Answers are grounded in this paper and cite their sources.
To the Editor: In a study of condom use and incident sexually transmitted diseases (STDs) among urban STD clinic attenders, Zenilman et al1 found that self-reported condom use was not associated with lower STD incidence. The investigators concluded that self-reports of condom use may be subject to substantial reporting bias and thus evaluation studies of human immunodeficiency virus risk-reduction interventions that rely on self-reported data may be seriously deficient. The methods used in the Zenilman et al study did not include adequate controls to protect and strengthen the validity of self-reported data, however, and thus their conclusions deserve further examination. We believe that their flawed methods of self-reported data collection may be largely responsible for the discrepancies found. As with clinical data, self-reported behavior data are subject to a wide range of threats to their validity.2,3 Such threats might come from the context in which the study takes place and whether assurances of confidentiality are convincing to participants, or they might come from specific characteristics of the study population, the study design, the interviewer, and the data collection instrument. Understanding these factors and how they can influence accuracy is crucial to estimating the degree of potential bias and to using effective methods to minimize bias. To ensure that valid self-reported behavior data are collected, specific procedures must be followed carefully. For example, interviewers must be chosen who can provide credible assurances of confidentiality, and they should be nonjudgmental, appropriately trained, and of the same sex and ethnicity as the respondent.4,5 Further, the data collection instrument must be designed with careful attention to question wording and order, and to the time frame in which the respondent is expected to recall or project events.6,7 In research on sexual behavior, having the participants keep coital logs or diaries is one method that has been used to minimize error due to recall bias (i.e., inaccurate reporting because of faulty memory.8 Decisions about these data collection procedures are as complex—and as critical—as decisions about the procedures that should be used to collect, store, transport, and test clinical specimens. Zenilman et al mention some of the known threats to the validity of self-report, such as recall bias, poor measurement of construct such that reported use may not mean correct use, and social desirability (i.e., respondents' efforts to give what they perceive as the desired response). They provide, however, no evidence that they attempted to control for these threats. For example, the researchers state that interviews were administered by a study clinician, rather than a trained interviewer appropriately matched to the study population, thus creating a strong potential response bias. Further, they did not describe their procedures for assuring patient privacy and confidentiality, and thus we can only assume that none were in place. To obtain valid self-reported data, especially in a clinic setting in which situational pressures to report condom use may be strong, an interviewer must be objective and convincingly assure the respondents that the information they provide will not be shared with clinicians and clinic staff. That condom use reports were inaccurate is not surprising—the Zenilman et al study shows how misleading patient responses to condom use questions can be if threats to validity are not well understood and carefully controlled in the study design and methods. Their conclusions, however, that self-reported sexual behavior data are generally inaccurate and should not be relied on to evaluate interventions, are unwarranted. When researchers assume that the multiple threats to the reliability and validity of self-reported data cannot be controlled, techniques that are available for controlling these threats are not used, or are not used appropriately. A substantial body of research assures us that valid self-report can be collected, even under difficult circumstances and from populations at high risk. For example, two decades of studies on alcohol and drug use demonstrate that, given proper methodology, valid self-report can be collected (as verified by collateral informants or biochemical test).9–14 In a study of cocaine and heroin use among methadone maintenance clients, Ehrman et al compared self-reports of drug use with laboratory reports of weekly, random, supervised urine samples taken over a 6-month period, and found the self-reports reliable and valid.15 Techniques used to control measurement error included having interviews conducted by nonclinic staff and assuring clients that the information they provided would be fully confidential and that clinic personnel would not have access to the interview records. The research on the validity of self-reported smoking cessation (as assessed by a variety of methods including biochemical test) is even more extensive.16 Velicer et al. (1992), in a summary of this research,17 provide estimates of the level of misreporting that can be expected under specific conditions and from specific study populations when data are collected by appropriate methods. Research on the validity of self-reported sexual behavior is less well-developed.7 Studies to date, however, generally confirm that reliable and valid self-reported data can be collected. Research on prostitutes18 and homosexual men suggests that sexual histories, taken either in repeated interviews or through self-administered test-retest procedures, are reliable.8,19,20 The observations that self-reported condom use is associated with decreased human immunodeficiency virus seroconversion in studies of discordant couples21,22 and with lower rates of endocervical gonorrhea and genital lesions in research on commercial sex workers23 support the validity of these self-reports. We are not suggesting that self-reported data, collected properly, will be entirely free of error, or that individuals will never misrepresent their behavior. Rather, what we want to emphasize is that under-reporting and over-reporting of behavior are not random,24–26 and that threats to the validity of self-reported data can be controlled. Because measurement error can be such a serious problem, it is important to minimize it by using appropriate techniques. To better understand the conditions under which sexual risk behaviors may be misreported and to generate estimates of measurement error, we must conduct methodological research that builds on the well-established body of knowledge on self-report. As we need sensitive, specific laboratory tests to advance our knowledge of medical conditions, we need precise and valid measures of self-reported behavior and of related cognitive, emotional, and psychosocial constructs to strengthen our understanding of human behavior.27
No takes yet. Share an insight, caveat, or question.
Galavotti et al. (1996) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: