Social Desirability

Abstract

Social desirability is a form of social behavior: the tendency to present oneself in a way that will be viewed favorably by others, and the response bias this introduces into self-report. MeSH defines it as a personality trait centered on a desire for approval and the avoidance of criticism. Cognitive psychology studies it along converging lines: Edwards's finding that how often a trait is endorsed tracks how desirable it is judged to be; the scales built to measure the disposition, from the Edwards and Marlowe–Crowne instruments to the Balanced Inventory of Desirable Responding; the two-component model splitting honest-but-inflated self-views from deliberate faking; and the survey-methods arsenal — randomized response, the bogus pipeline — for reducing the bias. A modern meta-analytic critique questions whether these scales measure bias at all. A worked example and three interactive demonstrations follow.

Keywords: social desirability, response bias, impression management, self-deceptive enhancement, randomized response technique

What Social Desirability Is

Social desirability is the tendency to respond to questions about oneself in a way that will be viewed favorably by others — over-reporting approved behaviors and attitudes, under-reporting disapproved ones — together with the systematic distortion this introduces into self-report data. MeSH defines it as a personality trait rendering the individual acceptable in social or interpersonal relations, centered on a desire for approval and a tendency to avoid criticism, and files the descriptor under social behavior, of which it is one species. The placement is apt: the disposition exists only in relation to a real or imagined audience whose approval the respondent seeks, even though its engine — the weighing of how an answer will be received against what is true — is cognitive.

A distinction organizes the whole field and is easy to blur. Social desirability can name a stable personality trait — a dispositional need for approval that some people carry more strongly than others — or a transient response bias, the contamination of a particular measurement by the pull toward looking good. The two are related but not identical: a person high in the trait will bias more of their answers, but even a person low in the trait will shade responses when a question is threatening enough. Much of the research below is an attempt to say, for a given score, how much reflects what the respondent is really like and how much reflects the management of an impression — a question that turns out to be harder than the early scale-builders assumed.

The concept also sits within a family of terms worth keeping apart. Faking good is the deliberate, motivated distortion of answers to secure an outcome, as when a job applicant inflates their conscientiousness. Acquiescence is the unrelated tendency to agree with items regardless of content. Self-deception is distortion the respondent does not know they are committing, an honestly held but unrealistically positive self-view. Social desirability proper spans the honest and the strategic ends of this range, which is exactly why a single score is ambiguous. The sections that follow take up the discovery of the social desirability variable and the scales built to capture it, the two-component model that resolves much of the ambiguity, the survey-methods toolkit for reducing the bias, and the sensitive-questions literature that studies when it bites hardest; Table 1 orients the reader to these converging lines before each is developed in turn.

The Desirability Variable and Its Scales

The systematic study of social desirability begins with Allen Edwards's observation that the content of a personality item is not the only thing that governs whether people endorse it. Edwards had judges rate a large set of personality statements for how socially desirable it would be to possess the characteristic each described, then compared those ratings with the proportion of a separate sample who actually endorsed each item as self-descriptive. The two lined up almost perfectly: the correlation between judged desirability and endorsement probability exceeded .80, so that the more desirable a trait was rated, the more people claimed to have it (Edwards, 1953). The finding was unsettling because it implied that a large share of the variance in any self-report inventory might reflect not the respondents' actual standing on the trait but their shared sense of which answer looks better — a pervasive response determinant riding underneath the intended content.

Edwards turned the observation into a measuring instrument, assembling the Edwards Social Desirability Scale from items drawn from the Minnesota Multiphasic Personality Inventory on which clinical judges agreed about the desirable answer. But building the scale from clinical items created a confound that his successors found unacceptable: because the items described psychopathology, a high score could mean either a disposition to answer desirably or genuinely good adjustment, and the scale correlated heavily with measures of maladjustment. Douglas Crowne and David Marlowe set out to measure the disposition cleanly, apart from psychopathology. Their Marlowe–Crowne Social Desirability Scale consists of thirty-three items describing behaviors that are culturally approved but improbable (claiming one never hesitates to help someone in trouble) or disapproved but nearly universal (admitting one sometimes resents being asked to return a favor), so that the desirable answer is in each case a small lie almost anyone could tell (Crowne & Marlowe, 1960). A high score therefore indexes a need for social approval rather than clinical status — the readiness to misrepresent oneself in the approved direction on matters where the truth is mundane.

