Countering Perfectionism in Research on Clinical Practice. III: Implications for Randomized Designs of the Presence of Uncontrolled Interactions and Spontaneous Changes

In a crossover study, we took pre-test, post-first-treatment, and post-second-treatment measures of self-expressiveness (assertiveness) and of acute emotional distress while we successively trained different groups of students in assertiveness and group therapy exercises. Our (unfulfilled) expectations were straightforward: We thought that assertiveness training would produce specific improvement in self-expression measures with little impact on overall distress and that group therapy would produce specific improvement in general distress with little impact on assertiveness. Both treatments improved assertiveness and decreased distress. The fact that our treatments had impure effects enhanced the likelihood of Type II statistical errors and compromised the effectiveness of our design. In addition, quite surprisingly, assertiveness training selectively improved the mental health of women students, and group therapy selectively improved the assertiveness of men students more than the alternative treatment did. This interaction between gender and randomly administered experimental treatment procedures raises serious, generally important questions about the logic of statistically controlled experimentation that relies on randomization to eliminate the effects of uncontrolled, but relevant, variables.