Skip to contents

Generate Posterior Predictive Samples for Multilevel Latent Time Series Models

Source: R/mlts_posterior_sample.R
mlts_posterior_sample.Rd

The mlts_posterior_sample() function generates replicated datasets from a fitted mlts model using draws from the posterior distribution. The function can simulate data under the population model or based on individual-specific (random effect) parameters.

Usage

mlts_posterior_sample(
  fit,
  draw_person_pars = FALSE,
  n_draws = 10,
  draws = NULL,
  as_matrix = TRUE
)

Arguments

fit

An object of class mlts.fit, as returned by a fitted model using mlts().

draw_person_pars

Logical. If TRUE, samples are generated using person-specific parameters (random effects). If FALSE, only population-level parameters are used. Defaults to FALSE.

n_draws

Integer. Number of posterior draws to use for simulating replicated datasets. Ignored if draws is provided. Defaults to 10.

draws

Optional integer vector indicating specific posterior draw indices to use. If NULL, n_draws draws are chosen with the maximum distance between posterior samples.

as_matrix

Logical. Return replications of each variable as a matrix with n_draw rows, ready to run graphical posterior predictive checks using the bayesplot package.

Value

A list of replicated datasets, each as a data.frame with columns:

Y_rep

Replication number.

ID

Subject/cluster ID.

time

Time point.

...

One column per time-series variable defined in the model.

Details

The function extracts posterior samples of population-level (and optionally individual-level) parameters from a fitted mlts model and simulates replicated datasets from the posterior predictive distribution. Each replication corresponds to a different posterior draw and reflects uncertainty in the model's parameters. See PPC for an overview on graphical posterior predictive checks and how they can be performed.

If draw_person_pars = TRUE, the function uses sampled person-specific random effects and covariate effects from the posterior to generate new data at the individual level. This requires that the model was fitted with monitor_person_pars = TRUE in mlts_fit. If this condition is not met, the function will throw an error.

Posterior draws are either selected with the maximum distance between posterior samples (n_draws) or specified manually using the draws argument. Optionally, left or right censoring is respected in the simulated data if such constraints were present in the model.

See also

mlts_pp_check for plotting posterior predictive checks.

Examples

if (FALSE) { # \dontrun{
# build a simple vector-autoregressive mlts model with two time-series variables
var_model <- mlts_model(q = 2)

# simulate data from this model with default true values
# (true values are randomly drawn from normal distribution)
var_data <- mlts_sim(
  model = var_model,
  N = 50, TP = 30, # number of units and number of measurements per unit
  default = TRUE # use default parameter values
)

# fit model
fit <- mlts_fit(
  model = var_model,
  data = var_data,
  id = "ID", ts = c("Y1", "Y2"),
  time = "time",
  monitor_person_pars = TRUE
)

# Simulate 20 replications from the posterior
yreps <- mlts_posterior_sample(fit = fit, n_draws = 20)

# Include person-specific parameters in simulation
yreps <- mlts_posterior_sample(fit = fit, draw_person_pars = TRUE)

# Use specific posterior draws
yreps <- mlts_posterior_sample(fit = fit, draws = c(10, 50, 100))
} # }