tag:blogger.com,1999:blog-987850932434001559.post3681635628123887523..comments2024-04-16T13:17:49.865+02:00Comments on The 20% Statistician: ROPE and Equivalence Testing: Practically Equivalent?Daniel Lakenshttp://www.blogger.com/profile/18143834258497875354noreply@blogger.comBlogger13125tag:blogger.com,1999:blog-987850932434001559.post-9077211098504105672017-09-29T09:22:29.161+02:002017-09-29T09:22:29.161+02:00And there we agree 100% ! :)And there we agree 100% ! :)Rasmus Bååthhttps://www.blogger.com/profile/16575386339856902265noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-41781932284738449002017-09-29T09:21:32.312+02:002017-09-29T09:21:32.312+02:00Ok - but I was only saying you don't need p-va...Ok - but I was only saying you don't need p-values. Daniel Lakenshttps://www.blogger.com/profile/18143834258497875354noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-65648412453288551312017-09-28T20:58:44.256+02:002017-09-28T20:58:44.256+02:00So I can easily come up with statistical models wh...So I can easily come up with statistical models where it's kind of tricky to come up with a CI, but the Bayesian credible interval is easy to get. An example of such a model would be the statistical model behind BEST.Rasmus Bååthhttps://www.blogger.com/profile/16575386339856902265noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-83287389104908368942017-09-28T20:22:25.545+02:002017-09-28T20:22:25.545+02:00Rasmus, no need to calculate p-values - just the 9...Rasmus, no need to calculate p-values - just the 90% CI around whatever estimate you have. Just as flexible as Bayesian approaches. Daniel Lakenshttps://www.blogger.com/profile/18143834258497875354noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-73946100180359827062017-02-20T12:49:03.157+01:002017-02-20T12:49:03.157+01:00Really nice blog shared. Keep sharing more updates...Really nice blog shared. Keep sharing more updates with us.<br />plots near tcs indorehttp://www.blfbhumi.com/about-us/noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-75588939157686237212017-02-16T22:25:37.173+01:002017-02-16T22:25:37.173+01:00Thanks for this interesting blog post, Daniel. I&#...Thanks for this interesting blog post, Daniel. I've created a follow-up that shows cases in which TOST+NHST yield conflicting decisions, which can never happen with the HDI+ROPE procedure. It's here: http://doingbayesiandataanalysis.blogspot.com/2017/02/equivalence-testing-two-one-sided-test.htmlJohn K. Kruschkehttps://www.blogger.com/profile/17323153789716653784noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-24363030584744124132017-02-16T22:24:49.379+01:002017-02-16T22:24:49.379+01:00This comment has been removed by the author.John K. Kruschkehttps://www.blogger.com/profile/17323153789716653784noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-30531319902663292017-02-14T10:00:57.852+01:002017-02-14T10:00:57.852+01:00Hi, that looks like it will be much easier to use ...Hi, that looks like it will be much easier to use in the future! Excellent!Daniel Lakenshttps://www.blogger.com/profile/18143834258497875354noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-898102092936956462017-02-14T02:58:13.756+01:002017-02-14T02:58:13.756+01:00agreed...
It is important to emphasize that in o...agreed... <br /><br />It is important to emphasize that in one instance you have a measurement of belief and the other you can only make a yes or no decision that may or may not update your belief but in the end provides no measurement of that belief. That is not a trivial pedantic distinction to ignore. People end up coming away thinking that the frequentist method provides a measure that it does not.Unknownhttps://www.blogger.com/profile/00227235335343168838noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-38609932056081186822017-02-14T02:40:20.953+01:002017-02-14T02:40:20.953+01:00Hi Daniel, Glad you find BEST useful. If you just ...Hi Daniel, Glad you find BEST useful. If you just want an HDI, use HDInterval::hdi; same as BEST::hdi but faster for large objects and you don't need to install JAGS. The next version of BEST will 'Depend' on HDInterval.Anonymoushttps://www.blogger.com/profile/09509778781847126034noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-22663138064826088582017-02-13T16:56:34.792+01:002017-02-13T16:56:34.792+01:00Hi Heather - you can use it for equivalence with a...Hi Heather - you can use it for equivalence with a non-zero range as well. For example, using the TOST for one-sample, you could test whether a score is equivalent to guessing average (e.g., 0.5). <br /><br />I'm not sure this is addressed in the Coursera course - but I will be updating the equivalence assignment in the future now my own paper on this is out, and will add a non-zero example!Daniel Lakenshttps://www.blogger.com/profile/18143834258497875354noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-7961250058665800192017-02-13T14:11:31.206+01:002017-02-13T14:11:31.206+01:00Nice post, Daniel. Thank you. Does TOST or ROPE re...Nice post, Daniel. Thank you. Does TOST or ROPE require including 0 in the interval, or can one or both be used to examine equivalence within nonzero ranges? (It's possible your Coursera course addressed this for TOST. If so, color me embarrassed that I can't remember.)Anonymoushttps://www.blogger.com/profile/13691556381219306479noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-78097488060212515752017-02-12T22:02:51.652+01:002017-02-12T22:02:51.652+01:00So I would argue that a region of practical equiva...So I would argue that a region of practical equivalence (ROPE) is both computationally and conceptually very different from a equivalence testing. <br /><br />A ROPE is a very simple concept, it's still a good concept, but it's also very simple. It's a range of differences between any two parameters where, if the "underlying" difference falls in that range, it isn't large enough to be of interest. This is a very general concept not tied to a specific model or specific parameters. You could use it for differences in means, scale parameters, or any other exotic parameters. You could use it for simple group models, or for more advanced models where it's not even clear how you would calculate a p-value. Even if it was originally introduced together with BEST you can use it with *any* Bayesian model, and once you have fitted a Bayesian model it's straight forward to calculate how much probability is in or out of the ROPE (or use an HDI if you want to).<br /><br />Equivalence testing is something different, it's a procedure that requires you to use a model and a parameter where you can calculate p-values. Using a ROPE can be seen as a way of summarizing a posterior distribution, while equivalence testing relies on p-values. And I would say that there is a big conceptual difference between posterior probabilities and p-values even if they, in a few select cases, are numerically similar.Rasmus Bååthhttps://www.blogger.com/profile/16575386339856902265noreply@blogger.com