Inspect a Julia joint missing-predictor fit

Methods for Gaussian response models with one additive mi() predictor, using a fixed-effect Gaussian or Bernoulli predictor model through engine = "julia". This experimental route uses the shared DRM.jl joint likelihood. Other missing-predictor families and structures remain separate implementation work.

Usage

# S3 method for class 'drmTMB_julia_joint'
coef(object, dpar = NULL, ...)
# S3 method for class 'drmTMB_julia_joint'
vcov(object, ...)
# S3 method for class 'drmTMB_julia_joint'
nobs(object, ...)
# S3 method for class 'drmTMB_julia_joint'
predict(object, newdata = NULL,
  dpar = c("mu", "sigma"), type = c("response", "link"), ...)
# S3 method for class 'drmTMB_julia_joint'
imputed(object, variable = NULL,
  rows = c("missing", "all"), se = TRUE, include_observed, ...)
# S3 method for class 'drmTMB_julia_joint'
summary(object, conf.int = FALSE,
  level = 0.95, method = c("wald", "profile"), ...)
# S3 method for class 'drmTMB_julia_joint'
confint(object, parm = NULL,
  level = 0.95, method = c("wald", "profile", "bootstrap"), ...)

Arguments

object

A fit returned by the joint Julia route.

dpar

Coefficient block to extract, or "mu"/"sigma" for prediction.

newdata

Complete new predictor rows. Omit to obtain stored conditional predictions.

type

Prediction on the response or link scale.

variable

Name of the modelled missing predictor.

rows

Return missing-predictor rows only, or every retained model row.

se

Include available conditional imputation standard errors.

include_observed

Optional compatibility selector for all or missing rows.

conf.int

Append Wald intervals to the summary table.

level

Confidence level strictly between zero and one.

method

Currently "wald" only. Profile and bootstrap requests fail explicitly.

parm

Optional compact dpar:term or fixef:dpar:term target names.

...

Reserved.

Details

Binary predictor labels follow the fitted encoding, regardless of the levels present in the new batch. Numeric new values match raw numeric fitted labels; for nonnumeric fitted labels, numeric 0/1 values specify the encoded states. If distinct labels become identical as numbers (for example, "01" and "1"), supply character or factor values to preserve their identity.

The sigma_mi_<variable> coefficient is the Gaussian predictor standard deviation on its natural scale. Its covariance includes the full delta-method transformation, including cross-covariances. Its Wald interval is constructed on the raw log-SD coordinate and then exponentiated, matching native R's positive-scale interval construction. This does not establish interval coverage or full cross-engine inference parity. Joint missing-predictor bootstrap is not yet implemented on either engine.

imputed() returns conditional summaries, not multiple-imputation draws. Gaussian imputation uncertainty includes conditional variance and parameter uncertainty; Bernoulli uncertainty is the conditional Bernoulli standard deviation. Observed predictors have no imputation standard error.

With response = "include", retained missing responses have unavailable residuals. nobs() counts observed responses, while predictions and imputed(rows = "all") retain every model row. The Julia adapter applies response = "drop" before joint preparation; the current native route can instead reject that combination. This preprocessing difference is explicit.

The initial route requires default fitting controls, ML, and complete exogenous fixed-effect designs. Weights, REML, offsets, random effects and structured predictor models are not admitted by this route.

Value

Coefficient blocks, a covariance matrix, an observation count, prediction vectors, an eight-column imputation table, or a summary/interval table, according to the generic used.

See also

drmTMB, mi, impute_model, imputed