topic Re: Can anyone help me with a contrast statement for interaction term? in Statistical Procedures

Can anyone help me with a contrast statement for interaction term?

mone — Sun, 16 Feb 2020 16:14:35 GMT

Re: Can anyone help me with a contrast statement for interaction term?

ChrisNZ — Sun, 16 Feb 2020 09:02:07 GMT

Welcome to the SAS community Simone.

I can't answer your question, but I have a comment about the format.

I reckon that you should protect your identity a bit better.

I would not publish my email address and phone number like this.

Just my (paranoid) 2 cents of course. 🙂

Also, you should format your code when posting it here. There are 2 icons to do that: running man and {i}.

Re: Can anyone help me with a contrast statement for interaction term?

Ksharp — Sun, 16 Feb 2020 07:49:02 GMT

Better post this kind of question at STAT forum. @StatDave @Rick_SAS could give you a hand.

Re: Can anyone help me with a contrast statement for interaction term?

Kurt_Bremser — Sun, 16 Feb 2020 08:52:16 GMT

I moved the question to the Statistical Procedures community.

Re: Can anyone help me with a contrast statement for interaction term?

Kurt_Bremser — Sun, 16 Feb 2020 08:56:19 GMT

Concur with @ChrisNZ , don't post your email like that, unless you have a strong interest in male enhancement drugs and are looking for a wife from Russia 😉

Re: Can anyone help me with a contrast statement for interaction term?

ChrisNZ — Sun, 16 Feb 2020 09:11:22 GMT

Or your phone number being on all the calling robots' lists.
Don't help the privacy invaders. Resist.

Re: Can anyone help me with a contrast statement for interaction term?

ChrisNZ — Sun, 16 Feb 2020 09:10:44 GMT

I reckon you can remove the last line of your signature Kurt, now that your (good) proposal has been mothballed.

I wonder how many Ballot ideas are closed with status IMPLEMENTED? Are there any?

Re: Can anyone help me with a contrast statement for interaction term?

Rick_SAS — Sun, 16 Feb 2020 11:12:57 GMT

There is a SAS Knowledge Base article about writing contrast statements that involve interactions.

I assume the levels of Migraine are 0 and 1.

What are the levels (and interpretation) of the Age variable?

Re: Can anyone help me with a contrast statement for interaction term?

PaigeMiller — Sun, 16 Feb 2020 11:58:55 GMT

@mone wrote:

Hello,

I have this code, but I do not know what to do where I have question marks. Can anyone give me a hand? t

proc logistic data=cchs1516study;
class migraine(ref='0') age(ref='0') sex(ref='0') race(ref='0') osa(ref='0') income (ref='1')/param=ref;
model htn=migraine age sex race osa income age*migraine sex*migraine/risklimits;
contrast 'migraine vs no migraine in young people' age ???? age*migraine ??? /estimate=both;
contrast 'migraine vs no migraine in menopause people' age ???? age*migraine ??? /estimate=both;
contrast 'migraine vs no migraine in old people' age ???? age*migraine ??? /estimate=both;

contrast 'female vs male in 20-40 group' age ???? sex*migraine ??? /estimate=both;
contrast 'female vs male in 40-60 group' age ???? sex*migraine ??? /estimate=both;
contrast 'female vs male in >60 group' age ???? sex*migraine ??? /estimate=both;
run;

I really appreciate it.

Simone

In my opinion, the whole idea of using a CONTRAST statement is for cases where you are making some unusual linear combination of results, such as -1.3 * level 1 + 3.4 * level2 ... etc., where the coefficients are not +1 or –1 or zero. It is not for the case where you are directly comparing one group to another, all the coefficients are +1 or -1 or zero.

In the case where there are two coefficients that are +1 or -1 and the rest are zero, you can use the LSMEANS command or the SLICE command to obtain the results, and then you don't have the difficulty of figuring out what the proper CONTRAST statement is, because SAS has already done this for you. See the first example at: https://documentation.sas.com/?docsetId=statug&docsetTarget=statug_introcom_sect064.htm&docsetVersion=15.1&locale=en

Re: Can anyone help me with a contrast statement for interaction term?

mone — Sun, 16 Feb 2020 17:35:11 GMT

I spent many hours trying to understand why my contrast statement gives a different OR compared to the proc sort statement. I just cannot figure it out by myself. I would appreciate your help.

Two interaction terms: age (YOUNG/ADULT/OLD) *migraine(Y/N)

sex (M/F) *migraine

Codes:

male=0

female=1

young=0

adult=1

old=2

CONTRAST STATEMENT:

proc logistic descending;
class migraine(ref='0') age(ref='0') sex(ref='0') race(ref='0') osa(ref='0') income (ref='1')/param=ref;
model htn(event="1")=migraine age sex race osa income migraine*age migraine*sex/risklimits;

contrast 'migraine vs no migraine in young' migraine 1 0 migraine*age 0 0 /estimate=both;
contrast 'migraine vs no migraine in menopause' migraine 1 0 migraine*age 1 0 /estimate=both;
contrast 'migraine vs no migraine in old' migraine 1 0 migraine*age 0 0 /estimate=both;

contrast 'migraine vs no migraine in female' migraine 1 0 migraine*sex 1 0 /estimate=both;
contrast 'migraine vs no migraine in male' migraine 1 0 migraine*sex 0 0 /estimate=both;
run;

PROC SORT STATEMENT:

proc logistic descending;
class migraine(ref='0') sex (ref='0') race(ref='0') osa(ref='0') income (ref='1')/param=ref;
model htn=migraine age sex race osa income/risklimits;
by age;
run;

proc logistic descending;
class migraine(ref='0') age(ref'0') race(ref='0') osa(ref='0') income (ref='1')/param=ref;
model htn=migraine age sex race osa income/risklimits;
by sex;
run;

In my contrast statement code, I am able to put the 2 interaction terms together, but I am not able to do the same with proc sort.

I appreciate your advice.

Please, if you know how to solve it, contact me: simone_derzi@hotmail.com