levante-metadata-scoring

This dataset contains scoring models and specifications for LEVANTE tasks.

The model_registry table indexes all the files of model records. Their filenames encode the specification of the model – task, itemtype (rasch/2pl), number of factors, invariance. For multigroup models they're further specified as fit to items that overlap between groups or to all items (anchored to overlapping items); for by language models they're further specified as which language's data they're fit to.

The scoring_models table specifies, for each combination of task and dataset, which model should be used to score data from that task and dataset, identified with: model_set (multigroup_site/by_language), subset (all_items/overlap_items/{language_code}), itemtype (rasch/2pl), nfact (f1/f2), invariance (scalar/metric/configural).

The item_parameters table contains the item parameters for each unique model specified in the scoring_models table, as difficulty and discrimination for each item_uid.