Ways to find interaction effects of 2 nominal variables on nominal dependant variable

#1
Hello

I am trying to doing analysis of interaction effects of 1-7 dichotomous or polychotomous dependent variables (nutrition type, either 2 or 3 levels) and one dichotomous variable (young or old age) on dichotomous dependent variable (hyptertension, yes or no).

My thought is that as you get older, your nutrition changes as a partial explanation of the increased prevalence of hypertension seen in aging.

One thought I had was to to assess the interaction of the 3 variables using a chi-square value calculated by the log-linear method.

I would use 1 of the 7 physical activity variables at a time in a 2 x 4 contingency table with hypertension and age as defined below

Rows (nutrition variable) (A)
Columns (Hypertension) (B)
Strata (age) (C)

The interaction would be the ABC combination.

Should I only test one nutrition variable at a time with hypertension and age or should I consider them all at once in a 9 way interaction.

One issue I have is I do not understand the difference between a 2x2x2 and 2x4 contingency table.

Should I bother with the log-linear method or can I answer everything I want with logistic regression.

I know i can use logistic regression to get individual effects of each independent variable on probability of having hypertension but I'm confused about how to test the interaction effects. Is it as simple as testing for differences among the beta coefficients?

Any thoughts, I have just become more confused trying to read up on these issues.


Thank you!
 
#2
Re: Ways to find interaction effects of 2 nominal variables on nominal dependant vari

There are many ways you can do this. I suggest using logistic regression simply because it seems a lot simpler than really complex contingency tables (as I think you agree). You can just specify interaction terms among the design variables created from your categorical variable.

I am confused on what your IV and DV are as in your first sentence you only have dependent variables listed. If your IV are all 2 level categorical variables (like gender) its fairly easy to analyze interaction. If there are more than two levels of your categorical variable than analyzing interaction could be a bit tricky.

But the way you do it is fairly straight forward. You just specify interaction terms between the variables you think interact and then see if they are statistically signficant. So if you have two nutrition types with two levels each (say A and B) than you specify A*B as an interaction effect and test that. If you have three two level nutrition variables (A, B, C) you would have three two way interaction terms (A*B A*C B*C) and one three way interactions A*B*C.

Note that as you get beyond 3 way interaction it gets really difficult to explain what they mean :p I actually have never seen such an analysis.
 
#3
Re: Ways to find interaction effects of 2 nominal variables on nominal dependant vari

Thanks for your answer
Sorry for the confusion
2 main IV's (but really 9)
nutrition type, but I have 7 different types of nutrition variables that are either (coded 0,1 or 0,1,2)
and
age (old or young)
1 DV
Hypertension (yes or no).

Logistic regression sounds more and more like the way to go.

Someone suggested using a weighted sum of all the nutrition variables and create a latent variable for nutrition. I was thinking about doing that, however I think it would become coded (0,1,2,3...19) or something like that.