pcalg (version 1.0-0)

beta.special: Compute set of intervention effects

Description

This function computes a set of intervention effects of one variable onto another variable.

Usage

beta.special(dat=NA, x.pos, y.pos, verbose=0, a=0.01, myDAG=NA, myplot=FALSE, perfect=FALSE, method="local", collTest=TRUE, pcObj=NA, all.dags=NA, u2pd="rand")

Arguments

dat

data matrix

x.pos

Column of x in dat

y.pos

Column of y in dat

verbose

0=no comments, 2=detail on estimates

a

significance level of tests for finding CPDAG

myDAG

needed if true correlation matrix shall be computed

myplot

plot estimated graph

perfect

True cor matrix is calculated from myDAG

method

"local" - local (all combinations of parents in regr.); "global" - all DAGs

collTest

True - Exclude orientations of undirected edges that introduce a new collider

pcObj

Fit of PC Algorithm (CPDAG); if this is available, no new fit is done

all.dags

All DAGs in the format of function allDags; if this is available, no new function call allDags is done

u2pd

Function for converting udag to pdag;"rand": udag2pdag;"relaxed": udag2pdagRelaxed;"retry": udag2pdagSpecial

Value

• estimates of intervention effects

Details

After estimating a CPDAG, intervention effects of x on y are computed. For details, see references.

References

M.H. Maathuis, M. Kalisch, P. B"uhlmann (2009), Estimating high-dimensional intervention effects from observational data; Annals of Statistics, 2009.

See Also

pcAlgo,dag2cpdag,beta.special.pcObj

Examples

##############################
## Global
##############################
set.seed(125)
p <- 10
n <- 100000

myDAG <- randomDAG(p, prob = 0.4)
plot(dag2cpdag(myDAG))
cmat <- cov2cor(trueCov(myDAG))
d.mat <- rmvDAG(n, myDAG, errDist = "normal")

x.pos <- 1
y.pos <- 5
## true value
true.eff <- causalEffect(myDAG,y.pos,x.pos)
## value using oracle (true value should be element of this)
est.eff.p <- beta.special(x.pos=x.pos, y.pos=y.pos, myDAG=myDAG, perfect=TRUE, method="global")
## value using data (true value should come close to some element of this)
est.eff.e <- beta.special(dat=d.mat, x.pos=x.pos, y.pos=y.pos, perfect=FALSE, method="global", myplot=TRUE)

true.eff
est.eff.p
est.eff.e

## Setting perfect&global always contains exact value of true value
## OK

##############################
## Local
##############################
set.seed(125)
p <- 10
n <- 100000

myDAG <- randomDAG(p, prob = 0.4)
cmat <- cov2cor(trueCov(myDAG))
d.mat <- rmvDAG(n, myDAG, errDist = "normal")

x.pos <- 1
y.pos <- 5
## true value
true.eff <- causalEffect(myDAG,y.pos,x.pos)
## value using oracle (true value should be element of this)
est.eff.p <- beta.special(x.pos=x.pos, y.pos=y.pos, myDAG=myDAG, perfect=TRUE, method="local")
## value using data (true value should come close to some element of this)
est.eff.e <- beta.special(dat=d.mat, x.pos=x.pos, y.pos=y.pos, perfect=FALSE, method="local")
plot(pcAlgo(d.mat, alpha=0.01, directed=TRUE))

true.eff
est.eff.p
est.eff.e