Sequential Conditional Paired-Sample Testing with Applications to Online Monitoring

We study sequential conditional paired-sample testing, where observations arrive over time and the goal is to continuously assess whether two conditional distributions are equal given a common set of covariates. This problem arises naturally in model monitoring, where distributional changes must be detected without a prespecified terminal sample size, and in online comparisons, where responses are compared conditional on subject-level characteristics. We propose SCPT, an anytime-valid testing procedure for conditional distributions based on sequentially constructed conditional kernel discrepancy statistics and test martingales. The proposed method is nonparametric and accommodates a broad class of data-generating processes beyond the i.i.d. setting. We establish time-uniform Type I error control and derive asymptotic power guarantees under suitable regularity conditions. We also develop several useful extensions to address different needs, including testing under broader dependence structures, comparisons involving multiple auxiliary streams, and conditional goodness-of-fit testing. Simulations and real-data experiments show that SCPT effectively detects conditional distributional differences in streaming data, demonstrating its practical utility in a wide range of real-world applications.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-17
DOI
https://doi.org/10.48550/arxiv.2611.22886
Primary Topic
Statistical Methods and Inference
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Sequential Conditional Paired-Sample Testing with Applications to Online Monitoring

Xiaoyi Zheng, Hong Zhang
arXiv (Cornell University)
Statistical Methods and Inference
preprint

Sequential Conditional Paired-Sample Testing with Applications to Online Monitoring

Xiaoyi Zheng, Hong Zhang
preprint en

Abstract

We study sequential conditional paired-sample testing, where observations arrive over time and the goal is to continuously assess whether two conditional distributions are equal given a common set of covariates. This problem arises naturally in model monitoring, where distributional changes must be detected without a prespecified terminal sample size, and in online comparisons, where responses are compared conditional on subject-level characteristics. We propose SCPT, an anytime-valid testing procedure for conditional distributions based on sequentially constructed conditional kernel discrepancy statistics and test martingales. The proposed method is nonparametric and accommodates a broad class of data-generating processes beyond the i.i.d. setting. We establish time-uniform Type I error control and derive asymptotic power guarantees under suitable regularity conditions. We also develop several useful extensions to address different needs, including testing under broader dependence structures, comparisons involving multiple auxiliary streams, and conditional goodness-of-fit testing. Simulations and real-data experiments show that SCPT effectively detects conditional distributional differences in streaming data, demonstrating its practical utility in a wide range of real-world applications.

arXiv (Cornell University)
Statistical Methods and Inference
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon