13.11.2012 Views

Introduction to Categorical Data Analysis

Introduction to Categorical Data Analysis

Introduction to Categorical Data Analysis

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

7.4 INDEPENDENCE GRAPHS AND COLLAPSIBILITY 223<br />

Such models focus on the effects of G and L on S and on I as well as the association<br />

between S and I.<br />

7.4 INDEPENDENCE GRAPHS AND COLLAPSIBILITY<br />

We next present a graphical representation for conditional independences in loglinear<br />

models. The graph indicates which pairs of variables are conditionally independent,<br />

given the others. This representation is helpful for revealing implications of models,<br />

such as determining when marginal and conditional odds ratios are identical.<br />

7.4.1 Independence Graphs<br />

An independence graph for a loglinear model has a set of vertices, each vertex representing<br />

a variable. There are as many vertices as dimensions of the contingency table.<br />

Any two vertices either are or are not connected by an edge. A missing edge between<br />

two vertices represents a conditional independence between the corresponding two<br />

variables.<br />

For example, for a four-way table, the loglinear model (WX, WY, WZ, YZ) lacks<br />

XY and XZ association terms. It assumes that X and Y are independent and that X and<br />

Z are independent, conditional on the other two variables. The independence graph<br />

portrays this model.<br />

Edges connect W with X, W with Y , W with Z, and Y with Z. These represent<br />

pairwise conditional associations. Edges do not connect X with Y or X with Z,<br />

because those pairs are conditionally independent.<br />

Two loglinear models that have the same conditional independences have the same<br />

independence graph. For instance, the independence graph just portrayed for model<br />

(WX, WY, WZ, YZ) is also the one for model (WX, WYZ) that also contains a threefac<strong>to</strong>r<br />

WYZ term.<br />

A path in an independence graph is a sequence of edges leading from one variable<br />

<strong>to</strong> another. Two variables X and Y are said <strong>to</strong> be separated by a subset of variables if<br />

all paths connecting X and Y intersect that subset. In the above graph, W separates X<br />

and Y , since any path connecting X with Y goes through W . The subset {W,Z} also<br />

separates X and Y . A fundamental result states that two variables are conditionally<br />

independent given any subset of variables that separates them. Thus, not only are X<br />

and Y conditionally independent given W and Z, but also given W alone. Similarly,<br />

X and Z are conditionally independent given W alone.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!