Quantifying household resilience with high frequency data: Temporal dynamics and methodological options

Resilience as a metric is of growing interest to development researchers and practitioners, particularly for those whose work concerns the effects of climate change, conflict and epidemics. The growing need for resilience measurements motivates this research, measurements that reflect the complex, dynamic features of welfare among populations living in shock-prone contexts. It presents insights from three measurement approaches to explore the dynamic, intra-annual effects of shocks on household well-being, mediated by household characteristics; one based on shock persistence, one based on the stochastic distribution of well-being, and one driven by machine learning algorithms and based on predictive power. The insights gained from the comparison of measurement approaches offer a fuller understanding of the factors driving resilience than any single approach on its own. The paper harnesses a novel data-set from the ‘Measuring Indicators for Resilience Analysis’ project, which, each month, tracks shocks and food security indicators for one year in a highly food insecure population in Malawi. Across approaches, our study consistently finds that shocks and food insecurity are very persistent, and that households living in the flood plains are more resilient. When focusing on specific shocks such as illness, the gender of the household head matters as-well. A broader search for predictors of food insecurity using LASSO and Random Forest algorithms uncovers other characteristics that affect resilience, such as the distance to drinking-water. The paper demonstrates how multiple complementary approaches can identify different but overlapping factors related to resilience, painting a fuller, richer picture. It also demonstrates the empirical benefits derived from using high-frequency data sets in the study of resilience.