Statistics for Comparing Means and Proportions

Abstract

This chapter covers methods to compare means across groups of study participants. Means can be the averages of continuous variables, e.g., hours of sleep when a drug is given or proportions of dichotomous variables, e.g., the proportions of students accepted by college admission. Nonparametric methods may be appropriate if the distribution of a continuous variable is skewed. In this case the nonparametric test (e.g., wilcox.test()) may be more appropriate than parametric test (e.g., t.test()). Comparisons of proportions can be made by prop.test() or chisq.test(). If the sample size is small, fisher.test() is more appropriate than chisq.test(). More generally, loglinear modeling using loglin() can address mutliway contingency tables. We use loglin() to test a mediation effect.