permute with reghdfe

[juliana pinto]

Dear All,
Please, I am using reghdfe absorbing a(counties years). I want to 'scramble' years, assigning the 'wrong' years for each observation and then run 500 repetitions to calculate the p-value of those repetitions. Then, in another exercise, I want to 'scramble' the counties. I am trying to use permute, but as counties and years are not included as independent variables, it does not work. Does anyone know how I could do this exercise?
Many thanks.

[Felix Bittmann]

I think you could use the user-written package shufflevar (ssc install shufflevar, replace) and then write a wrapper which combines shuffling the target vars and feeding them into reghdfe. Using simulate you can accumulate your results.

[juliana pinto]

Hi Felix, thank you for answering.
I am not sure how I code 'write a wrapper which combines shuffling the target vars and feeding them into reghdfe'. Do you have an idea?
Thank you.

[Felix Bittmann]

Well its not too complicated but I am not familiar with reghdfe and what kind of statistic you are interested in. So here is a generic example using regress. Say, you are interested in the beta coefficient of weight and you are shuffling whether cars are foreign built or not.

clear all
sysuse auto


*Write wrapper*
cap program drop shuffler
program define shuffler, rclass
    shufflevar foreign
    reg mpg weight foreign_shuffled
    return scalar beta_weight = _b[weight]
    drop foreign_shuffled
end


*Test program*
shuffler
return list


*Run simulation*
simulate beta=r(beta_weight), reps(100) seed(123) nodots: shuffler
sum beta, det
hist beta

[juliana pinto]

That's great help! Thank you very much Felix!!