Seminars for 2016/17

02 May 2013: Evaluation of institution performance in 2-D (Summer Term 2012/13)

Dr Aris Perperoglou from Department of Mathematical Sciences, University of Essex

At 14:00 in 6.314.

A variety of methods have been suggested for modelling institutional performance on cross-sectional data. Most of the suggested approaches are based on random effects models that account for variation between institutions.  In this work we start with a Penalized Regression with Individual Deviance Effects model (PRIDE)  (Perperoglou and Eilers, 2012) to evaluate institutional performance. This penalized approach may include covariates and adjust performance according to several risk factors. All remaining extra variation   is modelled by the deviance effects, which can be used to reveal which institutions are within control at the given time point. However, when data come from a longitudinal study, then one has to adjust also for variation within institutions.  In this work we will extend PRIDE models to fit data in two dimensions. A two-dimensional PRIDE model is a way of smoothing data that come into grids while at the same time account for extra variation that may be present.  A matrix of deviance effects is introduced into the model which is used to compute the effects of different institutions across time. To control the shrinkage of the estimates a penalty is subtracted from the likelihood. For estimation an efficient algorithm based on tensor products will be used. We will illustrate the methods and present an application to data from a longitudinal study evaluating performance of gynaecological practices in the Netherlands. 

This event is open to the general public.

<< Back to seminar listing