A methodological exploration. The use of inverted cluster analysis on MMPI profiles

The Minnesota Multiphasic Personality Inventory has been widely used to evaluate the problems of neuropsychiatric patients. The authors and othrs have stressed the importance of a pattern or profile interpretation of the test. Since there has been an emphasis on profile interpretation of the MMPI, it was felt that it would be useful to see if homogenous profiles could be obtained by means of an inverted cluster analysis. The subject of this study were fifty white male neuropsychiatric patients who wree first admissions to the Veterans Administration Hosptial, Houston, Texas. Each subject was given the individual form of the MMPI. The directions suggested by the authors for administering and scoring the MMPI were used. Each subject was given all the time he needed to complete the MMPI. It was necessary to shorten the work required to compute N(N-1)/2 correlations on an N of 50. An attempt was made to use the central tendency, variability and skewness as preliminary descriptive statistics of likeness. It was felt that score distributions, which were similar on these preliminary statistics, would warrant the time spent in computing correlations on them. These preliminary statistics were found to be too insensitve to give groupings of any value. From clinical experience with the test, it was decided to compute the rank-difference correlations on those score-distributions which were within plus or minus one class interval deviation on both hte Mf and Sc scales. These criteria of likeness considerably reduced the number of correlations which had to be computed. After computing the rank-difference r on those score distributions which were within plus or minus one class interval deviation on the Mf and Sc scale, a correlation matrix was devised. Of course, the correlation matrix was an incomplete one between correlations were not computed on all the score-distributions. A visual inspection was made of the profiles which correlation .60 and above. It was discovered that profiles of score-distributions which correlated .75 and above were consistently enough alike to warrant the time spent in a more rigorous visual inspection for profile similarity. By a rigorous visual inspection of the profiles with correlations of .75 and above, it was possible to get six clusters which were highly similar. Five of the clusters form rigidly homogeneous groups, while a sight is a less well-defined cluster. It is possible to get clusters of homogeneous profiles by an inverted cluster analysis technique.