Probability Matching, a classical violation of expected utility maximization, refers to people's tendency to randomize, or even match their choice frequency to the outcome probability, when choosing over binary lotteries that differ only in their probabilities. Why? I present an experiment designed to distinguish between several broad classes of explanations: (1) models of Correlation-Invariant Stochastic Choice --- randomizing due to factors orthogonal to the correlation between lotteries, such as non-standard preferences or errors, and (2) models of Correlation-Sensitive Stochastic Choice --- deliberately randomizing due to misperceived hedging opportunities, especially when lotteries are negatively correlated. My experimental design differentiates between their testable predictions by varying the correlation between lottery outcomes. The findings indicate that the first class, despite being home to most existing theories, has limited explanatory power. Using additional treatment, I rule out Similarity Heuristics as a competing explanation with the second class. The results indicate that a vast majority of individuals deliberately randomize due to misperceived hedging opportunities.

We experimentally investigate how individuals use and value different statistical characteristics of realized signals, referred to as sample features, for belief updating. While a large literature studies how people update beliefs when receiving signals with all sample features embedded, individuals in the real world often encounter information with some certain sample feature being highlighted. We find that the updating behavior differs by sample feature and does not monotonically respond to increasing the informativeness of sample features: closest to the Bayesian benchmark when utilizing Proportion (the relative frequency of realized outcomes) for belief updating, even though it is not the most informative one. Subjects’ perceived usefulness of the sample features also diverges from the predictions of informativeness: prefer the sample features which contain Proportion over those that do not. Combining preference and belief updating performance, we show that, on average, subjects make better use of the sample features they prefer, while there exist non-negligible inconsistencies between preference and performance. Taken together, our findings indicate that the biased use of sample features in belief updating is more likely to be intentional deviations rather than inattentive heuristics.