Status:Closed    Asked:Nov 15, 2017 - 03:16 PM

Are weights adjusted when using "select cases" in an extraction?

I am working with the 5 year ACS 2015 and want to limit my sample to a subset of observations, e.g. ages 18-67. If I employ the "select cases" option prior to extracting my sample is the perwt adjusted for this subsample? I saw a previous forum question that the repwts are not adjusted and the full sample needs to be downloaded to obtain appropriate standard errors when using svy commands, is this correct?

Do you have the same question? Follow this Question

Staff Answer

Replicate weights allow you to generate empirically derived standard error estimates. As mentioned in the Replicate Weight FAQ page, accurate calculation of standard errors using replicate weights requires all records from the sample to be present. So you will need to include all cases from a sample to use replicate weights.

However, if you are using PERWT to obtain nationally representative statistics for person-level analyses, then you can use the "select cases" option when creating your extract. Because the weighted frequencies are meant to represent the entire U.S. population, each person within your sample will represent the same proportion of the U.S. population regardless of the inclusion or exclusion of others.


Nov 16, 2017 - 10:47 AM

Report it


What kind of adjustments do you think are necessary? I can't think of any.

The technical answer is that you are working with domain, or subpopulation. For most surveys and most domains, that would mean that you indeed need to download the full data set and run the analysis with -svy, subpop()- in Stata, survey::svyby() in R, etc. The issue is in computing standard errors, and making sure that your subsetting the data does not interact in weird ways with the sampling design and calibration routines. See West et al (2008) for detailed examination and explanation of issues --

With multiyear ACS and adult age domain (which I am sure cuts through all the strata and clusters), you *could* be good just getting the data with replicate weights, and analyzing them with no other adjustments.

See if you could link to that other answer on this forum. You can find pretty much anything on the Internet, so "I read about this" is a very weak form of reference these days.


Nov 16, 2017 - 02:31 PM

Report it


How Do I Link Husbands & Wives in 1974 & 2008 Liberian IPUMS Samples to Study...
I am looking for the Kessler 6 variable in IPUMS-MEPS
Reason for missing MIGSTA1 values in ASEC files for 1985 (and 1995)
India Area-level data (geo2) inconsistent for total population, urban share, ...
Does anyone have a method for handling missing data in CPS "EDUC" variable? 2...
How do I find the data file:Ipumsi_00001.dat for the 1974
Login   |   Register

Recently Active Members

View More »

Share |