Prediction, Bootstrapping and Monte Carlo Analyses Based on Linear Mixed Models with QAPE 2.0 Package

Abstract

The paper presents a new R package qape for prediction, accuracy estimation of various predictors and Monte Carlo simulation studies of properties of both predictors and estimators of accuracy measures. It allows to predict any population and subpopulation characteristics of the response variable based on the Linear Mixed Model (LMM). The response variable can be transformed, e.g. to logarithm and the data can be in the cross-sectional or longitudinal framework. Three bootstrap algorithms are developed: parametric, residual and double, allowing to estimate the prediction accuracy. Analyses can also include Monte Carlo simulation studies of properties of the methods used. Unlike other packages, in the prediction process the user can flexibly define the predictor, the model, the transformation function of the response variable, the predicted characteristics and the method of accuracy estimation.