Nonparametric Measures of Association

When analyzing data, how should the relationship between two or more sets of observations be described, that is, values of two or more variables, when the variables are ordinal and not bivariate normal? Aimed at helping the researcher select the most appropriate measure of association for two or more variables, the author clearly describes such techniques as Spearman′s rho, Kendall′s tau, Goodman and Kruskals′ gamma and Somer′s d and carefully explains the calculation procedures as well as the substantive meaning of each measure. In addition, each technique is illustrated by one or more examples from recent social or behavioural science studies. Finally, Gibbons provides information on the strengths and weaknesses of leading statisti

This compact and highly readable volume presents Spearman′s and Kendall′s rank correlation and coefficients, Kendall′s coefficients of concordance and of partial correlation, and several association measures for ordered contingency tables. . . . This inexpensive and lucid text offers a good introduction, or a quick review, of methods of rank correlation. It should prove beneficial to the practitioner who selects from and interprets the many measures produced by modern statistical packages. --Journal of the American Statistical Association When analyzing your data, how should you describe the relationship (or, association) between two or more sets of observations, i.e., values of two or more variables, when the variables are ordinal and not bivariate normal? Aimed at helping the researcher select the most appropriate measure of association for two or more variables, Jean Dickinson Gibbons clearly describes such techniques as Spearman′s rho, Kendall′s tau, Goodman & Kruskals′ gamma, and Somer′s d. She also carefully explains the calculation procedures as well as the substantive meaning of each measure (such as that rho is based on rankings while tau is based on paired comparisons). In addition, each technique is illustrated by one or more examples from recent social or behavioral science studies. Lastly, Gibbons provides information on the strengths and weaknesses of leading statistical packages for calculating these measures.