Estimating and Correcting for Response Bias in Self-Reported HIV Risk Behavior

Journal of Sex Research, Feb, 1999 by David R. Gibson, Esther S. Hudes, David Donovan

One of the most difficult problems in behavioral research on human sexuality and drug use is its reliance on people's self-reports about their behavior. Given the essentially private nature of sexual and drug-taking practices, many aspects of these practices are difficult, if not impossible, to validate. Few studies exist on the validity of self-reported sexual behaviors (Catania, Gibson, Chitwood, & Coates, 1990). As an isolated example, Udry and Morris (1967) were able to validate the incidence of coitus in women using urine tests for sperm, but the tests were unable to establish the frequency of vaginal intercourse. Even more difficult to validate are behaviors that place people at risk of sexually transmitted diseases, such as AIDS. Biological markers can be used to corroborate self-reports of high-risk sexual and drug-taking practices in samples of respondents, but cannot validate behavior at the individual level. For example, self-reported increases and decreases in the prevalence of unprotected anal intercourse paralleled both increases and decreases in the number of cases of gonorrhea in gay men in San Francisco (Coates, Stall, Catania, Dolcini, & Hoff, 1989; Winkelstein et al., 1987). Similarly, sharp decreases in high-risk injection practices in a cohort of injecting drug users over four years coincided with a sharp decrease in HIV seroincidence (Wiebel et al., 1996).

In the absence of measures of validity, researchers have used a variety of other techniques for assessing the quality of self-reported information about drug-related and sexual behaviors. Several studies have compared agreement levels among heterosexual (Upchurch, 1991), homosexual (Coates et al., 1989; Padian, 1990; Seage, Mayer, Horsburgh, Cai, & Lamb, 1992), and drug-injecting (Gibson & Young, 1994) partners in the reporting of selected high-risk behaviors. Self-reported sexual activity has also been compared with data gathered prospectively using a daily coital log (Hornsby & Wilcox 1989). Methods such as these, however, constitute at best a weak form of validity, since partners' responses and repeated measurements are also subject to bias.

The lack of a gold standard for validating self-reports is particularly troubling with respect to HIV-related behavior. To the extent that self-reports are affected by response bias, the prevalence of high-risk practices will be underestimated, hindering efforts to slow the spread of HIV. There is extensive evidence (Sudman & Bradburn, 1974) that self-presentation bias can lead to substantial underreporting of behaviors when respondents find questions to be embarrassing or threatening. For example, Bradburn, Sudman, Blair, and Stocking (1978) found that subjects who rated themselves as very uneasy about sexual questions were more likely to refuse to answer questions about vaginal intercourse than subjects who felt very comfortable with sexual questions. Similarly, Catania, McDermott, and Pollack (1986) found that the frequency of refusing to answer sexual questions correlated significantly with self-reported willingness to disclose sexual information.

Self-presentation bias is a particularly common problem in behavioral research which can be attributed to the natural tendency of people to present themselves in a socially desirable light. At least since Edwards' influential 1957 monograph, researchers have been aware of this tendency, which manifests itself in a research subject claiming socially desirable but unlikely traits ("I never hesitate to go out of my way to help someone in trouble") and denying undesirable but highly probable qualities ("I sometimes try to get even, rather than forgive and forget.") Crowne and Marlowe (1964) interpreted these two tendencies as evidence of a broader need for social approval, and developed a 33-item scale designed to measure this personality characteristic in a way that would have minimal pathological implications (DeMaio, 1984). In the last 30 years, the scale has been used in thousands of studies in social and personality psychology and other areas of research.

Past research on social desirability indicates that this tendency is especially likely to manifest itself not only when respondents find subject matter to be threatening or embarrassing (DeMaio, 1984), but also when subjects are defensive, lacking in self-esteem, or sensitive to status differences. Among clinical populations, including people with drug abuse problems, subjects with a need for social approval often exhibit an intense and perhaps neurotic fear of disapproval which leads to continual "deception of the other" (Millham & Jacobson, 1978, p. 382).

In a recent study (Latkin, Vlahov & Anthony, 1993) conducted with 2,885 injecting drug users interviewed in the Baltimore metropolitan area, investigators examined the extent to which socially desirable responding might confound self-reported injection behavior. Two measures of social desirability, self-deception and impression management, were employed, both derived from the Marlowe-Crowne Social Desirability Scale. The results showed that scores on self-deception and impression management were inversely correlated with self-reported sharing of injection equipment and injecting at "shooting galleries" (venues where drug users rent syringes). They were unrelated, however, to other variables, such as self-reported cocaine use and HIV status, and had a negligible impact on the magnitude of relationships between risk behaviors and HIV serostatus. The range of HIV risk variables studied was limited, and included only one measure of sexual risk behavior, receptive anal intercourse, which was reported by less the two percent of this predominantly heterosexual sample. There were only minor differences in the extent to which the self-deception and impression management predicted risk behavior, suggesting that a single measure of social desirability would have been adequate to assess the extent of response bias.