Re: st: Weighty Matter

[Austin Nichols <austinnichols@gmail.com>]

Alistair Windsor (U of M) <alistair.windsor@gmail.com> :

Using pweights is equivalent to using aweights and requesting robust
SE, which are pretty much always a good idea.  Using aweights is
easier with some commands, though, e.g. to get a simple mean where one
is not concerned with correct SE.

So use pweights unless you don't care about the SE, in which case
using aweights is equivalent.

If you want to implement a reweighting scheme with negative weights
you will have to use iweights, at your own risk.

On Sun, May 15, 2011 at 11:11 AM, Alistair Windsor (U of M)
<alistair.windsor@gmail.com> wrote:
> Dear All,
>
> I am working with some propensity score reweighting schemes. My
> question concerns which type of weights to use. Originally I was using
> pweights (and some svy commands) with the justification that a
> propensity score is essentially a sampling probability . My current
> choice has been to use aweights based on Nichols 2008 "Erratum and
> discussion of propensity score reweighting". However I noticed that in
> the earlier 2007 Nichols "Causal inference with observational data"
> the data was weighted using pweight rather than aweights. I have also
> been working on a program similar to psmatch2 but with reweighting.
> Looking into the code of pstest I find that it uses iweights.
>
> The actual reweighting scheme results do not depend on the choice of
> pweight, aweight, or iweights and I am using bootstrapping to
> determine the standard error of the treatment effect estimator.
> However determining the quality of the match uses the standard errors
> and this depends on the type of weight chosen.
>
> Anyone with insight into which weight to use?
>
> Yours,
>
> Alistair
*
*   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/