The Marlowe–Crowne scale became the most widely used measure of socially desirable responding, but its length and dated wording prompted refinements. Joachim Stöber's Social Desirability Scale-17 modernized the content and shortened the instrument to seventeen items, reporting adequate internal consistency, convergent validity with the Marlowe–Crowne and related scales, and discriminant validity against substantive personality dimensions, while noting the characteristic correlation of desirable responding with age (Stöber, 2001). Across these instruments a common logic holds: the scale presents choices where one answer is flattering and improbable and the other honest and unflattering, and treats a pattern of flattering answers as a signature of the disposition. What none of them can do on its own is say why a respondent gives the flattering answer — whether they believe it or merely assert it — which is the problem the two-component model was built to solve.

The Two-Component Model

Delroy Paulhus resolved much of the single-score ambiguity by showing that socially desirable responding is not one thing but two, separable both conceptually and statistically (Paulhus, 1984). The first component is self-deceptive enhancement: an honestly held but unrealistically positive view of oneself, in which the respondent genuinely believes the flattering answer. This is not lying — the person is not aware of any distortion — and it is tied to adjustment and well-being, since a rosy self-image is characteristic of the psychologically healthy. The second component is impression management: the deliberate, conscious tailoring of answers to create a favorable impression on a specific audience, which is lying in the ordinary sense and rises and falls with the incentives of the situation. The two components are only weakly correlated, respond differently to instructions, and have different correlates, so collapsing them into a single desirability score discards exactly the information needed to interpret it.

The distinction has sharp practical consequences. Because impression management is responsive to context, it inflates sharply under conditions where looking good pays — job applications, evaluations, any setting the respondent knows is consequential — whereas self-deceptive enhancement stays relatively stable because the respondent is not managing anything. This predicts, correctly, that the desirability signal in anonymous research answered in private is dominated by self-deception, while the signal in high-stakes assessment is dominated by impression management. Paulhus operationalized the model in the Balanced Inventory of Desirable Responding, which scores the two components on separate subscales, so that a researcher can ask not merely whether a respondent answered desirably but which kind of desirability is at work and whether it is the kind that situational anonymity should remove.

The ResponseDecomposition demonstration below makes the two-component model interactive: setting a respondent's level of self-deceptive enhancement and of impression management, and toggling whether the setting is anonymous research or a high-stakes evaluation, shows how the same observed desirability score decomposes into an honest-but-inflated part that stays fixed and a strategic part that swells when looking good pays.

Detecting and Reducing the Bias

If social desirability contaminates self-report, the methodological response is to remove it at the point of measurement rather than merely measuring it afterward. Anton Nederhof's review organized the available techniques into a taxonomy that remains the standard map: some methods work on the rater (selecting respondents low in the disposition, or using others' reports), some on the instrument (neutral question wording, forced-choice formats that pit equally desirable options against each other, randomized response), and some on the testing conditions (guaranteed anonymity, self-administration, the bogus pipeline) (Nederhof, 1985). The most effective, he concluded, act on the conditions and the instrument rather than trying to correct a contaminated score after the fact — a judgment the subsequent literature has largely upheld.

Two instrument-level techniques are landmarks. Stanley Warner's randomized response technique makes a truthful answer to an incriminating question safe by ensuring the interviewer never knows which question the respondent answered. A randomizing device — a coin, a spinner — secretly assigns each respondent either the sensitive statement or its negation (or an unrelated innocuous one), and the respondent answers only 'yes' or 'no'; because the interviewer cannot tell which statement was answered, no individual answer reveals anything, yet the known probabilities of the randomizer let the analyst recover the population prevalence from the aggregate (Warner, 1965). Edward Jones and Harold Sigall's bogus pipeline attacks the problem from the opposite direction: respondents are connected to an apparatus they are told can detect their true attitudes physiologically, so that lying would be pointless and even embarrassing, which motivates candor through the belief that dishonesty is detectable (Jones & Sigall, 1971). Robert Fisher's indirect questioning takes a subtler route, asking respondents about the attitudes of others like them rather than about themselves; because people project their own private views onto similar others, the indirect estimate recovers the sensitive attitude while sparing the respondent the threat of a direct admission (Fisher, 1993). Each technique buys validity at a price — in statistical efficiency, in apparatus and deception, or in interpretive directness — and the worked example below quantifies that price for the randomized response case.

