Truncated component regression, multicollinearity and the MMPI's use in a police officer selection setting

The clinical scales of the MMPI are highly intercorrelated (multicollinear). Theoretical considerations suggest, therefore, that multiple correlations between these scales and criterion measures will be inflated; yet this possible inflation has not been examined empirically in MMPI research. A suggested method for use with multicollinear measures is truncated component regression (TCR). The technique involves use of principal components instead of original raw scales as predictors. Experiment I compared TCR and ordinary least squares regression (OLSR) of MMPI and field performance indices obtained from police officers during training, and in their early years of service. Multiple correlations derived from OSLR were often found to be arithmetically, but not statistically, much larger than TCR-derived multiple correlations. This suggests these OLSR measures are inflated. However, several MMPI-field performance correlations derived from TCR were significant, supporting the use of the MMPI in police selection. Experiment II indicated that the truncated components structure derived from an original small sample (n = 120) of police recruits were stable when applied to large police, security guard, and pain clinic patients samples. The criterion-related dimensions of the MMPI generally were unrelated to ethnicity and sex.