Bayesian approaches to non- and semiparametric density estimation [with a rejoinder to my discussants]

Nils Lid Hjort

arxiv: 2604.20238 · v1 · submitted 2026-04-22 · 🧮 math.ST · stat.TH

Bayesian approaches to non- and semiparametric density estimation [with a rejoinder to my discussants]

Nils Lid Hjort This is my paper

Pith reviewed 2026-05-09 23:12 UTC · model grok-4.3

classification 🧮 math.ST stat.TH

keywords Bayesian nonparametricdensity estimationDirichlet processHermite expansionsemiparametric modelsnonparametric priorslocal likelihood

0 comments

The pith

Bayesian density estimation proceeds by centering nonparametric priors on parametric models, using orthogonal basis expansions, or employing local parametric forms.

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

The paper organizes Bayesian approaches to estimating an unknown density into three categories without relying on a fixed parametric shape. One route builds a nonparametric prior around a parametric model using Dirichlet processes and their relatives. A second expresses the density as an additive or multiplicative expansion in orthogonal functions such as a robust Hermite series around the normal and places priors on the coefficients. The third treats the density locally as parametric, constructs local likelihoods to capture information, and places priors on the local parameters. These constructions aim to deliver Bayesian inference that quantifies uncertainty about the density form itself.

Core claim

Bayesian nonparametric and semiparametric density estimation is feasible through three families of constructions: nonparametric priors centered on parametric models via Dirichlet processes, orthogonal basis expansions with priors on the coefficients including a robust Hermite expansion around the normal, and local parametric approximations equipped with local likelihoods and local priors.

What carries the argument

The three categories of constructions that combine parametric structure with nonparametric flexibility: Dirichlet-process priors around a base model, coefficient priors on orthogonal expansions, and local parametric forms with local likelihoods.

If this is right

The Dirichlet-process route yields flexible mixtures that capture multimodality while retaining a parametric center.
The Hermite expansion supplies a concrete way to add controlled nonparametric correction around the normal distribution.
Local parametric modeling allows the shape parameters to vary smoothly across the support of the density.
All three routes furnish posterior uncertainty measures that parametric models alone cannot provide.

Where Pith is reading between the lines

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

These constructions could be adapted to semiparametric problems that combine a parametric component with a nonparametric density for the residuals.
The basis-expansion approach might prove especially convenient when computational tools for sampling from coefficient posteriors are already available.
Practical performance of the local-likelihood route would likely depend on how the locality bandwidth is chosen or integrated out.

Load-bearing premise

The proposed constructions produce valid posterior distributions that can support inference and estimation for the unknown density.

What would settle it

A simulation or theoretical example in which one of the three constructions yields an inconsistent posterior or fails to concentrate on the true density for a continuous non-normal target.

read the original abstract

This invited paper proposes and discusses several Bayesian attempts at nonparametric and semiparametric density estimation. The main categories of these ideas are as follows: 1) Build a nonparametric prior around a given parametric model. We look at cases where the nonparametric part of the construction is a Dirichlet process or relatives thereof. (2) Express the density as an additive expansion of orthogonal basis functions, and place priors on the coefficients. Here attention is given to a certain robust Hermite expansion around the normal distribution. Multiplicative expansions are also considered. (3) Express the unknown density as locally being of a certain parametric form, then construct suitable local likelihood functions to express information content, and place local priors on the local parameters.

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

This is an invited discussion paper that organizes three Bayesian strategies for nonparametric density estimation but adds no new theorems, proofs, or empirical checks.

read the letter

The key point is that this invited paper reviews three categories of Bayesian constructions for density estimation but does not deliver new formal results or validations. Hjort groups the ideas into nonparametric priors around parametric models via Dirichlet processes, orthogonal basis expansions with priors on the coefficients including a robust Hermite one, and local parametric forms using local likelihoods. The rejoinder to discussants is included, which can help address questions that come up. What works is the straightforward organization and the practical notes on things like robustness. It gives a sense of how these methods try to balance flexibility and structure. The soft spots are the absence of any derivations for posterior properties or consistency, and no examples or simulations to show how they perform. As a discussion piece this is normal, but it means the paper relies on prior literature for the heavy lifting. This is mainly for people in Bayesian statistics who want an overview of these approaches from an expert. A reader looking for a self-contained new contribution will not find it here. I think it deserves peer review for the discussion format, even though it is not a standard research paper with fresh theorems.

