This paper by @philipbstark is a must read for anyone who builds and fits statistical models as it shows how most stat models in several disciplines are insanely disconnected from reality 1/n
There are so many excellent points and examples in the paper so the rest of this thread will be screenshots with some minor annotations, first I focus on general points in the paper and then discuss the model examples which start around tweet #20 2/n
Issue: Quantifauxcation and forcing complex phenomenon into simplistic numbers for the sake of rigor and comparisons 3/n
Issues with many “cost-benefit analyses” 4/n
Can something as complex as uncertainty always be represented by probability? 5/n
Some general remarks about frequency and subjective based interpretations of probability 6/n
General remarks about probability continued 7/n
Are coin tosses really random? 8/n
Can we trust peoples intuitions and perceptions about complex phenomenon and how that translates into components of the model? 9/n
A common issue in statistical modeling and interpretation, confusing rates with probabilities 10/n
What does random even mean? What is unpredictable? 11/n
How inferential outputs assume a random component is being modeled in order to be valid 12/n
The relationship between probabilistic models and science 13/n
Is it true in practice that with more data the prior (a modeling choice) will eventually be swamped? 14/n
If you model something as being random when it is not does the probabilistic interpretation seem sensible? 15/n
What about situations in which you create randomness and account for it? 16/n
Simulations are completely dependent on what you bake into them, they don’t reflect reality itself 17/n
The role of sensitivity analyses and uncertainty quantification 18/n
What if you started calculating inferential statistics for phenomenon that have no stochastic component? 19/n
Some extremely helpful practical examples in seismology: PHSA models that are based on metaphors instead of actual physics 20/n
Helpful example: issues with a model used to explain wind turbine interactions with birds resulting in a type III error 21/n
Helpful example: statistical tests employed in a randomized clinical trial 22/n
Helpful example: modeling whether women are more likely to be interrupted in academic talks and how the chosen model resulted in a type III error 23/n
Helpful example: the issue with all the models used in the many analysts, one dataset study and how this was also a type III error 24/n
A quick thread on how the ASA statement on P-values from 2020 and the Nature Paper has had some "interesting outcomes” on interpreting results 1/8
As @jd_wilko has pointed out many times here, and from what I’ve seen myself, many researchers point to these papers about abandoning traditional NHST to argue the importance of the point estimate from a study and present it as being the true effect 2/8
What’s hilarious is that if anyone read any of those papers carefully, they’ll realize that almost all of them heavily emphasize inspecting and interpreting the interval estimate and all consistent values inside it and near it. 3/8