In fact, they add this as Step 2b to the pseudo algorithm for CAT. In this experiment, the probability of being administered after being selected depends on the popularity of the item. In an extensive simulations study, the probabilities are set in such a way that for all items the expected exposure rate is smaller than the maximum exposure rate (generally set at max r = 0.20 or max r = 0.25 ) allowed. The more popular items have a small probability of passing this hurdle and the less popular items have a probability of being administered close to 1.