Re: st: standardized % bias before and after matching using the "diff" command

["Ariel Linden, DrPH" <ariel.linden@gmail.com>]

I would suggest calculating the difference-in-difference (DID) estimator
manually and then running the propensity scoring approach, since the (DID)
is nothing more than: 

DID = Tx(per2 - per1) - Con(per2 - per1)

However, if you prefer to use the "canned" approach provided by -diff- (from
http://fmwww.bc.edu/RePEc/bocode/d), then you can perform the standardized
difference routine after running -diff- using -pbalchk- (a user-written
program by Mark Lunt, found at
http://personalpages.manchester.ac.uk/staff/mark.lunt)

For example (from the help file for -diff-):

use cardkrueger1994, clear
diff fte, t(treated) p(t) cov(bk kfc roys) kernel id(id)

* this runs a t-test in -diff. Gets same means as -pbalchk-, but doesn't
calculate std diffs.
diff fte, t(treated) p(t) cov(bk kfc roys) test id(id) kernel

* this will estimate std diffs, for treatment vs control at t=0 (baseline)
pbalchk treated bk kfc roys if t==0, wt( _weights)


As you can see, the canned approach does not result in covariate balance.
This is why I would suggest that you conduct these processes manually so
that you can find the "matching" approach that gets the best covariate
balance...

I hope this helps

Ariel





________________________________________
From
  Luis Aranda <laranda44@gmail.com>
To
  statalist <statalist@hsphsun2.harvard.edu>
Subject
  st: standardized % bias before and after matching using the "diff" command
Date
  Fri, 9 Aug 2013 16:13:15 +0200
________________________________________
Dear all,

I am using the user-written "diff" command (Villa, Juan M. 2011) to
perform a difference-in-differences (kernel) matching estimator (as in
Heckman, Ichimura, and Todd, 1998). My data is longitudinal in nature.

My question concerns the possibility of calculating the standardized
percentage bias before and after matching (see Rosenbaum and Rubin,
1985) using the output given by the "diff" command, which will then
allow for the calculation of the achieved percentage reduction in bias
and with it a way to give credibility to the propensity score matching
methodology I am using.

The percentage covariate bias as well as the model's mean and median
biases are automatically given by other matching commands (e.g.
"pstest" from "psmatch2") as summary indicators of the distribution of
the bias. Thus, I was wondering if there exists a way within the
"diff" framework for automatically calculating them using Stata.

Many thanks in advance for your advice and kind assistance,

Luis Aranda



*
*   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/