Choosing the Correct Statistical Test

With the individual tests in hand, the practical skill is choosing correctly, and it comes down to a few questions asked in order. First, what is the outcome variable type, continuous, ordinal, or categorical? Second, how many groups are you comparing, one, two, or three or more? Third, are the groups independent (different people) or paired/repeated (the same people)? Fourth, for continuous outcomes, are the data roughly normal (parametric) or skewed/ordinal (non-parametric)? Answering these points you straight to the test. For two independent groups with a continuous, normal outcome, use the independent t-test; if skewed or ordinal, Mann-Whitney U. Two paired measurements call for the paired t-test or the Wilcoxon signed-rank test. Three or more groups call for ANOVA or Kruskal-Wallis, followed by post-hoc tests. Two categorical variables call for the chi-square test (Fisher's exact if expected counts are small). A relationship between two continuous variables uses Pearson (normal) or Spearman (skewed or ordinal) correlation. The decision map in the figure lays this out at a glance. Two cross-cutting reminders. Match your descriptive statistics to your test: report mean (SD) with parametric tests and median (IQR) with non-parametric ones. And remember the test only addresses whether a difference or association exists, so always pair it with an effect size and a confidence interval to make the result clinically interpretable. The single most common mistake is reaching for a familiar test, usually the t-test, regardless of the data, running it on three groups, on paired data, or on skewed and ordinal outcomes. A two-minute check of the three questions prevents most analysis errors and most reviewer complaints. The same logic extends to relationships and prediction: when you model an outcome from several variables at once you move to regression (the next module), linear regression for a continuous outcome and logistic regression for a yes/no outcome. But the starting questions never change. Keep them on a sticky note, outcome type, number of groups, independent or paired, normal or not, and run through them before every analysis; that habit, more than memorizing test names, is what makes your statistics defensible.