a********a 发帖数: 346 | 1 %macro surv(dist=);
proc lifereg data=surv_w2;
class PERFSTAT;
model log_day*censor(1)=PERFSTAT base_ts_center PTRWK8 /dist=&dist;
run;
%mend;
%surv(dist=lnormal);
%surv(dist=exponential);
%surv(dist=weibull);
%surv(dist=gamma);
I am running proc lifereg with different distribution for the model. I got
Log likelihood for different distributions as following. Which distribution
I should use? How exponential distribution can be this worse with only 1
degree of freedom difference with weibu | j*****e 发帖数: 182 | 2 You can only compare weibull vs exponential, since the latter is nested in
the former. Exponential assumes constant hazard rate. Use nonparametric
method to estimate it, and see if this is the case.
For the rest, the likehood functions are different and they are not
comparable. |
|