Online Program

Friday, February 20
CS11 Bootstrapping Applications Fri, Feb 20, 2:00 PM - 3:30 PM
Napoleon C

Bootstrapping Confidence Intervals for Effect Sizes (and Other Weird Things) (302895)

Jason W. Osborne, University of Louisville 
*Erin Smith, University of Louisville 

Keywords: replicability, bootstrap, confidence intervals, effect sizes

In 1999, the American Psychological Association released a task force report that recommended researchers report and focus on effect sizes, confidence intervals, and replication in empirical articles as a complement to ubiquitous reliance on null hypothesis statistical testing (NHST). Other scholars, (e.g., Bruce Thompson, Peter Killeen, Geoff Cumming) have since pushed researchers further, even suggesting that confidence intervals around effect sizes (CIE4S) are desirable. The fu- ture of statistical practice is challenged, however, by tradition (NHST) and how complex and difficult it is to calculate CI4ES in many cases. However, with ubiquitous computing power and simple scripts, we can bootstrap almost any statistic, providing empirical CI4ES (e.g., R-squared, odds ratio) and CIs for other odd and useful statistics, like eigenvalues, Cronbach’s alpha, medians, etc. I will review the importance of evaluating replicability of analytic results, the history of NHST vs. ES debate, and using an SPSS macro. I will present examples of how bootstrapping of effect sizes and other statistics can provide valuable insight into your results.