The covert attitudes test: utilizing biased assimilation of evidence to measure attitudes

Abstract

Prejudice research faces a fundamental measurement dilemma: explicit measures capture deliberative attitudes but suffer from social desirability bias (SDB), whereas implicit measures avoid dishonest responding by tapping automatic processes but may reflect societal stereotypes rather than personal attitudes. This creates a critical gap: no existing measure captures deliberative attitudes while circumventing SDB. I developed the Covert Attitudes Test (CAT) to address this gap. The CAT measures attitudes through biased assimilation of evidence, using diverted attention to reduce SDB. Specifically, rather than asking participants to report attitudes directly, the CAT asks them to evaluate research on social groups, with the expectation that people will rate studies more favorably when the findings align with their existing attitudes--a design that eliminates social desirability concerns while preserving deliberative processing. The CAT's validity was established across five studies, with over 1,500 participants from three populations and three attitude domains. The measure demonstrated convergent validity by showing subjects displayed ingroup bias on it, as well as expected correlations with explicit attitudes and related constructs without being redundant (Studies 1-5). Crucially, Studies 3-5 demonstrated discriminant validity by finding no relationship between the CAT and implicit prejudice measures--confirming it captures deliberative rather than automatic processes. Study 4 found acceptable to good test-retest reliability, indicating stable measurement. Predictive validity (of behavior) showed mixed results: Study 4 demonstrated incremental validity beyond existing measures, but Study 5 failed to conceptually replicate these findings. Studies 4-5 confirmed the CAT's effective use of diverted attention, showing the vast majority of participants remained unaware it measured attitudes, and that convergent and predictive validity were not moderated by participants' awareness of the CAT measuring prejudicial attitudes. These findings establish the CAT as psychometrically sound and easily adaptable, filling an important gap by reducing SDB while capturing deliberative processes. Like all new measurement approaches, challenges remain in establishing optimal applications, particularly regarding when the CAT provides predictive validity advantages over existing measures. Nevertheless, the CAT offers a promising approach to measuring attitudes in socially sensitive domains where traditional measures struggle. Future work should continue refining the design and extending its applications to new contexts.