How Many Ratings per Item are Necessary for Reliable Significance Testing?

• URL: http://arxiv.org/abs/2412.02968v2
• Date: Tue, 28 Oct 2025 19:25:28 GMT
• Title: How Many Ratings per Item are Necessary for Reliable Significance Testing?
• Authors: Christopher Homan, Flip Korn, Deepak Pandita, Chris Welty,
• Abstract summary: A cornerstone of machine learning evaluation is the assumption that model and human responses are reliable enough to evaluate models against unitary, authoritative, gold standard'' data.<n>We adapt a method to determine whether an (existing or planned) dataset has enough responses per item to assure reliable null hypothesis statistical testing.<n>We show how our methods can help AI researchers make better decisions about how to collect data for AI evaluation.
• Score: 7.422152765037947
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: A cornerstone of machine learning evaluation is the (often hidden) assumption that model and human responses are reliable enough to evaluate models against unitary, authoritative, ``gold standard'' data, via simple metrics such as accuracy, precision, and recall. The generative AI revolution would seem to explode this assumption, given the critical role stochastic inference plays. Yet, in spite of public demand for more transparency in AI -- along with strong evidence that humans are unreliable judges -- estimates of model reliability are conventionally based on, at most, a few output responses per input item. We adapt a method, previously used to evaluate the reliability of various metrics and estimators for machine learning evaluation, to determine whether an (existing or planned) dataset has enough responses per item to assure reliable null hypothesis statistical testing. We show that, for many common metrics, collecting even 5-10 responses per item (from each model and team of human evaluators) is not sufficient. We apply our methods to several of the very few extant gold standard test sets with multiple disaggregated responses per item and show that even these datasets lack enough responses per item. We show how our methods can help AI researchers make better decisions about how to collect data for AI evaluation.

