# A hierarchical decomposition for explaining ML performance discrepancies

arXiv: https://arxiv.org/abs/2402.14254 · HTML: https://mlsanity.com/paper/2402.14254/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 23 of its 64 references that are in the map.
- First posted on arXiv 22 Feb 2024.

## Builds on (most cited first, 23 of 23)

- [A Unified Approach to Interpreting Model Predictions](https://mlsanity.com/paper/1705.07874/) (2017)
- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)](https://mlsanity.com/paper/1711.11279/) (2017)
- [Data Shapley: Equitable Valuation of Data for Machine Learning](https://mlsanity.com/paper/1904.02868/) (2019)
- [Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shift](https://mlsanity.com/paper/1810.11953/) (2018)
- [Domino: Discovering Systematic Errors with Cross-Modal Embeddings](https://mlsanity.com/paper/2203.14960/) (2022)
- [Semiparametric doubly robust targeted double machine learning: a review](https://mlsanity.com/paper/2203.06469/) (2022)
- [An introduction to domain adaptation and transfer learning](https://mlsanity.com/paper/1812.11806/) (2018)
- [Distilling Model Failures as Directions in Latent Space](https://mlsanity.com/paper/2206.14754/) (2022)
- [The Spotlight: A General Method for Discovering Systematic Errors in Deep Learning Models](https://mlsanity.com/paper/2107.00758/) (2021)
- [Efficient nonparametric statistical inference on population feature importance using Shapley values](https://mlsanity.com/paper/2006.09481/) (2020)
- [Diagnosing Model Performance Under Distribution Shift](https://mlsanity.com/paper/2303.02011/) (2023)
- [Why did the distribution change?](https://mlsanity.com/paper/2102.13384/) (2021)
- [Towards Explaining Distribution Shifts](https://mlsanity.com/paper/2210.10275/) (2022)
- [Feature Shift Detection: Localizing Which Features Have Shifted via Conditional Distribution Tests](https://mlsanity.com/paper/2107.06929/) (2021)
- ["Why did the Model Fail?": Attributing Model Performance Changes to Distribution Shifts](https://mlsanity.com/paper/2210.10769/) (2022)
- [Rethinking Distribution Shifts: Empirical Analysis and Modeling for Tabular Data](https://mlsanity.com/paper/2307.05284/) (2023)
- [An Interventionist Approach to Mediation Analysis](https://mlsanity.com/paper/2008.06019/) (2020)
- [Variable importance measures for heterogeneous treatment effects](https://mlsanity.com/paper/2204.06030/) (2022)
- [Explaining medical AI performance disparities across sites with confounder Shapley value analysis](https://mlsanity.com/paper/2111.08168/) (2021)
- [Towards a More Rigorous Science of Blindspot Discovery in Image Classification Models](https://mlsanity.com/paper/2207.04104/) (2022)
- [Efficient and Multiply Robust Risk Estimation under General Forms of Dataset Shift](https://mlsanity.com/paper/2306.16406/) (2023)
- [The Lifecycle of a Statistical Model: Model Failure Detection, Identification, and Refitting](https://mlsanity.com/paper/2202.04166/) (2022)

## Built on by (most cited first, 5 of 5)

- ["Who experiences large model decay and why?" A Hierarchical Framework for Diagnosing Heterogeneous Performance Drift](https://mlsanity.com/paper/2506.00756/) (2025)
- [Explaining Concept Shift with Interpretable Feature Attribution](https://mlsanity.com/paper/2505.20634/) (2025)
- [Data Heterogeneity Modeling for Trustworthy Machine Learning](https://mlsanity.com/paper/2506.00969/) (2025)
- [What's the next frontier for Data-centric AI? Data Savvy Agents](https://mlsanity.com/paper/2511.01015/) (2025)
- [ShapShift: Explaining Model Prediction Shifts with Subgroup Conditional Shapley Values](https://mlsanity.com/paper/2604.11200/) (2026)

## Abstract (arXiv)

Machine learning (ML) algorithms can often differ in performance across domains. Understanding textitwhy their performance differs is crucial for determining what types of interventions (e.g., algorithmic or operational) are most effective at closing the performance gaps. Existing methods focus on textitaggregate decompositions of the total performance gap into the impact of a shift in the distribution of features p(X) versus the impact of a shift in the conditional distribution of the outcome p(Y|X); however, such coarse explanations offer only a few options for how one can close the performance gap. textitDetailed variable-level decompositions that quantify the importance of each variable to each term in the aggregate decomposition can provide a much deeper understanding and suggest much more targeted interventions. However, existing methods assume knowledge of the full causal graph or make strong parametric assumptions. We introduce a nonparametric hierarchical framework that provides both aggregate and detailed decompositions for explaining why the performance of an ML algorithm differs across domains, without requiring causal knowledge. We derive debiased, computationally-efficient estimators, and statistical inference procedures for asymptotically valid confidence intervals.