research methods
Social desirability bias
What is social desirability bias?
Social desirability bias is the tendency to answer in ways that make us look good rather than ways that are true - over-reporting the admirable, under-reporting the embarrassing. In research it means self-report drifts toward the flattering: people say they read the terms, exercise more, and skip the ads more than they really do, especially when someone's watching.
Also known as: social desirability bias, social desirability, self-report bias
Prefer to watch?
Watch the recap 0:50
The demo
Same honest question, asked two ways. Flip between a named, face-to-face survey and a truly anonymous one, and watch how respectable everyone suddenly becomes when their name is attached.
“Do you always read the terms and conditions before agreeing?”
What this demo shows (text version)
A slightly embarrassing self-report question - "do you always read the terms and conditions?" - is shown asked two ways. In a named, face-to-face survey, the reported share who say "yes, always" is high, because admitting you skip them feels a bit shameful with your name attached. In a guaranteed-anonymous version, that figure collapses to a far more honest, much lower level. (The numbers here illustrate the effect rather than report real data.)
The gap between the two is social desirability bias: people give the respectable answer, especially when identifiable. It's a big reason self-report and real behaviour diverge - surveys catch the presentable self, analytics catch the actual one. Reduce it with real anonymity, neutral wording and indirect questions, but the surest defence is to measure what people do, not just what they say.
People tell you the version of themselves they'd like to be true. So self-reported behaviour skews toward what's socially approved - inflating the virtuous, hiding the awkward - and it gets worse the less anonymous the setting. Reduce it with genuine anonymity, neutral non-judgemental wording, and indirect phrasing - but the surest fix is to measure behaviour, not claims, because what people do beats what they say they do.
Ask it with a name attached and almost everyone "always reads the terms"; ask it anonymously and the honest, sheepish truth appears. Nobody lied exactly - they just gave the answer that felt respectable. That gap between the named and anonymous version is social desirability bias, and it's why "what people say" and "what people do" so rarely match.
We're wired to manage how others see us, so any time an answer could reflect on our character - habits, money, health, prejudice, how carefully we read that contract - we shade it toward the approved version. The effect is strongest face-to-face or wherever responses feel identifiable, and weakest when people are confident no one can trace the answer to them.
It's a major reason self-report and behaviour diverge. Surveys and interviews capture the presentable self; analytics capture the actual one. "I'd definitely pay for that," "I never skip onboarding," "I always check the permissions" routinely overstate reality - so designing around stated intentions, rather than observed behaviour, is a classic way to build the wrong thing.
Mitigations help but don't fully cure it: guarantee and clearly explain anonymity, word questions neutrally and without a "right" answer, allow honest outs, and sometimes ask indirectly ("how often do people like you…"). The strongest move, though, is to triangulate self-report with behavioural data - and when they disagree, trust what people did.