The Strategies Of Building Mixed Model And Application For The AIDS Clinical Data

Mixed effects model has become more and more popular, especially at the aspectof the analysis of longitudinal data. One difficult challenge is how to build goodenough mixed effects models. In this paper, we come up with a systematic strategyfor this challenge and introduce easily implemented practical advice to build mixedeffects models. General scientific strategies motivate the recommended five-stepprocedure for model fitting. Model needs both the mean structure (the fixed effects)and the covariance structure (the random effects and residual error) and creates thefundamental flexibility and complexity. Some very practical recommendations help toconquer the complexity. Centering, scaling and full-rank coding of all the predictorvariables improve the convergence, computing speed, and numerical accuracy.Applying computational and assumption diagnostics from univariate linear models tomixed model data helps to detect and solve the related computational problems.Applying computational and assumption diagnostics from the univariate linearmodels to the mixed model data can radically improve the convergence, computingspeed, and numerical accuracy. The approach helps to fit more general covariancemodels, a crucial step in selecting a credible covariance model needed for defensibleinference. A detailed demonstration of the recommended strategy is based on datafrom a hospital for the analysis of the treatments for the HIV/AIDS. The discussionhighlights the need for improving how scientists and statisticians teach and review theprocess of finding a good enough mixed model.