The Dynamics of “ Structured ” Personality Tests

PREFATORY COMMENT. This paper—my first publication—appeared a quarter-century ago, and I like to think that I have learned a little something since then. I consented to its reprinting because I still believe it has a message, and of course it was an influential contribution to its time. (Professor Wiggins, before proceeding to make his impressive case contra, flatteringly calls it “Meehl's now classic empirical manifesto,” Wiggins, 1969, p. 127). In 1945, a clinician or counselor who used structured verbal tests based on empirical keying against psychiatric or occupational criteria (MMPI, SVIB) was on the defensive against two dissimilar groups of opponents. The first group were academic psychologists in the “Bernreuter-Inventory” tradition, who tended to combine (a) unjustified trust in face-validity, (b) relative lack of psychiatric experience and clinical orientation, and (c) psychometric reliance upon fairly crude internal-consistency approaches in test-building. The other group were clinicians identified with the Rorschach and other projective techniques, who were (as subsequent critical research has shown) overly intoxicated with the “projective” idea, insufficiently aware of the unavoidable normative problem (having bought the fallacious X ray analogy in L. K. Frank's famous 1939 paper) and ideologically tendentious with respect to all “structured” instruments. In that situation, my 1945 paper was probably needed, and I do not have to apologize for its exaggerations. Being a neo-Popperian in my philosophy, I view the growth of science as a series of errors and corrections, of “conjectures and refutations.” But I cannot consent to the paper's reprinting without at least indicating— no convincing arguments being intended—the respects in which I now disagree with it. Mainly, I see it as overly “dust-bowl empiri cist,” insufficiently theoretical, and psychometrically simplistic. For example, the validity shrinkage we see in moving to new populations (“validity generalization” problem, not soluble by mere calculation of a random sampling error statistic) is, I am sure, partly a result of the presence of items that appeared in the criterion analysis mainly because they were correlates of unrecognized nuisance-variables. I now believe that “blind empirical keying,” where we retain an