Fuzzydid with large dataset

[Rachel Paya]

I am analysing the effect of a social transfer earmarked for energy consumption on the consumption of electricity and I have data aggregated at the neighbourhood level so that I know the percentage of people receiving the transfer in each neighbourhood and the total electricity consumption of each neighbourhood. I am using the fuzzydid command developed by de Chaisemartin and d'Haultfoeuille but I get an error saying "too many values" "an error occurred when bootstrap executed estim_wrapper". When I use the command on a subsample of my dataset however, it works well. The problem thus seems to come from the number of observations used in the analysis. As I am fairly new to Stata, I have never encountered such an issue and do not know how to go about it. Any help would be much appreciated !

[William Lisowski]

You give us very little to go on. The error message is the sort that Stata gives when a system limit of some sort is exceeded, something that programmers should avoid by checking assumptions before issuing a command, and if the limit is exceeded, issue a more meaningful error message that the system default

The fuzzydid command - a community contributed command available from SSC and from the Stata Journal - does suggest that having a large number of values for the treatment can cause problems, and this is described further in Section 3.3 of the paper. Perhaps this is such a problem in your case - using aggregated data - that the list of possible values exceeds some Stata limit, and the authors did not first check to see if this was going to be a problem, because they did not imagine someone straying that far from the basic assumption of a binary treatment.

So perhaps you need to use the newcatg() option to reduce the number of treatments. But again, this is a guess based on very little information.

Other than that, the names and email addresses of the authors are given at the end of the output of help fuzzydid - perhaps they can further enlighten you.

[Rachel Paya]

Dear William,

Thank you for your reply.

I did use the newcatg() option as I have indeed too many treatment values otherwise. The error I get is not referenced in the ado-file for the fuzzydid command so I thought it might be an error from Stata's internal bootstrap command, which is used by fuzzydid to estimate standard errors. Yet I did not find anything about the bootstrap command not being able to handle many observations in the documentation.

I will follow your advice and contact the authors !

[Xu Ke]

Hi, Rachel

I've also met the same problem. I was only able to run the regression in about 300*200 observations. Have you solved it now?

Thanks!