Talk:P-rep

Thanks for the image of the p-rep function, but some comments would be helpful. I believe the p-value is shown on the x-axis and the p-rep on the y-axis. Now for a p-value of close to zero, the plot gives a p-rep of close to one. Any comments on the meaning of this? 194.120.84.9 (talk) 11:36, 16 March 2009 (UTC)[reply]

"... a p-value indicates the probability of obtaining a result by chance alone, ..." incorrect statement - see the article on p-value

also: it would be best to delete the "approximation" part again: if there exists a mapping from p-rep to p-value, then p-rep would have the same shortcomings as p, because it would carry exactly the same information. over-seeing the linked article i believe this "approximation" is obtained for a certain preassumed family of probability distributions on the sample, probably the distribution the "inventors" of p-rep believe to be the most common. But then again for this family of distributions p and p-rep are as similar as the approximation is good.

I'm not saying p-rep is scientifically useless, but very sometimes it seems that proponents for p-rep prefer it over p just because they don't understand the concept p. It appears to me as if there are assumptions necessery for calculation of p-rep, that are inherently arbitrary. The classical p-value does not need any such assumptions, which makes it more meaningful in an abstract sense. Obviously the 0.05 bound for p is ridiculously arbitrary, though. —Preceding unsigned comment added by 91.19.105.203 (talk • contribs) 14:18, 19 September 2007

Clarification question: Can someone please add an example to the article to give a more plastic understanding of the p-rep? If I use the given formula to calculate the p-rep for a p-value of 0.05, I get p-rep = 0.88. Lower p-values give higher p-rep-values. What is the meaning of this? Do I have (at a p-value of 0.05) a probability of 88 % to reproduce the obtained result? 194.120.84.9 (talk) 15:07, 31 July 2008 (UTC)[reply]

I think it is supposed to be the probability of reproducing a result at most as extreme as the result obtained. Unfortunately, though, it is impossible to calculate replicability without knowing the prior distribution of what you are trying to measure. The p-rep values seem to me to be a joke. —Preceding unsigned comment added by Ummonk (talk • contribs) 22:43, 23 April 2009 (UTC)[reply]