predictability function

Predictability: Bayesian Variance Explained (R2)

Compute nodewise predictability or Bayesian variance explained \insertCite @R2 @gelman_r2_2019BGGM. In the context of GGMs, this method was described in \insertCite Williams2019;textualBGGM.

predictability(
  object,
  select = FALSE,
  cred = 0.95,
  BF_cut = 3,
  iter = NULL,
  progress = TRUE,
  ...
)

Arguments

• object: object of class estimate or explore
• select: logical. Should the graph be selected ? The default is currently FALSE.
• cred: numeric. credible interval between 0 and 1 (default is 0.95) that is used for selecting the graph.
• BF_cut: numeric. evidentiary threshold (default is 3).
• iter: interger. iterations (posterior samples) used for computing R2.
• progress: Logical. Should a progress bar be included (defaults to TRUE) ?
• ...: currently ignored.

Returns

An object of classes bayes_R2 and metric, including

• scores A list containing the posterior samples of R2. The is one element

  for each node.

Note

Binary and Ordinal Data :

R2 is computed from the latent data.

Mixed Data :

The mixed data approach is somewhat ad-hoc \insertCite @see for example p. 277 in @hoff2007extending;textualBGGM. This is becaue uncertainty in the ranks is not incorporated, which means that variance explained is computed from the 'empirical' CDF.

Model Selection :

Currently the default to include all nodes in the model when computing R2. This can be changed (i.e., select = TRUE), which then sets those edges not detected to zero. This is accomplished by subsetting the correlation matrix according to each neighborhood of relations.

Examples

# data
Y <- ptsd[,1:5]

fit <- estimate(Y, iter = 250, progress = FALSE)

r2 <- predictability(fit, select = TRUE,
                     iter = 250, progress = FALSE)

# summary
r2

References

\insertAllCited