True and estimated subject values, one row per subject

Puts every subject's generating value next to its posterior point estimate, one column pair per parameter, with the covariates alongside. This is the input for a structural equation model of true against estimated values, for example in lavaan, which recovers a correlation that the posterior means attenuate by shrinkage (see extract_correlations()). bmmtools does not fit that model itself and does not depend on lavaan.

Usage

subject_table(
  x,
  fit = NULL,
  point = c("mean", "median"),
  scale = c("link", "natural"),
  links = NULL
)

Arguments

x: A bmmtools_simulation, together with fit; a bmmtools_simulation_set from simulate_components(), together with its fits; or the result of recovery_grid(), whose attribute subject_means carries the posterior means, medians and true values of every cell.
fit: The fit of x when x is a simulation: a brmsfit, or any object with an extract_subject_draws() method. For a simulation set, the fits from fit_components(), one per component, whose terms appear prefixed with the component name. Not used with a grid result.
point: The point estimate: the posterior "mean" or "median" of each subject's value.
scale: "link" or "natural". On the natural scale the true values and the point estimates of each parameter go through inverse_link(); covariates are left as they are.
links: A named character vector mapping a term to a link name. NULL reads the simulation model's link table, or the one the grid stored; without one, the natural scale falls back to the link scale with a message.

Value

A tibble with one row per condition, replication and subject: condition (NA for a single simulation), replication (1 for a single simulation), id, then true_<term> and est_<term> for each parameter with subject-level draws, then one column per covariate, under its own name.

Details

A parameter that does not vary in the simulation has no subject truth; its true_<term> column is the population value for every subject.

On the natural scale the estimate is the inverse link of the point estimate on the link scale, not the mean of the inverse links, as for the point estimator of extract_correlations(). For the median the two are the same.

Examples

if (FALSE) { # \dontrun{
sim <- simulate_recovery(
  bmm::mixture2p(resp_error = "y"),
  pars = c(kappa = log(8), thetat = qlogis(0.75)),
  n_subjects = 100, n_trials = 50,
  sds = c(kappa = 0.3, thetat = 0.5),
  covariates = list(G = c(mean = 0, sd = 1)), seed = 1
)
fit <- bmm::bmm(recovery_formula(sim$model), sim$data, sim$model)
subject_table(sim, fit)

# one table over a grid, from its sidecars
out <- recovery_grid(model, grid, pars, dir = "fits", sds = sds)
subject_table(out, point = "median")
} # }