Test for d-separation in a DAG
Tests for d-separation of nodes in a DAG. dsepTest() is written to be easily used in skeleton, pc, fci.
dsepTest(x, y, S=NULL, suffStat)
x,y: (integer) position of variable and , respectively, in the adjacency matrix.
S: (integer) positions of zero or more conditioning variables in the adjacency matrix.
suffStat: a list with two elements,
"g": Containing the Directed Acyclic Graph (object of class "graph", see graph-class from the package graph), and"jp": Containing the shortest path distance matrix for all pairs of nodes as computed by johnson.all.pairs.sp from package RBGL.If x and y are d-separated by S in DAG G the result is 1, otherwise it is 0. This is analogous to the p-value of an ideal (without sampling error) conditional independence test on any distribution that is faithful to the DAG G.
The function is based on dsep. For details on d-separation see the reference Lauritzen (2004).
S.L. Lauritzen (2004), Graphical Models, Oxford University Press.
dsepAMTest for a similar function for MAGs. gaussCItest, disCItest and binCItest for similar functions for a conditional independence test for gaussian, discrete and binary variables, respectively.
Markus Kalisch (kalisch@stat.math.ethz.ch )
p <- 8 set.seed(45) myDAG <- randomDAG(p, prob = 0.3) if (require(Rgraphviz)) { ## plot the DAG plot(myDAG, main = "randomDAG(10, prob = 0.2)") } ## define sufficient statistics (d-separation oracle) suffStat <- list(g = myDAG, jp = RBGL::johnson.all.pairs.sp(myDAG)) dsepTest(1,6, S= NULL, suffStat) ## not d-separated dsepTest(1,6, S= 3, suffStat) ## not d-separated by node 3 dsepTest(1,6, S= c(3,4),suffStat) ## d-separated by node 3 and 4