tag:blogger.com,1999:blog-3240271627873788873.post295816854317629314..comments2024-03-26T06:46:11.752-04:00Comments on Doing Bayesian Data Analysis: Journal bans null hypothesis significance testsJohn K. Kruschkehttp://www.blogger.com/profile/17323153789716653784noreply@blogger.comBlogger15125tag:blogger.com,1999:blog-3240271627873788873.post-37766460642061979692015-10-10T23:45:20.680-04:002015-10-10T23:45:20.680-04:00Anony wrote "LOL" because, you see it...Anony wrote "LOL" because, you see it's him, not me, yet here we don't really disagree.MAYO:ERRORSTAThttps://www.blogger.com/profile/02967648219914411407noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-52999626220465754912015-02-26T15:13:35.197-05:002015-02-26T15:13:35.197-05:00LOLLOLAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-64469844621471937202015-02-26T06:04:32.196-05:002015-02-26T06:04:32.196-05:00a "fascist move"?
mayo is that you? :Da "fascist move"?<br /><br />mayo is that you? :Dmatushttp://simkovic.github.io/noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-42134080381796660822015-02-25T13:27:18.583-05:002015-02-25T13:27:18.583-05:00Well... psychology and social "sciences"...Well... psychology and social "sciences" are yet short of being REAL science. No wonder they can try this sort of fascist move! Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-19837339122624253802015-02-25T12:52:50.658-05:002015-02-25T12:52:50.658-05:00Wow! This is awesome!Wow! This is awesome!Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-70386868103474432522015-02-25T10:00:58.918-05:002015-02-25T10:00:58.918-05:00Standard errors for the effect size?!? Perhaps the...Standard errors for the effect size?!? Perhaps the editors would instead accept percentiles of the sampling distribution? Say the 2.5th and 97.5th percentiles. Whoops, no, that's a confidence interval.<br /><br />But seriously, this is all rather silly. The whole purpose of reporting effect size estimates, with or without standard errors, is to make inferences about the true but unknown effect size! If standard errors are also reported then we have exactly the same information as a Wald confidence interval. Do the editors not expect their readers to use that information to make inferences about the effect size?BioStatMatthttps://www.blogger.com/profile/00021826218804288140noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-8245660351301558022015-02-25T02:42:18.397-05:002015-02-25T02:42:18.397-05:00I don't think that a sense of uncertainty has ...I don't think that a sense of uncertainty has to be lost because on inferential statistics are used (Even if I personally prefer the Bayesian approach). Uncertainty can still be conveyed by standard deviations and standard errors. Moreover, in the applied field effect sizes should often be more relevant than the result of hypothesis test. I can only hope that researchers will report those more than :-).<br />Finally, the editors write that they will be stricter when it comes to the required sample sizes. I think this is actually the best news, because larger samples do more to protect from inaccurate or incorrect inferences than frequentist or Bayesian stats alone can do. So here is hope that they enforce the requirement of larger than nowadays typically seen sample sizes.guidohttps://www.blogger.com/profile/10981583489689200030noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-86495051518234874192015-02-24T17:37:34.342-05:002015-02-24T17:37:34.342-05:00I think that this is a scary precedent. The nomina...I think that this is a scary precedent. The nominal "ban on NHSTP" is really misleading, since the journal is also banning confidence intervals, and discouraging Bayesian intervals! Indeed, inferrential procedures are no longer required. I believe that the journal is making a grave mistake by implementing this policy, because it will be very difficult for their editorial staff and readership to get a sense of uncertainty in the statistics that are presented. The editorial seems to acknowledge this, but suggests that the remedy is larger sample sizes... <br /><br />I wonder if the editorial board simply dislikes statistics, or fails to understand its fundamental purpose.BioStatMatthttp://biostatmatt.comnoreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-30656700490031196262015-02-24T11:23:13.303-05:002015-02-24T11:23:13.303-05:00I'm as fanatical a pure Bayesian as it gets an...I'm as fanatical a pure Bayesian as it gets and I think this is a really really bad idea. People either need to be persuaded to avoid bad statistical methods or they're not persuaded and they keep on doing crap research. Those are the only two alternatives. <br /><br />Trying to win debates by outlawing one side has never worked in science's favor in the long run. Frequentists did this sort of thing to Bayesians decades ago. How did that work out for them and for science?Entnoreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-2388148668676857402015-02-24T11:16:05.522-05:002015-02-24T11:16:05.522-05:00After all, it is Basic and Applied Social Psycholo...After all, it is <i>Basic and Applied Social Psychology</i> ;-)Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-78292179519098880422015-02-24T04:38:09.715-05:002015-02-24T04:38:09.715-05:00I would ban all of hypothesis testing and model co...I would ban all of hypothesis testing and model comparison not just NHST, so I think this is great news. <br /><br />Since the editors do not recommend any alternative method, it will be interesting how will the authors and reviewers handle this inferential limbo. The problem of social psychology is that researchers use unvalidated measures and unvalidated, unstandardized manipulations which relation to research question is only hazy and unclear. Of course, when you don't know what you are measuring, then no amount of statistical sophistry will give you interpretable results. So I don't think this will remove junk research. But it may motivate some good research.matushttp://simkovic.github.io/noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-28416650143644497242015-02-24T02:43:13.107-05:002015-02-24T02:43:13.107-05:00"Unreal!"
Yes and sort of fun! Since th..."Unreal!"<br /><br />Yes and sort of fun! Since they ban NHST and doesn't really embrace Bayes it is as if they sort of discourage the use of statistical models altogether. I would say this is an exiting experiment! I guess one will have to resort to Tukey style exploratory data analysis, or similar...Rasmus Bååthhttps://www.blogger.com/profile/16575386339856902265noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-396744431706472522015-02-23T23:22:54.320-05:002015-02-23T23:22:54.320-05:00I would argue that you *cannot* use diffuse priors...I would argue that you *cannot* use diffuse priors (I can't bring myself to call them Laplacian) for Bayesian hypothesis test, at least not the way I do them.Jeff Roudernoreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-86070995938352760782015-02-23T22:23:35.156-05:002015-02-23T22:23:35.156-05:00Unreal!
For the time being, they don't even r...Unreal!<br /><br />For the time being, they don't even require *any* sort of inferential statistics, and I don't know how I feel about that.Patrickhttps://www.blogger.com/profile/11301659025178029016noreply@blogger.comtag:blogger.com,1999:blog-3240271627873788873.post-52170481392463135522015-02-23T21:02:59.725-05:002015-02-23T21:02:59.725-05:00And, of course, even within the realm of Bayesian ...And, of course, even within the realm of Bayesian hypothesis tests, you don't need to assume a Laplacian prior on the the alternative hypothesis...John K. Kruschkehttps://www.blogger.com/profile/17323153789716653784noreply@blogger.com