Hello everybody,
Currently, I am doing research on a mandatory disclosure, whereby several firms are mandated to disclose (treatment group) and some firms don't have to disclose (control group).
I have got a control group of 9,975 and a treatment group of 2,755 firm-year observations.
I would like to match my treatment group with the control group, based on similar characteristics (x1 x2 x3 x4 and x5). I have defined the T as the treatment, and the outcome as Y.
I am trying to perform a propensity score matching procedure, with the nearest neighbor matching (eventually with a caliper of 0.25 and with replacement). I want to do firm matching and not weighting.
I have tried to perform the matching with the commands psmatch2 and teffects nnmatch.
(1) The following codes I performed for psmatch2:
sort id
sort id Year
global controlvars "x1 x2 x3 x4 x5"
global outcomes "Y"
logit $treatment $controlvars
predict pscore1, p
gen psm_weights=.
replace psm_weights = 1/pscore1 if treatment==1
replace psm_weights = 1/(1 - pscore1) if treatment==0
psmatch2 $treatment, pscore(pscore1) outcome(Y) caliper(0.25) logit ate
2) I tried the Statistics - Treatment effects - function in STATA itself for the other one.
The following command came out: teffects nnmatch (Y x1 x2 x3 x4 x5) (T) dtolerance(0.25)
Both matchings resulted in either the same amount of companies in the full sample or an equal matching sample of 2,042 firm-year observations, split up in 1,021 treated firms and 1,021 control firms. I am pretty sure this is not completely right..., because that would mean that I would have lost around the 10,000 firm-year observations. I am trying to have the same amount of firm-year observations for the control group as I have for the treatment group (2,755).
I hope someone can help me with this problem!
Currently, I am doing research on a mandatory disclosure, whereby several firms are mandated to disclose (treatment group) and some firms don't have to disclose (control group).
I have got a control group of 9,975 and a treatment group of 2,755 firm-year observations.
I would like to match my treatment group with the control group, based on similar characteristics (x1 x2 x3 x4 and x5). I have defined the T as the treatment, and the outcome as Y.
I am trying to perform a propensity score matching procedure, with the nearest neighbor matching (eventually with a caliper of 0.25 and with replacement). I want to do firm matching and not weighting.
I have tried to perform the matching with the commands psmatch2 and teffects nnmatch.
(1) The following codes I performed for psmatch2:
sort id
sort id Year
global controlvars "x1 x2 x3 x4 x5"
global outcomes "Y"
logit $treatment $controlvars
predict pscore1, p
gen psm_weights=.
replace psm_weights = 1/pscore1 if treatment==1
replace psm_weights = 1/(1 - pscore1) if treatment==0
psmatch2 $treatment, pscore(pscore1) outcome(Y) caliper(0.25) logit ate
2) I tried the Statistics - Treatment effects - function in STATA itself for the other one.
The following command came out: teffects nnmatch (Y x1 x2 x3 x4 x5) (T) dtolerance(0.25)
Both matchings resulted in either the same amount of companies in the full sample or an equal matching sample of 2,042 firm-year observations, split up in 1,021 treated firms and 1,021 control firms. I am pretty sure this is not completely right..., because that would mean that I would have lost around the 10,000 firm-year observations. I am trying to have the same amount of firm-year observations for the control group as I have for the treatment group (2,755).
I hope someone can help me with this problem!
