Social Desirability Bias

Social desirability bias (SDB)—also termed socially desirable responding or faking good—is a pervasive form of systematic measurement and information bias in survey epidemiology, behavioral surveillance, clinical adherence monitoring, and nutritional biostatistics in which study participants distort self-reported answers to align with perceived cultural norms, legal standards, or clinician expectations. First psychometricized by Allen L. Edwards (1957) and Douglas P. Crowne and David Marlowe (1960) and decomposed into two distinct cognitive dimensions by Delroy L. Paulhus (1984, 1991), social desirability bias causes respondents to systematically over-report socially approved or health-promoting behaviors (such as handwashing, childhood vaccine adherence, condom use, physical activity, and fruit-and-vegetable consumption) while under-reporting stigmatized, illegal, or medically discouraged behaviors (such as cigarette smoking during pregnancy, alcohol and illicit substance use, antiretroviral non-adherence, high caloric intake, and unsafe sexual practices). In public health research, unmitigated social desirability bias inflates population coverage estimates of preventive interventions, masks transmission corridors in sexually transmitted and blood-borne outbreak investigations, and corrupts dose-response regressions in nutritional and behavioral cohorts.

Impression Management vs. Self-Deceptive Enhancement

While early psychometricians treated socially desirable responding as a unitary personality trait reflecting a need for social approval (Crowne & Marlowe, 1960), factor-analytic experiments by Delroy L. Paulhus (1984) demonstrated that social desirability bias comprises two statistically orthogonal mechanisms that respond differently to survey anonymity:

1. Impression Management (IM): The conscious, deliberate falsification of responses tailored to an external audience (an interviewer, physician, employer, or public health auditor) to avoid embarrassment, legal jeopardy, or social disapproval. Because Impression Management is an intentional public presentation strategy, its magnitude varies strongly across interview modes—peaking during face-to-face interviewer-administered questionnaires and dropping markedly when responses are collected via private, anonymous computerized interfaces.
2. Self-Deceptive Enhancement (SDE): An unconscious, narcissistic cognitive bias in which the respondent genuinely believes their overly positive self-assessment (for example, sincerely believing that they exercise 60 minutes daily or consume only 1,500 kilocalories per day while ignoring habitual snacking). Because Self-Deceptive Enhancement protects internal self-esteem rather than public reputation, it persists even when surveys are conducted in total anonymity and can be corrected only through objective physiological biomarkers or accelerometry.

The Crowne–Marlowe Scale and Balanced Inventory of Desirable Responding (BIDR)

To quantify an individual participant's propensity for socially desirable responding and adjust epidemiological models accordingly, investigators embed validated psychometric scales alongside health questionnaires:

• The Marlowe–Crowne Social Desirability Scale (MCSDS; Crowne & Marlowe, 1960): A 33-item true/false instrument (with validated 13-item short forms, Reynolds, 1982) presenting culturally approved behaviors that have a very low probability of universal occurrence (for example, "I have never intensely disliked anyone" or "I am always courteous, even to people who are disagreeable") paired with common minor faults that nearly everyone commits. High scores indicate a strong tendency to deny ordinary human shortcomings.
• The Balanced Inventory of Desirable Responding (BIDR; Paulhus, 1991): A 40-item Likert-scale instrument featuring two 20-item subscales that separately score Self-Deceptive Enhancement (SDE) and Impression Management (IM), allowing epidemiologists to test whether an observed association between two self-reported variables is confounded by personality-driven response distortion.

Nutritional Epidemiology and Doubly Labeled Water Validation (Hebert et al., 1995)

In nutritional epidemiology, self-reported dietary intake via 24-hour recalls and Food Frequency Questionnaires (FFQs) is notoriously vulnerable to social desirability bias. When James R. Hebert et al. (1995, 1997) and the NCI Observing Protein and Energy Nutrition (OPEN) Study (Subar et al., 2003) compared self-reported caloric intake against doubly labeled water (²H₂¹⁸O)—the metabolic gold standard that measures total carbon dioxide (CO₂) production and energy expenditure over 14 days—they found that respondents systematically under-reported energy intake by 15% to 35%, selectively omitting refined carbohydrates, saturated fats, and alcohol. Critically, every 1-point increase on the Marlowe–Crowne Social Desirability Scale predicted an additional downward distortion of 20 to 45 kilocalories per day, with stronger under-reporting among overweight and obese participants. This differential misclassification flattens or inverts observed regressions between dietary fat/sugar intake and cardiometabolic disease.

Maternal Smoking Validation via Serum Cotinine

Because cigarette smoking during pregnancy carries intense clinical and social stigma, perinatal surveys relying solely on maternal self-report miss a substantial fraction of fetal tobacco exposures. Using nicotine's primary metabolite cotinine (half-life ≈ 16–20 hours in non-pregnant adults, ~9 hours in pregnancy) measured in serum or urine in the U.S. National Health and Nutrition Examination Survey (NHANES), Patricia M. Dietz et al. (2011) demonstrated that 22.9% of pregnant smokers with biochemically confirmed active smoking (serum cotinine >10 ng/mL) reported being non-smokers during interviewer-administered questionnaires, compared with only 9.2% non-disclosure among non-pregnant women.

