pwSEM (version 1.0.0)

view.paths: view.paths

Description

This is a function, usually called after pwSEM, to allow you to visually see how two variables in the DAG relate to each other along all directed paths from one to the other and to see how the 1st derivative of this relationship changes as the "from" variable changes. For linear relationships, this is a constant (the path coefficient).

Usage

view.paths(
  from,
  to,
  sem.functions,
  data,
  minimum.x = NULL,
  maximum.x = NULL,
  scale = "response",
  return.values = FALSE,
  dag
)

Value

Graphs are produced for each directed path from-->to and (return.values=TRUE) a data frame containing the values of the "from" variable and approxiate values of the 1st derivative of the function linking from-->to (i.e.the path coefficient) if the relationship is linear.

Arguments

from

The name (character) of the variable at the beginning of the path

to

The name (character) of the variable at the end of the path

sem.functions

A list containing the gam or gamm4 functions that define the structural equations. This is normally the sem.functions returned from pwSEM

data

The data frame containing the empirical data used in the SEM functions

minimum.x

The minimum value of the "from" variable; defaults to NULL, in which case the minimum and maximum values of the "from" variable in the data set are used.

maximum.x

The maximum value of the "to" variable; defaults to NULL, in which case the minimum and maximum values of the "from" variable in the data set are used.

scale

The chosen scale in which to express the results; either "response" (which uses the original scale of the variable) or "link" (which uses the scale of the link function for the "to" variable)

return.values

A logical value (TRUE returns the values of "from" and an estimate of the 1st derivative of the function for from-->to)

dag

The directed acyclic graph (DAG) used; usually this is the causal.graph object returned from a call to pwSEM

Examples

# Example with correlated endogenous errors, Poisson distributed variables
#and no nesting structure in the data
# DAG: X1->X2->X3->X4 and X2<->X4
my.list<-list(mgcv::gam(X1~1,data=sim_poisson.no.nesting,family=gaussian),
         mgcv::gam(X2~X1,data=sim_poisson.no.nesting,family=poisson),
         mgcv::gam(X3~X2,data=sim_poisson.no.nesting,family=poisson),
         mgcv::gam(X4~X3,data=sim_poisson.no.nesting,family=poisson))
out<-pwSEM(sem.functions=my.list,marginalized.latents=list(X4~~X2),
          data=sim_poisson.no.nesting,use.permutations = TRUE,n.perms=10000)
# To see each of the effects of X1 on X4 (only one in this example), we
# use view.paths() while imputing the list of SEM functions (out$sem.functions)
# and the DAG (out$causal.graph) that are output from the pwSEM function.

view.paths(from="X1",to="X4",sem.functions=out$sem.functions,data=
sim_poisson.no.nesting,scale="response",dag=out$causal.graph)