Elucidating the Results of Multiple Linear Regression: A Systematic Literature Review and Monte Carlo Comparative Simulation Study of Relative Importance Methodologies

Theory-supported multiple linear regression models are often formed to assess the impact of HR practices on organizational performance and other desired outcomes. Several methods are available to determine which predictors contribute most to said outcomes. However, researchers may make agential cuts in the method selection phase, with their histories, experiences, or perspectives on relative importance measures influencing their decisions. When multiple measures are employed, discordant results leave room for differing interpretations, potentially leading to suboptimal guidance. Correlation-Adjusted coRrelations (CAR) scores offer a unique alternative to established relative importance methodologies. However, a systematic literature review indicated that no large-scale comparison of CAR scores with established measures of relative importance across simulated conditions representing plausible scenarios encountered by researchers had yet been conducted. To address this gap, a nested Monte Carlo simulation study was conducted to analyze how the various measures converged using general dominance weights and CAR scores as the baseline metrics. Regression and relative importance analyses were also performed. Relative weights yielded rank order results that were most consistent with general dominance weights, but CAR scores also yielded results highly consistent with both relative weights and general dominance weights. The findings include reaffirming the extent of rank order convergence between general dominance weights and other measures, expanding the results to include CAR scores, and identifying the effects of criterion-related validity and multicollinearity. Implications include consideration for the orientation of predictor variables and uncorrelated predictor subsets and selecting general dominance weights or CAR scores to assess the relative importance of predictors.