Keith and Reynolds (1990; see also Ramsay, 1997) have suggested path analysis as a means of assessing predictive bias. Figure 4.1 shows one of their models. Each arrow represents a path, and each oblong or rectangle represents a variable.
The Examiner-Examinee Relationship 85
The path from group membership to intelligence test score denotes bias. Its beta value, then, should be small. The absence of this path would represent bias of zero.
A limitation of this approach is that no true ability measures exist. Thus, a path model could not incorporate true ability unless it was measured by three or more existing variables. Figure 4.2 shows a proposed model that disposes of this limitation. Here, true ability drops out, and a path leads from the predictor, Achievement Test Score, to the criterion, School Achievement. The path from group membership to the predictor denotes bias; as before, its beta value should be small. The absence of this path would, again, reflect zero bias.
Was this article helpful?