Searching for reliable relationships with statistics packages: an empirical example of the potential problems

Many social scientists appear to possess an overconfidence in the reliability of research results from a single, small-sample, inferential study. In this article, the authors speculate that "user-friendly" statistics packages have the potential to exacerbate statistical misinterpretation by providing researchers with a tool to explore data easily and identify what is interpreted as "reliable" relationships. This article contains an empirical demonstration of the potential problems that arise when a large number of statistical tests are interpreted. Results show that statistically significant results may be unreliable. Also, a zero relationship can erroneously appear as a medium to large effect size relationship when a small sample is used (e.g., n = 30). The authors suggest the need for multiple replications as the criterion of a reliable finding.