What This Document Is
This document from East Carolina University’s Department of Psychology addresses the challenge of reporting effect sizes when using nonparametric statistical methods. It acknowledges the increasing emphasis on effect size reporting in psychological research, particularly standardized measures like Cohen’s d, and explores appropriate alternatives when parametric assumptions are violated – a common reason for choosing nonparametric analyses. The document focuses on estimators suitable for situations where the data don’t meet the requirements of traditional parametric tests.
Why This Document Matters
This resource is valuable for students and researchers in PSYC 2101 (Psychological Statistics) and beyond who utilize nonparametric statistical tests. It’s particularly relevant when dealing with data that deviates from normality or exhibits unequal variances. Properly reporting effect sizes strengthens research findings and allows for more meaningful interpretations, even when nonparametric methods are employed. The document responds to a noted gap in the field: the lack of widely known or readily available nonparametric effect size estimates.
Common Limitations or Challenges
This document doesn’t provide a comprehensive guide to *performing* nonparametric tests. It assumes a foundational understanding of these tests and focuses specifically on the issue of effect size estimation. It also acknowledges that the presented estimators aren’t universally implemented in standard statistical software, requiring users to potentially calculate them manually or utilize specialized tools. The document also clarifies that nonparametric tests address a different null hypothesis than parametric tests, requiring careful interpretation.
What This Document Provides
The document includes:
* Discussion of the issues with using parametric effect size estimates (like Cohen’s d) with nonparametric data.
* Several “d-like” and “eta-squared like” estimators suitable for two independent samples designs.
* An in-depth explanation of the rank biserial correlation (rb), also known as Cliff’s dominance measure, with computational formulas.
* References to further resources by Glass, Palij, Grissom, and Kinnear for deeper exploration.
* An illustrative example of calculating the rank biserial correlation.
This preview does *not* include step-by-step calculations, SPSS syntax, or a complete list of all possible nonparametric effect size estimators. It does not provide a tutorial on nonparametric statistics themselves.
