Build a candidate direct-anchor table from fitted estimates

Usage

make_anchor_table(
  fit,
  facets = NULL,
  include_person = FALSE,
  digits = NULL,
  readiness_policy = c("error", "review")
)

Arguments

fit: Output from fit_mfrm().
facets: Optional subset of facets to include.
include_person: Include person estimates as candidate anchors. Use only when cross-run person identity and the intended longitudinal constraint are substantively justified. Fixed Person values are supported only by a receiving JML fit; MML rejects Person anchors.
digits: NULL (default) preserves the fitted numeric values. A nonnegative integer explicitly rounds them for display or external use.
readiness_policy: How a source fit that is not inference-ready is handled. "error" (default) refuses anchor export or reuse. "review" permits explicit review-only extraction; those values must not be used as operational anchors.

Value

A data.frame with Facet, Level, and Anchor.

Details

This function performs a mechanical conversion from fitted estimates to the Facet/Level/Anchor schema accepted by fit_mfrm(). The returned rows are candidate direct constraints, not an approved anchor set. Anchor uses the internal, untransformed location metric. The table does not include category thresholds, GPCM slopes, model settings or source SEs; it is not a complete saved calibration. See extract_mfrm_calibration() for that separate workflow. Fixing an estimated value does not propagate its source uncertainty into the new fit.

By default, the function refuses export when the source fit is not inference-ready under the current readiness contract. Set readiness_policy = "review" only to inspect candidate values; this does not make them eligible for reuse. The function cannot verify cross-run element identity or invariance. Before reuse, document why selected elements retain the same meaning, check the observed design's connectedness, and run review_mfrm_anchors() on the receiving data. That review checks schema and receiving-data support; it does not validate the source fit or the substantive invariance assumption.

Keep digits = NULL for recalibration. Rounding individual anchors can break their joint sum or group-mean constraint, even at six decimal places. Explicitly rounded values are different constraints and must be reviewed before reuse; the fitter does not silently recenter supplied anchors.

Interpreting output

Facet: facet name to be anchored in later runs.
Level: specific element/level name inside that facet.
Anchor: fixed logit value (rounded only when digits is supplied).

Typical workflow

Fit and diagnose a defensible reference run with fit_mfrm().
Confirm source readiness, element identity, and the intended link.
Export candidates with make_anchor_table(fit) and review them.
Pass selected rows back into fit_mfrm(..., anchors = ...).

Examples

# Supplementary workflow: run explicitly to keep routine checks brief.
# example("make_anchor_table", run.dontrun = TRUE)
if (FALSE) { # \dontrun{
ratings <- load_mfrmr_data("example_operational")
fit <- fit_mfrm(
  ratings, "Person", c("Rater", "Criterion"), "Score",
  method = "MML", rating_min = 1, rating_max = 4,
  category_policy = "preserve"
)
# Select a facet deliberately; matching labels alone do not prove invariance.
anchors_tbl <- make_anchor_table(fit, facets = "Rater")
anchors_tbl
} # }