WHICH STATISTICAL TESTS ARE USEFUL FOR TESTING COGNITIVE WORKLOAD?
The most common statistical tests involving cognitive workload are the parametric F tests of means and variances based on the general linear model (i.e., all analysis of variance and regression models) and the nonparametric tests of distributions (i.e., chi square and Kolmogorov-Smirnov tests).
The first type—general linear model tests—are most appropriate when the workload measure is one of a number of factors to be tested. For example, tests involving several groups of subjects of different demographics with each subject tested on a number of different tasks. Statistical questions include whether there are workload differences among the groups regardless of task, whether there are workload differences among the tasks regardless of groups, and whether some groups on some tasks have higher workload than others (i.e., interactions).
The second type—tests of distributions—are appropriate when the objective is to look more closely at the performance of a single individual or the performance of several individuals on a single task. For example, one individual might carry out three tasks that are known to be of varying difficulty. Tests of distributions can be used to determine whether the individual’s workload differs significantly across the tasks. In this case, all observed workload estimates from a task are part of the analysis, not just a summary statistic such as the mean.
All statistical analyses described here are based on estimates of cognitive workload derived from the Index of Cognitive Activity (ICA).