Published in The Review of Economic Studies in 2005, this paper used household survey data from rural Tanzania to ask a question that belief alone cannot answer: not why people accuse women of witchcraft, but why an accusation turns into a killing when it does.
The finding was that extreme rainfall, drought and flood alike, and the income shock that follows, predicts a rise in the killing of women accused of witchcraft in the villages surveyed. Elderly women in a household were disproportionately the victims. Rainfall is the useful variable precisely because it is outside anyone's control: a village cannot choose its weather, so a correlation between weather and killings is difficult to explain by any prior difference between villages.
The argument that follows is about scarcity governing the moment of action. Accusation is widespread and constant; killing is not. What the paper identifies as varying with the shock is the point at which a household under acute economic pressure acts on an accusation against a dependent member. That reframes the phenomenon from a question about what people believe to a question about the conditions under which belief becomes violence, and it is the study the atlas rests on when it makes that claim.