Re: st: weighting regressions and clustering standard errors

[Richard Goldstein <richgold@ix.netcom.com>]

weighting for a variable number of matches (I think this is your
situation) is discussed in (1) Rosenbaum, PR (2010), _Design of
Observational Studies_, Springer, esp. chapter 8 and (2) Stuart, EA
(2010), "Matching methods for causal inference: a review and a look
forward," _Statistical Science_, 25: 1-21

Rich

On 3/25/11 2:06 PM, Zeke Hausfather wrote:
> I'm working on an analysis of the difference in trends between urban
> and rural temperature stations in the U.S. Specifically, I'm taking
> all permutations of urban and rural station pairs that fulfill certain
> requirements (e.g. are relatively close to each other, have the same
> instrument type, etc.). This results in a list of station pairs that,
> while unique, can include many permutations of the same urban station
> being paired with multiple nearby rural stations and vis versa.
> 
> I'm trying to regress the difference between the urban and rural
> stations in the pairs against a time variable (months) in a way that
> avoids overweighting cases where a spatial cluster of stations results
> in numerous permutations of the same urban or rural stations. I've
> tried using the cluster command, which gives me a better estimate of
> the standard errors given the non-unique data points, but this
> prevents me from also applying a weight to the regression based on the
> relative frequency of station occurrence in the pairs. I'm curious if
> anyone has thoughts on the best way to estimate the unique
> occurrence-weighted OLS fit while also reflecting this correctly in
> the standard errors.
> 
> --
> Zeke Hausfather
> Chief Scientist
> Efficiency 2.0
> 
> (o) 646 478 8509
> (m) 917 520 9601
*
*   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/