Talk:Canonical analysis

Hi, this article is not that clear. Is this medthod the same as Linear discriminant analysis? This link seems to say so.

Tony Bruguier 03:51, 17 April 2006 (UTC)[reply]

Hi Tony, the algorithm is similar, however, in the DA the data matrix is transposed, so one analyzes entities instead of attributes. Also, Wilkinson described the rotation for the DA. David Cruise 15:38, 17 April 2006 (UTC)[reply]

I have just tagged this article as 'confusing' and in need of expert attention. A few of my concerns:

"..solving the characteristic equation for its latent roots.." - no explanation of what characteristic equation.
"..formal structures in hyperspace.." - exactly which formal structures are we dealing with?
"..leaves many optimizing properties preserved.." - which optimizing properties?

These are just a sample, and I believe the whole article is confusing. Madmath789 12:53, 15 July 2006 (UTC)[reply]

Hi Mark,
Thank you for your support by restoring my articles, but here are the reasons why I try to dissociate myself from Wikipedia. For instance, if you'll ever run the SAS program for Canonical Analysis, you'll be using my theoretical reasoning and my algorithm, published in Psychometrika, one of the most prestigious journals in the discipline, founded by Karl Pearson of the Pearson product moment correlation fame. However, the self-appointed crirics of the Wikipedia kangaroo courts doubt the legitimacy of Psychometrika, and express their opinions in a fashion such as that matrix subtraction is "probably" mathematically correct, but ... blah blah. With people who use this type of arguments you do not talk with, you just leave their presence. As Oscar Wilde observed, the opposite of love is not hate, but indifference. David Cruise 18:13, 24 July 2006 (UTC)[reply]

Ahem. I asked about the legitimacy of Psychometrika. I doubted (and still doubt, to some extent) the legitimacy of David Cruise writing articles here which refer only to papers by "Krus, DJ". — Arthur Rubin | (talk) 20:26, 26 July 2006 (UTC)[reply]

Agreed! I have spent quite a lot of time looking at his articles and following the links to his Visualstatistics site, as well as reading his contributions to Homogeneity/Canonical Analysis/Matrix addition/matrix subtraction/Supermatrices etc. etc. and find that his 'work' seems to be at odds with all the standard texts and authors. The difficulty is to decide what to do with his contributions. Canonical analysis could probably be re-written to make sense, and to be less like original research. Much harder to deal with is Homogeneity (statistics). After looking at it pretty closely, I would say that if I were to try to improve it, I would first have to remove the diagrams (meaningless without an explanation that I think only David Cruise could provide) - then remove the 'truth tables' (variables and so forth not properly defined or explained) - but then there is virtually nothing left! Madmath789 20:40, 26 July 2006 (UTC)[reply]

FYI, there's also this article Canonical correlation. I don't know if it's related. Tony Bruguier

Still confused

The article does not describe the method well. The references are unfortunately useless, as they require journal subscription. The maximization at the end of the simple description gives the only hint of what is going on, but even there it is not clear if XU is features x features size matrix (probably, yes)? Some "see also" links, or some relevant reference available online would be helpful. Tomash (talk) 18:00, 6 November 2013 (UTC)[reply]