A two thousand and ten experiment reported that people who held expansive postures felt more powerful, took more risk, and showed higher testosterone and lower cortisol. The TED talk built on it argued that such poses can change the outcomes of your life. Dozens of published studies followed, all showing effects. Simmons and Simonsohn asked whether that record means anything.
Less than the count suggests. They applied to the thirty-three studies the original authors had cited as evidence, and the resulting curve came out flattish rather than . Their reading: either power posing has no effect overall, or the effect is too small for the existing samples to have meaningfully studied it. A separate curve on the measures leaned the right way without reaching significance.
The original experiment is specific about what it found. The authors reported that participants who held power poses felt more powerful, sought more risk, and showed higher testosterone and lower cortisol. The popular version went further, and the talk built on the work told audiences that power poses can significantly change the outcomes of your life.
By the time the argument got going there were thirty-three published studies, and all of them showed an effect. Simmons and Simonsohn put the central point first. Because studies may reach print only when they show an effect, the fact that all the published evidence shows one tells you nothing. What you need is a test of the distribution of the significant p values themselves.
That is what p-curve does. It asks whether can be ruled out as the sole explanation for a set of findings below the usual threshold. A true effect produces a right-skewed curve, and no effect produces a flat one. Run over the thirty-three studies the original authors had cited in their reply, the curve came out flattish, definitely not right-skewed. A sensitivity check that swapped in manipulation checks where studies reported both left it flat.
The conclusion is careful about what it rules out. Either power posing overall has no effect, or the effect is too small for the existing samples to have meaningfully studied it. One or two of the thirty-three may yet be real, since more than that would have shown up as skew. The evidence for the basic effect, they write, is too fragile to go looking for moderators or to advise people to pose for the good of their lives. The hormone claim rests on the original study and one replication, and Amy Cuddy, quoted in the same post, respectfully disagrees with the interpretation throughout.
wat. Thirty-three studies, every one positive, and a curve that still looks like nothing is there.
wat wat
Every one of those thirty-three studies found an effect. That is what a literature looks like when only the successful runs get written up, so the count on its own proves nothing.
How do we know?
A p-curve analysis of the thirty-three studies the original authors had cited in their own reply, published by Joe Simmons and Uri Simonsohn on Data Colada, with the disclosure table the method requires.
What is still unknown
A blog analysis, not a peer-reviewed paper, and the thirty-three studies are the ones the original authors chose to cite rather than a systematic search. The authors allow that one or two effects may be real, and Amy Cuddy disagrees with their conclusions in the post itself.
Learn something?
Where to next
Does the full moon really keep people awake?replication
What happened when the Dunning-Kruger effect was run on random data?replication
What was left of the marshmallow test when they ran it on ten times as many children?replication