There Is No ‘Normal’ Distribution In Human Performance
November 13, 2012 Leave a comment
Research published by O’Boyle Jr and Aguinis (2012) challenges an alleged assumption made by professionals in the human resource management, organizational behaviour and I/O psychology fields that individual performance follows a normal (Gaussian) distribution. Instead their extensive research which included 633,263 participants demonstrated fairly consistently that human performance follows a power law (Paretian) distribution.
In a normal (Gaussian) distribution we are likely to see performance clusters around a mean which then breaks off into symmetrical tails which contrasts with a Paretian (power law) distribution that is typically comprised of unstable means, infinite variance, and a greater proportion of extreme events.
The study is controversial in the extent to which it affects how performance is interpreted in the workplace context. Implications are likely to be far-reaching, impacting upon performance measurement, management, the utility of pre-employment testing, training and development, personnel selection, leadership and the prediction of performance.
During the assessment of performance across all of the groups involved in the research it was found that a Paretian distribution was a significantly better fit for the data as compared to a Gaussian distribution.
O’Boyle Jr and Aguinis (2012) refer to how the Paretian distribution compares with a Gaussian distribution when applying a utility analysis to pre-employment testing, training and development. With a normality assumption, productivity among elite workers is estimated at $33,981 ($11,327 ×3) above the median (using Gaussian distribution), but the productivity of these workers is actually $141,588 above the median (when following a Paretian distribution). This was computed based on the figures used by Schmidt et al. (1979).
Using the example of utility analysis above the value-added by new pre-employment tests should acknowledge that the differences between employees at the tails and employees at the median are much wider than previously thought.
The research is not restricted to utility analysis examples, evidence is presented across a variety of human performance activities, convincingly demonstrating that normal distribution is an ineffective means of describing and measuring performance in work settings and elsewhere.
Read the article in Personnel Psychology for free here: