The treatment sample size is small. What should I do

[Huyen Nguyen]

I am studying the effect of hydrodams on agriculture. I have 241 districts, but only 6 districts observe the variance in the number of hydrodams. What should I do? Can only 6 districts represent the effect of hydrodam for the whole sample.

Thank you.

[Carlo Lazzaro]

Huyen:
I do not have the rough idea about what hydrodams are, but limiting your analysis to a subsample of 6 districts that experienced a variation in the number of hydrodams means obtaining biased/ unreliable results.
I woud go with whole sample instead and provide an explanation of the results.

[Huyen Nguyen]

Huyen:
I do not have the rough idea about what hydrodams are, but limiting your analysis to a subsample of 6 districts that experienced a variation in the number of hydrodams means obtaining biased/ unreliable results.
I woud go with whole sample instead and provide an explanation of the results.

Thank Carlo.

I would like to make it more clear. I am not going to used only 6 districts for my regression. I am estimating the effect of number of hydro electric power dams on agriculture productivity in district level from 2000- 2010. The number of districts I have is 241 districts. But only 6 districts have hydro-power dams construction. So when I do the regression, the 235 districts will have 0 value (no variation in the number of hydro-power dams). I am not sure my result is consistent or not with only 6 districts having dam construction activity.

Huyen

[Carlo Lazzaro]

Huyen:
thanks for clarifying.
It would seem that you're dealing with a N>T (N=panles; T=time) panel data regression.
In all likelihhod, this is a job for -xtreg-.
However, it seems weird (to me, at least) that your research goal relies upon a sample with such a scant number of hydro electric power dams.
In brief, I do not think that your results will be informativet.

[Amin Sofla]

I totally agree with Carlo's posts. As he suggested, you should go with the whole sample. You might want to replace the 'number of hydroelectric power dams' with a binary variable taking the value 1 if the district has hydroelectric power dam, and 0 otherwise. However, your sample is still very small.