pcalg (version 1.1-4)

idaFast: Multiset of Possible Total Causal Effects for Several Target Var.s

Description

This function estimates the multiset of possible total causal effects of one variable (x) on a several (i.e., a vector of) target variables (y) from observational data.

idaFast() is more efficient than looping over ida. Only method="local" (see ida) is available.

Usage

idaFast(x.pos, y.pos.set, mcov, graphEst)

Arguments

x.pos

Position of variable x in the covariance matrix

y.pos.set

Vector containing the position of the target variables y in the covariance matrix

mcov

Covariance matrix that was used to estimate graphEst

graphEst

Estimated CPDAG from the function pc. If the output of pc is pc.fit, then the estimated CPDAG can be obtained by pc.fit@graph.

Value

• Matrix with length(y.pos.set) rows. Row $i$ contains the multiset of estimated possible total causal effects of x on y.pos.set[i]. Note that all multisets in the matrix have the same length, since the parents of x are the same for all elements of y.pos.set.

Details

This function performs ida(x.pos, y.pos, mcov, graphEst, method="local", y.notparent=FALSE, verbose=FALSE) for all values of y.pos in y.pos.set simultaneously, in an efficient way. See the help file of ida for more details. Note that the option y.notparent = TRUE is not implemented, since it is not clear how to do that efficiently without orienting all edges away from y.pos.set at the same time, which seems not to be desirable. Suggestions are welcome.

References

M.H. Maathuis, M. Kalisch, P. Buehlmann (2009). Estimating high-dimensional intervention effects from observational data. Annals of Statistics 37, 3122--3164.

M.H. Maathuis, D. Colombo, M. Kalisch, P. Buehlmann (2010). Predicting causal effects in large-scale systems from observational data. Nature Methods, to appear.

See Also

pc for estimating a CPDAG, and ida for estimating the multiset of possible total causal effects from observational data on only one target variable but with many more options (than here in idaFast).

Examples

## Simulate the true DAG
set.seed(123)
p <- 7
myDAG <- randomDAG(p, prob = 0.2) ## true DAG
myCPDAG <- dag2cpdag(myDAG) ## true CPDAG
covTrue <- trueCov(myDAG) ## true covariance matrix

## simulate data from the true DAG
n <- 10000
dat <- rmvDAG(n, myDAG)

## estimate CPDAG (see help on the function "pc")
alpha <- 0.01
indepTest <- gaussCItest
suffStat <- list(C = cor(dat), n = n)
pc.fit <- pc(suffStat, indepTest, p, alpha)

if(require(Rgraphviz))
  plot(myDAG)

(eff.est1 <- ida(2,5,cov(dat),pc.fit@graph,method="local",verbose=FALSE))
(eff.est2 <- ida(2,6,cov(dat),pc.fit@graph,method="local",verbose=FALSE))
(eff.est3 <- ida(2,7,cov(dat),pc.fit@graph,method="local",verbose=FALSE))
## These three computations can be combinded in an efficient way
## by using idaFast :
(eff.estF <- idaFast(2,c(5,6,7),cov(dat),pc.fit@graph))