LinkSpec.mean_correction

abstractmethod LinkSpec.mean_correction(posterior, output_var='y')

Per-draw factor converting median-scale outputs to the response mean.

Counterfactual contributions are computed on the conditional median of the response (the inverse link applied to mu). For links whose conditional mean differs from the median, multiplying by this factor rescales the median-based quantity to the conditional mean E[y | mu, ...].

Parameters:

posteriorxr.Dataset

Posterior group of the fitted model’s InferenceData.

output_varstr, default "y"

Name of the observed variable, used to locate the likelihood scale parameter in the posterior.

Returns:

xr.DataArray

The multiplicative correction with (chain, draw, ...) dims (broadcasting over date). It is identically 1 for links whose mean equals the median (e.g. the identity link).