NIDS-CRAM is widely used to investigate the impact of the COVID pandemic on individuals and households. However, because NIDS-CRAM is a survey of individuals it is difficult to make accurate statements about households. Nevertheless many issues of interest, such as the hunger questions in NIDS-CRAM, are about the household and not just the respondent. The problem with using the existing NIDS-CRAM weights for these analyses is that there is double-counting: there are potentially many individuals from the same household in the NIDS-CRAM survey. We show that overlapping membership affects between 40% to 50% of the observations. In this paper we lay out the theory for dealing with this problem and generate a set of “household weights” to reduce the double-counting. We use these weights to produce some initial estimates of how prevalent hunger might have been during the lockdown.

Paradoxically estimates of the fraction of households affected by hunger are not changed much by using the household weights rather than the person weights released with NIDS-CRAM. The reason for this is that hunger is only very weakly associated with household size, so the double-counting implicit in using the person weights does not skew the estimates much. However if one wants to generate estimates of the number of households or people affected by hunger the household weights make a much bigger difference. Indeed, we generate a first set of numbers that quantify the problem. For instance somewhere between 1.5 million and 3.1 million children were affected by hunger at the time of the field work for NIDS-CRAM wave 5. These estimates have to be treated with some caution, because our weights do not properly deal with changes in the distribution of households since 2017, in particular new household formation.