Argues that generalizing to message populations by treating messages as a random variable is inappropriate for complex messages, and proposes meta-analytic techniques for investigating biases in message samples and other methodological factors that can limit generalizability of communication research. (MS)