Recent research on the tests of cognitive validity have found that such tests are

Test anxiety refers to the “phenomenological, physiological and behavioral responses” (Zeidner, 2007, p. 166) that accompany testing. It is a subjective emotional state experienced before or during a specific evaluation relating to the act of completing the evaluation itself, the threat of failing, and perceived negative consequences. Thus, test anxiety is conceptualized as the manifestation of a situation-specific trait (Zeidner, 1998). Modern views of test anxiety conceptualize it as having two major components: Worry and Emotionality (Cassady and Johnson, 2001, Spielberger and Vagg, 1995; see also Liebert & Morris, 1967). Worry is the cognitive component of test anxiety reflecting the debilitating thoughts and concerns the test-taker has before or during the test. The Emotionality component (sometimes called Tension) refers to the heightened physiological symptoms stemming from arousal of the autonomic nervous system and associated affective responses.

As mentioned above, the negative correlation between test anxiety and various evaluative outcomes has been found across several domains. For example, Hembree’s (1988) meta-analysis showed negative correlations between test anxiety and performance on IQ, aptitude, memory and problem-solving tests. He also found negative correlations for several school-related outcomes such as overall grades, and performance in language and mathematics tests, among other outcomes. Similar results are comprehensively reviewed by Zeidner, 1998, Zeidner, 2007. Given these findings, it is not surprising that test anxiety has been posited as a potential biasing factor in test performance. Below we review the concept of differential predictive validity, and we integrate it with two perspectives of test anxiety and test performance—the deficit and the interference perspectives.

Educational institutions routinely use scores on cognitively-loaded exams like the SAT, or the Graduate Record Examination to select students given their predictive validity vis-à-vis educational outcomes (Kuncel et al., 2001, Sackett et al., 2009. Similarly, employers often use CATs as part of a selection system given their robust predictive validity vis-à-vis job performance (Schmidt & Hunter, 1998). Thus, a key concern is whether test scores demonstrate differential prediction as a function of some personal or social factor. For example, differential prediction by race and gender has been investigated with respect to CATs (e.g., Dunbar and Novick, 1988, Schmidt et al., 1981). However, differential prediction due to psychological factors such as test anxiety, though often discussed, has not been as frequently examined empirically.

Often discussed under the broader label of predictive bias, differential prediction occurs when a third variable (such as test anxiety) influences the predictor–criterion relationship (such as the relationship between CAT scores and job performance). Consequently, predictive bias is commonly assessed within a moderated multiple regression framework, where bias is said to exist if any coefficients within the regression equation relating the predictor and criterion differ across subgroups. The issue of differential validity, which is our focus, is tested by examining the interaction of the predictor (e.g., test scores) and a potential biasing variable (e.g., anxiety) (Sackett, Laczo, & Lippe, 2003). A significant interaction term indicates that the predictive relationship with the criterion (i.e., the slope) differs across sub-groups defined by the biasing variable (e.g., high and low anxiety test-takers). If only intercept differences are present (indicated by a significant regression coefficient associated with the biasing variable but no significant interaction), this indicates that the predictor test has equal validity across groups but that the use of a single regression line would over-predict performance of the lower scoring group. Although this can lead to selection bias if a common regression line is used, we do not focus on this issue given that few real-world settings would consider using test anxiety as a protected class of information by which to define groups.

The deficit perspective specifies that anxiety results from the test-taker being aware of a skills deficit which will be (accurately) reflected by poor test performance (Covington and Omelich, 1987, Tobias, 1985). According to this perspective, “anxiety in the test situation has no causal status, but is simply an epiphenomenon reflecting students’ lack of preparation for the test and their metacognitive awareness of their low probability of succeeding on the exam” (Zeidner, 1998, p. 70). Test anxiety occurs as a byproduct of a true ability deficit, which the test accurately measures. Thus, the deficit perspective would not predict any form of differential validity due to anxiety.

Conversely, the interference perspective implies that test anxiety should result in differential validity. This perspective posits that anxiety interferes with a person’s test performance by competing for cognitive resources. Specifically, cognitive resources are spent on off-task processing such as worrying, managing immediate physiological reactions, or focusing on negative self-evaluations (Eysenck, 1973, Spielberger and Vagg, 1995). These off-task cognitions prevent the test-taker from focusing solely on the actual test, and require them to spend valuable resources on managing divergent thoughts. Thus, interference models state that ability test scores should be less predictive of criterion performance (i.e., have lower criterion-related validity) for high test anxiety test-takers compared to low anxiety ones.

However, a framework proposed by Reeve et al. (2009) suggests that test anxiety will only decrease the criterion-related validity of predictor tests to the degree that anxiety is not found in the criterion as well. Given the evaluative nature of performance appraisals, it is likely that they evoke the same situation-specific trait (i.e., test anxiety) as do other testing situations. Thus, people may experience anxiety during the evaluation of criterion performance in much the same way they do during the predictor assessment. Hence, test anxiety may not result in differential validity if anxiety experienced during the criterion is included in the model. Rather, test anxiety would actually aid in the prediction of criterion performance because the criterion would now share an additional source of systematic variance with the predictor.

Given the nature of these two perspectives, two competing hypotheses can be generated. According to the interference perspective, we would expect to find evidence of differential validity due to individual differences in test anxiety (Competing Hypothesis 1a). However, the Reeve et al. (2009) framework suggests that test anxiety may simply be an additional predictor to the extent it is experienced at the time of criterion assessment and would not be expected to result in differential validity (Competing Hypothesis 1b). The latter conclusion of no differential validity would also be consistent with the deficit perspective of test anxiety.