Using Model/Data Simulations to Detect Streakiness

Abstract

AbstractA simulation-based approach is proposed for approximating a Bayesian analysis. Parameters and data are simulated from a Bayesian model and inference about a parameter is performed by exploring the set of simulated parameter values conditional on a set of values of a simulated statistic. The approach is used to learn about parameters of a streaky model on the basis of a statistic used to measure streakiness. The method is illustrated to detect streakiness in baseball hitting data and basketball shooting data.KEY WORDS: Bayesian inferenceCoin-tossingHothandMarkov switching modelOverdispersionQuasibinomial