How to run a regression on matched data?

[Rose Sinclair]

Hello. My apologies as I am not very well acquainted with Stata. I have read through the manual and searched the internet extensively to answer my question, yet unfortunately to no avail.

I am using the -psmatch2- command in order to conduct propensity score matching. i have run the first-stage of the propensity score matching using:

psmatch2 dummy N_dir Tenure Age N_nations lnMV CASH LEVERAGE VOL R_Ya TOBINQ, out(N_ACQ) common

where dummy = 1 when there is at least one female board member, and 0 otherwise. The outcome variable relates to the number of acquisitions a firm undertakes in a given year.

However, I cannot figure out for the life of me how to do the 2nd stage of the regression analysis, whereby I run a regression on my matched data.

Any guidance would be heavily appreciated. Thank you.

[Øyvind Snilsberg]

you've done both stages, the "Difference" from psmatch2's output is the estimated treatment effect

display r(att)

[David Radwin]

you've done both stages

That's not quite true. Sometimes there may be a good reason to run a regression on the matched sample.

Please see the sources cited in https://www.statalist.org/forums/for...opensity-score and https://www.statalist.org/forums/for...ion-covariates

For sample Stata code, see appendix SA2 of DuGoff et al. (2014).

[Jared Greathouse]

Under the hood, psmatch2 does the trick. However, it is possible as far as I remember from my matching days to use other parametric models (e.g., Poisson) on a matched sample.

[Øyvind Snilsberg]

That's not quite true. Sometimes there may be a good reason to run a regression on the matched sample.

Thank you for pointing this out. I was hinting at psmatch2 and regress producing the same treatment effect estimate,

sysuse auto, clear
psmatch2 foreign mpg rep78 ,out(price) common
regress price foreign [aw=_weight]

[Jane Quan]

you've done both stages, the "Difference" from psmatch2's output is the estimated treatment effect

display r(att)

Hi,

I am also wondering if there is a way to calculate treatment effect for each individuals, which means I use all the covariates to conduct a propensity score matching instead of using command "if",
and then use ttest over the subsamples. Well, I need treatment effect on treated or untreated for each observations is because I also need to use them for other purpose.

I found following command in "help psmatch2" but I don't quite understand what "groupvars" means.
> g att = . > egen g = group(groupvars) > levels g, local(gr) > qui foreach j of local gr { > psmatch2 treatvar varlist if g==`j', out(outvar) > replace att = r(att) if g==`j' > } > sum att Hope someone can help me out with this.
Thanks^^

[Matthew Williams]

Dear all,

May I have a question regarding the PSM estimation? Specifically, it seems that PSM has been designed for panel data, but does it work well with cross-sectional data? If so, I would appreciate it if anyone here could send me some references.

Thank you.

[David Radwin]

If you can't find these references by searching for "matching," please follow FAQ Extra Advice 1.5 Starting a new thread: "If you have what is essentially a new and different question, it is generally best to start a new thread. A cross-reference to the previous thread is helpful if there is a relation."

[Sattar Khan]

Thank you for pointing this out. I was hinting at psmatch2 and regress producing the same treatment effect estimate,

sysuse auto, clear
psmatch2 foreign mpg rep78 ,out(price) common
regress price foreign [aw=_weight]

Dear Snilsberg, can we do it with an unmatched sample too, just like you mentioned for matched sample"regress price foreign [aw=_weight]"

[Øyvind Snilsberg]

yes, just drop the weights

[Sattar Khan]

yes, just drop the weights

Kindly Sir if you don't mind provide an example for this end

[Øyvind Snilsberg]

sysuse auto, clear

*regress price on foreign in unmatched sample
regress price foreign 

*regress price on foreign in matched (on mpg and rep78) sample
psmatch2 foreign mpg rep78
regress price foreign [aw=_weight]

*alternatively
psmatch2 foreign mpg rep78, out(price)

[Sattar Khan]

sysuse auto, clear

*regress price on foreign in unmatched sample
regress price foreign

*regress price on foreign in matched (on mpg and rep78) sample
psmatch2 foreign mpg rep78
regress price foreign [aw=_weight]

*alternatively
psmatch2 foreign mpg rep78, out(price)

I got it,
Thank You Dear Snilsberg

[Sattar Khan]

sysuse auto, clear

*regress price on foreign in unmatched sample
regress price foreign

*regress price on foreign in matched (on mpg and rep78) sample
psmatch2 foreign mpg rep78
regress price foreign [aw=_weight]

*alternatively
psmatch2 foreign mpg rep78, out(price)

Dear Snilsberg, thanks for your comments and useful commends, Snilsberg, how we matched the sample based on a variable, although I have tried the code ematch and biasadj, but it didn't work.