New draft: "Decline effects, statistical artifacts, and a meta-analytic paradox". In this manuscript I show how a common practice in meta-analysis (eg the 2015 Open Science Collaboration) creates artifactual signatures of poor scientific behavior. PDF: raw.githubusercontent.com/richarddmore... 1/x
My front yard :)
Simonsohn has now posted a blog response to our recent paper about the poor statistical properties of the P curve. @clintin.bsky.social and I are finishing up a less-technical paper that will serve as a response. But I wanted to address a meta-issue *around* this that may clarify some things. 1/x
I love this post about science and metascience. A lot of quotables but I’ll lead with this:
“Those seeking a scientific method – one that can be written down and followed mechanically […] – betray a kind of childish impatience with a process they clearly don’t understand.”
AI slop and the destruction of knowledge irisvanrooijcogsci.com/2025/08/12/a...
1. "'Trusting the experts is not a feature of either a science or democracy," Kennedy said."
It's literally a vital feature of both science and of representative democracy.
I've written a fair bit about trust in expertise as a vital mechanism in the collective epistemology of science.
We proved they could not scale link.springer.com/article/10.1...
Definitely something worth digging into. I’ll give it some thought
here was our call for methodological standards in metaresearch four years ago. instead of getting fixated on a particular inference we'd like to make, we need to maintain scientific standards, do the hard work, respect the evidence. we can't keep jumping at self-serving solutions without question.
some of the discussion around the p-curve paper is depressing. i still see many missing the point clearly stated in the conclusion, and instead of demanding strong standards or questioning whether they're even asking good questions, they've immediately started asking for replacement methods.
I would add that while pcurve is comprised of tests, that indeed correspond to error rates, the actual hypotheses being tested have little to do with “evidential value”
Test statistics being used are just simple sums, no 3rd moment information enters the test.
Would this be grounds for dismissing the remaining 54 studies as lacking value? It makes no sense. Part of the problem is that the original p curve papers aren’t clear on what exactly is being tested. The authors claim they are tests of skew, but this is incorrect as the
Happy to clarify. Pcurve is used to test whether a set of studies have (or lack) “evidential value” (which is not really defined). But the actual hypotheses being tested by pcurve don’t permit this, as Richard and I show. Suppose one study WAS underpowered in a set of 55 studies -
All pcurve tests are just a simple sum of transformed pvalues. There is a fundamental disconnect between the null hypotheses being tested by p-curve and the claims being made.
What this means is that a significant result for either test only allows one to claim that “at least one” study (out of the set) doesn’t have the property being considered. Why does this happen? Because pcurve completely ignores the configuration of the pvalues being considered.
The test for evidential value simply examines whether the effect size is zero for all studies. The test for lack of evidential value tests whether all studies are “underpowered”, I.e., have small non-centrality parameters.
The developers of p-curve claim that p-curve can be used to make claims about the evidential value (or lack thereof) of whole sets of studies. We show that the actual hypotheses being tested do not allow for such strong conclusions.
The basic idea of p-curve rests on the idea that the skew of a set of p-values is informative about whether QRPs are occurring. As we show, the p-curve tests have nothing to do with skew. It is trivial to create left skewed pvalues that p-curve would confidently label as right skewed.
New paper with @richarddmorey.bsky.social now out in JASA, where we critically examine p-curve. Below is Richard’s excellent summary of the many poor statistical properties of p-curve (with link to paper). I wanted to add some conceptual issues that we also tackle in the paper.
New paper by my PhD student @semihaktepe.bsky.social now published 🚀
"Revisiting the effect of discrepant perceptual fluency on truth judgments" tinyurl.com/2eepue5y
--> Two experiments & a meta-analysis indicate that high visual contrast does not lead to higher truth judgments.
I'm so sorry this happened to you. There is no excuse for such bs.
wherein I'm quoted with "Science is essentially effective to the extent that it can remain independent, decentralized, and democratic." (gift link, email required)
🚀Postdoc position @unimarburg.bsky.social in the project:
"Bridging the Gap Between Verbal Psychological Theories & Formal Statistical Modeling with Large Language Models"
(funded by @volkswagenstiftung.de)
📅Start: 01.10.2025 | ⏳4 years
🔗 Apply now: uni-marburg.de/jhbCen
🔄 Thanks for sharing!
NEW paper! 💭🖥️
“Combining Psychology with Artificial Intelligence: What could possibly go wrong?”
— Brief review paper by @olivia.science & myself, highlighting traps to avoid when combining Psych with AI, and why this is so important. Check out our proposed way forward! 🌟💡
osf.io/preprints/ps...
Great pic
If you'll be at APS2025 in DC next week, I'll be talking about the terrible statistical properties of the p curve procedure in the "Current Issues in Meta-Science" session Saturday, May 24th at 3pm. This will likely be my last US conference in a very long time. A brief summary follows. 1/
I am looking forward to expanding the scope of my professorship by combining cognitive and statistical modeling with LLMs😊
There will be two job openings for postdoc positions soon - one starting in September 2025 and another one a year later.
Another reminder that this online event will take place this Friday. I might be alone representing the West Coast (it's at 7am PT) but hope colleagues from other time zones consider stopping by and joining the discussion. My talk will be not-so-subtly titled "Replication Is (Not)".
