Implicit discriminatory attitudes in surveys: validation of the semantic affect misattribution procedure for dealing with social desirability in surveys

Abstract We examine the use of the Affect Misattribution Procedure (AMP) for assessing implicit or latent attitudes in the context of general population surveys. Specifically, we test and validate a Semantic AMP (SAMP) which uses words as primes rather than images. We suggest that the SAMP could be incorporated into the repertoire of experimental methods to be used in broad omnibus surveys to deal with social desirability bias. We discuss findings of an empirical investigation with regard to the procedure’s validity and reliability, as well as in comparison with direct questions about potentially sensitive attitudes. The procedure shows promise: confirmatory factor analyses using structural equation models suggest satisfactory construct validity and acceptable measurement invariance for subsamples in terms of age, gender and education. Further, the implicit measures generally correlate with explicit measures of discriminatory attitudes. Correlations between implicit and explicit measures are, as expected, weak to moderate in magnitude, suggesting some amount of discrepancy, potentially due to social desirability. The procedure is also relatively easy to implement in online surveys. But we also find some signs of satisficing behaviour amongst respondents, suggesting the procedure may be overwhelming to some.