jjjj6666 发表于 2014-2-2 11:25 
read this article,
http://www.ats.ucla.edu/stat/mult_pkg/faq/general/complete_separation_logit_mo ...
I do not want to alter my original data, now i can only exclude the year dummy control and cluster(year), the year span in my sample is quite wide, almost 25 years. In some years, there're only 3 or 4 observations, that's the problem here. But if i still want to include the year dummy, any suggestions?