The Smile That Launched a Thousand Textbooks
In 1988, three psychologists ran one of the most elegant experiments ever devised. Fritz Strack, Leonard Martin, and Sabine Stepper wanted to know whether smiling could make you happier, but asking outright would tip participants off. So they invented a cover story: participants rated cartoon funniness while holding a pen teeth-gripped (a forced smile) or lip-gripped (a forced pout). The teeth group rated the cartoons 0.82 points funnier on a 10-point scale, suggesting the face could talk back to the mind; textbooks reprinted the finding for three decades.
Then the replication crisis came for it.
Seventeen Labs and a Funeral
In 2016, Eric-Jan Wagenmakers and colleagues ran the definitive replication: 17 independent labs, more than 1,800 participants, one preregistered protocol vetted in advance, and the result was a flatline. The smile-minus-pout gap came out at 0.03 units, with a 95 percent confidence interval from −0.11 to 0.16, a range that comfortably includes zero, sits more than twenty-five times below the original 0.82, and is so narrow the whole interval would fit inside the original effect three times over. Press coverage wrote the obituary, and the pen trick joined power posing and ego depletion in the museum of beautiful findings that were not true. Case closed. For six years, that was the settled story.
The Camera in the Room
But there was a loose thread. In 2018, Tom Noah, Yaacov Schul, and Ruth Mayo spotted the discrepancy: the 17 labs had filmed participants, while Strack’s 1988 lab had not. When Noah’s team reran the pen paradigm with and without a camera, the effect reappeared unobserved and vanished on film. Their title said it plainly: both the original study and its failed replication were correct, because feeling observed makes people self-conscious enough to short-circuit the feedback loop. The most famous failure of the replication crisis might have been a methods artifact. Somebody needed to run the definitive test.
Skeptics and Believers Walk Into a Lab
That somebody turned out to be everybody. Nicholas Coles assembled the Many Smiles Collaboration, in which skeptics and believers co-designed the protocol so neither side could later claim the test was rigged. The Stage 1 protocol was accepted as a Registered Report in April 2020, freezing every analysis before data collection began. The design crossed facial pose (happy or neutral) with three induction methods, mimicry of smiling photographs, the voluntary facial action technique, and the classic pen-in-mouth task, and with whether pleasant images were shown, which separates amplifying a feeling from initiating one out of nothing. In the end 3,878 participants unaware of the hypothesis took part across 19 countries: more than twice the 2016 replication that had buried the effect.
What 3,878 Faces Revealed
Posing a happy expression raised self-reported happiness to 2.51 against 2.16 for a neutral pose, a significant difference (p = .002). Two of the three induction methods worked, facial mimicry (p < .001) and the voluntary facial action technique (p = .03), while the pen-in-mouth task, the famous one, did not (p = .73). The effect appeared whether or not pleasant images were present (no interaction, p = .64): a posed smile could both amplify a feeling already underway and initiate one from a neutral baseline.
How big? The 0.35-point gap divided by the pooled standard deviation of about 1.50 gives Cohen’s d ≈ 0.23, small by standard benchmarks and close to the authors’ own comparison: similar to looking at mildly pleasant images. The lead author’s translation, from his 2019 meta-analysis of five decades of experiments: if 100 people smiled, roughly seven would feel happier than if they had not.
Here is the irony the textbooks must reckon with: the pen-between-the-teeth demonstration, printed in every psychology textbook for thirty years and whose 2016 failure became the replication crisis’s favorite anecdote, was the only method that failed in the largest and most careful test ever run. The famous demo is the worst demo of a real effect. The methods nobody wrote textbooks about are the ones that work.
The Strongest Case Against
The skeptic’s case is strong and deserves no softening. In the mimicry and voluntary-action conditions, participants knew they were making happy faces; the manipulation was completely transparent. The one covert manipulation, the pen task, in which participants had no idea their expression was the point, produced nothing. That is precisely the pattern you would expect if people report feeling happier because they infer it is the expected answer rather than because facial muscles feed back into emotion; a preregistered adversarial collaboration is still a room full of people who know they are in a smiling study.
Then there is the size. A d of 0.23, seven people in a hundred, makes the self-help promise of “smile your way to happiness” a rounding error with a p-value; nobody’s depression is being treated by a posed grin. Worse, Noah’s camera finding cuts against the effect even as it rescues it, because if merely feeling observed eliminates facial feedback, the effect dies exactly where real-world smiling happens: in social situations, around other people, under observation. The honest scoring: the effect is real, it is small, and the most likely mundane explanation, that people tell researchers what they think researchers want to hear, has not been conclusively ruled out.
What We Didn’t Prove
This study did not prove that smiling treats depression or improves well-being, only that it nudges a laboratory happiness rating by a third of a point. It did not validate the pen trick. It showed the opposite. It did not test how long the feeling lasts, or whether it survives outside the lab. And with the outcome self-reported on a Likert scale and no behavioral or physiological measure, the demand-characteristics objection above remains live. The adversarial collaboration made the test fair, but it did not make the outcome objective.
The Bottom Line
For six years, the smart money said the facial feedback hypothesis was dead, a casualty of the replication crisis, good riddance. The largest and most careful test ever conducted says otherwise: the effect is real, it is small, and the famous demonstration of it was the weakest possible version, a verdict delivered by skeptics who had spent years arguing the effect did not exist. Science did not just correct an old finding here; it corrected the correction. And the premature funeral is itself now part of the evidence, a reminder that “failed to replicate” describes a specific method in a specific context, not a verdict on the phenomenon.
What You Can Do
To try it yourself, skip the pen. The methods that worked involved actually forming the expression: mimic a genuine smile, or pull the corners of your mouth toward your ears and hold it. Expect a nudge, not a transformation: roughly the emotional equivalent of glancing at a pleasant photograph. If you teach psychology, the 2016–2022 arc is the real lesson to assign: not “smiling is debunked,” but “here is what it looks like when skeptics and believers design the test together and let the data settle it.” Keep the honest number in mind the next time someone sells you a grin as therapy: about seven in a hundred.
Related Reading
- Fact-Checking Doesn’t Backfire. 52 Issues, 10,100 People, Zero Boomerangs
- Half of Social Science Didn’t Survive a Seven-Year Retest. 274 Claims Redone, and the Survivors Came Back Half as Strong
- Hundreds of Studies Confirmed That the Least Competent People Are the Most Overconfident. A Psychological Review Paper Found They All Fell for the Same Statistical Illusion.