decisionfacets 0.2.0
- Designs with one examiner per case (or any partially overlapping
rating design) are now supported.
df_counterfactual()
compares each candidate’s observed examiners with random reassignments
that draw, for each case, an examiner qualified for that case, without
repeating an examiner for the same candidate. df_simulate()
gains design = "per_item" and
raters_per_item.
- Continuous scores (for example 0-100) are supported through a linear
many-facet model,
score = ability - case difficulty - examiner severity + error.
df_data() gains
scale = c("auto", "ordinal", "continuous");
df_fit() uses the new "linear" engine for
continuous scores; df_simulate() gains
scale = "continuous".
- New decision rule
df_cut(value, "raw_mean"): pass if
the mean observed rating reaches value (e.g. 70),
convenient when candidates have different numbers of ratings.
df_fit() stops with a clear message when the rating
network is disconnected, because rater severities would then not be
comparable.- New synthetic data set
oral_exam: 300 candidates, 12
cases, one examiner per case, 0-100 scores.
- Results for the existing crossed designs with ordinal scores are
unchanged.
decisionfacets 0.1.0
- Initial release.
- Many-facet Rasch rating scale model fitting via ‘TAM’ (marginal ML)
or built-in JMLE (
df_fit()).
- Explicit decision rules: raw total, fair average, measure
(
df_cut(), df_classify()).
- Counterfactual pass probabilities under observed, average and random
rater panels, with rater-dependence flags
(
df_counterfactual()).
- Attribution of expected misclassification to measurement error and
rater assignment (
df_attribute()).
- Known-truth simulation (
df_simulate()).