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

From |
"Ángel Rodríguez Laso" <angelrlaso@gmail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: How to create weight variable |

Date |
Wed, 5 Nov 2008 09:10:16 +0100 |

"Ángel, I am not exactly clear about the part where you suggest calculating psweights after pweighting the sample. Part of my question related to confirming if my thinking on how to pweight the sample was correct." Fran, Sorry for the confusion I've created with the term pweight. Post stratification (also known as calibration) is the last step in weighting. Post stratification weights are calculated after weighting the sample by its design weights (also known as probability of selection weights, that are the inverse to the probability with which each sample unit is selected) and its non-response weights (the inverse to non response rates in different parts of the population). If in your sample all individuals from one strata had the same probability of being selected (without clustering, for example) and response rates were similar across groups of the population, then you only need to calculate post stratification weights with the formulas that you or me are proposing. The weight you obtain then is the pweight you have to use in Stata. Angel Rodriguez-Laso 2008/11/4 fran brittan <franbrittan@googlemail.com>: > Thank you so much, Maarten and Ángel! > > Maarten, it was very helpful to be pointed to the term post stratification. > Unfortunately, I have Stata 8, and the poststratify add-on doesn't > seem to be supported in that version. > > Ángel, I am not exactly clear about the part where you suggest > calculating psweights after pweighting the sample. Part of my question > related to confirming if my thinking on how to pweight the sample was > correct. > > I only have one file (the sample), as well as descriptions of the > population from census data, but obviously not the census dataset > itself. So the only thing I thought I can do was to calculate weights > according to this formula: > > weight(stratum x) = % population (stratum x) / % sample (stratum x) > > I then followed this with > > svyset[pweight=educweight], strata[education] > > and followed this with svylogit. > > My question at this stage is: is what I did statistically acceptable? > > The results of my regressions are similar to the unweighted results, > except a few variables which drop or gain significance. > > Thank you very much once again! > > Fran > > > On Tue, Nov 4, 2008 at 12:19 PM, Ángel Rodríguez Laso > <angelrlaso@gmail.com> wrote: >> I have taken a look to the information provided by -findit post >> stratification- and found that Stata has options in -svy- (poststrata, >> postweight) that probably do automatically what Fran Brittan was >> proposing; I've also found a package (poststratify) that even merge >> cells with 0 counts (where the formula that Fran is implicitly using >> gives 0). The problem with these two resources is that they take away >> control of the weighting process by the researcher. What I would do is >> calculating the poststratification weights myself, using the formula: >> >> psweight of stratatum x=(population in stratum x * total sample >> size)/(total population * sample size in stratum x) >> >> after pweighting the sample. >> >> If there are cells with 0 sample size or some psweights are >> disconected from the rest, the offending strata should be merged with >> similar ones. >> >> >> HTH >> >> Angel Rodriguez-Laso >> >> >> >> 2008/11/3 Maarten buis <maartenbuis@yahoo.co.uk>: >>> --- fran brittan <franbrittan@googlemail.com> wrote: >>>> I have a dataset from a survey that over-sampled highly educated >>>> people. I'd like to weight it so that the data is closer to the >>>> values in the true population, which I have from census data. >>> >>> The term you are looking for is post stratification, see: >>> -findit post stratification- >>> >>> Hope this helps, >>> Maarten >>> >>> ----------------------------------------- >>> Maarten L. Buis >>> Department of Social Research Methodology >>> Vrije Universiteit Amsterdam >>> Boelelaan 1081 >>> 1081 HV Amsterdam >>> The Netherlands >>> >>> visiting address: >>> Buitenveldertselaan 3 (Metropolitan), room N515 >>> >>> +31 20 5986715 >>> >>> http://home.fsw.vu.nl/m.buis/ >>> ----------------------------------------- >>> >>> >>> >>> * >>> * 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/ >> > > * > * 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: How to create weight variable***From:*"fran brittan" <franbrittan@googlemail.com>

**Re: st: How to create weight variable***From:*Maarten buis <maartenbuis@yahoo.co.uk>

**Re: st: How to create weight variable***From:*"Ángel Rodríguez Laso" <angelrlaso@gmail.com>

**Re: st: How to create weight variable***From:*"fran brittan" <franbrittan@googlemail.com>

- Prev by Date:
**Re: st: Reshape limit** - Next by Date:
**Re: st: Interpretation of regressionmodel of ln-transformed variable** - Previous by thread:
**Re: st: How to create weight variable** - Next by thread:
**st: Yahoo! - Message d'absence** - Index(es):

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