The cyclic selection problem of conditional inference

Conditional selective inference adjusts an interval for the event that led a scientist to examine it. But the scientist may then select again, keeping only the adjusted intervals that exclude zero. A further conditional adjustment can change that selection, inviting another adjustment. This difficulty has been recognized in earlier work. We give an elementary analysis for independent Gaussian observations, using equal-tailed intervals and conditioning on the complete history of reporting decisions. If all coordinates remain available for reconsideration, then, for every finite mean vector—including every vector of positive means—every finite pattern of reporting sets recurs infinitely often almost surely. The procedure reaches an empty set almost surely if stopped there; if continued, empty and nonempty sets both return forever. If only retained coordinates are ever revisited, the report becomes permanently empty almost surely. The continuing procedure’s long-run average number of reports is independent of the signal strengths. An ordinary random draw illustrates four successive complete replacements. We then present the established marginal alternative: the Benjamini–Yekutieli false coverage rate adjustment, with selection designed jointly with the intervals, gives the largest compatible collection of zero-excluding intervals within a chosen marginal family, and a procedure that terminates.

Authors

Institutions

Publication Details

Journal
Stanford Digital Repository
Published
2026-11-14
DOI
https://doi.org/10.25740/hp622rs2767
Primary Topic
Statistical Methods and Inference
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

The cyclic selection problem of conditional inference

Aaditya Ramdas
Stanford Digital Repository
Statistical Methods and Inference
article

The cyclic selection problem of conditional inference

Aaditya Ramdas
article en

Abstract

Conditional selective inference adjusts an interval for the event that led a scientist to examine it. But the scientist may then select again, keeping only the adjusted intervals that exclude zero. A further conditional adjustment can change that selection, inviting another adjustment. This difficulty has been recognized in earlier work. We give an elementary analysis for independent Gaussian observations, using equal-tailed intervals and conditioning on the complete history of reporting decisions. If all coordinates remain available for reconsideration, then, for every finite mean vector—including every vector of positive means—every finite pattern of reporting sets recurs infinitely often almost surely. The procedure reaches an empty set almost surely if stopped there; if continued, empty and nonempty sets both return forever. If only retained coordinates are ever revisited, the report becomes permanently empty almost surely. The continuing procedure’s long-run average number of reports is independent of the signal strengths. An ordinary random draw illustrates four successive complete replacements. We then present the established marginal alternative: the Benjamini–Yekutieli false coverage rate adjustment, with selection designed jointly with the intervals, gives the largest compatible collection of zero-excluding intervals within a chosen marginal family, and a procedure that terminates.

Stanford Digital Repository
Stanford University (US)
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