Re: Q: dichotomous variables

On Apr 12, 10:51 am, Erkki.Komulai...@xxxxxxxxxxxxxxxxxxx wrote:
I think I found something. Like with df=1 Chi-square = N * r(phi) I
think (based on figures I calculated) that in k*2 tables Chi-square = N
* R (i.e. N * the multiple correlation you get using dummy coding in
k-categories and multiple correlating to the dicothomous variable?

Agree?

For an R x C contingency table,
Cramer's V = sqrt(chisquare/(N*min(R-1,C-1)))
= the root-mean-square canonical correlation
between the dummy-coded rows and dummy-coded columns.
If R = 2 or C = 2 then V = the multiple correlation
between a dichotomous d.v. and dummy-coded predictors.
And if R = C = 2 then V = |phi = the ordinary Pearson r|.
.