A Bayesian Model Averaging Approach to Predicting Social Psychological Phenomena

Authors: 
David S. Yeager and William Damon

Division 8: Society for Personality and Social Psychology The growing field of Positive Psychology has yielded many new constructs to operationalize, such as Damon's (2008) notion of a Purpose in Life, but with this field-building has emerged a need for reliable exploratory statistics to build new theories about constructs. Common frequentist methods include variables in a model if they meet a certain threshold of "significance," but these methods discard the uncertainty that the researcher has about variable inclusion. Several ad hoc criteria for selecting models have emerged, but choosing an information criterion comes with its own form of bias. This poster compares a frequentist method (step-wise regression) for predicting purpose to another method (Bayesian Model Averaging) and concludes that step-wise regressions would seriously mislead researchers if used uncritically.