pith. sign in

arxiv: 2603.27718 · v2 · submitted 2026-03-29 · 📊 stat.ME · math.ST· stat.TH

Induced replication and the assessment of models

Heather Battey , Nancy Reid This is my paper

Pith reviewed 2026-05-14 21:52 UTC · model grok-4.3

classification 📊 stat.ME math.STstat.TH
keywords semiparametric model assessmentinduced replicationancillaritysufficiencywithin-sample prediction errorproportional hazardssparse regression
0
0 comments X

The pith

Induced replication via ancillarity and sufficiency separations lets semiparametric models be assessed with within-sample prediction error.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that semiparametric models can be assessed from parametric foundational principles by inducing replication under the postulated model. This uses non-standard inferential separations akin to Fisherian ancillarity and sufficiency to generate within-sample replicates, replacing the need for out-of-sample prediction error. The approach sidesteps estimation of nuisance components through smoothing or expansions and their associated tuning choices. It is illustrated in settings such as the proportional hazards model, time-dependent Poisson processes, matched-pair designs, and sparse regression confidence sets. Numerical checks show nominal error rates when the model holds and sensitivity to semiparametric departures.

Core claim

Inducement of replication under the postulated model, achieved through non-standard inferential separations in the sense of ancillarity/co-ancillarity and sufficiency/co-sufficiency, replaces out-of-sample prediction error with a type of within-sample prediction error for assessing semiparametric and other highly parametrized models, without requiring estimation of nuisance functions.

What carries the argument

Induced replication through non-standard inferential separations (ancillarity/co-ancillarity and sufficiency/co-sufficiency) that generate valid within-model replicates for prediction-error assessment.

If this is right

  • Model assessment for the proportional hazards model proceeds without kernel or basis estimation of the baseline hazard.
  • Time-dependent Poisson processes with semiparametric intensity admit direct within-sample assessment.
  • Matched-pair and two-group designs yield assessment procedures based on induced replicates.
  • Confidence sets for sparse regression models follow from a post-reduction inference approach.
  • Nominal error rates are recovered under the postulated model while sensitivity to semiparametric alternatives is retained.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same separation-based replication might simplify assessment in other high-dimensional parametric families where direct out-of-sample validation is costly.
  • Conditional inference traditions could be re-examined through this lens to derive new within-sample diagnostics.
  • The framework suggests testing whether induced replicates remain stable when the model is slightly misspecified in directions not captured by standard semiparametric alternatives.

Load-bearing premise

The postulated model must permit non-standard inferential separations sufficient to induce valid replication under that model.

What would settle it

Simulations or data where the induced within-sample prediction errors fail to match their expected distribution or nominal coverage when the postulated model is true.

Figures

Figures reproduced from arXiv: 2603.27718 by Heather Battey, Nancy Reid.

Figure 1
Figure 1. Figure 1: Logarithm of E(Rj ) and V(Rj ) calculated from (22) and (4), showing how the sensitivity to detect model misspecification varies with ς and ψ ∗ . Dashed lines indicate regions of the parameter space for the true Weibull model where E(Rj ) and V(Rj ) coincide with their putative values under the hypothesised exponential model. a function of ς and ψ ∗ . This shows that (8) is most sensitive to departures fro… view at source ↗
read the original abstract

We study the assessment of semiparametric and other highly-parametrised models from the perspective of foundational principles of parametric statistical inference. In doing so, we highlight the possibility of avoiding the usual semiparametric considerations, which typically require estimation of nuisance components through kernel smoothing or basis expansion, with the associated difficulties of tuning-parameter choice that blur the distinction between estimation and model assessment. A key aspect is the inducement of replication under the postulated model. This can be cast in terms of some non-standard inferential separations, in the vein of Fisherian ancillarity/co-ancillarity and sufficiency/co-sufficiency separations, allowing the replacement of out-of-sample prediction error as a criterion for semiparametric model assessment by a type of within-sample prediction error. Framed in this light are new methodological contributions in multiple example settings, including model assessment for the proportional hazards model, for a time-dependent Poisson process with semiparametric intensity function, and for matched-pair and two-group examples. Also subsumed within the framework is a post-reduction inference approach to the construction of confidence sets of sparse regression models. Numerical work confirms recovery of nominal error rates under the postulated model and high sensitivity to departures in the direction of semiparametric alternatives. We conclude by emphasising open challenges and unifying perspectives.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes assessing semiparametric models via induced replication under the postulated model, achieved through non-standard Fisherian ancillarity/co-ancillarity and sufficiency/co-sufficiency separations. This replaces out-of-sample prediction error with within-sample error, avoiding nuisance estimation and tuning. The framework is illustrated for the proportional hazards model, time-dependent Poisson processes, matched-pair designs, two-group comparisons, and post-reduction inference for sparse regression models. Numerical experiments are reported to recover nominal error rates under the model and exhibit sensitivity to semiparametric alternatives.

