RCT proves masks work
-
Do the authors cluster at the village level or no?
The data are aggregated to a village level resolution and robustified. So in a way, yes. However, this is an issue because the research design says that the authors sample the data by cluster-pair. This means that the village data points are not independenly sampled and thus the standard errors need to be adjusted to account for clustering at the cluster-pair level.
-
Reading the published paper in Science now. It contains a map that clearly shows how the cluster design works. The villages are sampled in pairs and each pair is a next door neighbour. Clearly, these data are clustered and the standard errors should be adjusted in the usual way.
The "are these SEs clustered?" is the standard seminar bore question. However, there's a reason for this. Ensuring that our estimates of precision are correct is central to statistical inference.
Someone please tell me I am wrong here and there's a good reason why the standard errors are valid. If not, how on earth did a paper with such an obvious error get published in Science?!?