Reweighting Households for Person-level Estimates

USA
1

I have an analysis where I’m trying to estimate the number of children in a state who reside in a household that gets SNAP benefits (ie. what percent of children from state X live in a SNAP household). Ordinarily, this is straightforward: use or create an indicator for SNAP receipt at the HH level, assign it to the people within the HH, and then sum the person weights of observations that are children who have that indicator.

However, I’m interested in doing some reweighting to do some comparisons across geographies. So my question is more like “If Massachusetts looked more like Alabama demographically (but still kept MA’s relationships between those demographics and SNAP enrollment), how many children would be on SNAP?” I’d like to reweight MA’s IPUMS data to look like AL by raking across a number of demographic variables and then calculating the new estimate using the new weights.

My problem lies with the fact that the SNAP receipt that I care about is at the HH level, but the thing I want (count of children) uses the person level data/weights. Can I just count the raw number of children (persons below age 18) in the household and multiply that by the new HH weight? This seems problematic as is throws out the information in the person weights, but I’m stumped about how to propagate the reweighting that I do at the HH level down to the person level, whether that’s necessary, or whether there is another path that I’m missing.

2

This is outside the scope of what IPUMS User Support can provide guidance on. I’m not aware of any official guidance or documentation from the Census Bureau regarding this type of analysis. The information I can provide is strictly based on my own experience with analyzing IPUMS microdata. I would suggest reviewing the literature in your field for more information.

It sounds like you’re doing something analogous to entropy balancing. I’m familiar with entropy balancing at the level of the unit of analysis (e.g., reweighting households and then analyzing households). But I’m not sure how you would reweight a sample according to household characteristics and then incorporate those weights into a person-level analysis. My hunch is that you’d need to either conduct a household-level analysis (e.g., what share of households with children in them receive SNAP) or reweight at the person level. Perhaps you can reweight persons based on household characteristics.