Hello there,
I am working with a panel collected from the Panel Study of Income Dynamics (survey data), and have written several programs that manipulate the variables in the dataset and estimate three different models. I am interested in obtaining bootstrap standard errors of the estimated coefficients. In order to do that, I had thought of writing something similar to the example below, where I draw 50 samples with replacement by using bsample, and then I estimate the coefficients of my model in each of those samples. However, the PSID extract that I am working with also contains sampling weights, and I was wondering whether I should account for these weights in any way at the time of computing my bootstrapped standard errors. If so, is there any way to "tell" the bsample command to use those weights?
I am working with a panel collected from the Panel Study of Income Dynamics (survey data), and have written several programs that manipulate the variables in the dataset and estimate three different models. I am interested in obtaining bootstrap standard errors of the estimated coefficients. In order to do that, I had thought of writing something similar to the example below, where I draw 50 samples with replacement by using bsample, and then I estimate the coefficients of my model in each of those samples. However, the PSID extract that I am working with also contains sampling weights, and I was wondering whether I should account for these weights in any way at the time of computing my bootstrapped standard errors. If so, is there any way to "tell" the bsample command to use those weights?
Code:
webuse nlswork, clear
** Create and set the identifier cluster variables for the bootstrapped
** panels
cap generate newid = .
drop newid
generate newid = idcode
xtset newid year
** Obtain initial estimates and store them in a matrix
regress ln_wage age race collgrad tenure
matrix observe = (_b[age], _b[race], _b[collgrad], _b[tenure], _b[_cons])
** Write a program that samples the data with replacement and returns the
** statistic of interest
capture program drop my_xtboot_2
program define my_xtboot_2, rclass
* Preserve the data before sampling
preserve
* Sample with replacement
bsample, cluster(idcode) idcluster(newid)
xtset, clear
xtset newid year
* Run the regression model and output the statistics of interest
regress ln_wage age race collgrad tenure
return scalar bage = _b[age]
return scalar brace = _b[race]
return scalar bcollgrad = _b[collgrad]
return scalar btenure = _b[tenure]
return scalar bcons = _b[_cons]
* Return the data to the original state, prior to the bootstrapped
* sample
restore
end
** Collect the estimates of the bootstrapped sample
simulate bage = r(bage) brace = r(brace) bcollgrad = r(bcollgrad) ///
btenure = r(btenure) bcons = r(bcons), reps(50) seed(123): my_xtboot_2
** Summarize the results, including the inital estimates and the sample size
bstat, stat(observe) n(1000)
Comment