Asymptotic properties of the MLE in distributional regression under random censoring
Pith reviewed 2026-05-22 23:39 UTC · model grok-4.3
The pith
The maximum likelihood estimator for distributional regression is almost surely consistent and asymptotically normal under random right censoring.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
In distributional regression with randomly right-censored responses, the maximum likelihood estimator of the parameters that identify the conditional distribution from the given parametric family is almost surely consistent and asymptotically normal.
What carries the argument
The maximum likelihood estimator based on the censored-data likelihood, where each observation contributes either the density or the survival function according to whether the response is observed or censored.
If this is right
- The estimator converges almost surely to the true parameter value as the number of observations tends to infinity.
- The properly scaled estimation error converges in distribution to a multivariate normal limit.
- Large-sample confidence intervals and tests derived from the asymptotic normality are valid under the stated conditions.
- Standard likelihood-based model comparison remains justified for censored distributional regression.
Where Pith is reading between the lines
- The same estimator can be plugged directly into existing software packages for censored data without custom derivation.
- Similar consistency and normality arguments could be checked for left censoring or interval censoring under parallel regularity assumptions.
- Finite-sample bias corrections or bootstrap refinements may still be needed even when the asymptotic results hold.
Load-bearing premise
The parametric family, censoring mechanism, and covariate distribution satisfy unspecified regularity conditions that are standard in censored MLE theory.
What would settle it
A sequence of data sets generated from a correctly specified parametric conditional model with random right censoring in which the MLE fails to converge almost surely to the true parameter as sample size grows.
read the original abstract
Distributional regression aims to find the best candidate in a given parametric family of conditional distributions to model a given dataset. As each candidate in the distribution family can be identified by the corresponding distribution parameters, a common approach for this task is to use the maximum likelihood estimator (MLE) for the parameters. In this paper, we establish theoretical results for this estimator in case the response variable is subject to random right censoring. In particular, we provide proofs of almost sure consistency and asymptotic normality of the MLE under censoring. The empirical behavior is illustrated by a simulation study and a real data example.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript claims to derive almost-sure consistency and asymptotic normality of the maximum-likelihood estimator for the parameters of a parametric conditional distribution family in a distributional regression model when the response is subject to random right censoring. The claims are supported by proofs whose details are asserted to exist; the paper also contains a simulation study and a real-data illustration.
Significance. If the proofs are complete and the required regularity conditions hold uniformly in the covariate, the results would constitute a useful but incremental extension of classical censored MLE theory to the distributional-regression setting. The contribution is primarily technical; no new methodological device or falsifiable prediction is introduced.
major comments (1)
- [Abstract and proof sections (unspecified)] The manuscript asserts proofs of almost-sure consistency and asymptotic normality but neither states nor verifies the regularity conditions required for these limit theorems under random censoring (identifiability of the conditional family, twice continuous differentiability of the observed-data log-likelihood, positive-definiteness of the integrated Fisher information, and domination conditions permitting interchange of derivative and integral). These conditions are load-bearing for both central claims; without an explicit list and verification that they hold uniformly in the covariate and under the censoring mechanism, the proofs cannot be checked for correctness.
Simulated Author's Rebuttal
We thank the referee for the careful reading and the constructive comment on the regularity conditions. We address the major comment below.
read point-by-point responses
-
Referee: [Abstract and proof sections (unspecified)] The manuscript asserts proofs of almost-sure consistency and asymptotic normality but neither states nor verifies the regularity conditions required for these limit theorems under random censoring (identifiability of the conditional family, twice continuous differentiability of the observed-data log-likelihood, positive-definiteness of the integrated Fisher information, and domination conditions permitting interchange of derivative and integral). These conditions are load-bearing for both central claims; without an explicit list and verification that they hold uniformly in the covariate and under the censoring mechanism, the proofs cannot be checked for correctness.
Authors: We agree that the regularity conditions must be stated explicitly and verified for the proofs to be fully checkable. In the revised version we will insert a new subsection (immediately preceding the consistency and asymptotic normality theorems) that enumerates the full set of conditions—conditional identifiability of the parametric family, twice continuous differentiability of the observed-data log-likelihood, positive-definiteness of the integrated Fisher information matrix, and suitable domination conditions allowing interchange of derivative and integral—and we will supply brief arguments establishing that each condition holds uniformly in the covariate under the random right-censoring mechanism adopted in the paper. revision: yes
Circularity Check
No circularity detected in consistency and normality proofs
full rationale
The paper derives almost-sure consistency and asymptotic normality of the MLE directly from the observed-data likelihood under random right censoring. These limit theorems are obtained via standard arguments on the parametric conditional family and censoring mechanism; no step reduces by the paper's own equations to a fitted quantity, self-definition, or load-bearing self-citation. The central claims remain independent mathematical results rather than tautological renamings or imported uniqueness theorems.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.