How Much Does Your Data Exploration Overfit? Controlling Bias via Information Usage
Open accessThe DOI matches its Crossref record; 10 of the 10 references it cites were found in a registry or an index.
Abstract
Modern data is messy and high-dimensional, and it is often not clear a priori what are the right questions to ask. Instead, the analyst typically needs to use the data to search for interesting analyses to perform and hypotheses to test. This is an adaptive process, where the choice of analysis to be performed next depends on the results of the previous analyses on the same data. Ultimately, which results are reported can be heavily influenced by the data. It is widely recognized that this process, even if well-intentioned, can lead to biases and false discoveries, contributing to the crisis of reproducibility in science. But while any data-exploration renders standard statistical theory invalid, experience suggests that different types of exploratory analysis can lead to disparate levels of bias, and the degree of bias also depends on the particulars of the data set. In this paper, we propose a general information usage framework to quantify and provably bound the bias and other error metrics of an arbitrary exploratory analysis. We prove that our mutual information based bound is tight in natural settings, and then use it to give rigorous insights into when commonly used procedures do or do not lead to substantially biased estimation. Through the lens of information usage, we analyze the bias of specific exploration procedures such as filtering, rank selection and clustering. Our general framework also naturally motivates randomization techniques that provably reduce exploration bias while preserving the utility of the data analysis. We discuss the connections between our approach and related ideas from differential privacy and blinded data analysis, and supplement our results with illustrative simulations.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing
Benjamini, Y., & Hochberg, Y. (1995) · Journal of the Royal Statistical Society Series B: Statistical Methodology · doi:10.1111/j.2517-6161.1995.tb02031.x · Cited by 109,482 · cited 1×
“Much of this literature has focused on controlling the false discovery rate in multiple-hypothesis testing where the hypotheses are not adaptively chosen [2], [3].”
BackgroundThe control of the false discovery rate in multiple testing under dependency
Benjamini, Y., & Yekutieli, D. (2001) · The Annals of Statistics · doi:10.1214/aos/1013699998 · Cited by 10,490
Efron, B., Hastie, T., Johnstone, I., & Tibshirani, R. (2004) · The Annals of Statistics · doi:10.1214/009053604000000067 · Cited by 9,357
Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2011) · Psychological Science · doi:10.1177/0956797611417632 · Cited by 6,567
A significance test for the lasso
Lockhart, R., Taylor, J., Tibshirani, R. J., & Tibshirani, R. (2014) · The Annals of Statistics · doi:10.1214/13-aos1175 · Cited by 541
Statistical learning and selective inference
Taylor, J., & Tibshirani, R. J. (2015) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1507583112 · Cited by 398 · cited 2×
“One recent line of work [8], [9] proposes a framework for assigning significance and confidence intervals in selective inference, where model selection and significance testing are performed on the same dataset.”
BackgroundThe reusable holdout: Preserving validity in adaptive data analysis
Dwork, C., Feldman, V., Hardt, M., Pitassi, T., Reingold, O., & Roth, A. (2015) · Science · doi:10.1126/science.aaa9375 · Cited by 307
Blind analysis: Hide results to seek the truth
MacCoun, R., & Perlmutter, S. (2015) · Nature · doi:10.1038/526187a · Cited by 218 · cited 1×
“The raw, uncorrupted, dataset is only used in computing the final reported values [32].”
MethodGeneralization in Adaptive Data Analysis and Holdout Reuse
Dwork, C., Feldman, V., Hardt, M., Pitassi, T., Reingold, O., & Roth, A. (2015) · arXiv · doi:10.48550/arxiv.1506.02629 · Cited by 152
The Ladder: A Reliable Leaderboard for Machine Learning Competitions
Blum, A., & Hardt, M. (2015) · arXiv · doi:10.48550/arxiv.1502.04585 · Cited by 92
Cited by
Multiaccuracy: Black-Box Post-Processing for Fairness in Classification
Kim, M. P., Ghorbani, A., & Zou, J. (2019) · Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society · doi:10.1145/3306618.3314287 · Cited by 193
Self-citationDistributed Variational Representation Learning
Aguerri, I. E., & Zaidi, A. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2019.2928806 · Cited by 84
“Although it is not yet very well understood how to optimally control model complexity, recent works [16], [17] show that the generalization error can be upper-bounded using the mutual information between the input dataset and the output of the algorithm.”
BackgroundA translational perspective towards clinical AI fairness
Liu, M., Ning, Y., Teixayavong, S., Mertens, M., Xu, J., Ting, D. S. W., Cheng, L. T.-E., Ong, J. C. L., Teo, Z. L., Tan, T. F., et al. (2023) · npj Digital Medicine · doi:10.1038/s41746-023-00918-4 · Cited by 82
Information-theoretic analysis of stability and bias of learning algorithms
Raginsky, M., Rakhlin, A., Tsao, M., Wu, Y., & Xu, A. (2016) · 2016 IEEE Information Theory Workshop (ITW) · doi:10.1109/itw.2016.7606789 · Cited by 70
One-Class Anomaly Detection for Industrial Applications: A Comparative Survey and Experimental Study
Paolini, D., Dini, P., Soldaini, E., & Saponara, S. (2025) · Computers · doi:10.3390/computers14070281 · Cited by 21
Metadata
- Authors
- Russo, D., & Zou, J.
- Venue
- IEEE Transactions on Information Theory
- Published
- 2020-01
- Volume
- 66
- Issue
- 1
- Pages
- 302-323
- ISSN
- 0018-9448 (ISSN-L)
- 1557-9654
- OpenAlex
- W2979452771
- Cited by
- 119
- References
- 10
- FWCI
- 6.68
- Type
- journal article
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.