Jump to content

Talk:Multinomial distribution

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Multinomial test

[edit]

The "Statistical inference" section should include a link to the "Multinomial test" page in Wikipedia - and viceversa. Rigonz (talk) 08:49, 8 September 2023 (UTC)[reply]

Support

[edit]

It is unclear that the support is positive integers. The use of « x_i » instead of « n_i » as in the french version, or « k » as in the binomial distribution makes this further unclear. — Preceding unsigned comment added by Scharleb (talkcontribs) 23:44, 28 December 2023 (UTC)[reply]

I agree that using x_i instead of n_i to denote counts is generally confusing. Furthermore, there is a confusing change in conventions along the article: the sample size originally denoted by n (lowercase), becomes N (uppercase) in the section of asymptotics and concentration theorems, while n denotes there the number of categories (microstates in statistical physics), originally denoted by k which in the section of conditional sampling denotes the number of empirical constraints. I would suggest to use N for the sample size referring to large-N effective theories (large deviation theory in the standard language of statistics), n_i for sampled counts and L for the number of categories p_1,...,p_L OreLouk (talk) 11:20, 2 February 2024 (UTC)[reply]
Concerning the number of independent linear constraints imposed on multinomial sampling, I would use d or D alluding to the dimensionality of a "model" distribution q_1,...,q_L in the dual parametric formulation of I-divergence minimization. OreLouk (talk) 11:24, 2 February 2024 (UTC)[reply]