Sensitive Questions in Surveys

Survey methodology has mapped in detail when social desirability bias actually bites, and the answer is: when a question is sensitive and when the conditions of answering make honesty feel costly. Roger Tourangeau and Ting Yan's synthesis of the literature on sensitive questions distinguished three senses in which a question can be sensitive — intrusive (an unwelcome invasion of privacy regardless of the answer), threatening in its risk of disclosure to third parties, and socially undesirable in the answer it would truthfully elicit — and showed that each has distinct remedies (Tourangeau & Yan, 2007). Their central empirical conclusion is that the mode and format of data collection govern misreporting: respondents under-report socially undesirable behavior (illicit drug use, abortion) and over-report desirable behavior (voting, religious attendance), and the distortion shrinks reliably when the respondent answers a self-administered questionnaire rather than speaking to an interviewer, because removing the human listener removes much of the impression-management pressure.

Ivar Krumpal's review assembled the determinants of social desirability bias in sensitive surveys into a coherent account built on the respondent's cost–benefit calculation: a respondent misreports when the perceived cost of a truthful answer — embarrassment, legal jeopardy, loss of the interviewer's regard — outweighs the weak benefit of accuracy, and anything that lowers that perceived cost lowers the bias (Krumpal, 2013). The practical levers follow directly: credible guarantees of confidentiality and anonymity, self-administration, dedicated-question techniques such as randomized response, and careful wording that normalizes the undesirable behavior all reduce misreporting by reducing its perceived cost. The survey-methods tradition thus converges with the two-component model from the other side: impression management is a situational calculation, and the design of the survey is precisely the manipulation of that situation.

The SensitiveQuestionMode demonstration below makes this interactive: setting the true prevalence of a socially undesirable behavior and choosing the data-collection mode — face-to-face interview, self-administered questionnaire, or a confidentiality-assuring technique — shows how the reported prevalence falls short of the truth by an amount that shrinks as the mode lowers the perceived cost of an honest answer.

Worked Example: Recovering the Truth From Randomized Response

Figure 1

Warner's technique can be made precise with a little arithmetic that ties the privacy guarantee to the statistical price it exacts. A randomizing device assigns each respondent, with known probability p, the sensitive statement ('I have done X') and, with probability 1 − p, its negation ('I have not done X'); the respondent answers 'yes' or 'no' truthfully, and only they know which statement they answered. If the true population prevalence of the sensitive attribute is π, then a 'yes' arises in two ways — a respondent who has the attribute and received the direct statement, or one who lacks it and received the negation — so the expected proportion of yes answers is

λ = p · π + (1 − p) · (1 − π)

Take a design with p = 0.75 and a true prevalence π = 0.30. Then the expected yes-proportion is λ = 0.75 × 0.30 + 0.25 × 0.70 = 0.225 + 0.175 = 0.40. The analyst does not know π — that is what the survey is trying to learn — but observes λ ≈ 0.40 in the aggregate and inverts the relationship. Solving λ = p·π + (1−p)(1−π) for π gives the estimator

π̂ = (λ − (1 − p)) / (2p − 1) = (0.40 − 0.25) / 0.50 = 0.15 / 0.50 = 0.30

which recovers the true prevalence exactly, and does so without any individual answer ever revealing whether that person has the attribute — the entire point of the method. The estimator is unbiased for any p ≠ 0.5; at p = 0.5 the denominator vanishes and the design carries no information, because the randomizer has made the two statements equally likely and the answer is pure noise.

The privacy is not free, and the cost is visible in the estimator's variance. For a sample of size n, the variance of π̂ is the ordinary sampling variance of a proportion plus a second term contributed entirely by the randomization:

Var(π̂) = π(1−π)/n + p(1−p) / [n·(2p−1)²]

With π = 0.30, p = 0.75, and n = 1000, the first term is 0.30 × 0.70 / 1000 = 0.00021 and the second is 0.75 × 0.25 / (1000 × 0.25) = 0.00075, for a total of 0.00096 and a standard error of √0.00096 ≈ 0.031. Had respondents simply answered a direct question truthfully, the standard error would be √(0.00021) ≈ 0.0145 — less than half as large. The randomized response technique therefore roughly doubles the standard error, which is the quantitative statement of a general law for every bias-reduction method in the previous section: protecting the respondent costs statistical efficiency, and the design question is always whether the honest answers thereby obtained are worth more than the precision given up. When the direct-question answers would be badly biased by social desirability, they are; when they would not, the extra noise is pure loss.

