Our new paper develops and tests a network-based method for estimating death rates in complex humanitarian emergencies. Joint w/ @dennisfeehan.bsky.social and team from Geneva-based NGO @impact-initiatives.bsky.social academic.oup.com/aje/advance-...

Reliable estimates of death rates in complex humanitarian emergencies are critical for assessing the severity of a crisis and for allocating resources. However, in many humanitarian settings, logistical and security concerns make conventional methods for estimating death rates infeasible.

We adapt the network survival method to the unique setting of humanitarian emergencies. The idea behind the network survival method is that respondents can report on 1) the size of their social network and 2) deaths in their social networks. This info can be aggregated to estimate death rates.

We collected original data in Tanganyika Province, Democratic Republic of the Congo, a realistic setting where such emergencies have happened in the past. Study design involved two data collection efforts: a quota sample possible in a humanitarian emergency and probability comparator sample.

— Quota sample: sampled respondents from 3 health zones at major transit hubs (e.g., markets, taxi stops, ports, health clinics, and footpaths) who were coming into Kalemie City (N = 2,526) — Probability sample: sampled respondents probabilistically across 3 health zones in their HH (N = 2,785)

We did several weeks of in-person fieldwork (cognitive interviews, focus groups, etc.) to better understand the types of social networks people could report accurately on in this setting.

Qualitative fieldwork suggested testing two different types of personal networks as the basis for estimates: deaths among immediate neighbors and deaths among kin.

An advantage of the network approach is that each interview provides mortality information on many more people than a typical mortality survey that asks respondents only about their household or siblings.

To account for selection into our quota sample, we constructed survey weights under several scenarios reflecting different hypothetical levels of auxiliary data availability for weighting targets.

From the quota sample, our estimates based on different types of respondent reports (i.e., reports on neighbors, kin, and household) produce similar and plausible crude death rate estimates. As we reweighted to account for selection into the quota sample, the estimated death rates increased.

This new method can be used to monitor trends in crude death rates over time.

This design could be deployed remotely in settings where operational constraints prevent humanitarian groups from reaching insecure areas, meaning it could potentially be applied to estimate death rates in a wide range of humanitarian emergencies.