The paradox of reality constraints: distributional violations of normality change outcomes in human performance modeling

Abstract

A reality constraint in a simulation is a rule designed to remove or adjust some physically impossible but mathematically reasonable event. An example would be some modification to a human reaction time that is three standard deviations below the mean, which might be extremely fast or impossibly fast depending on the sampling distribution. The present research utilized small arms combat modeling to illustrate the potential pitfalls when reality constraints interact with the variability endemic to human performance. Particularly when non-normal distributions are involved, reality constraints may inadvertently affect simulation results due to the interaction between their boundary conditions and the shape of the sampling distribution representing human performance. Findings indicated that human performance modeling was largely robust to some violations of normality, such as skew, but reality constraints could shift the outcome by nearly 10% when dealing with other violations of normality, particularly kurtosis, because these violations can shift the mean of the distribution. These results demonstrate the importance of accounting for the distribution of human performance observations in simulation efforts and the need to consider reality constraints as a meaningful element of human performance modeling. Recommendations for the construction of reality constraints are discussed.