FSFDSTMP categorization

Hi - I am using the December issue of the CPS ASEC data + the Food Security Supplement for the years 2010-2021. I need to a variable on whether a household received food stamps. The answers “Yes” and “No” are clear. My understanding is that those with income above 185% of the poverty level were not asked the question if they did not report being short of money and they would appear under “NIU”. Would I count all of those in NIU as “No”?

Whether you should code CPS respondents not in universe for SNAP related variables as not receiving snap or as missing data points depends on your research goals and your own analytical discretion. Counting people NIU as “no” will likely change the denominator of your analysis, and I cannot determine if that is appropriate for your purposes.

First, note that the universe is not the same for all SNAP related variables in the CPS. If you are using SNAP related variables (such as FSFDSTMP or FSSTMPVALC) from the Food Security Supplement, the universe is in line with actual SNAP eligibility requirements. However, the ASEC question corresponding to the IPUMS variable FOODSTMP is asked of everyone (only vacant housing units are NIU). The remainder of my response assumes you are referring to variables from the Food Security Supplement. If you are in fact referring to FOODSTMP from the ASEC, your question does not apply, as everyone who does not qualify for and therefore does not receive SNAP benefits is already coded as “no.” It would be up to your discretion if you want to attempt to identify SNAP-eligible respondents in this case.

Sometimes respondents are out of universe for a question because it does not apply to them, which implies a particular answer. For example, a survey may ask if a respondent is working full-time, and restrict the universe to currently employed people. Someone who is out of universe (not employed) cannot be working full time, so their answer would always be “no” if asked. We do not observe the answers of NIU respondents but can be fairly certain they would say “no” unless they misreported their employment status.

In other cases, respondents are not in universe for a question because it usually does not apply to them. For example, some survey questions on retirement are only asked of older respondents who are out of the labor force. There are certainly younger respondents who are out of the labor force because they are retired.

The SNAP eligibility variables seem to be somewhere in between. While SNAP eligibility is baked into the universe of these variables, people misreport income frequently, and misreporting may be more common among people in some income groups than others. The rate of income misreporting, and whether errors of omission (someone eligible for SNAP is not asked the SNAP questions) or errors of comission (someone ineligible for SNAP is asked the SNAP questions) would affect your analysis are important factors to consider as you decide how to treat NIU respondents. It may or may not be appropriate to include NIU respondents in your analysis. For example, if you want to calculate the rate of SNAP receipt among SNAP-eligible households, you should not include NIUs in the denominator since they are (mostly) SNAP ineligible.