I am using optimal bandwidth selection: mserd using STATA command rdbwselect. I am then providing the bandwidth output manually to rdrobust command. However, this does not work.

Example:

Assume cut-off value is "C", "X" is forcing variable, "T" is treatment, "Y" is my dependent variable. I calculate cut_off_margin as log(X) - log(C) (i.e. distance between forcing variable and cut-off in terms of log difference)

Bandwidth selection command is run using following:

rdbwselect Y cut_off_margin, fuzzy(T) p(1) c(0) bwselect(mserd)

I get following output:

----------------------------------------------------------------------
| BW est. (h) | BW bias (b)
Method | Left of c Right of c | Left of c Right of c
-------------------+------------------------------+-----------------------------
mserd | 0.182 0.182 | 0.574 0.574
--------------------------------------------------------------------------------

I am then running rdrobust command by manually passing the bandwidth. Sample command:

rdrobust Y cut_off_margin, fuzzy(T) p(1) h(-0.182 0.182) c(0) vce(cluster firm_date_cluster)

I get error: not enough observations. I tried with various other dependent variables and get the same error.


Could someone advise? I have followed sample command usage guidelines in command manual.

Alternately, if I remove the manual bandwidth option (i.e., h(-0.182 0.182)) from the rdrobust command, it works fine. This automatically uses mserd bandwidth algorithm.