mgm (version 1.1-7)

rs.mgm:

Description

Estimates MGMs/mixed VAR models on a number of resampled datasets. Can be used as input to summary() and plot() to investigate parameter-stability.

Usage

rs.mgm(..., B = 100, N = NULL, replace = TRUE, pbar = TRUE)

Arguments

…

Arguments as in mgmfit or var.mgm.

B

Number of resampling iterations.

N

Size of each resample.

replace

Subsampling with (bootstrap) or without replacement.

pbar

Shows a progress bar if TRUE

Value

Returns a list with

models

A list with B entries containing the mgmfit() or var.mgm() output for each of the B bootstrap samples.

B_wadj

A p x p x B array containing the estimated weighted adjacency matrix for each of the B bootstrap samples.

B_samples

A list with B entries containing n x 1 indicator vectors, defining the used bootstrap samples.

call

Returns the function call.

Details

Returns B mixed graphical models, each fitted on one bootstrap sample. Use the summary() and plot() function to get information about the stability of (a selection of) edges. If N = nrow(data) and replace = TRUE the resampling scheme is the nonparametric bootstrap.

Examples

## Not run: ------------------------------------
# 
# # Fit Mixed Graphical Model to Bootstrap Samples of the autism dataset
# rs_obj <- rs.mgm(data = autism_data$data, 
#                  type = autism_data$type,
#                  lev = autism_data$lev,
#                  VAR = FALSE, # fit mgm model
#                  B = 10,
#                  N = nrow(autism_data$data), # for classical bootstrap
#                  replace=TRUE)
# 
# # Summary of bootstrapped edge weights
# summary(rs_obj)
# 
# # Plot bootstrapped edge weights using boxplots
# plot(rs_obj)
# 
# # Pairwise scatter plots between bootstrapped edge weights to detect trade-offs
# # ... in the case of correlated predictors (negative correlation between weights)
# 
# ind <- unlist(rs_obj$edgeNonZero) # only consider edges with nonzero estimates
# m_bweights <- do.call(cbind, rs_obj$edgeWeights)[, ind] # B x k matrix, with k edges
# colnames(m_bweights) <- unlist(rs_obj$edgeNames)[ind] # add edge names
# 
# library(psych) # psych package for pairs
# pairs.panels(m_bweights)
# # (recommended to print a PDF, because this can be a huge plot:
# # pdf('filename.pdf', 20, 20))
# # ... plotting
# # dev.off()
# 
# 
# 
## ---------------------------------------------