Related papers

• Fantastic Bugs and Where to Find Them in AI Benchmarks [28.604919035475188]
  We introduce a framework for systematic benchmark revision that leverages statistical analysis of response patterns to flag potentially invalid questions.<n>Our approach builds on a core assumption commonly used in AI evaluations that the mean score sufficiently summarizes model performance.<n>Across nine widely used benchmarks, our method guides expert review to identify problematic questions with up to 84% precision.
  arXiv  Detail & Related papers  (2025-11-20T22:49:21Z)
• What Does Your Benchmark Really Measure? A Framework for Robust Inference of AI Capabilities [0.773472615056109]
  Evaluations of generative models on benchmark data are now ubiquitous.<n>Yet growing skepticism surrounds their reliability.<n>How can we know that a reported accuracy genuinely reflects a model's true performance?<n>We make this step explicit by proposing a principled framework for evaluation as inference.
  arXiv  Detail & Related papers  (2025-09-23T21:29:04Z)
• Large Language Models Often Know When They Are Being Evaluated [0.015534429177540245]
  We investigate whether frontier language models can accurately classify transcripts based on whether they originate from evaluations or real-world deployment.<n>We construct a benchmark of 1,000 prompts and transcripts from 61 distinct datasets.<n>Our results indicate that frontier models already exhibit a substantial, though not yet, level of evaluation-awareness.
  arXiv  Detail & Related papers  (2025-05-28T12:03:09Z)
• Where is this coming from? Making groundedness count in the evaluation of Document VQA models [12.951716701565019]
  We argue that common evaluation metrics do not account for the semantic and multimodal groundedness of a model's outputs. We propose a new evaluation methodology that accounts for the groundedness of predictions. Our proposed methodology is parameterized in such a way that users can configure the score according to their preferences.
  arXiv  Detail & Related papers  (2025-03-24T20:14:46Z)
• How to Select Datapoints for Efficient Human Evaluation of NLG Models? [57.60407340254572]
  We develop a suite of selectors to get the most informative datapoints for human evaluation. We show that selectors based on variance in automated metric scores, diversity in model outputs, or Item Response Theory outperform random selection. In particular, we introduce source-based estimators, which predict item usefulness for human evaluation just based on the source texts.
  arXiv  Detail & Related papers  (2025-01-30T10:33:26Z)
• Evaluation Agent: Efficient and Promptable Evaluation Framework for Visual Generative Models [51.067146460271466]
  Evaluation of visual generative models can be time-consuming and computationally expensive. We propose the Evaluation Agent framework, which employs human-like strategies for efficient, dynamic, multi-round evaluations. It offers four key advantages: 1) efficiency, 2) promptable evaluation tailored to diverse user needs, 3) explainability beyond single numerical scores, and 4) scalability across various models and tools.
  arXiv  Detail & Related papers  (2024-12-10T18:52:39Z)
• SureMap: Simultaneous Mean Estimation for Single-Task and Multi-Task Disaggregated Evaluation [75.56845750400116]
  Disaggregated evaluation -- estimation of performance of a machine learning model on different subpopulations -- is a core task when assessing performance and group-fairness of AI systems. We develop SureMap that has high estimation accuracy for both multi-task and single-task disaggregated evaluations of blackbox models. Our method combines maximum a posteriori (MAP) estimation using a well-chosen prior together with cross-validation-free tuning via Stein's unbiased risk estimate (SURE)
  arXiv  Detail & Related papers  (2024-11-14T17:53:35Z)
• Compare without Despair: Reliable Preference Evaluation with Generation Separability [20.50638483427141]
  We introduce a measure, separability, which estimates how suitable a test instance is for pairwise preference evaluation. For a candidate test instance, separability samples multiple generations from a pair of models, and measures how distinguishable the two sets of generations are. Experiments show that instances with high separability values yield more consistent preference ratings from both human- and auto-raters.
  arXiv  Detail & Related papers  (2024-07-02T01:37:56Z)
• A Comprehensive Assessment Benchmark for Rigorously Evaluating Deep Learning Image Classifiers [4.768207906634657]
  This article advocates benchmarking performance using a wide range of different types of data.<n>It is found that current deep neural networks, including those trained with methods that are believed to produce state-of-the-art robustness, are vulnerable to making mistakes on certain types of data.
  arXiv  Detail & Related papers  (2023-08-08T08:50:27Z)
• Bring Your Own Data! Self-Supervised Evaluation for Large Language Models [52.15056231665816]
  We propose a framework for self-supervised evaluation of Large Language Models (LLMs) We demonstrate self-supervised evaluation strategies for measuring closed-book knowledge, toxicity, and long-range context dependence. We find strong correlations between self-supervised and human-supervised evaluations.
  arXiv  Detail & Related papers  (2023-06-23T17:59:09Z)
• Position: AI Evaluation Should Learn from How We Test Humans [65.36614996495983]
  We argue that psychometrics, a theory originating in the 20th century for human assessment, could be a powerful solution to the challenges in today's AI evaluations.
  arXiv  Detail & Related papers  (2023-06-18T09:54:33Z)
• Conservative Prediction via Data-Driven Confidence Minimization [70.93946578046003]
  In safety-critical applications of machine learning, it is often desirable for a model to be conservative. We propose the Data-Driven Confidence Minimization framework, which minimizes confidence on an uncertainty dataset.
  arXiv  Detail & Related papers  (2023-06-08T07:05:36Z)
• Efficient Shapley Values Estimation by Amortization for Text Classification [66.7725354593271]
  We develop an amortized model that directly predicts each input feature's Shapley Value without additional model evaluations. Experimental results on two text classification datasets demonstrate that our amortized model estimates Shapley Values accurately with up to 60 times speedup.
  arXiv  Detail & Related papers  (2023-05-31T16:19:13Z)
• Accounting for multiplicity in machine learning benchmark performance [0.0]
  Using the highest-ranked performance as an estimate for state-of-the-art (SOTA) performance is a biased estimator, giving overly optimistic results. In this article, we provide a probability distribution for the case of multiple classifiers so that known analyses methods can be engaged and a better SOTA estimate can be provided.
  arXiv  Detail & Related papers  (2023-03-10T10:32:18Z)
• Evaluating Machine Unlearning via Epistemic Uncertainty [78.27542864367821]
  This work presents an evaluation of Machine Unlearning algorithms based on uncertainty. This is the first definition of a general evaluation of our best knowledge.
  arXiv  Detail & Related papers  (2022-08-23T09:37:31Z)
• Sharing pattern submodels for prediction with missing values [12.981974894538668]
  Missing values are unavoidable in many applications of machine learning and present challenges both during training and at test time. We propose an alternative approach, called sharing pattern submodels, which i) makes predictions robust to missing values at test time, ii) maintains or improves the predictive power of pattern submodels andiii) has a short description, enabling improved interpretability.
  arXiv  Detail & Related papers  (2022-06-22T15:09:40Z)
• CARMS: Categorical-Antithetic-REINFORCE Multi-Sample Gradient Estimator [60.799183326613395]
  We propose an unbiased estimator for categorical random variables based on multiple mutually negatively correlated (jointly antithetic) samples. CARMS combines REINFORCE with copula based sampling to avoid duplicate samples and reduce its variance, while keeping the estimator unbiased using importance sampling. We evaluate CARMS on several benchmark datasets on a generative modeling task, as well as a structured output prediction task, and find it to outperform competing methods including a strong self-control baseline.
  arXiv  Detail & Related papers  (2021-10-26T20:14:30Z)
• A Statistical Analysis of Summarization Evaluation Metrics using Resampling Methods [60.04142561088524]
  We find that the confidence intervals are rather wide, demonstrating high uncertainty in how reliable automatic metrics truly are. Although many metrics fail to show statistical improvements over ROUGE, two recent works, QAEval and BERTScore, do in some evaluation settings.
  arXiv  Detail & Related papers  (2021-03-31T18:28:14Z)
• Improving Dialog Evaluation with a Multi-reference Adversarial Dataset and Large Scale Pretraining [18.174086416883412]
  We introduce the DailyDialog++ dataset, consisting of (i) five relevant responses for each context and (ii) five adversarially crafted irrelevant responses for each context. We show that even in the presence of multiple correct references, n-gram based metrics and embedding based metrics do not perform well at separating relevant responses from even random negatives. We propose a new BERT-based evaluation metric called DEB, which is pretrained on 727M Reddit conversations and then finetuned on our dataset.
  arXiv  Detail & Related papers  (2020-09-23T18:06:52Z)
• Two-Sample Testing on Ranked Preference Data and the Role of Modeling Assumptions [57.77347280992548]
  In this paper, we design two-sample tests for pairwise comparison data and ranking data. Our test requires essentially no assumptions on the distributions. By applying our two-sample test on real-world pairwise comparison data, we conclude that ratings and rankings provided by people are indeed distributed differently.
  arXiv  Detail & Related papers  (2020-06-21T20:51:09Z)
• Meta-Learned Confidence for Few-shot Learning [60.6086305523402]
  A popular transductive inference technique for few-shot metric-based approaches, is to update the prototype of each class with the mean of the most confident query examples. We propose to meta-learn the confidence for each query sample, to assign optimal weights to unlabeled queries. We validate our few-shot learning model with meta-learned confidence on four benchmark datasets.
  arXiv  Detail & Related papers  (2020-02-27T10:22:17Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.