Hi Ariel, I'm sure that explains it.

In an attempt to wrap up this thread, let me review my original question and

the solution that was provided.

Objective: To identify the matched pairs that are created by the psmatch2

command.

The code below performs 1:1 nearest neighbor matching without replacement

and creates a new dataset named "paired" that contains the variable "pair"

that identifies which matched set each observation belongs to.

//--------------------------------------------------------------------------

psmatch2 xvar, pscore(pvar) outcome(yvar) caliper(.001) noreplace

neighbor(1)

gen pair = _id if _treated==0

replace pair = _n1 if _treated==1

bysort pair: egen paircount = count(pair)

drop if paircount !=2

save paired, replace

//--------------------------------------------------------------------------

Having the variable "pair" enables me to use the dataset of matched pairs to

run matched analyses in other stata commands or other software programs.

Thanks again for everyone's help. I have other questions about psmatch2 and

as I begin to articulate them I will start a new thread.

Garth

-----Original Message-----

From:

[hidden email]
[mailto:

[hidden email]] On Behalf Of Ariel Linden,

DrPH

Sent: Monday, April 05, 2010 10:41 AM

To:

[hidden email]
Subject: st: RE: psmatch2-identifying matched pairs

Hi Garth,

I haven't looked at the specific lines of code in PSMATCH2, but it seems

likely to me that the algorithm runs the ATT, then ATU, then ATE last. Each

iteration will naturally change the matches, since ATT means finding

controls to match to treated, ATU means finding treated to match to

controls, and ATE looks to adjust all units to get at the population average

effect.

To test this, try running the code in order ATT, ATU, and ATE, and review

the matches after each iteration. See if they change, and if they are

dropped after ATE.

Finally, you can email the authors of PSMATCH2. I have contacted them in the

past and found them very helpful.

Ariel

Date: Sun, 4 Apr 2010 23:43:32 -0500

From: "Garth Rauscher" <

[hidden email]>

Subject: st: RE: RE: psmatch2-identifying matched pairs

Thanks to Ariel for the explanation. To estimate the average treatment

effect in the population (ATE) it makes sense that the 1:1 matching would

not be used. However, with the ATE option in PSMATCH2, the output also

includes estimates of ATT and ATU along with ATE, therefore it would seem

that the program would still need to know the matched pairs in order to

estimate ATT. Therefore I don't understand why the identified pairs change

and some appear to be "broken".

Garth Rauscher

Associate Professor of Epidemiology

Division of Epid/Bios (M/C 923)

UIC School of Public Health

1603 West Taylor Street

Chicago, IL 60612

ph: (312)413-4317

fx: (312)996-0064

em:

[hidden email]
- -----Original Message-----

From:

[hidden email]
[mailto:

[hidden email]] On Behalf Of Ariel Linden,

DrPH

Sent: Saturday, April 03, 2010 10:04 AM

To:

[hidden email]
Subject: st: RE: psmatch2-identifying matched pairs

Garth

Per your question about ATE and dropped matches: That makes intuitive sense

since the ATE represents the average treatment effect in the population.

Therefore, you would need to have the outcome values of the entire

population, not just the subset of matched controls. As a result, you would

no longer have 1:1 matches.

Ariel

*

* For searches and help try:

*

http://www.stata.com/help.cgi?search*

http://www.stata.com/support/statalist/faq*

http://www.ats.ucla.edu/stat/stata/*

* For searches and help try:

*

http://www.stata.com/help.cgi?search*

http://www.stata.com/support/statalist/faq*

http://www.ats.ucla.edu/stat/stata/