tag:blogger.com,1999:blog-987850932434001559.post3431966567980430580..comments2023-01-26T12:54:16.768+01:00Comments on The 20% Statistician: What is a p-value?Daniel Lakenshttp://www.blogger.com/profile/18143834258497875354noreply@blogger.comBlogger10125tag:blogger.com,1999:blog-987850932434001559.post-11882430214205703352014-09-28T07:20:27.526+02:002014-09-28T07:20:27.526+02:00But if your question was 'is there is a differ...But if your question was 'is there is a difference between the groups in real life?', can't you just get this by comparing means and SDs, and deciding for yourself if the difference is meaningful? Why bother using p values at all? Seems unnecessarily complex to me.<br /><br />Is it even logically correct to reject the null hypothesis just b/c your p value is so low it is unusual? In your example, the p value tells you that the diff between the two groups is statistically significant i.e it is unusual assuming that no difference exists. But you don't know if you are right to make this assumption in the first place, since you don't know anything about either group. So how then can you use the p value to answer a question?<br /><br />I guess p values would make sense if you had reason to believe that there was no difference between groups to begin with. Then you could use p values as a model to fit your data to tell you how close your data is to the null model. But you would still need to know what a suitable sig. level would be. You can't just arbitrarily pick 0.05 because it's convention, right? Why not 0.001? (This link on p values and error rates - http://www.dcscience.net/Sellke-Bayarri-Berger-calibration-of-P-2001.pdf)Daniel D'Mellohttps://www.blogger.com/profile/08715447561669052286noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-24184595874518787202014-09-27T22:41:24.180+02:002014-09-27T22:41:24.180+02:00Hi Daniel, the answer is extremely simple: You sho...Hi Daniel, the answer is extremely simple: You should use them, because they answer the question you are interested in. If you cannot think of a question a p-value answers, don't use them. Please read around on my blog for some answers to your other questions, such as your statement about the null never being true, which I hear a lot *yawn* but I've addressed here: http://daniellakens.blogspot.nl/2014/06/the-null-is-always-false-except-when-it.htmlDaniel Lakenshttps://www.blogger.com/profile/18143834258497875354noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-29670868146999352662014-09-27T22:23:55.353+02:002014-09-27T22:23:55.353+02:00You have used p values here to determine if the di...You have used p values here to determine if the differences in means between groups is statistically significant based on the assumptions of the null hypothesis. But this does not mean that this difference is significant in real life. Nor does it contribute to your hypothesis in any way. I mean, the p value you got tells you that the result is unusual if the null hypothesis is true (zero difference between groups), but since your samples are small and your null is never really going to be exactly zero, this isn't saying much. More importantly, it doesn't tell you anything about your original hypothesis given your data (personally I don't think that means of 7 and 8 are very significantly different). My question is, why use p values at all if they are not informative and do not tell you anything about your hypothesis?Daniel D'Mellohttps://www.blogger.com/profile/08715447561669052286noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-91382805784449345942014-09-27T22:22:10.671+02:002014-09-27T22:22:10.671+02:00This comment has been removed by the author.Daniel D'Mellohttps://www.blogger.com/profile/08715447561669052286noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-84650136067369545882014-08-25T09:00:17.690+02:002014-08-25T09:00:17.690+02:00Hi, thanks for the suggestion. I think visualizati...Hi, thanks for the suggestion. I think visualization is very important, especially when teaching statistics to students. That's why I tried to make some good figures. For now, I think they suffice - learning ggplot would be nice, but I have so much to learn, I don't think it will happen any time soon. But it's on the list!Daniel Lakenshttps://www.blogger.com/profile/18143834258497875354noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-36105226865900135912014-08-23T02:14:38.668+02:002014-08-23T02:14:38.668+02:00Hi Jake, that is a very good point about avoiding ...Hi Jake, that is a very good point about avoiding unnecessary abstractions. The difficulty of learning ggplot2 I believe to be a subjective quality, and the amount (and type) of questions about ggplot2 on StackOverflow could indicate its success and adaptability to a variety of problems and skill levels, and not the contrary as you suggest. <br /><br />You are absolutely right in that as Dr. Lakens is quite obviously making great plots with base graphics already, my call for switching to ggplot2 was not reasonable.<br /><br />I kid, but the fact that R is built with C (or Fortran?) shouldn't convince us to use C instead.Mattinoreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-10181524021946782082014-08-22T11:31:48.743+02:002014-08-22T11:31:48.743+02:00+1+1Anonymoushttps://www.blogger.com/profile/16340502392117259180noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-49571939291985965022014-08-22T01:59:17.543+02:002014-08-22T01:59:17.543+02:00As you know, I write my blog in IPython notebook. ...As you know, I write my blog in IPython notebook. There is a similar tool for R, check it out: http://ramnathv.github.io/rNotebook/<br /><br />Initially it may take some time to figure out how translate notebooks into blogposts but in the long the notebooks give unconstrained flexibility and of course they look good :)matushttp://simkovic.github.io/noreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-11677576515948644352014-08-21T21:55:47.726+02:002014-08-21T21:55:47.726+02:00I think it is a basic principle of good programmin...I think it is a basic principle of good programming that you should avoid needless, limiting abstractions unless they give you a sufficiently large payback in terms of usability. And frankly I don't think ggplot2 has enough bang for the buck to warrant the widespread adoption that it enjoys today. Learning the ins and outs of ggplot2's plotting system is not exactly trivial, and that is all time that could be spent just learning how to use the base graphics system well, which is of course what ggplot2 uses "under the hood." As soon as you need to make some absolutely simple tweaks to your plots that ggplot2 wasn't specifically designed to make easy for you, you're going to have a bad time: take a look at the questions tagged "ggplot2" at StackOverflow if you don't believe me. Daniel, it looks like you can already do some fairly sophisticated things in the base graphics system, so it seems to me that you're better off just staying the course.Jake Westfallhttp://jakewestfall.orgnoreply@blogger.comtag:blogger.com,1999:blog-987850932434001559.post-21477323582612069352014-08-21T19:19:40.382+02:002014-08-21T19:19:40.382+02:00Great post as usual. I'm excited to see you...Great post as usual. I'm excited to see you're moving to R. However, you need to immediately switch over to ggplot2 for plotting. Leave base graphics (the Klingon of graphics) for 1992.<br /><br />http://ggplot2.org/<br />http://docs.ggplot2.org/current/<br /><br />Mattinoreply@blogger.com