Making people smile without saying “smile”
Everyone knows that emotions show on our faces. The facial feedback idea reverses the direction: the shape of our face may also change what we feel a little. The difficulty of testing this is obvious: tell someone to “smile” and they will understand what you are measuring and may rate the cartoons as funnier just to help you out.
In 1988 Fritz Strack, Leonard Martin and Sabine Stepper found a clever solution. They asked participants to hold a pen in their mouth, using a cover story that hid the real purpose of the study. One group held the pen with their teeth, unknowingly bringing their face closer to a smile; the other held it with their lips, which made smiling harder.
1988: 5.14 versus 4.32
With the pen in their mouths, participants rated cartoons from 0 (not funny at all) to 9 (very funny). Those holding the pen with their teeth gave 5.14 on average, those holding it with their lips 4.32. The difference of 0.82 points entered textbooks as one of the most elegant pieces of evidence for facial feedback.
By 2016 the study had been cited more than 1,300 times, and the pen technique became the standard way to change facial expressions without telling people what to do. But for years nobody ran an exact repeat of the same experiment; later studies usually used other measures and other setups.
2016: 17 labs, 0.03
In 2016, led by Eric-Jan Wagenmakers, 17 labs repeated Strack’s first study as part of a Registered Replication Report. The protocol was written and approved in advance; cartoons were pre-tested and moderately funny ones were chosen; participants were video-recorded to check that they held the pen correctly. People who guessed the purpose of the study or said they did not understand the cartoons were excluded from the analysis.
When the data of 1,894 participants were combined, the difference was 0.03 points, with a 95% confidence interval from −0.11 to 0.16. The original 0.82 lay far outside that range. The result became one of the symbols of psychology’s replication debate.
The camera: just a detail?
In a commentary published alongside the replication, Strack pointed to two differences: in the replications participants were filmed with a video camera, and the cartoons used might have seemed dated or unfamiliar to participants. Knowing that you are being watched could push people to rely less on their own inner feelings.
Noah, Schul and Mayo (2018) tested this idea directly, running the study once with a camera and once without. Without a camera the facial feedback effect appeared; with a camera it disappeared. This is a single study and does not settle the matter, but it is a nice example of how a small difference in method can change a result.
Many Smiles: a big attempt at agreement
In 2019 Coles, Larsen and Lench published a meta-analysis pooling 138 studies: facial feedback was real, but its effect was small and varied with conditions. Marsh, Rhoads and Ryan (2019) found an effect close to the original in a classroom demonstration with over 400 students across nine semesters.
Then believers and sceptics joined forces to design the Many Smiles study. Data from 3,878 participants in 19 countries showed that mimicking a smiling face and smiling voluntarily could both amplify and initiate happiness. For the pen task, the evidence was less conclusive. So the idea lives on; the famous pen method just may not be the most reliable way to show it.
Our crowd’s share
In this experiment participants are at home and nobody watches them, which in this respect resembles the original more than the camera-monitored replications. On the other hand, without supervision we only know from your report whether you held the pen correctly, and the page title gives away one of the holds. That is why, as in the replication study, we separate people who correctly guessed the purpose and those who failed the check questions from the main comparison.
Our cartoons were drawn for balabs; the ratings cannot be compared directly with ratings of the original cartoons, but the difference between the two groups can. On the results screen we show this difference on the same chart as 1988’s 0.82 and 2016’s 0.03, with its confidence interval. Each new participant adds a small data point to this story.