Graphical Modeling of Spatial Health Data

Authored by: Adrian Dobra

Handbook of Spatial Epidemiology

Print publication date:  April  2016
Online publication date:  April  2016

Print ISBN: 9781482253016
eBook ISBN: 9781482253023
Adobe ISBN:


 Download Chapter



Graphical models (Whittaker, 1990; Lauritzen, 1996) that encode multivariate independence and conditional independence relationships among observed variables X = (X 1, …, Xp ) have a widespread use in major scientific areas (e.g., biomedical and social sciences). In particular, a Gaussian graphical model (GGM) is obtained by setting off-diagonal elements of the precision matrix K = Σ −1 to zero of a p-dimensional multivariate normal model (Dempster, 1972). Employing a GGM instead of a multivariate normal model leads to a significant reduction in the number of parameters that need to be estimated if most elements of K are constrained to be zero and p is large. A pattern of zero constraints in K can be recorded as an undirected graph G = (V, E), where the set of vertices V = {1, 2, …, p} represent observed variables, while the set of edges EV × V link all the pairs of vertices that correspond to off-diagonal elements of K that have not been set to zero. The absence of an edge between X v 1 and X v 2 corresponds with the conditional independence of these two random variables given the rest and is denoted by X u 1 ╨   X u 2   |   X V \ { v 1 , v 2 } (Wermuth, 1976). This is called the pairwise Markov property relative to G, which in turn implies the local as well as the global Markov properties relative to G (Lauritzen, 1996). The local Markov property plays a key role since it gives the regression model induced by G on each variable Xv . More explicitly, consider the neighbors of v in G, that is, the set of vertices v′V such that (v, v′) ∈ E. We denote this set by bd G (v). The local Markov property relative to G says that X v ╨   X V \ { { v } ∪ bd G ( v ) } |   X bd G ( v ) . This is precisely the statement we make when we drop the variables { X v ′   :   v ′   :   ∈ V \ b d G ( v ) } from the regression of Xv on { X v ′   : v ′   ∈ V \ { v } } .

Search for more...
Back to top

Use of cookies on this website

We are using cookies to provide statistics that help us give you the best experience of our site. You can find out more in our Privacy Policy. By continuing to use the site you are agreeing to our use of cookies.