test statistic for categorical outcome with >2 values

test statistic for categorical outcome with >2 values

by Rena -

Hi all, I am trying to figure out the best test of significance to use for the below scenario.

My predictor variable is the number of living children, which is ordinal and continuous (correct?) but not normally distributed, hence I want to come medians.  My outcome variable is the type of contraceptive method, i.e. IUDs, pills, depo, and condoms, which is categorical but not ordered (i.e. pills is not a higher value than IUD).

From my review of biostats, the best test of significance I came up with was ordered logistic regression, but I worry that my outcome variables' values are not "ordered."

Thanks,

Rena

Re: test statistic for categorical outcome with >2 values

by Valy Fontil -

Hi Rena

You can try a logistic regression with number of living children as a categorical variable: logistic i.contraceptivemethod i.#_of_livingchildren

Re: test statistic for categorical outcome with >2 values

by Rena -

Well, but simple logistic regression presumes that the outcome is binary (0 or 1); this is not the case for me, since my outcome variable is non-ordered categorical and my predictor is discrete numerical.

Any other thoughts? Anyone else?  Thanks, Rena

Re: test statistic for categorical outcome with >2 values

by Mark Pletcher -

I'm sorry, I had replied previously to the wrong post, let me revise my previous answer (which I have now deleted).  For an unadjusted test, you could simply use a chi-squared test (both outcome and predictor are categorical).  Adjustment will be much more difficult with a non-ordered categorical outcome.  There is a logistic regression variant that can be used (not ordinal, but "mlogit"), but I actually don't recommend it...maybe collapse/dichotomize the outcome?  I think that will get you a cleaner question and result...BUT this really deserves more discussion, so just take this as a teaser.