可分析連續，分類與個數三種不同資料型態的R(個序列)與C(個時間點)的交叉設計的強韌概似推論法(3/3)

Typical and popular means for analyzing crossover design data include Bayesian and random effects approaches. The semiparametric generalized estimating equations method is also widely used. The former two methods face one of the well-known disadvantages of being too subjective in choosing the prior distributions and in choosing the distributions for the random effects. The shortcomings also include multi-dimensional integration and other computational difficulties as well. The generalized estimating equations approach only provides the Wald statistic for inference that, unfortunately, is not invariant under parameter reparameterization. There are situations where the correlation is only a nuisance to the inference of interest. Under the circumstances, one does not need to model the nuisance correlation parametrically. We propose to employ the model for the parallel design that assumes independence as the working model. We convert the working likelihood function to become robust that can deliver legitimate likelihood inference for comparing treatment effects for continuous,categorical, and count data for general crossover designs with R sequences and C periods.