Re: st: A problem with -pstest- command

 From "Alvaro Zevallos" <[email protected]> To <[email protected]> Subject Re: st: A problem with -pstest- command Date Tue, 16 Jan 2007 17:45:30 -0500

Hi Nick!

Let�s say that the following variables are in my database:

y = Outcome
t = Treatment Indicator
x1 = Covariate
x2 = Covariate

Then, I formulate the command line, in Stata, to obtain the ATT estimate

. psmatch2 t x1 x2 , out(y) caliper(0.05) com ties

and immediately afterwards

. pstest x1 x2

This command, -pstest-, calculates the "Standardised Bias Before Matching" (SB
before) and the "Standardised Bias After Matching" (SB after) for x1 and x2
suggested by Rosenbaum and Rubin (1985). Following the equation as seen in
Caliendo and Kopeinig (2005):

"Standardised Bias Before Matching"

SBbeforej = (M1(xj) - M0(xj)) / (0.5(V1(xj) + V0(xj)))^(0.5)
for j=1,2

let�s say

SBbeforej= Aj / Bj

where

M1(xj) = Mean of variable xj for all treated individuals
M0(xj) = Mean of variable xj for all Non-treated individuals
V1 (xj) = Variance of variable xj for all treated individuals
V0 (xj) = Variance of variable xj for all Non-treated individuals

Aj = numerator of SBbeforej
Bj = denominator of SBbeforej

and,

"Standardised Bias After Matching"

SBafterj = (M1m(xj) - M0m(xj)) / (0.5(V1m(xj) + V0m(xj)))^(0.5)
for j=1,2

Let�s say

SBafterj=Cj / Dj

where

M1m(xj) = Mean of variable xj for matched treated individuals
M0m(xj) = Mean of variable xj for matched Non-treated individuals
V1m(xj) = Variance of variable xj for matched treated individuals
V0m(xj) = Variance of variable xj for matched Non-treated individuals

Cj = numerator of SBafterj
Dj = denominator of SBafterj

I thing that there are a problem with -pstest- command because it calculates
SBafter as :

SBafterj = (M1m(xj) - M0m(xj)) / (0.5(V1(xj) + V0(xj)))^(0.5)

Let�s say

SBafterj=Cj / Bj

As we can observe, the denominator of the SBafterj should be Dj.

Would somebody help me to understand why this happens?

Alvaro Z.

----- Original Message ----- From: "n j cox" <[email protected]>
To: <[email protected]>
Sent: Tuesday, January 16, 2007 10:08 AM
Subject: Re: st: A problem with -pstest- command

This posting raises questions on various levels.

from SSC and the reference given is supplied in full
in psmatch2.hlp. As in a concurrent thread, posters are
asked to provide explanation of sources.

I don't see how anyone can be expected to comment on your Excel calculations.

twice might be expected to give different answers the second time.

In short, I think you need to say more for experts in this
topic to be able to comment more fully.

Nick
[email protected]

Alvaro Zevallos

I�m testing the quality of matching estimators of five ATT parameters and i�m using the - pstest- command after - psmatch2 -, but i stopped because i have problems with the results, specifically with the �Standardised Bias� Column.

When i replicated the Standardised Bias equation of Rosenbaum and Rubin
(1985) for all my covariates in Excel the results were the same as
the -pstest-�s only for unmatched means.

I was trying to find an explanation for the differences between the matched
means and I found that the -pstest- command uses the square root of the
average of sample variances before matching as the denominator of the
�standardised bias� after matching equation. I downloaded the command again but that didn�t solve the problem.
What am I doing wrong?

*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/

--
No virus found in this incoming message.
Checked by AVG Free Edition.
Version: 7.1.410 / Virus Database: 268.16.12/628 - Release Date: 15/01/2007

```*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```