Definition 1.1 has a multivariate Gaussian distribution with parameters and if the joint density is where . Note that we usually consider or
Theorem 1.1 Let has a multivariate Gaussian distribution with concentration matrix . Then iff , where is the corresponding entry in the concentration matrix.
2. Gaussian Graphical Model (GGM)
If is a matrix whose rows and columns are indexed by , we write to indicate the matrix indexed by (i.e., it has rows and columns) whose -entries are and with zeroes elsewhere. For example, if then where is used as an abbreviation for in the subscript.
Lemma 2.1 Let be a graph with decomposition , and . Then is Markov w.r.t. iff and and are Markov w.r.t. and respectively.
Proof. Since is a decomposition and is Markov, then , which implies for all , Then i.e., Therefore,
The converse holds similarly.
3. Maximum Likelihood Estimation
Let . The sufficient statistic for is the sample covariance matrix In addition, is the MLE for under the unrestricted model (i.e., when all edges are present in the graph). Let denote the MLE for under the restriction that the distribution satisfies the Markov property for , and denote the inverse of .
For a decomposable graph with cliques , the MLE can be written in the form
Example 3.1 Whittaker (1990) analyzed data on five maths test results administered to 88 students, in analysis, algebra, vectors, mechanics and statistics. Some of the entries in the concentration matrix are quite small, suggesting that conditional independence holds. We want to fit a graphical model and check if it gives an excellent fit.
Figure 1: A graph for the maths test data.
By computation, is:
Mechanics
Vectors
Algebra
Analysis
Statistics
Mechanics
Vectors
Algebra
Analysis
Statistics
We carry out a likelihood ratio test to see whether this model is a good fit to the data. We want to test The test statistic is and the -value is , i.e., the model is a good fit.
Here is the relevant code in R:
library(ggm) data(marks)
# MLE of the covariance matrix under the unrestricted model. S = cov(marks)