On the Distribution of Job Performance: The Role of Measurement Characteristics in Observed Departures from Normality


  • We would like to thank Steffanie Wilk for providing the call center data, Maria Rotundo for providing the NBA data, and Nathan Bowling and Gary Burns for providing the I-O psychologist publication data. We would also like to thank Fred Oswald for his comments on an earlier draft of the manuscript.

  • Adam Beatty is now at the Human Resources Research Organization (HumRRO).


In a recent article, O'Boyle and Aguinis (2012) argued that job performance is not distributed normally but instead is nonnormal and highly skewed. However, we believe the extreme departures from normality observed by these authors may have been due to characteristics of performance measures used. To address this issue, we identify 7 measurement criteria that we argue must be present for inferences to be made about the distribution of job performance. Specifically, performance measures must: (a) reflect behavior, (b) include an aggregation of multiple behaviors, (c) include the full range of performers, (d) include the full range of performance, (e) be time bounded, (f) focus on comparable jobs, and (g) not be distorted by motivational forces. Next, we present data from a wide range of sources—including the workplace, laboratory, athletics, and computer simulations—that illustrate settings in which failing to meet one or more of these criteria led to a highly skewed distribution providing a better fit to the data than a normal distribution. However, measurement approaches that better align with the 7 criteria listed above resulted in a normal distribution providing a better fit. We conclude that large departures from normality are in many cases an artifact of measurement.