Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Majordomo results: Re: st: Variance estimation with cluster


From   Majordomo@hsphsun2.harvard.edu
To   statalist@hsphsun2.harvard.edu
Subject   Majordomo results: Re: st: Variance estimation with cluster
Date   Thu, 8 Nov 2007 11:01:56 -0500 (EST)

--

>>>> Steven makes some good points.  I have a slightly different take:
**** Command 'steven' not recognized.
>>>> 
>>>> 1.  Use the -fpc- option, but understand what it means.   Imagine you
**** Command '1.' not recognized.
>>>> "sampled" w/o replacement 100% of establishments and workers in a
**** Command '"sampled"' not recognized.
>>>> population; with the fpc's, all standard errors would be zero.  This
**** Command 'population;' not recognized.
>>>> is as it should be; the svy SEs in a regression using the population
**** Command 'is' not recognized.
>>>> are zero, because svy SEs represent deviations around the population
**** Command 'are' not recognized.
>>>> value (not Fisher-Neyman notions of deviations about what might have
**** Command 'value' not recognized.
>>>> been observed in the population with a different random sprinkling of
**** Command 'been' not recognized.
>>>> regressors on individuals).
**** Command 'regressors' not recognized.
>>>> 
>>>> 2. svy + panel = trouble.  If you want to run a fixed-effect
**** Command '2.' not recognized.
>>>> regression, consider -areg- which allows pweights that vary over time
**** Command 'regression,' not recognized.
>>>> and a -cluster- option.
**** Command 'and' not recognized.
>>>> 
>>>> 3.   I would use the time-specific weights which measure the number of
**** Command '3.' not recognized.
>>>> person-years each observation represents in the population of workers
**** Command 'person-years' not recognized.
>>>> in the two years.  The population is then not people, but people*time.
**** Command 'in' not recognized.
>>>> 
>>>> On 11/8/07, Steven Joel Hirsch Samuels <sjhsamuels@earthlink.net> wrote:
**** Command 'on' not recognized.
>>>> > --
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > Maury:
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > I would would only add to Austin's good advice:
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > 1. If you are doing regressions and hypothesis tests, do not use the
**** Command '>' not recognized.
>>>> > fpc terms. Imagined you had studied 100% of establishments and
**** Command '>' not recognized.
>>>> > workers in a population; with the fpc's, all standard errors would be
**** Command '>' not recognized.
>>>> > zero.
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > 2. Stata's panel data and multi-level model -xt- commands will not
**** Command '>' not recognized.
>>>> > respond to -svyset-.  For panel data analysis, the options
**** Command '>' not recognized.
>>>> > accommodating the survey design vary by command.
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > 3. You should probably use the survey weights from year 1; but the
**** Command '>' not recognized.
>>>> > study documentation may have other advice. Obviously these weights
**** Command '>' not recognized.
>>>> > will not sum to the population size in either year 1 or year 2. If
**** Command '>' not recognized.
>>>> > the survey deliberately over-sampled a class of workers which is the
**** Command '>' not recognized.
>>>> > subject of your analysis (e.g. you wish to compare a minority to a
**** Command '>' not recognized.
>>>> > majority group, and the survey over-sampled the minority group), you
**** Command '>' not recognized.
>>>> > should probably ignore the survey weights altogether.
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > -Steven
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > On Nov 8, 2007, at 10:16 AM, Austin Nichols wrote:
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > > Maury Gittleman <Gittleman.Maury@bls.gov>:
**** Command '>' not recognized.
>>>> > > Just clustering on establishment is probably sufficient.
**** Command '>' not recognized.
>>>> > >
**** Command '>' not recognized.
>>>> > > You can also specify two levels of clustering with -svyset- e.g.
**** Command '>' not recognized.
>>>> > >
**** Command '>' not recognized.
>>>> > > webuse stage5a
**** Command '>' not recognized.
>>>> > > svyset su1 [pweight=pw], fpc(fpc1) || su2
**** Command '>' not recognized.
>>>> > >
**** Command '>' not recognized.
>>>> > > where su1 is your establishment id, fpc1 the number of distinct
**** Command '>' not recognized.
>>>> > > employees in both years, and su2 is a person id.
**** Command '>' not recognized.
>>>> > >
**** Command '>' not recognized.
>>>> > > Usually the second level of clustering is largely irrelevant.  But
**** Command '>' not recognized.
>>>> > > not always...
**** Command '>' not recognized.
>>>> > >
**** Command '>' not recognized.
>>>> > > svyset su1 [pweight=pw], fpc(fpc1) strat(strat)
**** Command '>' not recognized.
>>>> > > svy: reg yreg x?
**** Command '>' not recognized.
>>>> > > est sto c1lev
**** Command '>' not recognized.
>>>> > > svyset su1 [pw=pw], fpc(fpc1) str(str) || su2, fpc(fpc2)
**** Command '>' not recognized.
>>>> > > svy: reg yreg x?
**** Command '>' not recognized.
>>>> > > est sto c2lev
**** Command '>' not recognized.
>>>> > > esttab *, mti
**** Command '>' not recognized.
>>>> > >
**** Command '>' not recognized.
>>>> > >
**** Command '>' not recognized.
>>>> > > On 11/8/07, Gittleman, Maury - BLS <Gittleman.Maury@bls.gov> wrote:
**** Command '>' not recognized.
>>>> > >> Hello,
**** Command '>' not recognized.
>>>> > >>
**** Command '>' not recognized.
>>>> > >> I'm have a question concerning stata's approach to estimating
**** Command '>' not recognized.
>>>> > >> standard
**** Command '>' not recognized.
>>>> > >> errors in the presence of clustered survey data.  The survey I'm
**** Command '>' not recognized.
>>>> > >> using
**** Command '>' not recognized.
>>>> > >> collects information on individual wages, by first selecting
**** Command '>' not recognized.
>>>> > >> establishments at random, and then collecting information on multiple
**** Command '>' not recognized.
>>>> > >> workers within each establishment.  So, it is clear that, when I'm
**** Command '>' not recognized.
>>>> > >> running regressions, I need to cluster on establishment.
**** Command '>' not recognized.
>>>> > >>
**** Command '>' not recognized.
>>>> > >> My question arises when I use two years of data from the same survey.
**** Command '>' not recognized.
>>>> > >> For about 4/5 of the individuals, there will be data for two
**** Command '>' not recognized.
>>>> > >> years, and
**** Command '>' not recognized.
>>>> > >> I would expect that the correlation between the errors for any given
**** Command '>' not recognized.
>>>> > >> individual will be higher than the correlation between the errors for
**** Command '>' not recognized.
>>>> > >> two different individuals at the same establishment.  My thinking is
**** Command '>' not recognized.
>>>> > >> that I still want to define clusters by establishments, as the
**** Command '>' not recognized.
>>>> > >> variance
**** Command '>' not recognized.
>>>> > >> estimation is said to be robust to any arbitrary intra-cluster
**** Command '>' not recognized.
>>>> > >> correlation.
**** Command '>' not recognized.
>>>> > >>
**** Command '>' not recognized.
>>>> > >> Is this the right way to go or is there an alternative approach that
**** Command '>' not recognized.
>>>> > >> might be superior?
**** Command '>' not recognized.
>>>> > >>
**** Command '>' not recognized.
>>>> > >> Thanks very much.
**** Command '>' not recognized.
>>>> > >>
**** Command '>' not recognized.
>>>> > >> Maury
**** Command '>' not recognized.
>>>> > > *
**** Command '>' not recognized.
>>>> > > *   For searches and help try:
**** Command '>' not recognized.
>>>> > > *   http://www.stata.com/support/faqs/res/findit.html
**** Command '>' not recognized.
>>>> > > *   http://www.stata.com/support/statalist/faq
**** Command '>' not recognized.
>>>> > > *   http://www.ats.ucla.edu/stat/stata/
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > Steven  Samuels
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > sjhsamuels@earthlink.net
**** Command '>' not recognized.
>>>> > 18 Cantine's Island
**** Command '>' not recognized.
>>>> > Saugerties, NY 12477
**** Command '>' not recognized.
>>>> > Phone: 845-246-0774
**** Command '>' not recognized.
>>>> > EFax: 208-498-7441
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> > *
**** Command '>' not recognized.
>>>> > *   For searches and help try:
**** Command '>' not recognized.
>>>> > *   http://www.stata.com/support/faqs/res/findit.html
**** Command '>' not recognized.
>>>> > *   http://www.stata.com/support/statalist/faq
**** Command '>' not recognized.
>>>> > *   http://www.ats.ucla.edu/stat/stata/
**** Command '>' not recognized.
>>>> >
**** Command '>' not recognized.
>>>> *
**** Command '*' not recognized.
>>>> *   For searches and help try:
**** Command '*' not recognized.
>>>> *   http://www.stata.com/support/faqs/res/findit.html
**** Command '*' not recognized.
>>>> *   http://www.stata.com/support/statalist/faq
**** Command '*' not recognized.
>>>> *   http://www.ats.ucla.edu/stat/stata/
**** Command '*' not recognized.
**** No valid commands found.
**** Commands must be in message BODY, not in HEADER.

