Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RES: st: RES: examples propensity score matching


From   "Henrique Neder" <hdneder@ufu.br>
To   <statalist@hsphsun2.harvard.edu>
Subject   RES: st: RES: examples propensity score matching
Date   Wed, 29 Oct 2008 22:16:25 -0200

Ana Gabriela

Now, I think that I understand your question. First off all, you have to organize your data as the following:


Group   time  Response_var covariate1 covariate2 .... covariatep

....


Where group is equal a binary variable (= 1 if treated = 0 if untreated); time is equal to t1 or t2; the response variable that is your variable related to the measurement of the treatment impact; covariates - the variables used in right side of logit (or probit) equation. 
Now you have the psmatch2 Stata command. An example of the syntax command is:

psmatch2 group covariate1 covariate2 ... covariatep, outcome(Response_var)  neighbor(1) common logit 

This command estimates the ATT (average treatment effect on treated) using nearest neighbor matching with common support and specifying the logit model to estimate the propensity scores used in matching. In the help (help psmatch2) you can see that several matching algorithms are provided. This command assures that the matches are formed by a couple of treated and untreated individuals. The response variable may be a difference in t2 minus t1 and the ATT provided for the command is a double difference estimator. You need to install this program using ssc install psmatch2, all replace.
 
But before the matching you have to test the balancing property using the pstest Stata command.

One alternative to this is the pscore Stata command. This is a comparable command and produce similar results as psmatch2, but balancing test I think   
that is more rigorous.  

Henrique Neder


-----Mensagem original-----
De: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] Em nome de Ana Gabriela Guerrero Serdan
Enviada em: quarta-feira, 29 de outubro de 2008 20:41
Para: statalist@hsphsun2.harvard.edu
Assunto: Re: st: RES: examples propensity score matching

Thanks Henrique, 

I want to match individuals (treatment and control groups) in t2 (post treatment) with individuals  in t1 (pre-treatment) sing several covariates (observable characteristics).  So in this case I would be able to apply double difference.  

So how can I be assured that the matched individuals are different for the treatment and control groups? 



--- On Tue, 10/28/08, Henrique Neder <hdneder@ufu.br> wrote:

> From: Henrique Neder <hdneder@ufu.br>
> Subject: st: RES: examples propensity score matching
> To: statalist@hsphsun2.harvard.edu
> Date: Tuesday, October 28, 2008, 2:00 PM
> Ana Gabriela
> 
> Your data have to look like this:
> 	 
>        Year group Varresult1 varresult2 covariate1
> covariate2 ....
> covariatek 
> Obs 1   t1  treat   ....       ....      ....  
> Obs 2   t1  contr
> Obs 3
> 
> .....   t2  treat 
> Obs n   t1  contr 
> 
> Where varresult is result variable and covariate are the
> covariates used in
> logit/probit model.
> 
> If you have no counterfactual information for the period 1
> refering about
> result variables, you can´t estimate double difference but
> you estimate only
> a cross section estimate for the period t2. Your covariates
> in the logit
> model have to be dated in t1 (or more precisely prior to
> treatment). So, if
> your treatment is dated in t1 or before and if you don’t
> observe the control
> individuals in t1, it is impossible to estimate the
> propensity score. You
> have to observe the covariates (used in the logit/probit
> model) prior the
> treatment date unless all your covariates are fixed in
> time.
> 
> Henrique Neder       
> 
> -----Mensagem original----- 
> De: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] Em nome de
> Ana Gabriela
> Guerrero Serdan
> Enviada em: terça-feira, 28 de outubro de 2008 16:08
> Para: statalist@hsphsun2.harvard.edu
> Assunto: st: examples propensity score matching
> 
> Dear Stata users, 
> 
> A very basic question. I want to take a look at how to do a
> propensity score
> matching in stata (commands psmatch2, pscore, I think). 
> 
> I have two cross sectional surveys (t1, t2)  where I
> observe treatment and
> control individuals in the second period but I cant know
> the conterfactual
> in t1. I want to use propensity score matching to
> indivudals in t2 to
> individuals in t1 based on some observable characteristics.
> 
> 
> I have technical readings and literature but I would like
> to know the how to
> in Stata. how the data should look? should I pool
> individuals for both
> cross-sections?  
> 
> thanks, 
> Gaby 
> 
> 
>       
> *
> *   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/
> 
> No virus found in this incoming message.
> Checked by AVG - http://www.avg.com 
> Version: 8.0.175 / Virus Database: 270.8.4/1752 - Release
> Date: 28/10/2008
> 10:04
> 
> 
> *
> *   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/

No virus found in this incoming message.
Checked by AVG - http://www.avg.com 
Version: 8.0.175 / Virus Database: 270.8.5/1755 - Release Date: 29/10/2008 17:27


*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index