Association btw. dichotomous and ordinal variables, controlling for dichotomous var.

#1
Hi y'all!

I've got three variables: A (3-category ordinal), B (dichotomous), C (dichotomous). Any ideas on how to find a measure of association between A and B, while controlling/covarying for C?

I'm currently dropping everything into the partial correlation procedure in SPSS, which I'm assuming isn't terribly accurate.

How about using a logistic regression procedure, with B as the DV, C being forced in as a covariate, and A as a predictor variable? (Or should I use a ordinal regression procedure?) Would I use a pseudo-R square as a reportable figure?

I'm using SPSS; please let me know if there is some procedure in there.

Thanks so much!
 

Karabiner

TS Contributor
#2
Re: Association btw. dichotomous and ordinal variables, controlling for dichotomous v

You could do the logistic regression and treat the variable "A"
as categorical predictor variable.

Kind regards

K.
 

Karabiner

TS Contributor
#4
Re: Association btw. dichotomous and ordinal variables, controlling for dichotomous v

You could dichotomize your variable "A" and perform
a Mantel-Haenszel test of association, i.e. analysis of
a 2x2 table (here: AxB) while controlling for a third
variable (here: C).

Kind regards

K.
 
#5
Re: Association btw. dichotomous and ordinal variables, controlling for dichotomous v

Thanks again!

What about with these three variables?
A (3-category ordinal),
B (3-category ordinal),
C (dichotomous).

Any ideas on how to find a measure of association between A and B, while controlling/covarying for C?

The binary logistic procedure no longer works for this, because I can't put in a 3-category variable in the DV... I can only put in the covariate variable, which makes no sense.
 

Karabiner

TS Contributor
#6
Re: Association btw. dichotomous and ordinal variables, controlling for dichotomous v

Ordinal regression analysis with A as dependent variable
and B and C as categorical predictors/covariates. If
SPSS doen't do it automatically for you, then you'll have
to covert variable C into 2 dummy-variables first, since
ordinal variables cannot be predictors in such analyses.

Kind regards

K.