By Alan Reifman (Texas Tech University)
Information
Statistical power analysis is essential for efficiently conducting quantitative research. Small sample sizes afford little chance of detecting even sizable effects, so expending resources for such studies is questionable. According to Aberson et al. (2002), however, “statistical power remains among the most difficult topics to teach," drawing upon "null and alternative distributions, Type I and Type II error rates, sample size, and effect size.” Despite the importance of statistical power, therefore, its complexity likely inhibits teaching about it (Walmsley & Brown, 2017). As most of our students (social science Master's) are new to statistics, we define power straightforwardly (probability of “detecting something that’s there”), discuss familiar decision errors (e.g., pulling a fire-alarm when there is no fire vs. not pulling it amidst flames), and use graphics extensively. We will demonstrate a new activity (with grading rubric), in which students receive a dataset, state correlational hypotheses, specify an anticipated effect-size (correlation) based on past findings with their variables, indicate what to enter into a power calculator and why, calculate power, compute correlations with well-powered and smaller sample sizes, and interpret their findings. Changes in students' understanding is assessed via grades on the activity and pre-post comparisons on an online self-assessment quiz.
https://reifmanintrostats.blogspot.com/2010/11/this-week-well-be-covering-statistical.html