Marshall Drake, Fernando Payró, and Neil Thakral)
Job market alert: Marshall Drake is a truly exceptional Ph.D. Candidate from Boston University, who is on the job market this year.
This package helps researchers implement and run their own Bayesian Adaptive Choice Experiments (BACE). BACE allows researchers to elicit preferences quickly and efficiently using a dynamic experimental framework. Researchers specify a model that they want to estimate, prior beliefs over the model’s parameters, and questions (or “designs”) that can be shown to respondents.
At each stage, BACE selects the maximally informative question according to the mutual information criterion. This question is shown to a survey respondent. Based on the respondent’s answer, the posterior likelihood of the individual’s preference parameters is updated using Bayes’ rule and Monte Carlo techniques. After incorporating this new information, the next question is chosen using Bayesian Optimization, and the process repeats.
Computing the most informative question in real-time is computationally intensive. BACE helps you set up a back-end server to handle computation remotely, allowing BACE to handle many questions and respondents at once and scale to situations where respondents may have poor computing resources. You can then set up your favorite front-end survey interface (e.g., Qualtrics or SurveyMonkey) to query your BACE application, display designs to survey respondents, and record individuals’ responses.
A unified package for two-stage differences in differences and other differences-in-differences packages.