Unbalanced groups in nonparametric survival tests

Abstract

It is fairly common to find medical examples with survival data with unequal sample size among the groups. There are several tests for those cases, but in practice, the use of one test instead of another is done without justifying the election. Sometimes, the choice of one test or another can lead to different conclusions, so it is important to have some guidelines to help to choose the suitable test in unbalanced groups. The computation of the tests is done with the statistical software (BMDP, SAS, SPSS, Stata, Statgraphics, and S-Plus). However the commercial software only covers tests for the family of the weigthed tests, none of the score tests, and the nomenclature is not unified, using different names for the same test. We perform several simulations to give some pieces of advice for picking out the right test. Due to the fact that there are situations where it is advisable to use a test from the family of the score tests against a weighted one, we have developed a new software in JavaScript for Internet that computes score and weighted tests versions (10 tests) that unifies the nomenclature (this software is available from the authors upon request). We include real examples where we apply, using the new JavaScript programs, the recommendations suggested by the simulations.