Identifying Households in Panel Data

Question

For my thesis, I'm working with a panel dataset that tracks individuals within households. To allow for merging, the dataset provides the household ID in the previous round for each individual. What I'm wondering is how to decide based on this information that a household is "the same" in two different rounds of data collection.

As I'm looking at the socio-economic outcomes of various farming practices, my concern is that there are unobservable household characteristics that influence the farmer's returns to farming but then also correlate with various household outcomes, such as education, assets, etc. To control for this, I'd like to use household fixed effects.

However, I'm wondering which features of a household would justify me saying "this is the same household and these effects are fixed over time". For instance, I don't want to use the household head because if the head changes and every other aspect of the household is identical, the impacts on farming outcomes shouldn't be too large. My question hence is whether there are commonly accepted characteristics for identifying households over time?

For reference, the data I'm using is the Living Standards Measurement Survey - Integrated Surveys on Agriculture, by the World Bank.

Answer 1

I don't get it, you say you have a household ID. If it is the same across waves, then it is the same household over waves. Most likely the people who collected the data will be able to answer the question of whether this is the same household better than you.

The exact definition of who constitutes a household is necessarily arbitrary and varies to some degree between countries. However, the large surveys tend to gravitate around a definition that includes everybody who has lived in the same "building" (whatever that may be) and at least one meal together for more than 6 months in the last year. Then they tend to narrow this down e.g. excluding workers being paid while living there etc. You can check out a questionnaire from a large survey, for instance Living Standard Measurement Survey by the World Bank, at the beginning of the household roster they usually have a section describing who is part of the household.

Households will be given a unique id the first time they are interviewed and followed using this id over the subsequent waves (there is no kind of probability-based matching going on). To re-identify the same households, surveys will use address / GPS, name, and contact details.

In sum: There is really not much point worrying about this. Just group based on hhid and spend your time on other problems in the data (which likely there will be enough of ;))