LIFETEST TEST with multiple variables, strange results

[stuart_allen]

I'm fairly new to SAS, using LIFETEST with the TEST statement, looking at association between covariates and survival time. I've noticed that if I run PROC LIFETEST several times, testing only a single variable each time (say 'age', then 'sex', then 'weight') I get a different set of results than if I use multiple variables within a single PROC LIFETEST (TEST age sex weight;). I am talking about results for only "Univariate Chi-Squares for the Log-Rank Test".

The results are similar, but not the same - I was under the impression that by using multiple variables in the TEST statement it would simply run a univariate test for each variable in turn, but the differing results suggest otherwise.

Any advice would be greatly appreciated!
Thanks.

[Doc_Duke]

the first thing to look at are missing values. If you have any missing values for ANY variable in the TEST statement, that observation will be deleted and all the results change a little.

[stuart_allen]

Ah, so observations are deleted where ANY of the variables are missing, and then the results computed.

I suspect that is probably the answer! Thanks!

[Doc_Duke]

Yup. It is in the reference manual for LIFETEST under Missing data.

It is the normal behavior for most stat procedures in SAS (as well as other stat packages). Pairwise deletion is the exception (as in PROC CORR).

Doc

[stuart_allen]

I see it now:

"If any variable specified in the TEST statement has a missing value, that observation is not used in the calculation of the rank statistics."

Just need to know where to look! Very helpful, thanks again.