select.explore: Graph selection for 'explore' Objects

In BGGM: Bayesian Gaussian Graphical Models

Graph selection for explore Objects

Description

Provides the selected graph based on the Bayes factor \insertCiteWilliams2019_bfBGGM.

Usage

## S3 method for class 'explore'
select(object, BF_cut = 3, alternative = "two.sided", ...)

Arguments

object	An object of class explore.default
BF_cut	Numeric. Threshold for including an edge (defaults to 3).
alternative	A character string specifying the alternative hypothesis. It must be one of "two.sided" (default), "greater", "less", or "exhaustive". See note for further details.
...	Currently ignored.

Details

Exhaustive provides the posterior hypothesis probabilities for a positive, negative, or null relation \insertCite@see Table 3 in @Williams2019_bfBGGM.

Value

The returned object of class select.explore contains a lot of information that is used for printing and plotting the results. For users of BGGM, the following are the useful objects:

alternative = "two.sided"

• pcor_mat_zero Selected partial correlation matrix (weighted adjacency).

• pcor_mat Partial correlation matrix (posterior mean).

• Adj_10 Adjacency matrix for the selected edges.

• Adj_01 Adjacency matrix for which there was evidence for the null hypothesis.

alternative = "greater" and "less"

• pcor_mat_zero Selected partial correlation matrix (weighted adjacency).

• pcor_mat Partial correlation matrix (posterior mean).

• Adj_20 Adjacency matrix for the selected edges.

• Adj_02 Adjacency matrix for which there was evidence for the null hypothesis (see note).

alternative = "exhaustive"

• post_prob A data frame that included the posterior hypothesis probabilities.

• neg_mat Adjacency matrix for which there was evidence for negative edges.

• pos_mat Adjacency matrix for which there was evidence for positive edges.

• neg_mat Adjacency matrix for which there was evidence for the null hypothesis (see note).

• pcor_mat Partial correlation matrix (posterior mean). The weighted adjacency matrices can be computed by multiplying pcor_mat with an adjacency matrix.

Note

Care must be taken with the options alternative = "less" and alternative = "greater". This is because the full parameter space is not included, such, for alternative = "greater", there can be evidence for the "null" when the relation is negative. This inference is correct: the null model better predicted the data than the positive model. But note this is relative and does not provide absolute evidence for the null hypothesis.

References

\insertAllCited

See Also

explore and ggm_compare_explore for several examples.

Examples

#################
### example 1 ###
#################

#  data
Y <- bfi[,1:10]

# fit model
fit <- explore(Y, progress = FALSE)

# edge set
E <- select(fit,
            alternative = "exhaustive")

BGGM documentation built on Sept. 11, 2024, 5:19 p.m.