Referee Report

0 major / 1 minor

Summary. This invited paper proposes and discusses several Bayesian approaches to nonparametric and semiparametric density estimation, organized into three categories: (1) nonparametric priors around parametric models, with emphasis on Dirichlet process constructions and relatives; (2) orthogonal basis expansions (including a robust Hermite expansion around the normal distribution) with priors on the coefficients, along with multiplicative expansions; and (3) local parametric forms combined with local likelihood functions and local priors. The manuscript also contains a rejoinder to discussants.

Significance. If the outlined constructions can be shown to yield valid posteriors and consistent estimators, they would offer useful conceptual frameworks for Bayesian density estimation that blend parametric structure with nonparametric flexibility. The paper receives credit for its clear categorization of approaches and attention to practical considerations in an invited discussion format. However, because the manuscript provides no derivations, consistency results, convergence rates, or empirical validations, its significance is primarily as a stimulus for further research rather than a self-contained methodological advance.

minor comments (1)

The abstract and introduction would benefit from a brief statement clarifying that the paper focuses on conceptual proposals rather than formal proofs of posterior validity or consistency, to set reader expectations appropriately.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful summary of the manuscript and for recommending acceptance. The paper is an invited discussion piece whose primary goal is to categorize and outline several Bayesian approaches to nonparametric and semiparametric density estimation, together with a rejoinder to discussants. We agree with the referee's characterization of its role as a stimulus for further research.

read point-by-point responses

Referee: The manuscript provides no derivations, consistency results, convergence rates, or empirical validations, so its significance is primarily as a stimulus for further research rather than a self-contained methodological advance.

Authors: We agree. As an invited paper in discussion format, the manuscript deliberately focuses on conceptual frameworks, categorization of approaches (Dirichlet-process constructions, orthogonal expansions, and local parametric models), and practical considerations without supplying full technical proofs or numerical studies. Such developments are appropriate for follow-up research and are outside the scope of this invited contribution. revision: no

Circularity Check

0 steps flagged

No significant circularity

full rationale

The manuscript is an invited discussion paper that outlines three categories of Bayesian constructions for density estimation (Dirichlet-process-based nonparametric extensions, orthogonal basis expansions with coefficient priors, and local parametric forms with local likelihoods) and includes a rejoinder. It advances no formal theorems, consistency proofs, or predictive claims whose validity is asserted via equations or self-citations; the text remains at the level of conceptual motivation and practical considerations. No derivation chain exists that reduces outputs to inputs by construction, and the paper does not rely on load-bearing self-citations or fitted parameters renamed as predictions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Based solely on the abstract, the paper relies on established Bayesian nonparametric concepts such as Dirichlet process priors and orthogonal expansions; no new free parameters or invented entities are explicitly introduced in the summary.

axioms (2)

domain assumption Dirichlet process priors can be used to construct nonparametric priors around parametric models
Mentioned in the first category of approaches in the abstract.
domain assumption Orthogonal basis expansions like Hermite polynomials can represent densities with priors on coefficients
Described in the second category of the abstract.

pith-pipeline@v0.9.0 · 5415 in / 1292 out tokens · 35230 ms · 2026-05-09T23:12:39.297796+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

Antoniak, C.E. (1974). Mixtures of Dirichlet processes with application to Bayesian nonparametric problems.Annals of Statistics2, 1152–1174. Brunk, H.D. and Jones, P.W. (1990). Fitting conditional distributions using orthogonal expansions. Journal of Statistical Planning and Inference26, 325–337. Bunke, O. (1987). Bayesian inference in semiparametric mode...

work page 1974
[2]

West, M. (1992). Modelling with mixtures. InBayesian Statistics 4(J.M. Bernardo, J.O. Berger, A.P. Dawid and A.F.M. Smith, eds.), 503–524. Oxford University Press, Oxford. 27 Rejoinder to the discussion I am grateful for the many comments that were oﬀered on my paper, those that materialised as written discussion contributions as well as several points of...

work page 1992

[1] [1]

Antoniak, C.E. (1974). Mixtures of Dirichlet processes with application to Bayesian nonparametric problems.Annals of Statistics2, 1152–1174. Brunk, H.D. and Jones, P.W. (1990). Fitting conditional distributions using orthogonal expansions. Journal of Statistical Planning and Inference26, 325–337. Bunke, O. (1987). Bayesian inference in semiparametric mode...

work page 1974

[2] [2]

West, M. (1992). Modelling with mixtures. InBayesian Statistics 4(J.M. Bernardo, J.O. Berger, A.P. Dawid and A.F.M. Smith, eds.), 503–524. Oxford University Press, Oxford. 27 Rejoinder to the discussion I am grateful for the many comments that were oﬀered on my paper, those that materialised as written discussion contributions as well as several points of...

work page 1992