Talk:Correspondence analysis

It may be confusing to say that CA differs from PCA in that it "applies to categorical rather than continuous data". The CA is applied to a matrix of numbers, just as PCA is. I think the text should say it "is intended for count data rather than continuous data". RMGunton (talk) 21:24, 4 August 2011 (UTC)

I believe that the preprocessing of CA works specifically under the assumptions of a contingency table (counts, as in CA or disjunctive coding, as in MCA). If applied to continuous data, it should be assumed to be proportions, or in the very least, something that can be analyzed with a chi-squared test. Plain old continuous data, while it can go into a CA, is not always appropriate to analyze with CA. --Dfbeaton (talk) 16:19, 17 February 2013 (UTC)

(1C1)

Is there a reason for using (1C1) for the notation instead of n? n is much more common.Njfzest (talk) 16:15, 18 March 2015 (UTC)

incomprehensible

I have added a "disputed" tag to this article because the edit by user:Dfbeaton makes the article incomprehensible. What does "1C1" mean? Michael Hardy (talk) 18:46, 2 April 2016 (UTC)

Agreed. 1C1 is meaningless. The 1 cannot mean 1, or it could not be a matrix multiplication. The intent is probably vectors of ones, eg 1_c x C x (1_r)^-1, where 1_c is a vector of ones with 1 row and the same number of columns as C, and 1_r is a vector with as many ones as there are rows in C. Philgoetz (talk) 23:27, 25 March 2017 (UTC)

subscripts i,j for general row and column weights?

The notation shows rows ordered $i=1,...,m$. Should the row weights be notated $w_{i,:}$ and $w_{:,j}$ ?