PLACE: 319 Snedecor

SPEAKER:
Mark Kaiser
Department of Statistics
Iowa State University

TITLE:
A Goodness of Fit Procedure for Spatial Markov Random Field Models
 

ABSTRACT:
In one sense, all spatial data sets are 'small', consisting of at most one complete observation of a random field.  Probability models are often formulated based on auxiliary knowledge rather than exploratory, nonparametric, or robust examination of observed data. Given these two characteristics of many spatial problems, assessing whether a formulated probability structure is appropriate to describe the observed spatial field is an acute problem.  I will examine a number of issues for spatial data that render the usual approaches to determining the goodness of fit of a statistical model difficult to implement, such as indeterminacy of large-scale and small-scale model components and definition of residual quantities.  The bulk of the talk will then focus on a proposed diagnostic for overall goodness of fit of Markov random field models.  This diagnostic is based on a conditional probability integral transform and versions appropriate for both continuous and discrete random variables will be presented.
 

COFFEE: 3:45 p.m., 104 Snedecor Hall