How to get the P-trend from proc genmod procedure

[critiasun]

Good evening! Everyone! I am working on  Generalized Linear Model，Variable X2 has 4 value(1,2,3,4), now I can see a trend from Variable X2 , but I can not find the default parameter  to get the P-trend, How to do it?

[StatDave]

If you want to test for trend over the levels of X2, that implies that its levels are ordered. If the values 1,2,3,4 reflect the spacing between its levels, then specify X2 in the MODEL statement but not in the CLASS statement. The single p-value that you get for X2 will be a test of its linear effect on the response function.

[critiasun]

I had gotten rid of X2 from class statement, the results showed as follow:

what you mean is this results(ChiSquare=120.96, P<0.0001) is the  linear effect(P-trend) ?

[StatDave]

No, those are results from CONTRAST or ESTIMATE statements. I am referring to the test of X2 in the Parameter Estimates table.

[critiasun]

Could you show me how to write the statement， can you give me an example?

[StatDave]

proc logistic;
model y(event="1") = x2;
run;

The test of X2 is a test of its linear effect on the logit (log odds). You could have additional variables in the model and that is still true.

[critiasun]

If I use the 'proc genmod' procedure, can it be like this:

proc genmod;
model y("event=1")=X2;
run;

[StatDave]

If you are fitting a logistic model, the best tool is PROC LOGISTIC. PROC GENMOD can also be used, but since GENMOD fits a broader class of models than just logistic models, you need to specify the DIST=BINOMIAL option to tell GENMOD to fit a logistic model:

model y(event="1") = x2 / dist=binomial;

Note that this assumes that the response variable, Y, takes on values 0 or 1 and that Y=1 is considered the event of interest.

[critiasun]

I fitted GENMOD model as follow:
proc genmod desc data=tot;
model GDM= X2/ dist = binomial;
run;

the results as follow:

Is this the linear results like trend?

[StatDave · Accepted solution]

Yes. The large Wald chi-square for X2, 355.62, is significant (p<0.0001) indicating a non-zero trend on the log odd of the event level which is the first level of GSM as shown in the Response Profile table. 

[critiasun]

Question has been solved with your help! Thank you very much !