Statistical Tests for Categorical Features vs Binary Target

Quick Answer

This question is really asking you to match statistical tests to the type of variables involved, and all three variables here, the categorical features and the binary target, are categorical in nature, which rules out any test built for continuous data. The chi-square test of independence is the foundational tool for this situation because it directly tests whether two categorical variables are statistically associated, comparing observed frequencies in a contingency table against what would be expected if the variables were independent. Cramér's V builds on that same chi-square statistic but rescales it into a bounded measure of association strength, which chi-square alone doesn't provide, so it answers not just whether an association exists but how strong it is. Mutual information adds a third, complementary lens: it's a non-parametric measure that can capture more complex or non-linear dependency patterns between categorical variables that chi-square-based methods might not fully characterize. ANOVA and Pearson correlation are both designed for continuous data, ANOVA compares means of a continuous variable across groups, and Pearson measures linear correlation between two continuous variables, so neither is appropriate when every variable involved is categorical. Whenever a question specifies categorical features and a categorical (including binary) target and asks about testing association, expect the correct answers to cluster around chi-square-family and information-theoretic measures rather than tests meant for continuous variables.