Informative priors results in a more accurate item parameter recovery for both
the parameters. Especially for medium ( a∈[0.60,1.00]) and high ( 1.00) a >
discriminating items, the use of empirical priors increased measurement precision
considerably. The average Root Mean Squared Error (RMSE) was reduced by more
than fifty percent. This is especially important for CAT, because high discriminating
items are selected more often to be administered. For the difficulty parameter,
the effects were less drastic. Only for the very easy items ( 1.00) b < − , the use of
empirical priors reduced RMSE more than fifty percent.
The lesson learned from this example is that one might reduce the sample size
considerably when empirical priors are used, without any loss in measurement
precision. This is one way to reduce the costs of CAT.