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: Svy poststratification VS Pweighting


From   francesco manaresi <manaresi@gmail.com>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   Re: st: Svy poststratification VS Pweighting
Date   Mon, 21 Jun 2010 19:22:21 +0200

Thank you very much for your answer, Stas. Maybe I was not clear, sorry:
I did not use the postweight to' multiply pweight. I just Either used
postweight only (with svy postweight capabilities) or pweight with
weights calculated as N_h/n_h (the inverse of the prob of being
selected if strata where design strata - while in fact they Are not).
In the FIRST case it yields that absurd small St.error which I agree
with u makes no sense.
The issue is why using
svy , poststrata(stratavar) postweight(number_of_obs_in_strata_in_orig_pop)
yields that micro-s.err.



On Monday, June 21, 2010, Stas Kolenikov <skolenik@gmail.com> wrote:
> On Mon, Jun 21, 2010 at 11:42 AM, francesco manaresi <manaresi@gmail.com> wrote:
>> I've seen several questions on the issue of poststratification in
>> Statalist, but would like to ask you some clarifications on the
>> estimate of standard errors. Thank you for your kindness and
>> availability.
>> I have got a sample of firms which have been (supposedly) randomly
>> drawn from a reference population, and would like to post-stratify
>> based on two observable characteristics for which all cross-tables are
>> available.
>
> The pweight has to be the inverse probability of selection, if
> available. Period. If you have additional information on the
> composition of the population, you should use post-stratification
> capabilities. Using post-stratification adjustments to multiply the
> original pweights and running anlaysis as if this composite weight
> were the pure probability weight is a poor man's strategy, and should
> be discouraged when more appropriate tools are available.
>
> I would never trust a standard error of 5e-17 which is roughly
> c(epsdouble). I don't know what you've done there, but you obviously
> eliminated the variance in the sample, and you know this cannot be
> right (unless you sampled all the units with probability of 1, at
> which point it is not a random sample anymore).
>
> As for matching, you are on your own there. I don't trust ANY standard
> errors that come out of matching estimators, so they are all equally
> bad, in my eyes. There is no clear population analogue of the matching
> procedure for the finite population, so -svy:- mode of inference is
> hardly applicable.
>
> --
> 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/
>

-- 
-------------------------------
Francesco Manaresi
Department of Economics
University of Bologna
P.zza Scaravilli 2
40126 - Bologna (Italy)

Tel: +39-051-209-8887
Cell: +39-320-112-7417
Skype: f.manaresi
URL: www2.dse.unibo.it/francesco.manaresi

*
*   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