zlgsx 发表于 2015-6-17 21:25 
1、乘积项是否合理?有文献支持吗?
2、怀疑与样本量有关。
乘积项是想针对county-year进行双向cluster,我是在EJMR论坛上看到的这个办法:
reg y x z,cluster (state)
usually this is classic for papers on us...
you can also cluster at the state year level, gen yearstate = 50*state + year
then reg y x z, cluster (stateyear)
google thomas lemieux and check his notes on this...
就是说冲击是以county-year为单位的
样本量的确不够大,按照每个cell来分的确很少...