Discussion

The cognitive psychology of social desirability has converged, from several starting points, on a coherent account. At its center is a finding — Edwards's demonstration that the desirability of an item predicts its endorsement above .80 — that established the response bias as a pervasive contaminant of self-report rather than a nuisance confined to a few sensitive items (Edwards, 1953). The disposition behind the bias was given measuring instruments that progressively disentangled it from confounds: the Marlowe–Crowne scale from psychopathology, the SDS-17 from dated content, each treating a pattern of flattering-but-improbable answers as the signature of a need for approval (Crowne & Marlowe, 1960; Stöber, 2001). Paulhus's two-component model then resolved the deepest ambiguity in those scores by splitting honest self-deception from strategic impression management, explaining why the desirability signal is dominated by self-deception in anonymous research and by impression management in high-stakes assessment (Paulhus, 1984).

The applied response has been to attack the bias at the point of measurement rather than to correct it afterward. Nederhof's taxonomy organized the options, and the landmark techniques — Warner's randomized response, Jones and Sigall's bogus pipeline, Fisher's indirect questioning — each make honesty either safe or seemingly unavoidable, buying validity at a measurable cost in efficiency or apparatus (Nederhof, 1985; Warner, 1965; Jones & Sigall, 1971; Fisher, 1993). The survey-methods tradition specified when these remedies are needed, showing that misreporting is a function of a question's sensitivity and of a respondent's cost–benefit calculation, and that mode, format, and credible confidentiality are the levers that move it (Tourangeau & Yan, 2007; Krumpal, 2013). The practical upshot is that social desirability is neither an irreducible flaw in self-report nor a fixed trait to be subtracted out, but a tunable product of what is asked, of whom, under what guarantees — and the design of a study is largely the design of how much of it there will be.

Current Directions

The most consequential recent development is a direct challenge to the assumption underpinning the whole enterprise of social desirability scales: that they measure a bias at all. In a meta-analysis of the relationship between social desirability scales and prosocial behavior, Laura Lanz, Isabel Thielmann, and Fabiola Gerpott found that scores on these scales predict actual prosocial conduct — cooperation, honesty, generosity in behavioral tasks — in the way genuine personality traits do, rather than behaving like a content-free distortion (Lanz et al., 2022). If a 'social desirability' score reflects a substantive disposition toward prosociality, then using it to detect faking or, worse, partialling it out of a correlation as if it were pure noise removes real variance and can bias the very estimates it is meant to clean. The result formalizes a long-standing suspicion and cautions researchers against the routine use of these scales as bias controls.

A second, more applied line continues to refine bias reduction for specific research settings. Randall Larson's comparison of control methods in marketing research evaluated how item wording, response format, and administration mode trade off against one another in practice, offering survey designers concrete guidance on which controls actually reduce desirability bias in consumer self-report (Larson, 2019). Nancy Bergen and Ronald Labonté extended the problem to qualitative interviewing, where the usual quantitative remedies do not apply, showing how the 'everything is perfect' response — the interview analogue of the flattering questionnaire answer — can be detected and limited through interviewer technique and probing (Bergen & Labonté, 2020). And in environmental psychology, Stephan Vésely and Christian Klöckner's three meta-analyses asked whether pro-environmental self-reports are inflated by social desirability and found the contaminating effect smaller and more domain-specific than often assumed, a reminder that the magnitude of the bias is itself an empirical question that varies by content area (Vésely & Klöckner, 2020). Across these efforts the field is moving away from treating social desirability as a single universal nuisance to be scrubbed out, and toward asking, for each measure and setting, whether the bias is present, how large it is, and whether the standard correction does more harm than good.

Common Misconceptions

Social desirability is always deliberate lying.

Not necessarily. Paulhus's two-component model distinguishes impression management — conscious, strategic faking — from self-deceptive enhancement, an honestly held but inflated self-view the respondent does not know is distorted. Only the first is lying in the ordinary sense, and the two have different correlates and respond differently to anonymity (Paulhus, 1984).

A high social desirability score invalidates a person's other answers.

A high score does not by itself mark the rest of a protocol as fake. Because the scales predict real prosocial behavior as substantive traits do, a high score may reflect a genuine disposition, and partialling it out of other measures as if it were pure noise can remove valid variance and bias the result (Lanz et al., 2022).

Guaranteeing anonymity removes the bias entirely.

