ABSTRACT
In this chapter we explore a particular tool of statistical analysis that follows logically from the difference of means (and Mann-Whitney U statistic) discussed in Chapter 10: Analysis of Variance (or ANOVA), and its nonparametric version, the Kruskal Wallis H test. Basically, the term ANOVA applies to any statistical technique that compares the variance of two or more variables when testing hypotheses. More technically, the specific techniques we associate with ANOVA are an extension of the t-test introduced in Chapter 10. Simple, bivariate ANOVA models, known as one-way ANOVA models, require that the independent variable have at least three categories (in contrast to only two, as in the t-test). Thus, the variance patterns assume a different characteristic with an ANOVA model. As such, the researcher can ascertain with more confidence whether the bivariate relationship in question is characterized as being linear, curvilinear, or exponential. ANOVA, therefore, is a combination of tabular analysis and correlation analysis.
