Skip to contents

Compute Relative-Likelihood Weights for Model Averaging

Source: R/model_weights.R
compute_weights.Rd

Selects the top-\(k\) models by information criterion value and assigns relative-likelihood weights using the Akaike / Schwarz weight formula: $$w_i = \frac{\exp\!\bigl((\mathrm{IC}_{\min} - \mathrm{IC}_i)/2\bigr)} {\sum_{j=1}^{k} \exp\!\bigl((\mathrm{IC}_{\min} - \mathrm{IC}_j)/2\bigr)}.$$

Usage

compute_weights(ic_vals, top_k = 3L, criterion = "AIC")

Arguments

ic_vals

Named numeric vector of information criterion values, e.g. from extract_ic.

top_k

Integer. Number of top-ranked models to retain (default 3).

criterion

Character scalar passed through for labelling only. Typically "AIC" or "BIC".

Value

A tibble with four columns:

model

Distribution name (from names of ic_vals).

weight

Relative-likelihood model weight (sums to 1).

ic

Raw information criterion value.

criterion

Label passed via the criterion argument.

Examples

ic <- c(exp = 240.1, weibull = 238.7, gompertz = 241.5,
        llogis = 239.3, lnorm = 243.0, gamma = 238.9)
compute_weights(ic, top_k = 3, criterion = "AIC")
#> # A tibble: 3 × 4
#>   model   weight    ic criterion
#>   <chr>    <dbl> <dbl> <chr>    
#> 1 weibull  0.378  239. AIC      
#> 2 gamma    0.342  239. AIC      
#> 3 llogis   0.280  239. AIC