Bayesian bivariate survival estimation

C. Messan; J.K. Ghosh; Nils Lid Hjort; R.V. Ramamoorthi

arxiv: 2604.11819 · v1 · submitted 2026-04-10 · 🧮 math.ST · stat.TH

Bayesian bivariate survival estimation

J.K. Ghosh , Nils Lid Hjort , C. Messan , R.V. Ramamoorthi This is my paper

Pith reviewed 2026-05-10 16:06 UTC · model grok-4.3

classification 🧮 math.ST stat.TH

keywords bivariate survivalBayesian nonparametricDirichlet processBeta processposterior consistencyright censoringnonparametric estimation

0 comments

The pith

A Beta process prior with selective likelihood updating yields consistent Bayesian estimators for bivariate survival distributions while Dirichlet process priors produce inconsistent posteriors.

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

The paper shows that standard Dirichlet process priors lead to inconsistent posterior distributions for bivariate survival data by simplifying and extending an earlier counterexample. It then constructs a nonparametric prior using Beta processes and introduces an updating scheme that incorporates only the most relevant portions of the likelihood function. This construction avoids the negative mass problems that affect direct nonparametric estimators such as Dabrowska's and delivers consistency under typical right-censoring mechanisms. A reader would care because no simple extension of the Kaplan-Meier estimator exists for the bivariate case, so reliable Bayesian methods matter for joint survival analysis in medical and reliability studies.

Core claim

While the posterior from a Dirichlet process prior is inconsistent for estimating a bivariate survival function, as demonstrated by an extended Pruitt-type example, a Beta process prior together with a selective updating scheme that uses only the most relevant parts of the likelihood produces a consistent estimator for the bivariate survival distribution.

What carries the argument

Beta process prior combined with selective updating scheme that retains only the most relevant likelihood contributions.

If this is right

Dirichlet process priors produce inconsistent posteriors for bivariate survival estimation.
The Beta-process construction avoids assigning negative mass to subsets.
Consistency holds for the estimator under standard censoring schemes.
The method supplies a Bayesian route that sidesteps difficulties of direct nonparametric estimators like Dabrowska's.

Where Pith is reading between the lines

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

The selective updating approach may reduce computational cost in higher-dimensional survival problems by ignoring irrelevant likelihood terms.
The same prior and scheme could be tested for consistency in trivariate or competing-risks survival settings.
The construction suggests that other Lévy-process priors might be adapted with similar selective updates for multivariate censored data.

Load-bearing premise

The Beta process prior and its selective updating rule are assumed to guarantee consistency for bivariate survival data under standard censoring without further regularity conditions being stated.

What would settle it

A concrete simulation in which the proposed Beta-process estimator fails to converge in probability to the true bivariate survival function as sample size grows under independent right censoring.

read the original abstract

There is no easy extension of Kaplan-Meier and Nelson-Aalen estimators to the bivariate case, and estimating bivariate survival distributions nonparametrically is associated with various non-trivial problems. The Dabrowska estimator will for example associate negative mass to some subsets. Bayesian methods hold some promise as they will avoid the negative mass problem, butare also prone to difficulties. We simplify and extend an example by Pruitt to show that the posterior distribution from a Dirichlet process prior is inconsistent. We construct a different nonparametric prior via Beta processes and provide an updating scheme that utilizes only the most relevant parts of the likelihood, and show that this leads to a consistent estimator.

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.

Desk Editor's Note private letter to a colleague

Dirichlet priors turn out inconsistent for bivariate survival, but a tailored Beta process with selective updating delivers consistency.

read the letter

The main thing to know is that Dirichlet process priors produce inconsistent posteriors for bivariate survival under censoring, and the authors fix it by switching to a Beta process prior plus a selective updating rule that only pulls in the relevant likelihood pieces. They simplify and extend Pruitt's earlier example to make the inconsistency concrete, then build the new prior and show the posterior converges properly. This is genuinely new: the Beta-process construction and the selective scheme are not just a restatement of standard Dirichlet results but are adapted to the paired censored setting. It handles the negative-mass issue that hits frequentist estimators like Dabrowska without extra fixes. The approach is constructive and stays within the nonparametric Bayesian toolkit that people already use for survival data. The soft spot is that consistency rests on the selective updating working under the joint censoring law and on the Beta process having enough support; the abstract leaves the exact regularity conditions implicit, so the full proofs need to be checked for hidden restrictions that might limit use in messy real data. No circularity or invented quantities appear. This paper is for people working on multivariate survival or reliability who care about Bayesian nonparametric consistency. A reader already comfortable with process priors will get the most out of the technical details. It deserves a serious referee because the claims are specific, the fix is explicit, and the problem it solves is real. Send it to peer review with the usual request for proof verification and any simulation support.

Referee Report

0 major / 2 minor

Summary. The paper addresses the challenges of nonparametric estimation for bivariate survival distributions, noting that frequentist estimators like the Dabrowska estimator can assign negative mass. It simplifies and extends Pruitt's example to demonstrate that the posterior under a Dirichlet process prior is inconsistent. The authors then construct a nonparametric prior using Beta processes, introduce a selective updating scheme that incorporates only the most relevant parts of the likelihood, and establish that the resulting estimator is consistent.

