Abstract: | Two ways of conducting the search for generalizations about messages are considered: Morley's (this volume) proposal that single-message research designs be used, with subsequent meta-analytic summaries, and Jackson and Jacobs's (1983) proposal that multiple-message designs be used, with messages treated as a random factor in the statistical analysis. Jackson and Jacobs's approach is shown to provide a more dependable, efficient, and practical means for gathering the requisite evidence for dependable generalizations. The charge that multiple-message designs suffer from irreparable problems of experimenter bias is refuted. The treatment of messages as a random factor is defended as statistically appropriate and as clearly preferable to the statistical alternatives. |