[Forum SIS] Avviso di seminario - "Model-based recursive partitioning for preference data" - Università di Brescia
Marica Manisera
marica.manisera a unibs.it
Mar 22 Set 2015 10:13:38 CEST
*/Lunedì 5 ottobre 2015/,/alle ore 15.00/*//
*/
Thomas Rusch
/*
Competence Center for Empirical Research Methods
Wirtschaftsuniversität Wien
Vienna University of Economics and Business
terrà il seminario dal titolo:
*/Model-based recursive partitioning for preference data/*
//
presso il Dipartimento di Economia e Management dell’Università degli
Studi di Brescia
Sala della Biblioteca, sede di San Faustino, via San Faustino 74/b
(iniziativa finanziata dal Fondo di Ateneo per attività a carattere
internazionale).//
*/Abstract/*
For the analysis of preference data, the family of discrete choice
models became the de facto standard statistical modelling approach.
Today's challenging data sets, however, often call for more flexible
procedures. To that end, we present an approach that combines logit
models with recursive partitioning, coined LORET (Logistic Regression
Trees). Recursive partitioning algorithms separate a feature space into
a set of disjoint segments. In our proposal, a separate pre-specified
logit model is fitted into each of these segments. This has a number of
advantages over any of the separate methods: a) it off ers enhanced
interpretability of classical trees, b) provides an exploratory way to
assess variables, c) cuts the data set into segments based on di fferent
parameters of the logit model, and d) provides more stable predictions
than from classical trees and more flexible predictions as compared to
logit and other discrete choice models. In this talk I will discuss the
general idea of recursive partitioning of discrete choice models,
present an algorithm to fit them, and explain the strategy of
implementing it in R. I will discuss matters of existence and uniqueness
of the parameter estimation, how we incorporate this into our
implementation and present a way to assess the stability of the
segmentation. Furthermore, I will provide an outlook on how to
incorporate CUB models into the framework. The methodological aspects
will be supplemented and illustrated by a data set from a discrete
choice experiment in marketing.
//
/Tutti gli interessati sono cordialmente invitati a partecipare.
/
--
Informativa sulla Privacy: http://www.unibs.it/node/8155
-------------- parte successiva --------------
Un allegato HTML è stato rimosso...
URL: <http://www.stat.unipg.it/pipermail/sis/attachments/20150922/a276b3fa/attachment-0001.html>
Maggiori informazioni sulla lista
Sis