Evaluating alternatives to the general linear model in applied personality assessment

The study represents the first large scale investigation into whether or not alternatives to the General Linear Model (GLM) exist in personality-job performance relationships. Even though many experts have proposed models of non-linear personality-job performance relationships, only a handful of small sample studies have explored the topic further. Because the presence of non-linearity could have substantial practical and theoretical implications, a thorough examination into the nature of personality-job performance relationships was conducted.; Five steps were undertaken to investigate potential non-linearity in personality-job performance relationships. First, SMEs were asked to make judgments about the shape of personality-job performance relationships that they believed were most prevalent in jobs with similar J. Holland (1986) designations. Second, correlation and quadratic and cubic power polynomial regression were used to derive linear and non-linear validity coefficients for the Hogan Personality Inventory (HPI) scales in relations with task and contextual performance in 49 studies. Third, meta-analytic techniques were used to aggregate linear validities and partial correlations derived through power polynomial regression. These meta-analyses were conducted across all studies and by studies categorized by J. Holland job type. Fourth, within-study significant non-linear personality-performance relationships were graphed using the beta coefficients from quadratic and cubic regression results. Fifth, multivariate adaptive regression spline analyses (MARS) were conducted to detect the possibility of additional types of non-linear relationships beyond quadratic and cubic forms.; The study yielded a number of interesting findings regarding non-linearity in personality-based prediction. While SMEs believed that the inverted-U relationship was the non-linear relationship most likely to describe personality-performance relationships, the evidence overwhelmingly supported derivations of U shaped curves throughout the analyses. Even though no single definable non-linear model emerged, the results supported the existence of unique non-linear relationships for specific personality-performance relationships and a frequently occurring “threshold effect.” The threshold effect represents a range of scores where there is no relationship between an HPI-scale and job performance followed by a shift to a positive linear relationship. The threshold effect most commonly occurred in the low to moderate-low predictor-criterion score range. Practical and theoretical implications related to the threshold effect and other unique non-linear relationships are discussed.