Significance. If the claimed exact separations hold without approximation, the work offers a principled route to model assessment in highly parameterized settings that sidesteps kernel or basis nuisance estimation and its attendant tuning issues. The unification of multiple examples under Fisherian principles and the reported numerical confirmation of error rates constitute potential strengths, though the approach's validity hinges on the exactness of the inferential separations in infinite-dimensional cases.

major comments (2)
  1. [Abstract and framework description] The central claim requires that the induced replication distribution be exactly free of the infinite-dimensional nuisance (e.g., baseline hazard or intensity function). No explicit derivation or conditional distribution is supplied in the abstract or described sections to confirm this separation is exact rather than approximate in the semiparametric examples.
  2. [Numerical work] Numerical confirmation of nominal error rates is cited, but without the explicit construction of the replication distribution or code, it is impossible to verify that the simulations truly use the induced (nuisance-free) distribution rather than an approximation that reintroduces estimation.
minor comments (2)
  1. The abstract refers to 'post-reduction inference' for sparse models; the main text should explicitly locate this within the induced-replication framework and state the precise reduction step.
  2. Notation for the induced replication distribution and the associated prediction error should be introduced with a single running example before the applications.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed review and valuable feedback on our manuscript. We address each major comment below and have made revisions to strengthen the presentation of the framework and numerical validation.

read point-by-point responses

  1. Referee: [Abstract and framework description] The central claim requires that the induced replication distribution be exactly free of the infinite-dimensional nuisance (e.g., baseline hazard or intensity function). No explicit derivation or conditional distribution is supplied in the abstract or described sections to confirm this separation is exact rather than approximate in the semiparametric examples.

    Authors: The induced replication distributions are exactly free of the nuisance parameters by construction, relying on the exact ancillarity and sufficiency separations in the semiparametric models considered. For the proportional hazards model, the replication is induced conditionally on the observed failure times and censoring indicators, which are sufficient for the baseline hazard, rendering the conditional distribution independent of it. Similar exact separations hold for the other examples. We have expanded the manuscript with explicit derivations of these conditional distributions in a new section to demonstrate the exactness, rather than relying on the abstract alone. revision: yes

  2. Referee: [Numerical work] Numerical confirmation of nominal error rates is cited, but without the explicit construction of the replication distribution or code, it is impossible to verify that the simulations truly use the induced (nuisance-free) distribution rather than an approximation that reintroduces estimation.

    Authors: We agree that explicit construction is necessary for verification. The simulations in the original manuscript were performed using the exact induced distributions derived from the ancillarity/sufficiency separations, without any nuisance estimation. In the revision, we have added detailed pseudocode for the simulation procedure in each example, explicitly showing how the nuisance-free replications are generated. Additionally, we will provide the accompanying R code as supplementary material upon acceptance to allow full reproducibility. revision: yes

Circularity Check

0 steps flagged

Minor reliance on established Fisherian principles; no load-bearing reduction to fitted inputs or self-citation chains

full rationale

The derivation invokes non-standard ancillarity/co-ancillarity and sufficiency/co-sufficiency separations to induce within-sample replication, replacing out-of-sample prediction error. These rest on classical parametric inference concepts rather than paper-specific definitions or fits. Numerical confirmation of nominal error rates under the postulated model provides external checkability. No equations reduce a prediction to a fitted parameter by construction, and self-citations (if present) are not load-bearing for the central claim. The framework remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on the domain assumption that parametric inference principles extend to induce replication via non-standard separations in semiparametric settings; no free parameters or new entities are introduced in the abstract.

axioms (1)
  • domain assumption Foundational principles of parametric statistical inference permit non-standard ancillarity and sufficiency separations that induce replication under the postulated model.
    This separation is invoked to replace out-of-sample prediction with within-sample assessment.

pith-pipeline@v0.9.0 · 5526 in / 1162 out tokens · 50500 ms · 2026-05-14T21:52:31.016198+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

  1. [1]

    and Janson, L

    Barber, R. and Janson, L. (2022). Testing goodness of fit and conditional independence with approximate cosufficient subsampling.Ann Statist., 50, 2514–2544

    work page 2022

  2. [2]

    Barndorff-Nielsen, O. E. and Cox, D. R. (1994).Inference and Asymptotics. Chapman and

    work page 1994

  3. [3]

    Battey, H. S. (2023). Inducement of population-level sparsity.Canad. J. Statist., 51, 760— 768 (Festschrift for Nancy Reid)

    work page 2023

  4. [4]

    Battey, H. S. (2024). Maximal co-ancillarity and maximal co-sufficiency.Information Ge- ometry, 7, 355–369

    work page 2024

  5. [5]

