Recall Bias Revisited: Measure Farm Labor Using Mixed-Mode Surveys and Multiple Imputation

Smallholder farming dominates agriculture in poorer countries. Yet, traditional recall-based surveys on smallholder farming in these countries face challenges with seasonal variations, high survey costs, poor record-keeping, and technical capacity constraints resulting in significant recall bias. We offer the first study that employs a less-costly, imputation-based alternative using mixed modes of data collection to obtain estimates on smallholder farm labor. Using data from Tanzania, we find that parsimonious imputation models based on small samples of a benchmark weekly in-person survey can offer reasonably accurate estimates. Furthermore, we also show how less accurate, but also less resource-intensive, imputation-based measures using a weekly phone survey may provide a viable alternative for the more costly weekly in-person survey. If replicated in other contexts, including for other types of variables that suffer from similar recall bias, these results could open up a new and cost-effective way to collect more accurate data at scale.