Talk:Overdispersion
 | This article is rated Start-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | ||||||||||
| |||||||||||
I found this page useful, but would it be worth adding something about estimating dispersion parameters and their use in defining quasi-likelihood models? A little more on the causes of overdispersion would also be helpful. It's beyond me to add these things, but I'd certainly find it useful if someone else did! McB50 (talk) 10:00, 19 April 2013 (UTC)
Incompetent[edit]
A statistician should rewrite this. The page has stuff backwards.88.111.239.43 (talk) 13:38, 2 August 2018 (UTC)
No!!![edit]
A statistician should rewrite this by taking the backwards view. Statisticians want to talk about the stuff they know how to do; what we want is to know is what is the best practice based on our use-cases. The problem is that there is only minimal over lap between the two. If I need to make a decision, I can bullshit it or I can find out what the stats say. IMHO the second is the right path. What I need to know is if the stats people have an opinion but when they don't, that is great; I know that I have no chance of finding something definitive and can relax and just use "best practice," whatever that is. I thus found this page extremely useful and as McB50 said, more discussion of overdispersion would be great. So lets Talk. It seems the variance for a coin toss _ought_to_ follow 1/(2^n) where n is the size of the sample. Thus, for a single coin toss, the probablility of all tosses resulting in a head is 0.5. For n=2, an all heads outcome has probability 0.25 and so on. If I confound the coin toss outcome with a second variable -- lets say a second coin --- the "over"dispersion (the variance) is going to be different.. suggesting that ... This article seems to references work that takes this approach? Does that mean we should view a Gaussian distribution is some kind of "pure" version of a binomial distribution?!!! ... we know that a coin toss should follow 1/(2^n) and when it doesn't we need to start looking for confounding factors like a second coin? Now that would be addressing the question I have about how to spot when a sample is not unimodal. — Preceding unsigned comment added by 31.125.39.26 (talk) 18:45, 21 January 2021 (UTC)