Hi,
I'm analyzing a panel dataset from an RCT using difference-in-difference estimation, with around 160 samples and 4480 observations (daily data for 4 weeks).
The dependent variable, calculated based on consumption data, is positive (no 0s) with a staircase-like structure: I transformed the consumption data(a set of 0-1 variables) with a set of coefficients to get the dependent variable.
My questions are, what kind of model should I use to conduct DiD estimation? Is Log(Y) a good choice? Or should I use Poisson or negative binomial? Should I choose robust standard errors or cluster at individual level? I tried both, but they produced different results. Are these models compatible with DiD estimation so that I could interpret the estimation results as intended?
Thanks in advance!
I'm analyzing a panel dataset from an RCT using difference-in-difference estimation, with around 160 samples and 4480 observations (daily data for 4 weeks).
The dependent variable, calculated based on consumption data, is positive (no 0s) with a staircase-like structure: I transformed the consumption data(a set of 0-1 variables) with a set of coefficients to get the dependent variable.
My questions are, what kind of model should I use to conduct DiD estimation? Is Log(Y) a good choice? Or should I use Poisson or negative binomial? Should I choose robust standard errors or cluster at individual level? I tried both, but they produced different results. Are these models compatible with DiD estimation so that I could interpret the estimation results as intended?
Thanks in advance!
