topic Re: Specifying the number of canonical variates in a MANOVA in Statistical Procedures

Specifying the number of canonical variates in a MANOVA

xtc283x — Tue, 17 Sep 2019 20:04:29 GMT

Not having used the MANOVA/canonical option available from Proc GLM in some time, I'm surprised that my version of SAS only returns a single canonical variate regardless of the number of dependent variables are in the model statement.

Nothing in the documentation suggests that there is an option for requesting or specifying the number of canonical variates.

Is anyone aware of a workaround to this? In other words, how do you get Proc GLM/MANOVA to return more than one canonical variate?

Thank you.

Re: Specifying the number of canonical variates in a MANOVA

PaigeMiller — Tue, 17 Sep 2019 20:27:20 GMT

This example has 5 Y variables, and 5 canonical variables.

https://documentation.sas.com/?cdcId=pgmsascdc&cdcVersion=9.4_3.4&docsetId=statug&docsetTarget=statug_glm_examples06.htm&locale=en

Maybe there's something in your data that limits the results to 1 canonical variable.

Re: Specifying the number of canonical variates in a MANOVA

xtc283x — Tue, 17 Sep 2019 21:11:44 GMT

Thanks for your response. However, I don't see five variates, I see one which comprises 100% of the variance. If the example used the OUTSTAT= option, you would see what this means.

Re: Specifying the number of canonical variates in a MANOVA

PaigeMiller — Tue, 17 Sep 2019 21:59:27 GMT

I see five canonical variables (two of which account for 0% of the variability).

Re: Specifying the number of canonical variates in a MANOVA

PaigeMiller — Wed, 18 Sep 2019 10:43:01 GMT

Adding:

Perhaps this is what you are seeing:

but you have to read what it says. In the comparison of Llanederyn versus other sites, there is effectively only one real canonical vector, in other words comparing Llanederyn to other sites the variation of Y lies effectively in a single dimension.