Dimitrios Vagenas1, Vasiliki Totsika2. 1. Institute of Health and Biomedical Innovation, Queensland University of Technology, Australia. Electronic address: dimitrios.vagenas@qut.edu.au. 2. Centre for Educational Development, Appraisal, and Research (CEDAR) and Centre for Education Studies (CES), University of Warwick, UK; Centre for Developmental Psychiatry and Psychology, Department of Psychiatry, Monash University, Australia.
Abstract
BACKGROUND: The use of Multilevel Models (MLM) and Generalized Estimating Equations (GEE) for analysing clustered data in the field of intellectual and developmental disability (IDD) research is still limited. METHOD: We present some important features of MLMs and GEEs: main function, assumptions, model specification and estimators, sample size and power. We provide an overview of the ways MLMs and GEEs have been used in IDD research. RESULTS: While MLMs and GEEs are both appropriate for longitudinal and/or clustered data, they differ in the assumptions they impose on the data, and the inferences made. Estimators in MLMs require appropriate model specification, while GEEs are more resilient to misspecification at the expense of model complexity. Studies on sample size seem to suggest that Level 1 coefficients are robust to small samples/clusters, with any higher-level coefficients less so. MLMs have been used more frequently than GEEs in IDD research, especially for fitting developmental trajectories. CONCLUSIONS: Clustered data from research in the IDD field can be analysed flexibly using MLMs and GEEs. These models would be more widely used if journals required the inclusion of technical specification detail, simulation studies examined power for IDD study characteristics, and researchers developed core skills during basic studies.
BACKGROUND: The use of Multilevel Models (MLM) and Generalized Estimating Equations (GEE) for analysing clustered data in the field of intellectual and developmental disability (IDD) research is still limited. METHOD: We present some important features of MLMs and GEEs: main function, assumptions, model specification and estimators, sample size and power. We provide an overview of the ways MLMs and GEEs have been used in IDD research. RESULTS: While MLMs and GEEs are both appropriate for longitudinal and/or clustered data, they differ in the assumptions they impose on the data, and the inferences made. Estimators in MLMs require appropriate model specification, while GEEs are more resilient to misspecification at the expense of model complexity. Studies on sample size seem to suggest that Level 1 coefficients are robust to small samples/clusters, with any higher-level coefficients less so. MLMs have been used more frequently than GEEs in IDD research, especially for fitting developmental trajectories. CONCLUSIONS: Clustered data from research in the IDD field can be analysed flexibly using MLMs and GEEs. These models would be more widely used if journals required the inclusion of technical specification detail, simulation studies examined power for IDD study characteristics, and researchers developed core skills during basic studies.