A walk through random fields: robust estimation and small sample asymptotics
Prof. Davide La Vecchia, Research Center for Statistics and Geneva School of Economics and Management, University of Geneva
The talk is a gentle introduction to the statistical analysis of random fields on a lattice, with focus on some inferential issues. I organize the presentation in three parts.
In the first part, I discuss the problem of conducting reliable inference on large datasets, where some of the observations are anomalous records (say, outliers). To motivate and illustrate the topic, I consider the statistical analysis of functional magnetic resonance imaging (fMRI) data, where the data analysis involves a large number of records (about 1.65X10^9 observations) which may contain some outliers---e.g. due to patient movements and/or scanner failures. I illustrate how robust techniques are able to detect the outliers and to down-weight their impact on the final estimates.
In the second part, I discuss the problem of conducting inference on small datasets, where the limited number of observations makes the use of standard asymptotics (which implicitly assumes that the sample size is large) questionable. To motivate and illustrate the topic, I consider a macroeconomic spatial panel data model, where 14 yearly observations for 24 countries belonging to the Organisation for Economic Co-operation and Development (OECD) are available. With this setting, I illustrate how small sample techniques can be derived and implemented to improve the accuracy of the routinely-applied asymptotics.
In the third part, I present a topic for future research and potential collaboration with the Department of Physics. To motivate the investigation, I consider a data set made available by the US National Oceanic and Atmospheric Administration (NOAA). The aim is to define a model for the time-varying dynamics of maximal temperatures recorded in several meteo stations.
GAP, Département de physique Appliquée
22 ch. de Pinchat, 1227 Carouge
Seminar room 3