Build a Bayesian VAR(1) network (unregularized, Mplus-targeted)
Source:R/var_bayes.R
fit_var_bayes.RdNative, pure-R Bayesian VAR(1) that reproduces Mplus's Bayesian
(DSEM/time-series) estimates without needing Mplus. It is the unregularized
Bayesian counterpart of fit_graphical_var(): instead of a graphical-lasso /
EBIC sparse fit, it estimates a full VAR(1) with a flat prior on the
temporal coefficients and an inverse-Wishart prior on the residual
precision, then reports the temporal network B and the contemporaneous
partial-correlation network derived from the residual covariance. With more
than one subject the data are within-person centred and pooled (as in
fit_graphical_var()).
Usage
fit_var_bayes(
data,
vars,
id = NULL,
day = NULL,
beep = NULL,
lags = 1L,
scale = TRUE,
center_within = TRUE,
n_iter = 4000L,
n_burnin = NULL,
n_chains = 2L,
thin = 1L,
seed = NULL,
min_obs = NULL,
subject = NULL,
verbose = FALSE
)Arguments
- data
A
data.frameor matrix.- vars
Character vector of variable names (length >= 2).
- id
Character. Person-ID column, or
NULLfor a single series.- day
Character. Day/session column, or
NULL.- beep
Character. Beep/measurement column, or
NULL.- lags
Integer lag order; only
1is supported.- scale
Logical. Global standardization of each variable. Default
TRUE.- center_within
Logical. Within-person centre when >1 id (removes between-person variance, as in
fit_graphical_var()). DefaultTRUE.- n_iter, n_burnin, n_chains, thin
MCMC controls. Defaults
4000,n_iter/2,2,1.- seed
Integer or
NULL. Base seed (chaincusesseed + c).- min_obs
Integer or
NULL. Keep only subjects with at least this many observations.- subject
Optional vector naming the exact subject(s) to analyse.
- verbose
Logical. Progress messages. Default
FALSE.
Value
A var_bayes_result object (a cograph group with temporal and
contemporaneous netobjects) carrying beta, temporal, kappa, PCC,
PDC, posterior draws, and a tidy coefs() table (posterior median, SD,
95% CI, one-tailed p, significance by CI excluding 0).
See also
fit_graphical_var() (regularized GLASSO/EBIC), fit_var() (OLS),
fit_mlvar_bayes() (multilevel Bayesian VAR).
Examples
# Two independent AR(1) series with autoregressions of 0.4 and 0.3.
set.seed(1)
d <- data.frame(
A = as.numeric(stats::filter(rnorm(200), 0.4, method = "recursive")),
B = as.numeric(stats::filter(rnorm(200), 0.3, method = "recursive"))
)
fit <- fit_var_bayes(d, vars = c("A", "B"), n_iter = 500, seed = 1)
print(fit)
#> Bayesian VAR(1) result (unregularized, Mplus-targeted)
#> Variables: 2 (A, B)
#> Observations: 199
#> MCMC: 2 chains x 500 iter, 500 draws | max PSR = 1.010
#> Temporal 95% CIs excluding 0: 2 / 4
#>
#> Temporal [directed]
#> weights [-0.046, 0.373] | +2 / -2 edges
#> A B
#> A 0.37 -0.03
#> B -0.05 0.25
#>
#> Contemporaneous [undirected]
#> weights [-0.021, -0.021] | +0 / -1 edges
#> A B
#> A 0.00 -0.02
#> B -0.02 0.00
#>
#> coefs(x) | matrices(x) | edges(x) | nodes(x) | summary(x)
coefs(fit)
#> outcome predictor estimate posterior_sd ci_lower ci_upper p
#> 1 A A 0.37317738 0.06736368 0.2495538 0.50669480 0.000
#> 2 A B -0.04624642 0.06989868 -0.1748218 0.09766115 0.254
#> 3 B A -0.02512889 0.07216782 -0.1587258 0.11975205 0.346
#> 4 B B 0.24702492 0.07017144 0.1153309 0.38703367 0.000
#> significant
#> 1 TRUE
#> 2 FALSE
#> 3 FALSE
#> 4 TRUE