Our goal is to have the outcome from the investigation match the empirical outcome that is considered to really exist by the scholars. The comparison is between what the investigation produces using the significance rules and the outcome considered to be real or “true.” There are four possible outcomes of an experiment or survey, regardless of the relation assessed. Of the four outcomes, two are consistent and two involve errors. No errors have been committed if the investigation finds an effect (rejects the null hypothesis) and there is in fact a relation. Similarly, no error has been made if the investigation concludes there is no relation (fails to reject the null hypothesis) and in fact no relation exists. The other two outcomes are considered errors because the outcome of the investigation is inconsistent with what really exists (Allen 1998).
As social scientists, many scholars wish to offer conclusions that address group tendencies. If the question is “Do men or women initiate relationships?” a meta-analysis on this topic would not assert that all men or all women initiate expressions of interest. The conclusion is simply that one of the two groups is first to begin the conversation. Predictions about individuals are not made, because the level of analysis is the group. This observation often leads to the assertion that the social sciences are “soft” or unable to offer robust generalizations. Hedges (1987) explored this assertion by variability found in the natural sciences (the “hard” sciences) and the social sciences. His meta-analysis found that there is actually slightly less variability in investigation outcomes in the social sciences. Variability, then, it not something unique to the social sciences, but rather is something that occurs in all sciences investigating group-level outcomes. The difference is that the hard sciences have, for many years, been using some form of data aggregation to compare and contrast the variability in findings (e.g., smokers vs. nonsmokers, drug trial vs. placebo, bombarding one element with electrons vs. bombarding a different element).
Once the results of interpersonal communication research are tested for error, scholars can begin to treat the findings as a much closer approximation to truth. Consider the example of the interpersonal consequences of self-disclosure. Allen (1998) draws on earlier meta-analyses, summarizes key theoretical issues (sex differences, self-disclosure and liking, and reciprocity), and concludes that self-disclosure is indeed a foundational issue in relational development and management. When interpersonal communication research findings can be demonstrated as consistent across a large set of investigations, the confidence in the findings grows, as does the predictability of generating various outcomes.