Diagnose an OpenMx Model

Evaluates convergence and numerical diagnostics for an OpenMx model.

Usage

mx_diagnostics(
  model,
  check = c("convergence", "fit", "parameters", "standard_errors", "boundaries",
    "hessian"),
  tol = 1e-06
)

Arguments

model

A fitted mxModel object.

check

Character vector specifying which diagnostics to perform out of c("convergence", "fit", "parameters", "standard_errors", "boundaries", "hessian"). Default: all diagnostics.

tol

Numeric atomic vector, indicating the tolerance for checking whether estimates are near their boundaries.

Value

A list containing:

checks_pass

Logical indicating whether all requested checks pass.

convergence_ok

Logical indicating acceptable optimizer convergence.

fit_ok

Logical indicating a finite model fit value.

parameters_ok

Logical indicating finite parameter estimates.

boundaries_ok

Logical indicating whether parameter boundary checks pass.

standard_errors_ok

Logical indicating whether standard errors are available and valid.

hessian_ok

Logical indicating whether the Hessian satisfies the implemented diagnostic criteria.

warnings

Character vector describing detected problems.

Details

Convergence diagnostics do not guarantee that the global maximum of the likelihood has been found. Boundary estimates and non-positive-definite Hessians may indicate weak identification. Boundary estimates do not necessarily imply optimization failure.

Examples

if (FALSE) { # \dontrun{
library(OpenMx)
library(metamixture)

set.seed(123)
dat <- sim_mix_rma(n = 100)

fit <- remix(dat$yi, dat$vi, classes = 2)

# Run all -diagnostic checks
mx_diagnostics(fit)

# Check only optimizer convergence and model fit
mx_diagnostics(fit, check = c("convergence", "fit"))

# Overall result
mx_diagnostics(fit)$checks_pass
} # }