Talk:Frequentist inference
This level-5 vital article is rated Start-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | |||||||||||
|
Constraining type 1 error
editThe second bullet in the second section has the line containing "either of these conclusions has a given probability of being correct," which is usually not true. Frequentist methods typically prove a bound on type 1 or 2 error. We can say things like "using this testing procedure, the probability of accidentally rejecting the null hypothesis is at most 5%." Very very few methods provide exact error probabilities, and we can only control one type of error in normal situations, so constraining type 1 error probability does not give us a probability that our conclusion would be correct. — Preceding unsigned comment added by 152.3.43.48 (talk) 18:44, 16 February 2020 (UTC)
Definition of Pivot
editThe paragraph about "pivots" and confidence intervals makes no sense: "A pivot is a probablity such that for a pivot, p, which is a function...". I think it should be deleted or significantly expanded.
I agree with the previous anonymous opinion. Here are some more specifics:
- is called a probability (presumably a number), a function, and occurs in . The latter is never defined.
- is introduced with no explanation of how it relates to the immediately preceding or the inferential problem we are trying to solve. It is simply "a random vector".
- " is a range of outcomes": it is not; it is a number, as is , which is called a "two-sided limit".
- "This rigorously defines the confidence interval": it doesn't define it at all.
- Earlier an expression " is a upper limit for " was introduced, which seems to be taking the thing we are trying to define, the confidence limit, as part of its own definition.
Why so much about Bayesian inference?
editThis article seems to have as much about Bayesian inference as it does about frequentist inference, as if the only way to explain frequentist inference is in contrast to Bayesian inference. I'm not sure if that's the case, but it seems like it wouldn't be. Furthermore, the article currently links to another very in-depth article on Bayesian inference as well as several different pages dealing with interpretations of probability. It doesn't seem like Bayesian inference really needs to be referenced here as anything more than a side note. Joshisanonymous (talk) 17:10, 10 June 2018 (UTC)
New sections
editHello, I was interested in what new sections should be added to this. I am interested in a section on asymptotic theory. I also believe that an example of frequentist inferences' prediction on model selection, similar to the example in the likelihood principle's page. I am curious to see about feedback on the construction of confidence intervals from a pivot perspective. I am unsure as to what other kinds of rigorous definitions to add, but model selection is definitely useful. Also, could anyone add anything about software to use? Probably just Stata, R, but not sure after that. I don't like writing articles that lack a tangible component to them. Adamopoulos (talk) 16:49, 14 September 2021 (UTC)