Realistically it isn’t really possible to apply bonferroni type correction to most data mining. Too many things are looked at, many not consciously. And models are far worse. Pre-registration is only solution, coupled w cheaper publishing to allow publishing of post hoc findings.
A surprisingly large number of these frustrating claims would go away if people would just correct their alpha level for multiple comparisons as taught in intro stats. A handful of statisticians argue against these corrections. Their papers are cited 1000s of times. Go figure ;)https://twitter.com/deevybee/status/1222772370362044416 …
-
-
-
You should also correct for multiple comparisons in your preregistration.
Kraj razgovora
Novi razgovor -
-
-
I worry that researchers engage in self-deception when deciding how many comparisons are in play. The practice of articulating things in pre-registration awakens them to this.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Here’s a ?. Not about if we should, but where? People classically taught about groups and such. What about when you run a regression with lots of vars, especially with dummy codes? MC no? What about corr matrices? MC or do permutation/resampling test?
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.