Pooling

#statistics

In McElreath's literature, he talks about pooling of information which is a benefit of multilevel models.

First, there are usually some parameters, such as σ, that the model says do not depend in any way upon an African identity for each nation. By splitting the data table, you are hurting the accuracy of the estimates for these parameters, because you are essentially making two less-accurate estimates instead of pooling all of the evidence into one estimate. In effect, you have accidentally assumed that variance differs between African and non-African nations. Now, there’s nothing wrong with that sort of assumption. But you want to avoid accidental assumptions.