April 01, 2022 - A robust estimate of the probability of agreement between two measurement systems using P-splines - Speaker: Felipe Osorio (Universidad Técnica Federico Santa María)
UFSCar/USP joint Seminar
Announcement file.
Scheduled for:
Apr 01, 2022, at 2:00 pm
(GMT-03:00) Brasilia Standard Time - Sao Paulo
Video call link here.
Speaker:
Felipe Osorio (Universidad Técnica Federico Santa María)
Title:
A robust estimate of the probability of agreement between two measurement systems using P-splines
Abstract:
Assessing agreement between two measurement instruments which varies in pricing, fastness and other features, has been of growing interest in several areas of knowledge. Some methods assume that the differences of the paired measurements come from a normal distribution. In this work, we develop the probability of agreement to compare measurement systems using a semiparametric approach. Because the assumption of normality of these differences, is unrealistic in many situations, we focus on the use of penalized splines considering the scale mixture of normal distributions to model non-normal measurements. Based on asymptotic theory for penalized maximum likelihood estimators a method for constructing a confidence band for the probability of agreement is provided. This class of models is flexible and allows to model datasets with high levels of kurtosis, and potential outliers.
Bio:
Felipe Osorio received the degree in statistical engineering from the Universidad de Valparaíso, Chile, in 2001, and the D.Sc. degree in statistics from the Universidade de São Paulo, Brazil, in 2006. He is currently an Assistant Professor with the Department of Mathematics, UTFSM, Chile. He is the coauthor of the book Spatial Relationships Between Two Georeferenced Variables With Applications in R (Springer, in 2020). His research interests include models for data with longitudinal structure (mixed-effects models, GEE) and diagnostic methods and the computational implementation of such techniques. He is the creator and the maintainer of several packages for the R statistical environment.