Five studies behind popular dating advice were retracted. The advice wasn’t.
I spent about a year reading the primary literature behind mainstream dating advice, mostly because I’d repeated a lot of it confidently and wanted to check I hadn’t been wrong. I had been.
A recognisable shelf of that advice — the confident, citation-flavoured kind, the kind with a chart — rests on studies that have been retracted, placed under formal expressions of concern, or crushed by large preregistered replications. The advice did not get quietly withdrawn when the evidence collapsed. It got repackaged. It is still on sale this week.
Here are the bodies. Every claim below is checkable; sources are named at the point of use.
1. The touch experiments Retracted
If you were ever taught that a brief touch on a woman’s forearm dramatically increases compliance with a courtship request, you were taught a result from the field-experiment corpus of Nicolas Guéguen — an extraordinarily prolific producer of small, charming, headline-friendly studies. That corpus drew formal expressions of concern in 2023–24 and five retractions in March 2025, after independent critics documented implausible data patterns and missing ethics approvals.
There is no preregistered replication of the courtship touch-compliance effect. That means there is no trustworthy effect size for it — anyone quoting a number is quoting one whose source no longer exists. It is still taught this week.
2. Wear red Failed twice
The red effect — wear red, be perceived as more attractive — was a media favourite for a decade. The meta-analysis puts it at about d = 0.13, with evidence of publication bias, and it has failed two large replications. (Lehmann, Elliot & Calin-Jageman, 2018)
A d of 0.13 is not a strategy. It is wardrobe superstition with a p-value. Wear red because you like red.
3. Power posing Disavowed
Stand expansively in a bathroom for two minutes, alter your testosterone and your risk appetite. The large-sample replications failed on the physiological and behavioural outcomes, and a co-author of the original paper publicly withdrew confidence in the effect.
What partly survived is the subjective bit: expansive posture makes people feel more powerful. Worth knowing, and a much smaller claim than the one you were sold. Watch for that pattern — the felt effect surviving while the mechanism collapses. It recurs constantly in this field.
4. Ovulation targeting Killed by preregistration
Two meta-analyses published in the same year reached opposite conclusions about whether women’s mate preferences shift across the cycle. (Gildersleeve, Haselton & Fales, 2014, yes; Wood et al., 2014, no) That should have been the first clue.
Then the large, preregistered, hormonally validated studies arrived and found no compelling evidence that preferences track hormonal status. (Jones et al., 2018; Jünger et al., 2018) A decade of confident findings evaporated the moment somebody had to declare their analysis in advance.
5. “Deviance in the dark” A 1973 magazine story
“Dim lighting disinhibits people” traces to a 1973 Psychology Today piece — a magazine report, never peer-reviewed, no published data tables. A 2012 attempt to replicate it failed. (Gergen, Gergen & Barton, 1973)
Every “set the mood with lighting” recommendation you have read descends from a fifty-year-old magazine anecdote. That is the whole evidence base. All of it.
6. “70% of women have responsive desire” Nobody found that
This statistic is everywhere — coaching decks, wellness content, the better sort of dating advice. It is a misattribution: measured endorsement in the underlying work runs closer to 25–29%. The phenomenon is real and important. The famous number is roughly two and a half times too large, and it got that way because nobody checked.
Which is the actual lesson: a percentage quoted without a citation is usually wrong.
Why this matters more than the individual myths
None of these were quietly withdrawn when the evidence collapsed. They were repackaged. The retraction notice and the advice live in completely separate information ecosystems — and only one of them has a marketing budget.
The replication crisis gets covered as an academic story. It is also a consumer-protection story, and almost nobody covers that half.
Five questions to run on any “research says” claim
- Has it been retracted? Search Retraction Watch before you search for the conclusion. The touch corpus was cited in textbooks for years after its problems were public.
- Was it preregistered? Preregistration removes analytic wiggle room. An unregistered small study is a hypothesis, not a result.
- Did a second lab find it? Collapses cluster in single-lab findings — power posing, red. Require two independent labs before you change your behaviour.
- What is the effect size? Demand the number. If nobody will tell you the number, there isn’t one worth telling.
- Who were the participants, and what were the stakes? Undergraduates rating hypothetical strangers on a screen generalise poorly to a live Tuesday.
Most popular claims fail the first or third question. A fair number fail all five — including some I once repeated to people who trusted me, which is how this project started.
What actually survives
The findings that hold up under replication are duller and structural: ask better questions. Meet in person sooner. Make declining cheap. Remove pressure. Say the true thing.
None of that fits on an infographic — which is roughly why nobody’s selling it to you.
I ended up writing the full audit as a book — every claim graded for evidence strength, including the ones that came out weak, with the dead findings named in their own chapter. The first three chapters are free below, and the five questions above are printed in it so you can run them on the book itself.