The variability in behavior has frequently been assessed using a measure known as the U-value. Of concern in this article were the limits and constraints on U-value as a measure of variability. The relation between the U-value and aspects of variability was examined using three sets of simulated data. Our analysis demonstrates that the U-value as a measure of variability on its own fails to capture repetitive patterns in the sequence of responding. The U-value was shown to reflect the evenness of the distributions of responses across the categories/options used; however, when the number of categories actually used by the participant differed from the total number available, the relation between U-values and the number of categories allocated with responses was shown to be nonlinear. It was also shown that the same value of U can represent different levels of evenness in response distributions over categories, depending on the number of categories/options actually used. These constraints and limitations are discussed in relation to how researchers might report on behavioral variability.

This study examined whether trained variability would generalize across dimensions of the target response. Two experiments used a computerized rectangle drawing task that required participants to click and drag a mouse cursor to create rectangles on a computer screen. In Experiment 1, one group received points when successive rectangles varied in their size, shape and location (VAR), another group were yoked to the VAR group and received points that were allocated to them using a yoking procedure (YOKE), regardless of the variability in the size, shape or location of the rectangle drawn. Variability was higher for a dimension when variability on that dimension was directly reinforced. In Experiment 2, three groups of participants received points when rectangles varied on two dimensions; each group differed in the two dimensions that required variation. Variability was again higher for the reinforced dimensions for two of the three groups. Comparison with the YOKE group showed that the variability on those dimensions where variability was not directly reinforced was affected by reinforcement for variability on the other dimensions. Specifically, the variability in Shape and Location was significantly higher when these two dimensions occurred with other dimensions where variability was reinforced (as in Experiment 2) compared to when they were not required to vary (as in the YOKE group). This suggests that, for these two groups, the reinforced variability on the other two dimensions generalized to the third dimension. Implications of this finding to our understanding of factors that promote behavioral variability are discussed.