**** Help for Majordomo@hsphsun2.harvard.edu:

This is the "Majordomo" mailing list manager, version 1.94.1. 

In the description below items contained in []'s are optional. When
providing the item, do not include the []'s around it.  Items in angle
brackets, such as <address>, are meta-symbols that should be replaced
by appropriate text without the angle brackets.

It understands the following commands:

    subscribe <list> [<address>]
	Subscribe yourself (or <address> if specified) to the named <list>.
	
    unsubscribe <list> [<address>]
	Unsubscribe yourself (or <address> if specified) from the named <list>.
	"unsubscribe *" will remove you (or <address>) from all lists.  This
	_may not_ work if you have subscribed using multiple addresses.

    get <list> <filename>
        Get a file related to <list>.

    index <list>
        Return an index of files you can "get" for <list>.

    which [<address>]
	Find out which lists you (or <address> if specified) are on.

    who <list>
	Find out who is on the named <list>.

    info <list>
	Retrieve the general introductory information for the named <list>.

    intro <list>
	Retrieve the introductory message sent to new users.  Non-subscribers
	may not be able to retrieve this.

    lists
	Show the lists served by this Majordomo server.

    help
	Retrieve this message.

    end
	Stop processing commands (useful if your mailer adds a signature).

Commands should be sent in the body of an email message to
"Majordomo@hsphsun2.harvard.edu". Multiple commands can be processed provided
each occurs on a separate line.

Commands in the "Subject:" line are NOT processed.

If you have any questions or problems, please contact
"Majordomo-Owner@hsphsun2.harvard.edu".

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   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   |   What's new   |   Site index