Prof. Anders Ahlbom
Institute of Environmental Health
Special Courses in Statistics:
Department of Statistics and Quantitative Methods
PhD program in Statistics and Mathematical Financ
The main objective of this course is to introduce attendees to the most important methods for analyzing clustered categorical data. Such data are common in practice: in longitudinal studies (for which each cluster is a set of observations on a person over time), with survey data that have multivariate responses such as several variations of the same question with the same outcome categories, and with sampling methods that use clustering, such as samples that produce multilevel data. The course’s main emphasis is on introducing appropriate models and their interpretations, emphasizing generalizations of logistic regression. The course will show examples of the use of R and SAS for performing the analyses. Through examples, the attendees will learn how to use the models and weigh the advantages and disadvantages of the various model types.
Models for Matched Pairs (Comparing dependent proportions, McNemar test and generalizations, conditional vs marginal models for binary matched pairs, comparing margins of square contingency tables) Marginal Models (Marginal logit models for repeated binary response, maximum likelihood (ML) and its limitations, generalized estimating equations (GEE) approach, cumulative logit modeling of repeated ordinal responses)
GLMs with Random Efects (conditional logistic regression of clustered binary data, generalized linear mixed models (GLMMs), ML tting and inference, logistic GLMMs for clustered binary data) Additional topics about Mixture Models (GLMMs for clustered ordinal data, random intercepts and random slopes, beta-binomial model for clustered binary data, multilevel models).
Il giorno 4 ottobre alle ore 11.30 presso la Sala del Consiglio della Scuola di Economia e Statistica, al IV piano dell’edificio U7, il prof. Ruey Tsay della University of Chicago Booth School of Business, terrà un seminario su:
Nonlinear Models for High-Frequency Financial Data
Nonlinearity is commonly observed in high-frequency data, including financial data. We discuss some recent developments in econometric modeling of nonlinear high-frequency financial data. We discuss both parametric and nonparametric methods, approaches for handling count data, and statistical methods for analyzing big dependent data. Real examples are used to demonstrate the analysis and to compare different methods.
Tutti gli interessati sono invitati a partecipare. Per ulteriori informazioni:
Il giorno martedì 4 ottobre alle ore 15.00 presso la Aula Seminari al IV piano dell’edificio U7, il prof. Masanobu Taniguchi della Waseda University di Tokyo terrà un seminario su
High Order Asymptotic Theory of Shrinkage Estimation for General Statistical Models
In this paper we develop the high order asymptotic theory of shrinkage estimators for general statistical models, which include dependent processes, multivariate models and regression models, i.e., non-i.i.d. models.
Introducing a shrinkage estimator of MLE, we compare it with that of MLE by third-order mean squares error (MSE).
A sufficient condition for the shrinkage estimator to improve the MLE will be given in a very general fashion.
Our model is described as a curved statistical model p(·;\theta(u)), where \theta is a parameter of larger model and u is a parameter of interest with dim u < dim \theta.
This setting is especially suitable for estimation of portfolio coefficients u based on mean and variance parameters \theta.
We also mention the advantage of our shrinkage estimators when the dimension of parameter becomes large.
Numerical studies are given, and illuminate an interesting feature of the shrinkage estimator. (joint work with: Hiroshi SHIRAISHI, Yoshihiro SUTO, Takashi YAMASHITA)