly on the history. History dependence links the choices of the decision-maker across periods in a more fundamental sense standard dynamic discrete choice models typically assume. We consider both exponential discounting as well as the quasi-hyperbolic discounting models of time preferences. We show that if the utility function or the choice set depends on the current states as well as the past choices and/or states, then time preferences are non-parametrically point-identified separately from the utility function under mild conditions on the data and we may also recover the instantaneous utility function without imposing any normalization on the utility across states.

Ranking the Criteria Used in the Appraisal of Drugs for Reimbursement: A Stated Preferences Elicitation With Health Technology Assessment Stakeholders Across Jurisdictional Contexts

ObjectivesOur goal was to estimate the relative importance assigned to health technology assessment (HTA) criteria by stakeholders involved in the HTA process. HTA is an increasingly common framework used in the appraisal of drugs for public reimbursement. It identifies clinical, economic, social, and organizational criteria to be considered. The criteria can vary across jurisdictions and are typically appraised by multidisciplinary expert committees. Guidance on the relative weighing of criteria is often absent. MethodsWe elicited stakeholders’ preferences using a single-scenario discrete choice experiment and a best-worst scaling model with conviction scores to assess the weights assigned to selected criteria by HTA stakeholders. We recruited 111 HTA stakeholders across multiple jurisdictions, including members of expert committees, clinical and economic experts, patients, and public payer representatives. Each judged twelve hypothetical cancer drug profiles for suitability for public funding and identified which characteristics were best and worst. In addition to standard discrete choice experiment and best-worst scaling models, we estimated a hybrid model to obtain a ranking of criteria by importance they played in the appraisal. ResultsA strong clinical benefit proved the most important criterion, followed by cost considerations, presence of adverse events, and availability of other treatments. The importance of clinical benefit was moderated by unmet need, adverse events, and number of patients. ConclusionPolicymakers might want to consider providing an explicit weighing scheme, or moving to a 2-stage selection process with an assessment of the quality of clinical evidence as a gatekeeping step for a full HTA review.

From Individual to Population Preferences: Comparison of Discrete Choice and Dirichlet Models for Treatment Benefit-Risk Tradeoffs

Introduction. The Dirichlet distribution has been proposed for representing preference heterogeneity, but there is limited evidence on its suitability for modeling population preferences on treatment benefits and risks. Methods. We conducted a simulation study to compare how the Dirichlet and standard discrete choice models (multinomial logit [MNL] and mixed logit [MXL]) differ in their convergence to stable estimates of population benefit-risk preferences. The source data consisted of individual-level tradeoffs from an existing 3-attribute patient preference study (N = 560). The Dirichlet population model was fit directly to the attribute weights in the source data. The MNL and MXL population models were fit to the outcomes of a simulated discrete choice experiment in the same sample of 560 patients. Convergence to the parameter values of the Dirichlet and MNL population models was assessed with sample sizes ranging from 20 to 500 (100 simulations per sample size). Model variability was also assessed with coefficient P values. Results. Population preference estimates of all models were very close to the sample mean, and the MNL and MXL models had good fit (McFadden’s adjusted R2 = 0.12 and 0.13). The Dirichlet model converged reliably to within 0.05 distance of the population preference estimates with a sample size of 100, where the MNL model required a sample size of 240 for this. The MNL model produced consistently significant coefficient estimates with sample sizes of 100 and higher. Conclusion. The Dirichlet model is likely to have smaller sample size requirements than standard discrete choice models in modeling population preferences for treatment benefit-risk tradeoffs and is a useful addition to health preference analyst’s toolbox.

Double Decoys and a Possible Parameterization: Empirical Analyses of Pairwise Normalization

We implement a “Double Decoy” experiment designed to separate two competing accounts of the asymmetric dominance effect in choice behaviour. In our experiment, we place an additional decoy alternative within the range of existing alternatives, therefore a theory which weights attributes by their range would predict a null effect on relative choice probabilities. Instead, we observe a significant decrease in the relative proportion of targets chosen (on average) in our sample. We also observe considerably more variation in individual behaviour than expected under the null hypothesis. To address these features of the data, we consider an alternative theory in which attributes values are compared two by two and normalized. Using a hierarchical Bayesian framework, we apply this pairwise normalization model both to our Double Decoy data and a standard discrete choice experiment. We find that it captures the variation in behaviour that we observe in both datasets better than range normalization and the standard linear additive Logit model, both in-sample and in an out-of-sample prediction exercise. We therefore propose this model as a useful empirical tool for researchers in applied settings.