Anonymity removes much of the impression-management component, because there is no audience left to impress, but it leaves self-deceptive enhancement largely intact — the respondent is still not aware of that distortion. Anonymity shrinks the bias; it does not abolish it (Paulhus, 1984; Tourangeau & Yan, 2007).

Bias-reduction techniques are free improvements.

Every technique exacts a price. The randomized response technique roughly doubles the standard error of an estimate in exchange for its privacy guarantee, and the bogus pipeline requires apparatus and deception. A technique is worth using only when the bias it removes exceeds the efficiency or practicality it costs (Warner, 1965; Nederhof, 1985).

Glossary

Acquiescence.

The tendency to agree with questionnaire items regardless of their content; a response style distinct from social desirability, though both distort self-report.

Balanced Inventory of Desirable Responding (BIDR).

Paulhus's instrument scoring self-deceptive enhancement and impression management on separate subscales, operationalizing the two-component model.

Bogus pipeline.

A procedure in which respondents believe an apparatus can detect their true attitudes, motivating candor through the conviction that lying would be detected.

Edwards Social Desirability Scale.

An early scale built from MMPI items on which judges agreed about the desirable answer; confounded with psychopathology because its items described clinical content.

Impression management.

The deliberate, conscious tailoring of answers to create a favorable impression on an audience; the strategic component of socially desirable responding, which rises when looking good pays.

Indirect questioning.

Asking respondents about the attitudes of similar others rather than themselves; because people project their own views, it recovers a sensitive attitude while sparing a direct admission.

Marlowe–Crowne Social Desirability Scale.

A thirty-three-item scale measuring the need for social approval through approved-but-improbable and disapproved-but-universal behaviors, independent of psychopathology.

Need for social approval.

The dispositional motive the Marlowe–Crowne scale indexes: a readiness to misrepresent oneself in the approved direction to secure others' regard.

Randomized response technique.

A method in which a randomizing device secretly determines which of two questions a respondent answers, so no individual answer is revealing yet the population prevalence is recoverable from the aggregate.

Response bias.

A systematic tendency to answer questions on some basis other than their content; social desirability is the response bias toward the favorably viewed answer.

Self-deceptive enhancement.

An honestly held but unrealistically positive self-view; the non-conscious component of socially desirable responding, tied to adjustment and relatively insensitive to anonymity.

Sensitive question.

A question that is intrusive, carries a risk of disclosure, or would truthfully elicit a socially undesirable answer; such questions provoke the misreporting that bias-reduction methods target.

Social desirability variable.

Edwards's term for the item property — judged desirability — that predicts endorsement probability above .80, independent of the trait the item is meant to measure.

Social desirability.

The tendency to present oneself favorably, and the resulting distortion of self-report; classified by MeSH under social behavior and measurable as a trait or observable as a bias.

Key Researchers

Allen L. Edwards

(University of Washington, 1914–1994). Identified the social desirability variable, showing that the probability a personality item is endorsed correlates above .80 with how desirable the item is judged to be, and built the Edwards Social Desirability Scale from MMPI items. Wikidata

Ivar Krumpal

(University of Leipzig). Reviewed the determinants of social desirability bias in sensitive surveys and tested dedicated-question techniques such as the randomized response technique for reducing it under real survey conditions. ORCID

Delroy L. Paulhus

(University of British Columbia). Developed the two-component model of socially desirable responding, separating self-deceptive enhancement from impression management, and built the Balanced Inventory of Desirable Responding to measure both. Wikipedia

Isabel Thielmann

(Max Planck Institute for the Study of Crime, Security and Law, Freiburg). With Laura Lanz and Fabiola Gerpott, meta-analytically tested whether social desirability scales index a response bias, finding they predict prosocial behavior as substantive traits do and so function poorly as bias detectors or controls. ORCID

Roger Tourangeau

(Westat). With Ting Yan, synthesized the survey-methods literature on sensitive questions, showing how data-collection mode, question format, and perceived confidentiality govern the under-reporting of socially undesirable behavior. Google Scholar

Frequently Asked Questions

What is social desirability?

Social desirability is the tendency to answer questions about oneself in a way that will be viewed favorably by others (over-reporting approved behaviors and attitudes and under-reporting disapproved ones), together with the distortion this introduces into self-report. MeSH defines it as a personality trait centered on a desire for approval and the avoidance of criticism, and classifies it as a form of social behavior.

What is the difference between the Edwards and Marlowe-Crowne scales?

