Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Longitudinal Counterfactuals: Constraints and Opportunities

Alexander Asemota, Giles Hooker

TL;DR

The paper addresses the problem of plausibility in counterfactual explanations for recourse by introducing a longitudinal distance metric that compares proposed counterfactual changes to prior observed trajectories. It couples this metric with a two-stage plausibility scoring approach and a Genetic Longitudinal Counterfactuals framework to generate counterfactuals constrained by historical changes. Through experiments on MIMIC-III ARF data and simulated Adult-Income scenarios, the authors demonstrate that longitudinal plausibility can reduce changes to immutable features and improve the realism of recourse paths, albeit with trade-offs in validity and search space. The work advances practical recourse by providing a data-driven proxy for plausibility and a scalable method to generate plausible counterfactuals, while acknowledging persistent challenges such as varying domain constraints, data quality, and the balance between plausibility and achieving desired outcomes.

Abstract

Counterfactual explanations are a common approach to providing recourse to data subjects. However, current methodology can produce counterfactuals that cannot be achieved by the subject, making the use of counterfactuals for recourse difficult to justify in practice. Though there is agreement that plausibility is an important quality when using counterfactuals for algorithmic recourse, ground truth plausibility continues to be difficult to quantify. In this paper, we propose using longitudinal data to assess and improve plausibility in counterfactuals. In particular, we develop a metric that compares longitudinal differences to counterfactual differences, allowing us to evaluate how similar a counterfactual is to prior observed changes. Furthermore, we use this metric to generate plausible counterfactuals. Finally, we discuss some of the inherent difficulties of using counterfactuals for recourse.

Longitudinal Counterfactuals: Constraints and Opportunities

TL;DR

The paper addresses the problem of plausibility in counterfactual explanations for recourse by introducing a longitudinal distance metric that compares proposed counterfactual changes to prior observed trajectories. It couples this metric with a two-stage plausibility scoring approach and a Genetic Longitudinal Counterfactuals framework to generate counterfactuals constrained by historical changes. Through experiments on MIMIC-III ARF data and simulated Adult-Income scenarios, the authors demonstrate that longitudinal plausibility can reduce changes to immutable features and improve the realism of recourse paths, albeit with trade-offs in validity and search space. The work advances practical recourse by providing a data-driven proxy for plausibility and a scalable method to generate plausible counterfactuals, while acknowledging persistent challenges such as varying domain constraints, data quality, and the balance between plausibility and achieving desired outcomes.

Abstract

Counterfactual explanations are a common approach to providing recourse to data subjects. However, current methodology can produce counterfactuals that cannot be achieved by the subject, making the use of counterfactuals for recourse difficult to justify in practice. Though there is agreement that plausibility is an important quality when using counterfactuals for algorithmic recourse, ground truth plausibility continues to be difficult to quantify. In this paper, we propose using longitudinal data to assess and improve plausibility in counterfactuals. In particular, we develop a metric that compares longitudinal differences to counterfactual differences, allowing us to evaluate how similar a counterfactual is to prior observed changes. Furthermore, we use this metric to generate plausible counterfactuals. Finally, we discuss some of the inherent difficulties of using counterfactuals for recourse.
Paper Structure (19 sections, 4 equations, 1 figure, 2 tables, 1 algorithm)

This paper contains 19 sections, 4 equations, 1 figure, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Counterfactual Explanations and Recourse
  3. Offering Recourse Through Counterfactuals
  4. Longitudinal Data as a Proxy for Plausibility
  5. Methods
  6. A Longitudinal Distance Metric
  7. Scoring Plausibility in Counterfactuals
  8. Genetic Longitudinal Counterfactuals
  9. Benefits and Limitations of Our Approach
  10. Explorations in Evaluating plausibility
  11. Ranking Explanations from DiCE and MIMIC-III
  12. Setup
  13. Results
  14. Generating Explanations for Simulated Adult-Income
  15. Setup
  16. ...and 4 more sections

Figures (1)

  • Figure 1: A shows the proportion of individuals with explanations below each threshold of longitudinal distance (x-axis). The solid lines represent the proportion of individuals with at least once explanation below the threshold, and the dotted line is the average proportion of explanations below the threshold. The vertical dotted line shows the distance value we would expect a feature to have if we observed it changing for one individual. B and C compare the Geometric and Longitudinal distances between an input and a counterfactual. In all plots, blue refers to VITAL and orange refers to ALL.