Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

re: st: Propensity Score Matching


From   "Ariel Linden. DrPH" <ariel.linden@gmail.com>
To   <statalist@hsphsun2.harvard.edu>
Subject   re: st: Propensity Score Matching
Date   Tue, 21 Feb 2012 10:08:30 -0800

See the responses to your questions below: 

" For the above table i don't understand why the mean for treated group is same for unmatched and matched sample"

Because matching is by default an ATT estimator. As such the treated group's covariates will not change, but the controls' will be chosen to look as similar to those treated subjects on those covariates. Conversely, if you used a ATC estimator, the control group (unmatched and matched) would be identical and the matched treated subjects would chosen to look as similar to those controls.

" and how do i get number of observations."

-psmatch2- ( a user written program found on ssc) generates a new variable called _wt. If you run -tab treat _wt- (for 1:1 matching), you should see the count of treated vs controls who were matched. You may have to stipulate - if _support==1 - if there were those who fell off of common support.

Is there any way i can identify which firm is matched against which firm?

If you read the help file for psmatch2, you'll see that the authors state:

_nk In the case of one-to-one and nearest-neighbors matching, for every
        treatment observation, it stores the observation number of the k-th
        matched control observation. Do not forget to sort by _id if you want
        to use the observation number (id) of for example the 1st nearest
        neighbor as in

        sort _id
        g x_of_match = x[_n1]



Date: Mon, 20 Feb 2012 02:53:02 -0800 (PST)
From: vikramfinavker <vikramfinavker@gmail.com>
Subject: st: Propensity Score Matching

Dear Statalisters,

First i apologise for not being clear and I thank you all for your replies.
I will try to explain what i want to do and where i am facing problem.

I am using psmatch2 with nearest neighbours matching with replacement.
Following is the command which i have used. 

psmatch2 allheg10 vw1zscore vw1lnmva vw1mvgroslev vw1quickratio vw1fsaleds
d2-d11 yr_00-yr_10, n(1)

And following is the results of above command

Probit regression                                 Number of obs   =      
3406
                                                  LR chi2(24)     =    
548.41
                                                  Prob > chi2     =    
0.0000
Log likelihood = -1007.6135                       Pseudo R2       =    
0.2139

- ------------------------------------------------------------------------------
    allheg10 |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
Interval]
- -------------+----------------------------------------------------------------
   vw1zscore |   .0075111    .002529     2.97   0.003     .0025543   
.0124678
    vw1lnmva |   .3311453   .0249946    13.25   0.000     .2821568   
.3801339
vw1mvgroslev |   1.457169   .1968683     7.40   0.000     1.071315   
1.843024
vw1quickra~o |  -.0910223   .0231545    -3.93   0.000    -.1364042  
- -.0456404
  vw1fsaleds |   .0026704   .0009382     2.85   0.004     .0008316   
.0045091
          d2 |  -.5984644   .2177925    -2.75   0.006     -1.02533   
- -.171599
          d3 |   -.401803   .1907166    -2.11   0.035    -.7756007  
- -.0280053
          d4 |  -.6046699   .1852172    -3.26   0.001     -.967689  
- -.2416508
          d5 |  -.2739949   .1801271    -1.52   0.128    -.6270375   
.0790477
          d6 |  -.1714531   .1972053    -0.87   0.385    -.5579683   
.2150621
          d7 |  -.3407124    .190602    -1.79   0.074    -.7142855   
.0328606
          d8 |  -.4092481   .1824948    -2.24   0.025    -.7669314  
- -.0515648
          d9 |  -.5232287   .3189424    -1.64   0.101    -1.148344   
.1018868
         d11 |   -.292377   .1768571    -1.65   0.098    -.6390105   
.0542566
       yr_00 |  -.1229459    .131999    -0.93   0.352     -.381659   
.1357673
       yr_01 |   .0223229   .1407681     0.16   0.874    -.2535775   
.2982233
       yr_02 |   .1623231   .1470985     1.10   0.270    -.1259846   
.4506308
       yr_03 |  -.1825395   .1357505    -1.34   0.179    -.4486057   
.0835267
       yr_04 |  -.1473961   .1367151    -1.08   0.281    -.4153527   
.1205606
       yr_05 |  -.1815719   .1393327    -1.30   0.193     -.454659   
.0915152
       yr_06 |  -.0916688   .1448511    -0.63   0.527    -.3755717   
.1922342
       yr_07 |  -.1135523   .1495029    -0.76   0.448    -.4065726   
.1794681
       yr_08 |  -.0781585   .1607199    -0.49   0.627    -.3931636   
.2368467
       yr_09 |  -.1666735   .1575016    -1.06   0.290    -.4753711   
.1420241
       _cons |  -2.987349   .3787435    -7.89   0.000    -3.729673  
- -2.245026
- ------------------------------------------------------------------------------

Now let me explain my variables. The dependent variable is Hedging dummy
(i.e. 1 if a firm is uses foreign currency contracts or interest rate swaps
and 0 otherwise). Other independent variables are determinants of hedging.
THis regression will tell me what influences firms to hedge. So, Now i
assume that the above method has created some kind of score which identifies
firms that are similar on all independent factors but their choice of
hedging. So, after identifying these firms i want to do following.

I have different risk measures and i want to do a mean difference test for
matched firms on these risk measures.
I want to do like this;

Control Gropup	Treatment Group	
Variable	N	Mean	Median	Mean 	Median	Diff in Means
Risk 1	XYZ	0	0	0	0	0
Risk2	yxz	0	0	0	0	0


​
Earlier some one told me to use pstest after running psmatch2. BY doing that
i get the following results.

. pstest vw1edf1jun vw1eqvol eqvolomvol

- ----------------------------------------------------------------------------
                        |       Mean               %reduct |     t-test
    Variable     Sample | Treated Control    %bias  |bias| |    t    p>|t|
- ------------------------+----------------------------------+----------------
  vw1edf1jun  Unmatched | .98219   2.3953    -31.0         |  -7.61  0.000
                Matched | .98219     2.31    -29.1     6.0 | -11.24  0.000
                        |                                  |
    vw1eqvol  Unmatched | .39455   .49404    -41.3         |  -8.83  0.000
                Matched | .39455   .44392    -20.5    50.4 |  -9.13  0.000
                        |                                  |
  eqvolomvol  Unmatched | 2.3454   3.0729    -44.5         | -10.12  0.000
                Matched | 2.3454   2.7199    -22.9    48.5 | -10.34  0.000
                        |                                  |
- ----------------------------------------------------------------------------
For the above table i don't understand why the mean for treated group is
same for unmatched and matched sample and how do i get number of
observations.

Is there any way i can identify which firm is matched against which firm?

If its still confusing please forgive me. I am referring the following paper
to do this " The Effects of Derivatives on Firm risk and Value" by  Sohnke
Bartram, Gregory Brown and Jennifer Conrad.


- -- 
Regards,

Vikram Finavker



*
*   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   |   Site index