Significance. If the consistency result holds, the work provides a useful Bayesian nonparametric framework for bivariate survival analysis that sidesteps negative-mass issues in frequentist methods and the inconsistency problems of standard Dirichlet process priors. The selective updating approach is a notable technical contribution for handling the likelihood in censored bivariate data.

minor comments (2)

The abstract contains a typographical error: 'butare' should read 'but are'.
The manuscript would benefit from a brief statement of the precise regularity conditions (e.g., on the joint censoring distribution) under which the Beta-process consistency holds, even if these appear in the technical sections.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our manuscript and the recommendation for minor revision. We are pleased that the technical contribution of the selective updating scheme for the Beta process prior is recognized as a way to achieve consistency in bivariate survival estimation, in contrast to the Dirichlet process.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper derives inconsistency of the Dirichlet process posterior by simplifying and extending an external example from Pruitt, then constructs a Beta process prior with a selective likelihood updating scheme and proves consistency for bivariate survival estimation. These steps are theoretical derivations of posterior behavior under stated nonparametric priors and censoring mechanisms. No self-definitional reductions, fitted parameters renamed as predictions, load-bearing self-citations, or ansatzes smuggled via prior work are present. The consistency claim rests on the explicit construction and updating rule rather than reducing to the input assumptions by definition.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that Beta processes form a suitable nonparametric prior class for bivariate survival functions and that the selective updating rule preserves the necessary martingale or convergence properties.

axioms (1)

domain assumption Beta processes can serve as nonparametric priors for bivariate survival functions.
Invoked when the paper constructs the prior to replace the Dirichlet process.

pith-pipeline@v0.9.0 · 5402 in / 1203 out tokens · 55745 ms · 2026-05-10T16:06:38.327087+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

9 extracted references · 9 canonical work pages

[1]

and Keiding, N

Andersen, P.K., Borgan, Ø., Gill, R.D. and Keiding, N. (1993).Statistical Models Based on Counting Processes.Springer-Verlag, New York

work page 1993
[2]

Dabrowska, D.M. (1988). Kaplan–Meier estimate on the plane.Annals of Statistics15, 1475–1489

work page 1988
[3]

Hjort, N.L. (1990). Nonparametric Bayes estimators based on Beta processes in models for life history data.Annals of Statistics18, 1259–1294

work page 1990
[4]

Judd, K.L. (1985). The law of large numbers with a continuum of IID random variables.Journal of Econometric Theory35, 19–25

work page 1985
[5]

and Shaked, M

Langberg, N.A. and Shaked, M. (1982). On the identifiability of multivariate life distribution functions.Annals of Probability10, 773–779. BAYESIAN BIVARIATE SURVIVAL ESTIMATION 15

work page 1982
[6]

Peterson, A.V. (1977). Expressind the Kaplan–Meier estimator as a function of empirical subsurvival functions.Journal of the American Statistical Association 72, 854–858

work page 1977
[7]

Pruitt, R.C. (1988). An inconsistent Bayes estimate in bivariate survival curve analysis. Preprint 529, University of Minnestota

work page 1988
[8]

Pruitt, R.C. (1991). On negative mass assigned in the bivariate Kaplan–Meier estimator.Annals of Statistics19, 443–453

work page 1991
[9]

Pruitt, R.C. (1993). Identifiability of bivariate survival curves from censored data.Journal of the American Statistical Association88, 573–579

work page 1993

[1] [1]

and Keiding, N

Andersen, P.K., Borgan, Ø., Gill, R.D. and Keiding, N. (1993).Statistical Models Based on Counting Processes.Springer-Verlag, New York

work page 1993

[2] [2]

Dabrowska, D.M. (1988). Kaplan–Meier estimate on the plane.Annals of Statistics15, 1475–1489

work page 1988

[3] [3]

Hjort, N.L. (1990). Nonparametric Bayes estimators based on Beta processes in models for life history data.Annals of Statistics18, 1259–1294

work page 1990

[4] [4]

Judd, K.L. (1985). The law of large numbers with a continuum of IID random variables.Journal of Econometric Theory35, 19–25

work page 1985

[5] [5]

and Shaked, M

Langberg, N.A. and Shaked, M. (1982). On the identifiability of multivariate life distribution functions.Annals of Probability10, 773–779. BAYESIAN BIVARIATE SURVIVAL ESTIMATION 15

work page 1982

[6] [6]

Peterson, A.V. (1977). Expressind the Kaplan–Meier estimator as a function of empirical subsurvival functions.Journal of the American Statistical Association 72, 854–858

work page 1977

[7] [7]

Pruitt, R.C. (1988). An inconsistent Bayes estimate in bivariate survival curve analysis. Preprint 529, University of Minnestota

work page 1988

[8] [8]

Pruitt, R.C. (1991). On negative mass assigned in the bivariate Kaplan–Meier estimator.Annals of Statistics19, 443–453

work page 1991

[9] [9]

Pruitt, R.C. (1993). Identifiability of bivariate survival curves from censored data.Journal of the American Statistical Association88, 573–579

work page 1993