Events and Meetings of Italian Statistical Society, Advances in Latent Variables - Methods, Models and Applications

Font Size: 
Regression to the mean
Annibale Biggeri, Dolores Catelan, Michela Baccini

Last modified: 2013-06-14

Abstract


The regression to the mean effect has been described in several contexts but still it continues to emerge. It is a kind of  selection bias and it is the consequence of measurement error. It was described in clinical studies and epidemiological investigations whenever a selection of high/low responders is  part of the study design; second it is present when baseline  measurement is considered as confounder of covariate of interests. The general setting of the problem can be formalized via a latent variable and more than one imperfect measurements. Generalized linear mixed models are proposed. We present a comprehensive formulation of the problem and a simple explorative analysis using the correlation coefficients (Pearson’s, Lin’s and Bland-Altman’s mean-difference correlation).

Full Text: PDF