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]

From |
Mike Perry <m5552k@gmail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: autocorrelated panel data with pweights |

Date |
Thu, 1 Jul 2010 15:48:51 -0700 |

Thanks for that response. I'm still digesting the second paragraph, but let me mention something that might clarify the issue a little. The data is hourly electricity usage data over several hundred customers over the course of several months. I agree that clustering above the customer level could be important, but I think that the autocorrelation in the usage within customers is far more important. So this isn't standard survey data. We'd like to use probability weights because we'd like to oversample particular types of customers because they are rare but influential. We'd like to estimate their impact precisely. In the end, we'd like regression parameters that are scaled up to the whole population, rather than our stratified sample. I may be misunderstanding your second paragraph, but are you saying that there's no role for fixed effects in models with pweights? Isn't there a role for weighting the contributions of each panel after the mean values have been removed? thanks, Mike On Thu, Jul 1, 2010 at 3:08 PM, Stas Kolenikov <skolenik@gmail.com> wrote: > If you have complex survey data, then in all likelihood you have > clustering that you need to account for at the level way above the > panels. As a result, such clustering subsumes autocorrelations within > panels. Essentially, you'll be running -regress [pw=], cluster()-. The > downside is that you need many clusters to trust asymptotic results, > and there is a logical inconsistency in the number of parameters if > you introduce dummy fixed effects: you can only estimate (# of > clusters) parameters with their variances that way. > > Estimators that work with -pweights- apply them to the contribution of > the unit to the objective function. In Steve's example, you can > compute the contribution of the panel to the GEE objective function, > and you can weight that with -pweight-. Fixed effect estimators > circumvent this: they apply linear transformations (contrasts) to the > data and end up with estimating equations defined at the level of > observations rather than at the level of panels. So there's again a > logical inconsistency between fixed effects and survey inference, as > they have different unit of analysis perspectives. > > On Thu, Jul 1, 2010 at 4:32 PM, Mike Perry <m5552k@gmail.com> wrote: >> Thank you for that suggestion. It's very close to what I need, but in >> the context I'm working, I can't justify the assumptions for the "pa" >> option. I need to be able to use fixed effects. I'm guessing that >> this is a difficult problem since many Stata commands come close to >> what I want, but none that I've seen actually do it. >> >> Mike >> >> On Thu, Jul 1, 2010 at 2:07 PM, Steve Samuels <sjsamuels@gmail.com> wrote: >>> How about: >>> >>> **********************CODE BEGINS**** >>> webuse nlswork >>> xtreg ln_w grade [pweight=birth_yr], pa corr(ar) >>> **********************CODE ENDS**** >>> >>> Steve >>> >>> On Thu, Jul 1, 2010 at 4:24 PM, Mike Perry <m5552k@gmail.com> wrote: >>>> I would like to be able to run a panel regression that corrects for >>>> autocorrelation within panels and allows the use of pweights. >>>> Essentially, I would like either something like xtregar, but with the >>>> ability to use pweights, or else a survey command that corrects for >>>> autocorrelation. Does anyone know of such a command in Stata? If not >>>> Stata, does anyone know of any command like that in SAS, SPSS or R? >>>> Also, is there an easy way to sum up why xtregar can handle fweights, >>>> but not pweights? >>>> >>>> thanks much, >>>> Mike >>>> * >>>> * 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/ >>>> >>> >>> >>> >>> -- >>> Steven Samuels >>> sjsamuels@gmail.com >>> 18 Cantine's Island >>> Saugerties NY 12477 >>> USA >>> Voice: 845-246-0774 >>> Fax: 206-202-4783 >>> >>> * >>> * 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/ >>> >> >> * >> * 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/ > * * 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/

**References**:**st: autocorrelated panel data with pweights***From:*Mike Perry <m5552k@gmail.com>

**Re: st: autocorrelated panel data with pweights***From:*Steve Samuels <sjsamuels@gmail.com>

**Re: st: autocorrelated panel data with pweights***From:*Mike Perry <m5552k@gmail.com>

**Re: st: autocorrelated panel data with pweights***From:*Stas Kolenikov <skolenik@gmail.com>

- Prev by Date:
**Re: st: autocorrelated panel data with pweights** - Next by Date:
**st: quantile regression with IV** - Previous by thread:
**Re: st: autocorrelated panel data with pweights** - Next by thread:
**st: spmap & water** - Index(es):