[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: Testing for program effectiveness with heckman

From   "Riemer, Richard A CIV DMDC" <>
To   <>
Subject   RE: st: Testing for program effectiveness with heckman
Date   Fri, 30 May 2008 07:36:38 -0700

treatreg looks like a good option.  I will also look at references.
iv1 iv2 and iv3 are independent variables such as gender race and education level.  I will probably not find an independent variable that is related to self-selection that is not also related to afqt.
Is there any reason that effects coding (1's and -1's; or 1's, 0's, and -1's) can not be used for independent variables instead of dummy coding when using treatreg?


From: on behalf of Austin Nichols
Sent: Thu 5/29/2008 5:03 PM
Subject: Re: st: Testing for program effectiveness with heckman

Riemer, Richard <>:
I think you want -treatreg- instead of -heckman-
(but what are iv1 iv2 and iv3? do you have excluded instruments?)
See also

On Thu, May 29, 2008 at 3:25 PM, Riemer, Richard A CIV DMDC
<> wrote:
> Maarten, Thank you for your reply.  I can see the distinction you are
> making.  However, I wanted to use heckman because I thought it would do
> a better job at explaining self-selection of test-preparation rather
> than simple moderated regression where there could be correlated errors
> between the two equations.  Following the example of wage of women, we
> could say that 'afqt after test prep' is missing on sample members who
> do not engage in test prep and that those sample members would have
> scored lower than average if they would have engaged in test prep.
> Rich
> -----Original Message-----
> From:
> [] On Behalf Of Maarten buis
> Sent: Thursday, May 29, 2008 11:38 AM
> To:
> Subject: Re: st: Testing for program effectiveness with heckman
> This does not seem to be a case for -heckman-. -heckman- is for the case
> when there is selection on the y variable in your case afqt. The classic
> example is wage of women, where for some women we don't know their wage
> because they don't work, and we expect that those women that don't work
> would have earned less then average if they would have worked. The
> reason that such selection on the y causes problems is disccused
> graphically in a recent post:
> . Your use
> of the -heckman- syntax, with a selection of YesPrep and NoPrep suggests
> that that is not the case. In your case you should just use
> -regress-:
> regress afqt iv1 iv2 iv3 YesPrep
> where the coefficient of YesPrep tells you how much better students do
> when they prepare, while controlling for iv1 iv2 iv3.
> Hope this helps,
> Maarten
> --- "Riemer, Richard A CIV DMDC" <>
> wrote:
>> I want to use heckman on a program evaluation of the effectiveness of
>> test preparation behavior on aptitude scores (afqt).  Would I have to
>> run heckman twice; once for test-prep (YesPrep) and once for
>> non-test-prep (NoPrep) and then compare confidence intervals on the
>> constant term (afqt _cons) to determine if there were a significant
>> treatment effect on afqt scores?  On the other hand, is there a direct
>> way to test for a treatment effect, such as taking the output from
>> heckman and running it through another procedure?  Are there any good
>> references for this question or examples in the literature where stata
>> was used to test for program effectiveness.
>> heckman afqt iv1 iv2 iv3, select(YesPrep=iv1 iv2 iv3) heckman afqt iv1
>> iv2 iv3, select(NoPrep =iv1 iv2 iv3)
*   For searches and help try:


© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index