checkIdent | R Documentation |
Checks four sufficient conditions for identifiability of a Gaussian DAG model with one latent variable.
checkIdent(amat, latent)
amat |
a square matrix with dimnames, representing the adjacency matrix of a DAG. |
latent |
an integer representing the latent variables among the nodes, or the name of the node. |
Stanghellini and Wermuth (2005) give some sufficient conditions for checking if a Gaussian model that factorizes according to a DAG is identified when there is one hidden node over which we marginalize. Specifically, the function checks the conditions of Theorem 1, (i) and (ii) and of Theorem 2 (i) and (ii).
a vector of length four, indicating if the model is identified
according to the conditions of theorems 1 and 2 in Stanghellini
& Wermuth (2005). The answer is TRUE
if the condition holds
and thus the model is globally identified or FALSE
if the
condition fails, and thus we do not know if the model is identifiable.
Giovanni M. Marchetti
Stanghellini, E. & Wermuth, N. (2005). On the identification of path-analysis models with one hidden variable. Biometrika, 92(2), 337-350.
isGident
, InducedGraphs
## See DAG in Figure 4 (a) in Stanghellini & Wermuth (2005)
d <- DAG(y1 ~ y3, y2 ~ y3 + y5, y3 ~ y4 + y5, y4 ~ y6)
checkIdent(d, "y3") # Identifiable
checkIdent(d, "y4") # Not identifiable?
## See DAG in Figure 5 (a) in Stanghellini & Wermuth (2005)
d <- DAG(y1 ~ y5+y4, y2 ~ y5+y4, y3 ~ y5+y4)
checkIdent(d, "y4") # Identifiable
checkIdent(d, "y5") # Identifiable
## A simple function to check identifiability for each node
is.ident <- function(amat){
### Check suff. conditions on each node of a DAG.
p <- nrow(amat)
## Degrees of freedom
df <- p*(p+1)/2 - p - sum(amat==1) - p + 1
if(df <= 0)
warning(paste("The degrees of freedom are ", df))
a <- rownames(amat)
for(i in a) {
b <- checkIdent(amat, latent=i)
if(TRUE %in% b)
cat("Node", i, names(b)[!is.na(b)], "\n")
else
cat("Unknown.\n")
}
}
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.