Calculating Households with Social Security Income with INCSS Variable

While the total number of households is correct, the estimate for households with social security income is much lower in IPUMS than in the ACS SF. Normally, PUMS data are within the margin of error of SF data. Could there be a problem with my syntax or is there an issue with INCSS?

Here is the SPSS syntax I used:
COMPUTE filter_=(RELATE = 1). FILTER BY filter_.


I am not sure exactly which summary file table you are referencing, but suspect that the issue is related to trying to estimate household values from a single person’s reported value for social security income. Note that INCSS reports Social Security pensions, survivors benefits, permanent disability insurance, and U.S. government Railroad Retirement insurance payments at the individual level. By restricting your analysis to householders only (RELATE = 1), you are omitting all other reports of social security income in the household. If you are interested in the number of households receiving social security income, you will need to consider INCSS values for all members of the household.

Thank you for your quick response. The summary file I refer is the ACS Summary File (SF) table B19055 (one of the pretabulated ACS products). Even when I include all members of the household with social security income, not just the householder as in the syntax I posted, the total is much higher than the SF table. Any thoughts as to why this would happen?


When using the 2014-2018 5-year and eliminating group quarters (GQ = 3 or GQ = 4), my estimates fall within the published margins of error for B19055 for the 2014-2018 ACS 5-year. My estimates were a little high before I removed those in group quarters. Are you summarizing information across all members of the household to create a household-level indicator of receipt of social security income, and then restricting to a single observation per household? If you are weighting with HHWT and include more than one observation per household, the estimate will be too high because of double-counting. Similarly, if you retain all observations and use PERWT, you won’t replicate the table as you will be generating a person count rather than household count.

1 Like