Chapter 5 Group Differences

This section presents a statistical tests of comparison. Which test to use depends on the structure of the data. Below is a guide to which test to use.

Continuous (interval or ratio) and Ordinal Outcomes

• The Independent samples t-test (5.1) is the main way to compare a continuous dependent variable between the two levels of a binomial independent categorical variable. Revert to the nonparametric Wilcoxon Rank Sum Test (5.2) if the t-test assumptions fail. A special case arises when samples are paired. Paired samples are more like one-sample tests where the dependent variable is the difference between the pairs. Use the Paired Samples t-test (5.3) or the nonparametric Wilcoxon signed-rank test (5.4).
• If the independent categorical variable is multinomial, conduct an ANOVA (5.5) test or the nonparametric Kruskal-Wallis test (5.6).

Discrete (count) Outcomes

• The Chi-square test of homogeneity (5.7) is the main way to compare a discrete dependent variable among the levels of a binomial or multinomial independent categorical variable. Revert to the nonparametric Fisher’s Exact Test (5.8) if the sample size is small. Handle the special case of paired samples with the Pairwise Prop Test (5.9) or the nonparametric McNemar’s test (5.10).