Prediction Analysis in Political Research

Procedures for empirical tests of political theory should be designed to evaluate the success of the specific prediction being tested. This paper introduces; (1) a formal language, termed “prediction logic,†for stating an important class of predictions, including predictions that imply there will be relatively few cases in certain cells of a cross-classification; (2) a population measure, (“del†), that indicates the degree of success achieved by a statement in the language; (3) partial measures for the multivariate case; (4) bivariate statistical inference methods when the data arise from a sample rather than a population, both for an a priori prediction statement and for a statement selected post hoc. A number of well-known measures of “association†are measures for specific prediction logic statements. Research applications are indicated through the use of contingency tables appearing in APSR articles by Eulau and Eyestone, Goldberg, Muller, Riker and Zavoina, Rosenthal, Sawyer and MacRae, Sickles, Wolfinger and Field, and Wolfinger and Heifetz.