RE: st: nnmatch error in PSM

[Joe Canner <jcanner1@jhmi.edu>]

Katherine,

-psmatch2- creates several new variables that should help you with this:

_weight: weight of matched controls
_id: an identifier created by the program (you can probably specify one instead if you wish)
_n1: ID of nearest neighbor
_nn: # of matched neighbors

In the case of a 1:1 match you can save all cases with _weight==1.  The _id and _n1 variables can be used to see exactly who matched with whom if necessary.

Regards,
Joe

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Katherine Picho
Sent: Thursday, September 12, 2013 2:38 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: nnmatch error in PSM

Hello again Ariel, Joe

(and anyone else who knows about psmatch2)

I'm using stata12 IC.

I have 3500 observations in my treatment and 10k control observations.
I want to conduct 1:1   matching to get the best matches for my 3500
treatments this pool of 10,000 controls, and then save the new dataset (a subset of the original) that would contain only the 1:1 matched cases (thus ending up with a total sample of 7k observations; 1/2 trtment, 1/2 ctrl).

 My question is:

I have looked at the psmatch2 detailed syntax but do not see any options for creating such a new dataset from the original. Does it exist and am I missing something, or this feature not yet available?
thanks!

Katherine.


On Thu, Sep 12, 2013 at 1:30 PM, Katherine Picho <thestatsbabe@gmail.com> wrote:
> Hi Ariel
>
> Im only 2 days new to the stata list (& still familiarizing myself 
> with the etiquette) so if my reference to stata as 'STATA' proved 
> offensive to you and everyone else on the listserv, I apologize. your 
> correction is duly noted.
>
> I am using stata12 IC, and I obtained nnmatch by  typing ssc install 
> nnmatch in stata 12.
>
> thank you for your recommendations and for the link above. I will take  
> a look and follow your leads.
>
> Best,
>
> On Thu, Sep 12, 2013 at 11:20 AM, Ariel Linden, DrPH 
> <ariel.linden@gmail.com> wrote:
>> Katherine,
>>
>> Posters to the Statalist are asked to indicate which program they are 
>> using and where they got it from. In this case, you reference 
>> -nnmatch- which is a user written program found at http://www.stata-journal.com/software/sj4-3.
>>
>> This is not intended to be an academic exercise or "busy-work". In 
>> fact, it serves an important purpose. In Stata v13, there is a 
>> built-in program - teffects nnmatch- that is based on the original 
>> -nnmatch- but with some differences. Thus, others on this list have 
>> no idea which of these programs you are referring to (let alone that 
>> you did not indicate which version of Stata you are using (other than stating IC).
>>
>> Moreover, as a note of reference, it is "Stata" not "STATA".
>>
>> As to your problem, you should review your memory settings. In Stata 
>> v13, the settings are as follows (they are slightly different than in 
>> prior
>> versions):
>>
>> set maxvar      #   [, permanently]
>> set niceness    #   [, permanently]
>> set min_memory  amt [, permanently]
>> set max_memory  amt [, permanently]
>> set segmentsize amt [, permanently]
>>
>> As for -psmatch2- (http://fmwww.bc.edu/RePEc/bocode/p), it can indeed 
>> perform matching with and without replacement. Note that -nnmatch- 
>> performs variable matching whereas -psmatch2- will allow you to 
>> perform both variable matching and propensity score matching.
>>
>> I hope this helps
>>
>> Ariel
>>
>> Date: Wed, 11 Sep 2013 10:25:25 -0400
>> From: Katherine Picho <thestatsbabe@gmail.com>
>> Subject: st: nnmatch error in PSM
>>
>> Hi all
>>
>> I'm currently attempting propensity score matching with nearest 
>> neighbor using the nnmatch.
>>
>> However, at some point in running my ado file, STATA returns an error
>> message: "Insufficient memory to create more variables:Either 
>> increase memory or decrease m"
>>
>> I only have about 290 variables in the current dataset (21k observations).
>> I'm only matching using 3 variables. I'm using STATA IC and the max # 
>> var is 2048 (which is way more that I could ever use for this 
>> specific purpose).
>>
>> Even when I reduce the dataset to only 40 variables, and still match 
>> on the 3 variables of interest, the problem/ error code remains the 
>> same.
>> I have 3615 treatment observations that I'd like to match from a pool 
>> of about 16,500 controls (so theoretically this should not be a
>> problem)
>> My matching variables are categorical
>>
>> the nnmatch command that I run (where I run into trouble) is:
>>
>> nnmatch Depvar trtmt pscore, tc(att) m(1) keep(matched_att) replace
>>
>> this is where it generates the error message.
>>
>> I tried to remedy potential memory problems like- compressing the 
>> data/ variables. expanding the memory. (help query, within stata 
>> itself), etc, Nothing works. A search on the statalist archives 
>> indicates that a few people have run into the same problems using
>> nnmatch-- but none of the queries I found had received any responses 
>> on how to fix the problem.
>>
>> If anyone here has run into a similar issue that they were able to 
>> fix, please let me know. Thanks!
>>
>> Katherine
>> *
>>
>>
>>
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/faqs/resources/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/faqs/resources/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/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/