The earth is flat (p > 0.05): significance thresholds and the crisis of unreplicable research

View article
RT @vamrhein: The null hypothesis cannot be 'confirmed' nor 'strengthened' using a p-value, because very likely there are many better hypot…
reply repost on X like
RT @vamrhein: The null hypothesis cannot be 'confirmed' nor 'strengthened' using a p-value, because very likely there are many better hypot…
reply repost on X like
@keholsinger @DanielBolnick @StatModeling There are just so many Gelman papers one could cite! We cited some more, for example, here: https://t.co/y6ZeHs4xSm
reply repost on X like
RT @eimeardol: @paulrconnor I like the idea of considering p-values as "graded levels of strength of evidence against the null" (https://t.…
reply repost on X like
1637 days ago
@paulrconnor I like the idea of considering p-values as "graded levels of strength of evidence against the null" (https://t.co/1plLVjXzK1). I've previously used the term "weak evidence" to describe 0.05 <p<0.1, and tried to discuss accordingly.
reply repost on X like
RT @vamrhein: @BenSlaterNeuro Something along the lines is here :-) https://t.co/y6ZeHs4xSm (I actually wrote it for my students). Always w…
reply repost on X like
@BenSlaterNeuro Something along the lines is here :-) https://t.co/y6ZeHs4xSm (I actually wrote it for my students). Always worth a look is https://t.co/sf78ButBku. Both are not psychology of course.
reply repost on X like
@DevoEvoMed @_julien_roux Not sure I understand what you mean – our obsession with null hypothesis significance testing has been criticized for a century and by many, many statisticians. See, e.g., https://t.co/3fZGjQJSUj https://t.co/y6ZeHs4xSm https://t.co/NAlQGWg6I1
reply repost on X like
@MaartenvSmeden @dnunan79 ... it is less likely that all studies will suffer from the same type of bias; consequently, their composite picture may be more informative than the result of a single large trial” https://t.co/MceA0G9J1g
reply repost on X like
@PabloRichly @norabar Pablo, if I may reply in English: statistical significance cannot rule out that a result is a chance event, and a P-value alone says nothing about effect size and clinical importance. See, e.g., this figure: https://t.co/f7WmiHXnM8
reply repost on X like
@badamczewski01 @Scooletz @konradkokosa @SitnikAdam I do not recommend statistical tests for perf measurements 1) It has tons of problems that can easily mislead you if you don't have enough experience. It worth reading https://t.co/C6RqfCVvzU and https://t.co/B6XcLahdBb
reply repost on X like
RT @PasquierDiego: To emphasize how the statistics are fragile: one less symptomatic patient in the Day1 group and the result would have be…
reply repost on X like
To emphasize how the statistics are fragile: one less symptomatic patient in the Day1 group and the result would have been significant according to the arbitrary rules of medical research statistics. One should be careful with p-value tyranny. https://t.co/HaGtJ42Ao5
reply repost on X like
@JeremyAnso @Zulu18360299 @philippefroguel Bêtement interpréter des données de façon dichotomique en fonction du test de Fisher peut justement conduire à ce genre d'abbération. Lecture très instructive et pertinente à mon avis: https://t.co/HaGtJ42Ao5
reply repost on X like
Probably one of my favorite articles of all time. https://t.co/y53y3Uq2Gl
reply repost on X like
2192 days ago
RT @vamrhein: The null hypothesis cannot be 'confirmed' nor 'strengthened' using a p-value, because very likely there are many better hypot…
reply repost on X like
RT @vamrhein: The null hypothesis cannot be 'confirmed' nor 'strengthened' using a p-value, because very likely there are many better hypot…
reply repost on X like
だからなにっていう話だけど、 "The earth is round (p < .05)”（https://t.co/YteeUpDzQp）というP値論争？で有名な論文があるけど、”The earth is flat (p > 0.05)”（https://t.co/N6nskEHlWm）という論文もあるらしい。個人的には"(p ≥ 0.05)”のほうがいいんじゃないかなと思う（どうでもいい）
reply repost on X like
RT @vamrhein: @LinAung26696501 "In a series of studies, it is less likely that all studies will suffer from the same type of bias; conseque…
reply repost on X like
@LinAung26696501 "In a series of studies, it is less likely that all studies will suffer from the same type of bias; consequently, their composite picture may be more informative than the result of a single large trial." Of course, all of those studies should be published. https://t.co/MceA0G9J1g
reply repost on X like
2345 days ago
@hollymathnerd You may find this reference helpful: https://t.co/QZgsyTXJtz
reply repost on X like
2379 days ago
@lobrowR @cassie_olmsted @jmheberling @spiderdayNight @BrandonHoenig section "Uncertainty, probability, and statistical significance" here https://t.co/AasgDdORta and the last bit of this paper here for advice https://t.co/tHHCzVd6UI
reply repost on X like
2409 days ago
RT @biogeochemSCU: Important read for students & experts alike: "...mistrust in nonsignificant results that leads to publication bias is ca…
reply repost on X like
Important read for students & experts alike: "...mistrust in nonsignificant results that leads to publication bias is caused by confusion about interpretation of larger p-values that goes back to historical disputes among the founders of modern statistics" https://t.co/gj1APCJSYD
reply repost on X like
RT @Forensecology: “putting more emphasis on interpreting effect sizes and interval estimates, using non-automated informed judgment” https…
reply repost on X like