Designs for generalized linear models with random block effects via information matrix approximations

摘要: SUMMARY The selection of optimal designs for generalized linear mixed models is complicated by the fact thattheFisherinformationmatrix,onwhichmostoptimalitycriteriadepend,iscomputationally expensive to evaluate. We provide two novel approximations that reduce computational cost evaluating information matrix complete enumeration response outcomes, or Monte Carlo thereof: an asymptotic approximation accurate when there strong dependence between observations in same block; and via kriging interpolators. For logistic random intercept models, we show how interpolation can be especially effective finding pseudo-Bayesian incorporate uncertainty values model parameters. new results are used evaluate efficiency, estimating conditional from closed-form derived marginal models. Correcting attenuation parameters binary-response yields much improved designs, typically with very high efficiencies. However, some experiments exhibiting dependence, may still inefficient forconditionalmodelling.Ourasymptoticresultsprovidesometheoreticalinsightsintowhysuch inefficiencies occur.