I am analysing US family size trends using the ‘Children ever born’ data in censuses from 1900 onwards and am unclear about correct weights to use for 1940 and 1950. The ‘children ever born’ is a sample line variable for those two years and I have selected samples for those years using the SELFWTSL variable. The guidance on this variable in IPUMS seem to imply that no further weighting is needed for samples selected in that way. However, the guidance on the PERWT weight states that it should be used for all individual-level analysis for all census samples. This creates an uncertainty – should the SELFWTSL samples be weighted by PERWT or not? Using PERWT makes a big difference to results on family size patterns for 1940 and 1950 (especially 1950), so it’s important to get it right.
As mentioned on the Sample Designs page, the most appropriate weight to use when analyzing sample line data from the 1940 and 1950 samples is the SLWT variable. This variable gives Sample Line persons the appropriate weight to represent the entire population. However, if you have selected cases from the Self-weighted subset in 1940 and 1950 no weights are required as the Self-weighting subset is considered a “flat” sample.
I hope this helps.