Workplace relocation from suburb to city center: A case study of Rome, Italy

As metropolitan areas are generally characterized by decentralized jobs because of a car-based living, in this paper we focus on the feasibility of a “reverse” workplace relocation. The case study of a firm relocation from Rome’s suburbs to the historic city centre is presented. The new location, where the employees will have to move, is in a restricted traffic zone. A behavioural-based method was developed to forecast the future mode split of the employees of the firm being relocated. Next, the model’s estimates were compared with the results from a focus group interview. This approach is expected to provide reliable estimates of the mode choice, as it combines the standard discrete choice modelling for the mid-term forecast with the results from a focus group interview for the short-term forecast. The results from the discrete choice model overestimated the choice of public transport, whereas from the focus group interview it emerged that park & ride is expected to be the most chosen mode in the short-term. Finally, this case study suggests that a workplace relocation from the suburbs to the center could encourage the employees to lessen car-dependent habits.

PCN41 THE HEALTH UTILITY BOOK (HUB): A REGISTRY OF HEALTH STATE UTILITY VALUES IN ONCOLOGY

Health state utility values (HSUVs) are used to measure health-related quality of life in cost-utility analyses (CUAs). The formal adoption and endorsement of CUAs by decision makers motivates methodological advancement in utility measurement and application. Different preference elicitation methods (e.g. standard gamble, time trade-off, and discrete choice experiments) have been developed and applied. There is also a growing interest in developing generic and condition-specific utility-based instruments. As a result, the number of studies that use these methods to estimate HSUVs has been increasing substantially over recent decades. This is imposing a challenge for users to identify and select HSUVs for their CUAs. To promote the systematic identification, critical appraisal, and appropriate use of published HSUVs, we have established the Health Utility Book (HUB) project, a multi-institution initiative based out of McMaster University (Canada). One of the components of the HUB is the development of an online, searchable registry of published HSUVs. The registry will catalogue studies which use a direct, indirect, or combination approach to eliciting preferences, as identified through a systematic literature review (CRD42018095049) of medical literature databases. To accomplish this, we follow systematic review best practices, including the recent ISPOR Task Force’s recommendation on searching for HSUVs. No restrictions will be placed on date or language of publication. A team of reviewers, working independently and in duplicate, will evaluate abstracts and full-text publications for eligibility. Data extraction will capture details of study methodology, participants, health states, and corresponding HSUVs. We will present a pilot review focusing on thyroid cancer. The health utility registry will provide investigators a means of rapidly and confidently identifying relevant HSUVs for their CUAs in lieu conducting a bespoke systematic review. Through this, we anticipate saving resources and time for repeating tedious and resource intensive reviews for HSUVs.

Mergers in a Model with Complementarity

Standard discrete choice models used to evaluate mergers assume that different product varieties are substitutes. However, legal defences in some recent high-profile mergers rested on demand complementarity (e.g., GE/Honeywell). Since complements tend to be priced lower by a monopolist than by a duopoly, standard models will overstate consumer harm in these mergers. We use consumer level data from Nielsen to look at two products with natural demand complementarities and a history of regulatory activity - potato chips and carbonated soda pop. We estimate a discrete choice model that allows for demand complementarity across pop and chips, using a novel estimator that leverages information on bundle specific purchases from consumer micro data, as well as aggregate scanner data. We then simulate a number of anti-trust counterfactuals in the chips and carbonated soda market. Once demand complementarity is taken into account, a merger between the chips and soda producer PepsiCo/Frito-Lay and the soda producer Dr. Pepper will reduce chip prices. Soda prices will either increase or fall, depending on the market. By contrast, the standard discrete choice model predicts that soda prices would always increase following the merger. An additional counterfactual breaking up the PepsiCo/Frito-Lay conglomerate suggests that both chip and soda prices will increase as a result.

