weightedAverageModel: BIC-Weighted Model Averaging Across Consistent Candidate...

In cNORM: Continuous Norming

BIC-Weighted Model Averaging Across Consistent Candidate Models

Description

Computes a weighted average of the regression coefficients across the consistency-screened candidate models returned by regsubsets and screenSubset. Weights are information-theoretic model weights w_j \propto exp(-\Delta BIC_j / 2), restricted to models within deltaBIC of the best model. Each candidate is refitted on the complete sample via (weighted) least squares; coefficients of terms not included in a candidate are treated as zero.

Usage

weightedAverageModel(results, data, raw, weights = NULL, deltaBIC = 10)

Arguments

results	A (screened) summary.regsubsets object, including a consistent flag per row.
data	The preprocessed norm data.
raw	Name of the raw score variable.
weights	Optional numeric vector of case weights.
deltaBIC	Only models within this BIC distance of the best candidate are averaged (default 10; Burnham & Anderson, 2002).

Details

Because all candidates passed the monotonicity screening (same direction), and the averaging weights are positive and sum to one, the averaged model is itself monotone and thus consistent. In contrast to averaging coefficients over subsamples (deprecated subsample_lm), this approach targets the actual source of variance - model selection - while each component estimate remains a full-sample least squares fit.

Value

An lm-type object with averaged coefficients over the union of the candidate terms; fitted.values and residuals are updated accordingly.

References

Burnham, K. P., & Anderson, D. R. (2002). Model Selection and Multimodel Inference. Springer.

See Also

Other model: bestModel(), checkConsistency(), cnorm.cv(), derive(), modelSummary(), print.cnorm(), printSubset(), rangeCheck(), regressionFunction(), summary.cnorm()

cNORM documentation built on July 13, 2026, 5:08 p.m.