Skip to contents

Compute Shapley values for the longitudinal causal decomposition

Source: R/causal_shapley.R
causal_shapley.Rd

Computes the average decomposition across all 6 possible orderings of the three component types (behavioral, compositional, pretreatment). Since orderings are paired (between-group and within-group use the same structural ordering), there are exactly 6 evaluations.

Usage

causal_shapley(params, ref = NULL)

Arguments

params

An ineqx_params object with multiple time periods

Value

An object of class "ineqx_causal_longit" with order = "shapley" and additional fields:

shapley

data.frame with Shapley-averaged values for each component at each time period

all_orderings

List of 6 ineqx_causal_longit results, one per ordering

ranges

data.frame showing min/max for each component across orderings

Details

Shapley values provide a robustness check against path dependence in the sequential parameter-switching decomposition. When changes over time are small relative to levels, the ordering has little practical impact. When changes are large, Shapley values provide a useful robustness check.