PRM193 - FROM INDIVIDUAL PATIENT TO POPULATION PREFERENCES: MULTINOMIAL LOGIT MODEL VS DIRICHLET DISTRIBUTION

To compare how the Dirichlet and standard discrete choice (multinomial logit, MNL) model differ in their convergence to stable estimates of population benefit-risk preferences. Individual-level trade-offs from a three-attribute patient preference study (n=560) were used in a simulation study where MNL model was applied to individual choice data, and Dirichlet distribution applied to individual trade-off data. Discrete choice experiments (DCE) were conducted for sample sizes from 20 to 540 (100 simulations per sample size), where in each one subjects drawn from the empirical distribution were simulated to answer 6 DCE questions. MNL models were fit to the answer data and Dirichlet distribution estimated for the trade-off weights. Convergence was measured with Euclidean distance of the obtained estimate and the estimate with the full sample. Logit model variability was additionally assessed with coefficient p-values. The full-sample mean estimate of the Dirichlet model ([0.52, 0.17, 0.31]) was close to the experimental mean ([0.54, 0.14, 0.32]), whereas MNL was considerably different (normalized mean [0.57, 0.06, 0.36]). Both methods converge towards the full-sample estimate, and have <0.05 distance from the full-population estimate with a sample size of at least 260 (MNL) and 160 (Dirichlet) in 95% of the simulations. With MNL, in many simulations the preference estimate of the least important attribute is still insignificant (p>0.05). For example, with a sample size of 200, in 9% of the simulations, the analyst would not be able to conclude significance of the estimate. Dirichlet distribution is likely to have smaller sample size requirements than the MNL model, and is a promising approach for modeling population preferences.

Estimating switching costs with market share data: an application to Medicare Part D

Choice inertia and switching frictions are well-documented features of the demand for health insurance. In this paper, we estimate switching costs in the Medicare Part D market with aggregate market share data using standard discrete choice models for differentiated products. We consider various modelling assumptions: myopic and forward-looking consumers, and with and without random coefficients. Both myopic and forward-looking consumer models with no random coefficients yield switching cost estimates that closely match the actual average switching frequency, with implied dollar-valued switching costs of $1600 to $2000. We find the inclusion of random coefficients to the myopic consumer model results in smaller estimates of switching costs, but only at the expense of the model’s fit to the switching frequency. The estimated welfare losses from switching frictions are large, but they are smaller under the forward-looking consumer model, amounting to around $500 per enrollee annually, compared to over $1000 under the myopic model.

A laboratory experiment on the heuristic switching model

We present results from the first laboratory experiment on the seminal heuristic switching model introduced by Brock and Hommes (1997, 1998). Subjects choose between two alternatives, a sophisticated and stabilizing, but costly, heuristic, and a destabilizing, but cheap, heuristic, and are paid according to the performance of the chosen heuristic. Aggregate choices determine the evolution of a state variable and, consequently, the performance of both heuristics. Theoretically, an increase in the costs for the stabilizing heuristic generates instability and leads to endogenous fluctuations in both the state variable and the fraction of agents using that heuristic.We vary the costs of the stabilizing heuristic in the experiment and find that the predictions of the heuristic switching model are partially confirmed. For low costs the dynamics are stable. For high costs, the dynamics initially are unstable and exhibit the type of bubbles and crashes emblematic for the heuristic switching model. However, over time the pattern of bubbles and crashes disappears and the dynamics become more stable. We estimate a standard discrete choice model on aggregate choice data and observe that subjects have a tendency to become less sensitive to payoff differences when the environment is less stable, which has important implications for the application of heuristic switching models.

Preferences for Colorectal Cancer Screening Tests: Differences Between Insured and Uninsured Beneficiaries of Iranian Health Transformation Plan

Background: Study of individual preferences about different aspects of colorectal cancer screening tests can improve the compliance of people for these tests. Objectives: This study aimed to explore the differences in the preferences of average risk populations with different insurance coverage regarding the colorectal cancer (CRC) screening tests. Methods: Individual preferences were identified by a standard questionnaire that was completed by 500 people within the age of 50 to 75 and referred to a large government teaching hospital in Tehran between September 2016 and January 2017. Standard discrete choice models and conditional logit were used for analysis. Results: The final attributes were the test process, pain, place, sensitivity, interval, preparation, complications risk, reduced mortality rate, and cost. The result of conditional logit regression demonstrated that having complementary insurance coverage had a significant effect on individuals’ preferences. Conclusions: This study suggests that different insurance coverage may lead to heterogeneity of preferences for CRC screening tests.

Nonparametric Demand Estimation in Differentiated Products Markets

I develop and apply a nonparametric approach to estimate demand in differentiated products markets. Estimating demand flexibly is key to addressing many questions in economics that hinge on the shape - and notably the curvature - of market demand functions. My approach applies to standard discrete choice settings, but accommodates a broader range of consumer behaviors and preferences, including complementarities across goods, consumer inattention, and consumer loss aversion. Further, no distributional assumptions are made on the unobservables and only limited functional form restrictions are imposed. Using California grocery store data, I apply my approach to perform two counterfactual exercises: quantifying the pass-through of a tax, and assessing how much the multi-product nature of sellers contributes to markups. In both cases, I find that estimating demand flexibly has a significant impact on the results relative to a standard random coefficients discrete choice model, and I highlight how the outcomes relate to the estimated shape of the demand functions.