topic Re: Test for the effect of race adjusting for age, using a likelihood ratio test in SAS Programming

Test for the effect of race adjusting for age, using a likelihood ratio test

mblack-732 — Sat, 04 Apr 2020 16:31:00 GMT

Hi!

I'm struggling a little bit with a SAS Code... I'm trying to test for the effect of race adjusting for age, using a likelihood ratio test via the PROC PHREG method. My Code so far is as follows:

PROC PHREG DATA=HW9;
CLASS age;
MODEL time*status(0) = age;

RUN;

PROC PHREG DATA=HW9;
CLASS age race;
MODEL time*status(0) = age race;
RUN;

How do I get SAS to realize that I only want the difference in the model to be the race variable?

Re: Test for the effect of race adjusting for age, using a likelihood ratio test

PaigeMiller — Sat, 04 Apr 2020 17:48:19 GMT

@mblack-732 wrote:

How do I get SAS to realize that I only want the difference in the model to be the race variable?

What does that mean?

Re: Test for the effect of race adjusting for age, using a likelihood ratio test

mblack-732 — Sat, 04 Apr 2020 17:50:29 GMT

Hi, Paige!

I just want SAS to compare the full model with age and race to the smaller model with just age but it doesn't seem to be doing that (I have the correct answers from R).

Re: Test for the effect of race adjusting for age, using a likelihood ratio test

PaigeMiller — Sat, 04 Apr 2020 17:54:05 GMT

Typically, SAS doesn't compare models for you. You'd have to do the comparison using the output of each model. For ordinary regression, you can use the Extra Sum of Squares principle. I'm not sure what the equivalent is for PROC PHREG.

Re: Test for the effect of race adjusting for age, using a likelihood ratio test

FreelanceReinh — Sun, 05 Apr 2020 15:35:35 GMT

Hi @mblack-732,

Use the TYPE1 option of the MODEL statement to request that likelihood ratio test statistic and p-value:

MODEL time*status(0) = age race / type1;

This will create the additional output table "Likelihood Ratio Statistics for Type 1 Analysis" as shown below (using my random data):

           Likelihood Ratio Statistics for Type 1 Analysis

                                                      LR
Source                    -2 Log L      DF    Chi-Square    Pr > ChiSq

(Without Covariates)     1361.3610       .         .             .
age                      1350.7645       4       10.5965        0.0315
race                     1346.8845       3        3.8801        0.2747

The last row of the table contains the test results you're after.

Re: Test for the effect of race adjusting for age, using a likelihood ratio test

mblack-732 — Sun, 05 Apr 2020 15:37:12 GMT

Thank you so much!