    Battey, H. S. and Cox, D. R. (2018). Large numbers of explanatory variables: a proba- bilistic assessment.Proc. Roy. Soc. Lond. A: Math. Phys. Sci., 474, 20170631

    work page 2018

  6. [6]

    Battey, H. S. and Cox, D. R. (2020). High-dimensional nuisance parameters: an example from parametric survival analysis.Info. Geom, 3, 119–148

    work page 2020

  7. [7]

    S., Cox, D

    Battey, H. S., Cox, D. R. and Lee, S. H. (2024). On partial likelihood and the construction of factorisable transformations.Information Geometry, 7, 9–28

    work page 2024

  8. [8]

    S., Rasines, D

    Battey, H. S., Rasines, D. G. and Tang, Y. (2025). Post-reduction inference for confidence sets of models.arXiv:2507.10373

    work page arXiv 2025

  9. [9]

    Birnbaum, A. (1954). Combining independent tests of significance.J. Amer. Statist. As- soc., 49, 559–574

    work page 1954

  10. [10]

    Box, G. E. P. and Cox, D. R. (1955). An analysis of transformations (with discussion). J. R. Statist. Soc. B, 26, 211–252

    work page 1955

  11. [11]

    Cox, D. R. (1955). Some statistical methods connected with series of events (with discus- sion).J. R. Statist. Soc. B, 17, 129–164. 30

    work page 1955

  12. [12]

    Cox, D. R. (1972). Regression models and life-tables (with discussion).J. R. Statist. Soc. B, 34, 187–220

    work page 1972

  13. [13]

    Cox, D. R. (1975). Partial likelihood.Biometrika, 62, 269–276

    work page 1975

  14. [14]

    Cox, D. R. and Lewis, P. A. W. (1966).The Statistical Analysis of Series of Events

    work page 1966

  15. [15]

    Cox, D. R. and Oakes, D. (1984).The Analysis of Survival Data. Chapman and Hall, London

    work page 1984

  16. [16]

    Cox, D. R. and Reid, N. (1987). Parameter orthogonality and approximate conditional inference (with discussion).J. R. Statist. Soc. B, 49, 1–39

    work page 1987

  17. [17]

    Cox, D. R. and Wong, M. Y. (2010). A note on the sensitivity to assumptions of a gener- alized linear mixed model.Biometrika, 97, 209–214

    work page 2010

  18. [18]

    L., Bien, J

    Dharamshi, A., Neufeld, A., Gao, L. L., Bien, J. and Witten, D. (2026). Decomposing Gaussians with unknown covariance.Biometrika, 113, article number asaf057

    work page 2026

  19. [19]

    Dudley, R. M. (2002).Real Analysis and Probability. Cambridge University Press, New York

    work page 2002

  20. [20]

    and Lilleg˚ ard, M

    Engen, S. and Lilleg˚ ard, M. (1997). Stochastic simulations conditioned on sufficient statis- tics.Biometrika, 84, 235–240

    work page 1997

  21. [21]

    Fisher, R. A. (1915). Frequency distribution of the values of the correlation coefficient in samples from an indefinitely large population.Biometrika, 10, 507–521

    work page 1915

  22. [22]

    Fisher, R. A. (1932).Statistical Methods for Research Workers. Oliver and Boyd, Edin- burgh

    work page 1932

  23. [23]

    Fisher, R. A. (1950). The significance of deviations from expectation in a Poisson series. Biometrics, 6, 17–24

    work page 1950

  24. [24]

    Heard, N. A. and Rubin-Delanchy, P. (2018). Choosing between methods of combining p-values.Biometrika, 105, 239–246

    work page 2018

  25. [25]

    H., Taraldsen, G., Lilleg˚ ard, M

    Lindqvist, B. H., Taraldsen, G., Lilleg˚ ard, M. and Engen, S. (2003). A counterexample to a claim about stochastic simulations.Biometrika, 90, 489–490

    work page 2003

  26. [26]

    Lindqvist, B. H. and Taraldsen, G. (2005). Monte Carlo conditioning on a sufficient statis- tic.Biometrika, 92, 451–464

    work page 2005

  27. [27]

    A., O’Reilly, F

    Lockhart, R. A., O’Reilly, F. J. and Stephens, M. A. (2007). Use of the Gibbs sampler to obtain conditional tests, with applications.Biometrika, 94, 992–998

    work page 2007

  28. [28]

    Rasines, D. G. and Young, G. A. (2023). Splitting strategies for post-selection inference. Biometrika, 110, 597–614

    work page 2023

  29. [29]

    Wong, W. (1982). Theory of partial likelihood.Ann. Statist., 14, 88–123. 31

    work page 1982