interaction terms in proc logistic

Ashwini_uci — Sat, 29 Oct 2011 01:16:38 GMT

In the program below, I want to check the interaction between gender*race and gender*income.

In the model, I have 4 individual race variables and 3 individual income variables.

I also have a another form of “RACE” and “Income” variables, with 5 and 4 different levels, such as 1= white, 2=black, 3= hispanic, 4 =asian and 5=others and Income as 1= first quartile 2-= second quartile, 3= 3rd quartile and 4= 4th quartile.

I created the below yellow highlighted dichotomous variables from the above nominal variables to use them in the model;

Such as for race;

black , Hispanic, asian, otherraces, and Caucasian is a reference (these are dummy coded variables) from main variable “RACE”

for income;

secquart, thirquart, fourquart, (First quartile is the reference, again dummy coded from main variable “INCOME”

So which form of race and income variables should I use in interaction? I came up with following 2 types of models with interaction terms, please have a look at them and guide me as to which one of these is the correct one?

proc logistic data=library.nismicathcabg1 descending;

model died= age gender black hispanic asian otherraces secquart thirquart fourquart bldloss chf chrnlung coag depress drug hypothy liver lymph gender*black gender*hispanic gender*asian gender*otherraces gender*secquart gender*thirquart gender*fourquart;

title 'Logi Reg in-hosp mortality vs gender in MI patients with race + income';

run;

quit;

This includes the nominal RACE AND INCOME variable in the interaction terms.

proc logistic data=library.nismicathcabg1 descending;

model died= age gender black hispanic asian otherraces secquart thirquart fourquart bldloss chf chrnlung coag depress drug hypothy liver lymph gender*race gender*income;

title 'Logi Reg in-hosp mortality vs gender in MI patients with race + income';

run;

quit;

proc logistic data=library.nismicathcabg1 descending;

class gender race (ref= first) income (ref=first) /param=ref;

model died= age gender race income bldloss chf chrnlung coag depress drug hypothy liver lymph gender*race1 gender*income;

title 'Logi Reg in-hosp mortality vs gender in MI patients with race + income';

run;

quit;

which one of these is right?

Appreciate your suggestions. i have never worked with 'interactions' before.

After I run these models, I get different numbers and the output doesnot show "ODDS estimate for gender, and the variables that are used in interaction terms. I am not sure if i am doing it right, as it is strange not to get the ODDS estimate for only these variables.

interaction terms in proc logistic

Doc_Duke — Sat, 29 Oct 2011 20:35:50 GMT

The last one is the best one. However, you should use race in the interaction rather than race1. The reason that you were getting different results in the other ones is that each test is done after controlling for all the only other variables (including the other pieces of the IA terms).

You need to use the ODDSRATIO statement in LOGISTIC to get the ORs for the interaction variables. Check the documentation for its interpretation.

Doc Muhlbaier

Duke

interaction terms in proc logistic

Ashwini_uci — Tue, 01 Nov 2011 01:40:11 GMT

Thanks much for your response Doc! I still have to look for more details about using the odds ratio statement as I havenot done it before. I will appreciate if you have any idea about how to go about getting the odds ratio.

interaction terms in proc logistic

StatDave — Fri, 04 Nov 2011 18:28:32 GMT

See this usage note:

http://support.sas.com/kb/24455

and Example 3 of this usage note:

http://support.sas.com/kb/24447

topic interaction terms in proc logistic in Statistical Procedures

interaction terms in proc logistic

interaction terms in proc logistic

interaction terms in proc logistic

interaction terms in proc logistic