Antiretroviral Adherence in HIV Prevention Trials (FEM-PrEP and VOICE)

One of the most consequential manifestations of social desirability bias in clinical trials occurred in the FEM-PrEP (Van Damme et al., 2012) and VOICE (Marrazzo et al., 2015) randomized placebo-controlled trials of oral and topical tenofovir pre-exposure prophylaxis (PrEP) among young women in sub-Saharan Africa. Across monthly clinic visits, participants reported 90% to 95% daily adherence by self-report and returned empty pill bottles; yet the trials found zero efficacy against HIV acquisition. Subsequent pharmacokinetic assays of plasma tenofovir revealed that fewer than 25% to 30% of participants had actually taken the medication within the preceding week. Motivated by polite deference to trial clinicians and fear of losing access to ancillary primary healthcare provided by the study clinics, participants discarded unused pills before visits—proving that subjective adherence metrics cannot substitute for dried-blood-spot or hair drug-level biomarkers.

Audio Computer-Assisted Self-Interviewing (ACASI)

Because Impression Management is triggered by interpersonal interaction with a human interviewer, removing the interviewer from sensitive questions substantially reduces bias. In Audio Computer-Assisted Self-Interviewing (ACASI), instituted across national behavioral surveys by Charles F. Turner et al. (1998) in Science, the respondent wears private headphones, listens to pre-recorded questions (eliminating literacy barriers), and keys responses directly into a tablet or computer screen hidden from field staff. In randomized split-ballot trials among adolescent and adult cohorts, switching from face-to-face interviews to ACASI increased the reporting of male-to-male sexual contact, injection drug use, and induced abortion by 200% to 400% (Turner et al., 1998).

Warner's Randomized Response Technique (RRT) and Item Count Lists

When querying illegal or highly stigmatized behaviors (such as illicit drug use, tax evasion, or wildlife poaching), Stanley L. Warner's (1965) Randomized Response Technique (RRT) provides mathematical privacy so that not even the investigator can know any individual participant's true status:

1. The respondent privately flips a physical randomization device (such as a fair coin or spinner) out of sight of the interviewer.
2. Under the Unrelated-Question or Forced-Response RRT design, if the coin lands Heads (probability p = 0.50), the respondent is instructed to answer "Yes" automatically, regardless of their true behavior. If the coin lands Tails (1 − p = 0.50), the respondent answers the sensitive question ("Did you use intravenous heroin in the past 30 days?") truthfully ("Yes" with unknown population prevalence π, or "No").
3. Because any individual "Yes" response could simply reflect a Heads coin flip, the respondent enjoys complete plausible deniability. Yet across a sample of N respondents observing a total fraction λ of "Yes" answers:

λ = p + (1 − p)π ⟹ π̂ = (λ − p) / (1 − p) = 2λ − 1 Thus, the epidemiologist recovers an unbiased point estimate and variance of the true population prevalence π̂ without ever knowing any single person's true behavior! A non-probabilistic alternative is the Item Count Technique (List Experiment), in which a control group counts how many of 4 benign statements apply to them while a treatment group counts how many of 5 statements (the 4 benign statements plus 1 sensitive item) apply; the difference in group mean counts equals the population prevalence of the sensitive behavior.

How does social desirability bias differ from recall bias and the Hawthorne effect?

Recall bias is an involuntary memory distortion in retrospective studies where individuals who have suffered an adverse disease outcome ruminate on and remember past exposures with higher sensitivity than healthy controls. The Hawthorne effect is a true behavioral modification in which observed subjects actually change their physical actions (such as washing their hands more frequently) while under observation. By contrast, social desirability bias is a reporting distortion: the participant's actual behavior remains unchanged, but they misreport that behavior on a survey to appear compliant or virtuous.

Why can't researchers simply exclude participants who score high on the Marlowe–Crowne scale?

Excluding high-scoring respondents on the Marlowe–Crowne Social Desirability Scale (MCSDS) reduces sample size and can introduce selection (collider) bias because MCSDS scores correlate with age, socioeconomic status, conscientiousness, and religious participation. Instead, biostatisticians include the participant's continuous MCSDS or BIDR score as a calibration covariate in regression models alongside biomarker validation sub-studies (regression calibration) to adjust the slope of self-reported exposures.

1. Crowne, D. P., & Marlowe, D. (1960). "A new scale of social desirability independent of psychopathology." Journal of Consulting Psychology, 24(4), 349–354. https://doi.org/10.1037/h0047358
2. Warner, S. L. (1965). "Randomized response: a survey technique for eliminating evasive answer bias." Journal of the American Statistical Association, 60(309), 63–69. https://doi.org/10.1080/01621459.1965.10480775
3. Paulhus, D. L. (1984). "Two-component models of socially desirable responding." Journal of Personality and Social Psychology, 46(3), 598–609. https://doi.org/10.1037/0022-3514.46.3.598
4. Hebert, J. R., Clemow, L., Pbert, L., Ockene, I. S., & Ockene, J. K. (1995). "Social desirability bias in dietary self-report may compromise the validity of dietary intake measures." International Journal of Epidemiology, 24(2), 389–398. https://doi.org/10.1093/ije/24.2.389
5. Turner, C. F., Ku, L., Rogers, S. M., Lindberg, L. D., Pleck, J. H., & Sonenstein, F. L. (1998). "Adolescent sexual behavior, drug use, and violence: increased reporting with computer survey technology." Science, 280(5365), 867–873. https://doi.org/10.1126/science.280.5365.867