Classification of Statistical Analyses and Tests by Types of Variables OUTCOME CONTINUOUS CATEGORICAL Dichotomous EXPOSURE CONTINUOUS Univariate Polychotomous Ordinal Pearson correlation, linear regression Spearman rank correlation Multiple linear regression Logistic regression, discriminant analysis, Wilcoxon rank sum Logistic regression, discriminant analysis t-test, anova, linear regression Wilcoxon rank sum N-way anova, anlaysis of covariance multiple linear regression Matched pairs t-test Wilcoxon rank sum, Wilcoxon signed rank test Chi-square, logistic regression McNemar’s test (univariate) Conditional logistic regression (multivariate) Does not exist* Discriminant analysis Nominal regression Does not exist* ANOVA, linear regression Spearman rank correlation (ordinal) Kruskall-Wallis (nominal) Chi-square, logistic regression Spearman rank correlation (ordinal) Chi-square test for trend Ordinal regression Spearman rank correlation Chi-square, Discriminant analysis Nominal regression Multivariate n-way ANOVA, multiple regression Logistic regression Ordinal regression Matched repeated measures ANOVA multiple linear regression Conditional logistic regression Does not exist* Discriminant analysis Nominal regression Does not exist* Multivariate CATEGORICAL Dichotomous Univariate Multivariate Matched Polychotomous Ordinal or nominal Univariate Mantel-Haenszel, logistic regression Ordinal regression Spearman rank correlation Ordinal regression Nominal Wilcoxon rank sum Chi-sqaure test for trend Ordinal regression Ordinal regression Discriminant analysis Kruskall-Wallis Discriminant analysis Nominal regression Chi-square * does not exist: convert the polychotomous responses to dichotomous and do conditional logistic regression. NOTE: Multiple regression is used here to refer to using the least squares algorithm to do an analysis of variance, analysis of covariance, regression or any combination. Regression, theoretically, is used for the situation when one continuous variable is regressed on other continuous variables. Analysis of variance is used for a continuous outcome and one or more categorical exposures; analysis of covariance is used for comparing two or more regression lines such as arises with a continuous outcome regressed against one ore more continuous exposure (or predictor) variables and one other categorical variable. In practice, the term multiple regression is applied when referring to any one of these situations. Similarly, other multivariate regression models (logistic, ordinal and nominal regression) can be used in these “mixed mode” situation. The corresponding non-parametric tests are given in italics. Prepared by: Nancy E. Mayo, PhD (Revised April 2004) Prepared by: Nancy E. Mayo, PhD (Revised April 2004)