Bernard,
I think that the problem is a bug in the estimation of median survival proportions when there are no survivors in one group (for plaisir=3, all 8 records have arret=1). If you invert plaisir, by the way, (recode 1 to 2, 3 to 1), it may work, but not consistently. If you want the p50, then you can run the analysis for level 3 of plaisir alone:
lifetable arret durall /p50 if plaisir=3
This seems to work all the time.
Jamie
Bernard wrote:
For your interest, I join a file with four variables and 239 lines. "Arret" is outcome variable (1=interest), and "durall" time variable in weeks. Plus two variables "plaisir" and "moment" in two categories (1 and 2). The result of lifetable command with "lifetable arret durall /by=plaisir /p50 /t" is curious (with version 159 and anterior) instead of "7" or "8" for the categorie "plaisir=2" (probability of 0.50). It's OK with variable "moment"