xtreg and areg, time-variant weights, instrument variable

[Jason Su]

Hi everyone, in my empirical analysis, I firstly use areg to do the regressions as below:

areg dif_wage dif_output x1 x2 x3 x4 if (year==2006|year == 2004|year == 2002) [aweight=pop], absorb(province_year) cluster(province)

areg dif_wage dif_output x1 x2 x3 x4 i.province_year if (year==2006|year == 2004|year == 2002) [aweight=pop], absorb(city) cluster(province)

where the weighted variable pop is the city's population which changes yearly; dif_frac_PE_output is the main independent variable, x1-x4 are other control variables

The regression results are shown as attachments.

My next step is to use the variable dif_tariff as the instrumental variable for dif_output, but the command areg can't do the instrumental variable regression. It seems that I should do the IV with the command xtreg, but xtreg can't use the time-varied variable population as the regression weight.

So how can I do the instrumental variable with the time-varied population as regression weight in the panel data model?

[Joro Kolev]

Check whether you cannot do what you want to do with the user contributed -reghdfe-.

findit reghdfe

and then follow the instructions to install.

[Jason Su]

Check whether you cannot do what you want to do with the user contributed -reghdfe-.

findit reghdfe

and then follow the instructions to install.

Hey, Joro, thanks for your reply. As I know, reghdfe is used for high-dimension fixed effects and I'm not sure whether it would apply to my case. But I would check the help file of reghdfe as you suggested.

When I check other Q&A, others suggest that xtivreg2 may solve my problem because it allows both IV and the time-varied variable as regression weight. Do you agree with that?

[Joro Kolev]

Yes, of course, Jason. If -xtivreg2- accepts time varying weights, it can do the job.

-reghdfe- can perform IV regression as well.

My intuition is that -xt- commands might not accept time varying weights, mimicking the behaviour of commands that are written by Stata Corp, while "absorb" commands do accept time varying weights.

Hey, Joro, thanks for your reply. As I know, reghdfe is used for high-dimension fixed effects and I'm not sure whether it would apply to my case. But I would check the help file of reghdfe as you suggested.

When I check other Q&A, others suggest that xtivreg2 may solve my problem because it allows both IV and the time-varied variable as regression weight. Do you agree with that?

[Jason Su]

Yes, of course, Jason. If -xtivreg2- accepts time varying weights, it can do the job.

-reghdfe- can perform IV regression as well.

My intuition is that -xt- commands might not accept time varying weights, mimicking the behaviour of commands that are written by Stata Corp, while "absorb" commands do accept time varying weights.

Good, I would check both xtivreg2 and reghdfe. Thanks for your suggestion and I would let you know if I make the progress, Joro.