Inspect a Jeffreys-penalized fit — summary.mfrm_jeffreys

These methods inspect the separate result returned by fit_mfrm() with method = "JML", jml_penalty = "jeffreys". The result does not inherit ordinary JML uncertainty or FACETS diagnostics.

Usage

# S3 method for class 'mfrm_jeffreys_fit'
summary(object, ...)

# S3 method for class 'mfrm_jeffreys_fit'
print(x, ...)

# S3 method for class 'mfrm_jeffreys_fit'
coef(object, ...)

# S3 method for class 'mfrm_jeffreys_fit'
fitted(object, ...)

# S3 method for class 'mfrm_jeffreys_fit'
vcov(object, ...)

# S3 method for class 'mfrm_jeffreys_fit'
confint(object, ...)

# S3 method for class 'mfrm_jeffreys_fit'
logLik(object, ...)

Arguments

object, x: A mfrm_jeffreys_fit result.
...: Unused; additional arguments are rejected.

Value

summary() returns Person, facet and step point-estimate tables, checks, category-support evidence and notes. coef() returns the three point-estimate tables. fitted() returns category probabilities in prepared response-row order, with columns from rating_min to rating_max. print() displays the estimator and numerical status. vcov(), confint() and logLik() on this fit stop explicitly: parameter uncertainty and likelihood-based model comparison are not implemented. To estimate common-plan expected-score intervals, use fair_average_table() followed by confint.mfrm_fair_average().

Details

See fit_mfrm() for the penalty, assumptions and restrictions. checks$ordinary_nll is the ordinary negative log likelihood evaluated at the penalized point; checks$objective also includes the information penalty. Neither is an ordinary maximized likelihood for AIC or testing. category_support describes observed category usage, including missing categories; its ordinary-MLE warnings are not penalty-rank diagnostics.

Examples

# Long-running workflow; run explicitly when needed.
# example("summary.mfrm_jeffreys_fit", run.dontrun = TRUE)
if (FALSE) { # \dontrun{
ratings <- load_mfrmr_data("example_core")
# Each row is one Person's rating by a Rater on a Criterion.
fit <- fit_mfrm(ratings, "Person", c("Rater", "Criterion"), "Score",
  method = "JML", jml_penalty = "jeffreys", rating_min = 1, rating_max = 4,
  category_policy = "preserve")
summary(fit)$persons # Finite penalized Person points, without SEs
summary(fit)$checks$max_gradient # Penalized score-equation diagnostic
plan <- unique(ratings[c("Rater", "Criterion")])
scores <- fair_average_table(fit, reference_plan = plan)
scores$reference_scores # Expected averages under the same plan for all Persons
intervals <- confint(scores, parm = c("P001", "P002"), method = "delta")
intervals[, c("Target", "Estimate", "SE", "Available", "CIEligible")]
# Numerical availability is not a qualification of small-sample coverage.
} # }