Re: st: multi-way clustering in stcox

[Austin Nichols <austinnichols@gmail.com>]

Herbert Schuetze <hschuetz@uvic.ca> :
The duplication should not matter if you are clustering at a higher
level.  E.g. compare SEs here:

webuse psidextract, clear
g fake=id-mod(id,5)
reg lwage exp ed
reg lwage exp ed, cl(fake)
expand 2
reg lwage exp ed
reg lwage exp ed, cl(fake)

On Mon, Mar 21, 2011 at 2:34 PM, Herbert Schuetze <hschuetz@uvic.ca> wrote:
> Thanks Austin,
>
> The problem is that the Lunn and McNeil approach requires that the data layout be altered such that each individual subject enters the data twice (once for each type of failure).
>
> I don't think this is the same as the nested setting that you suggest.
>
> Hopefully this works - signed up to digest version
>
> Herb
>
>
> Herbert Schuetze <hschuetz@uvic.ca>:
> Are individuals nested within states? And do you have 50 or so states?
> If so, just cluster on state--the clustering on id comes for free.
>
> On Mon, Mar 21, 2011 at 1:55 PM, Herbert Schuetze <hschuetz@uvic.ca> wrote:
>> Hi all,
>>
>> I am attempting to estimate a competing risks model in a Cox proportional hazard setting, similar to that suggested by Lunn and McNeil (1995). The problem I am encountering is that the procedure proposed by Lunn and McNeil requires one to utilize robust standard errors clustered on individual id numbers. At the same time, my data are generated such that I also need to cluster on State. The variable of interest is EI benefit levels similar to Chetty (2008 JPE). Cameron, Gelbach and Miller (2006) suggest a method for non-nested multi-way clustering in simple OLS, logit and probit models.
>>
>> Has anyone applied this methodology in an stcox setting?
>>
>> Is there another approach that I am unaware of?
>>
>> Thanks,
>>
>> Herb

*
*   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/