Both measure socially desirable responding, but the Edwards Social Desirability Scale was built from clinical MMPI items, so a high score is confounded with genuine good adjustment. The Marlowe-Crowne scale instead uses commonplace behaviors that are approved but improbable or disapproved but near-universal, so it measures a need for social approval independent of psychopathology.

What are the two components of socially desirable responding?

Paulhus distinguished self-deceptive enhancement, an honestly held but inflated self-view the respondent does not know is distorted, from impression management, the deliberate tailoring of answers to look good to an audience. The two are only weakly correlated, and impression management rises sharply when looking good pays while self-deception stays relatively stable.

How does the randomized response technique work?

A randomizing device, such as a coin, secretly assigns each respondent either a sensitive question or an innocuous or negated one, and the respondent answers only yes or no. Because the interviewer cannot tell which question was answered, no individual answer is revealing, yet the known randomizer probabilities let the analyst recover the population prevalence from the aggregate of answers.

Does anonymity eliminate social desirability bias?

It reduces but does not eliminate it. Anonymity and self-administration remove most of the impression-management component by removing the audience, which is why self-administered questionnaires produce more honest reports of sensitive behavior. But self-deceptive enhancement persists because the respondent is unaware of that distortion.

Why can it be a mistake to control for social desirability?

Because social desirability scales predict real prosocial behavior the way substantive personality traits do, a score may capture a genuine disposition rather than pure distortion. Partialling it out of a correlation as if it were noise can remove valid variance and bias the estimate it was meant to clean, as a 2022 meta-analysis cautioned.

Which questions are most affected by social desirability?

Sensitive questions: those that are intrusive, that risk disclosure to third parties, or that would truthfully elicit a socially undesirable answer. Illicit drug use, abortion, and income are under-reported; voting and religious attendance are over-reported. The distortion is largest when a human interviewer is present and shrinks under self-administration.

What is the bogus pipeline?

A technique in which respondents are connected to an apparatus they are told can physiologically detect their true attitudes. Believing that lying would be detected and embarrassing, respondents answer more candidly. It reliably reduces socially desirable responding but requires deception and equipment, which limits its practical use to the laboratory.

References

Bergen, N., & Labonté, R. (2020). "Everything is perfect, and we have no problems": Detecting and limiting social desirability bias in qualitative research. Qualitative Health Research, 30(5), 783–792. https://doi.org/10.1177/1049732319889354

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

Edwards, A. L. (1953). The relationship between the judged desirability of a trait and the probability that the trait will be endorsed. Journal of Applied Psychology, 37(2), 90–93. https://doi.org/10.1037/h0058073

Fisher, R. J. (1993). Social desirability bias and the validity of indirect questioning. Journal of Consumer Research, 20(2), 303–315. https://doi.org/10.1086/209351

Jones, E. E., & Sigall, H. (1971). The bogus pipeline: A new paradigm for measuring affect and attitude. Psychological Bulletin, 76(5), 349–364. https://doi.org/10.1037/h0031617

Krumpal, I. (2013). Determinants of social desirability bias in sensitive surveys: A literature review. Quality & Quantity, 47(4), 2025–2047. https://doi.org/10.1007/s11135-011-9640-9

Lanz, L., Thielmann, I., & Gerpott, F. H. (2022). Are social desirability scales desirable? A meta-analytic test of the validity of social desirability scales in the context of prosocial behavior. Journal of Personality, 90(2), 203–221. https://doi.org/10.1111/jopy.12662

Larson, R. B. (2019). Controlling social desirability bias. International Journal of Market Research, 61(5), 534–547. https://doi.org/10.1177/1470785318805305

Nederhof, A. J. (1985). Methods of coping with social desirability bias: A review. European Journal of Social Psychology, 15(3), 263–280. https://doi.org/10.1002/ejsp.2420150303

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

Stöber, J. (2001). The Social Desirability Scale-17 (SDS-17): Convergent validity, discriminant validity, and relationship with age. European Journal of Psychological Assessment, 17(3), 222–232. https://doi.org/10.1027//1015-5759.17.3.222

Tourangeau, R., & Yan, T. (2007). Sensitive questions in surveys. Psychological Bulletin, 133(5), 859–883. https://doi.org/10.1037/0033-2909.133.5.859

Vésely, S., & Klöckner, C. A. (2020). Social desirability in environmental psychology research: Three meta-analyses. Frontiers in Psychology, 11, 1395. https://doi.org/10.3389/fpsyg.2020.01395

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