Linear-cost unbiased estimator for large crossed random effect models via couplings

In the following we show how it is possible to obtain parallelizable, unbiased and computationally cheap estimates of Crossed random effects models with a linear cost in the number of datapoints (and paramaters) exploiting couplings.

Crossed random effects models (CREM)
CREM model a continuous response variables (Y) as depending on …

<img src=’ https://youngstats.github.io/post/2023-09-27-linear-cost-unbiased-estimator-for-large-crossed-random-ef …