Adjusting for Measurement Error and Misclassification in Longitudinal Multilevel Models
Maria Eugénia Ferrão
Department of Mathematics & CEMAPRE, University of Beira Interior, Covilhã, Portugal

Measurement error is found in most of the variables used in social, behavioral and health sciences. Statistical analyses tend to omit it, and ignoring the measurement error can lead to biased estimates.

Statistical methods based on the MCMC algorithm were recently proposed to adjust for measurement errors in multilevel models with normally distributed predictor and response variables. The estimation of multilevel model parameters involving polynomial terms with errors-in-variables calls for a nonlinear framework.

We will present an extension of that MCMC algorithm involving a quadratic term. Value-added models applied to Portuguese data are used to illustrate the method. The survey design is longitudinal and consists of three waves - 2005/6, 2006/7 and 2007/8. Data were collected at the beginning and at the end of each academic year.

Keywords: Multilevel model; Measurement error; Longitudinal data; MCMC

Biography: Maria Eugénia Ferrão is Assistant Professor at University of Beira Interior and researcher at CEMAPRE, the Centre for Applied Mathematics and Economics, in Portugal. She got her Habilitation in Quantitative Methods (2008).