Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: mi and svyset brr


From   Stas Kolenikov <skolenik@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: mi and svyset brr
Date   Tue, 14 Sep 2010 11:21:59 -0500

I would run -mi- once for each set of imputed weights, using those
weights to calibrate the imputation models; and run the analysis with
matching weights/multiply imputed data, using BRR formulae only, not
the MI formulae. And keep my fingers crossed that the reviewers will
find this suitable.

Generally speaking, the interface of MI and design-based variance
estimation is quite opaque. There is no convincing technical
literature, unfortunately: there was a pretty heated debate in JASA 91
(434) [http://www.jstor.org/stable/i314318], but somehow research came
to a halt on the topic after that (or at least I have not seen much in
JASA and Survey Methodology, the most natural outlets for this kind of
stuff). I'd be very happy to hear otherwise though.

My take on this is that at the minimum, the MI procedures lead to
violations of the assumptions stating that data in different PSUs are
independent: as long as your imputation models use information from
the whole data set, your imputed data depend on data in all strata and
PSU. You can avoid this problem by fitting the imputation models
within each PSU, but you will rarely have enough data for these
imputation models to be of passable accuracy.

On Tue, Sep 14, 2010 at 10:28 AM, Hillary Arnold <arnoldhr@mcmaster.ca> wrote:
> I am using survey data for regression analysis. This data includes brr
> weights but no PSU or strata variables. If I use mi to impute I
> recover about 25% of my cases relative to listwise deletion. Stata
> does not allow brr weights with multiple imputation - I am left with
> linearized standard errors. Is there a substantive reason for not
> allowing mi and brr standard errors? If not is there a way to use
> both? Or alternately must I choose between the two.
>
> Thanks
>
> Hillary Arnold
> PhD Candidate
> McMaster University
> *
> *   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/
>



-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   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/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index