Statistical Diagnostics in Multilevel Models

Project leader

Funding source

Swedish Research Council - Vetenskapsrådet (VR)

Project Details

Start date: 01/01/2011
End date: 31/12/2013
Funding: 1572000 SEK


Multilevel modelling is a widely accepted tool for analyzing data which are organized into hierarchies or other complex classification structures. The overall purpose of the project is to extend existing statistical diagnostic tools and develop new ones for in multilevel models and provide data analysts with routine methods to assess the influence of observations on the model parameters and model fit. The project has the following specific aims: (i) to define single and multiple outliers and influential observations in multilevel models with respect to the dependent variable and covariates, respectively. (ii) to develop statistical diagnostics to detect outliers and influential observations defined in (i). (iii) To study properties of the statistical diagnostics obtained in (ii) and to demonstrate their performance using simulation studies and data from the National Social Insurance Agency (Försäkringskassan), NSIA. (iv) To write appropriate computer programs for the developed statistical diagnostic tools using such softwares as R and SAS. Nowadays the available statistical diagnostics are not adjusted to the context of multilevel models. Statistical diagnostics developed in this project should help practitioners to resolve problems connected to the identification of outliers and influential observations (single and multiple) within and between levels. Although, the proposed methodology will be applied to real data from NSIA, it can be used in many other applications.

Last updated on 2017-31-03 at 12:58