Bayesian Auxiliary Variable Model for Birth Records Data with Qualitative and Quantitative Responses

Julia Gohlke; Lulu Kang; Shyam Ranganathan; Xiaoning Kang; Xinwei Deng

arxiv: 2008.06525 · v1 · submitted 2020-08-14 · 📊 stat.ME

Bayesian Auxiliary Variable Model for Birth Records Data with Qualitative and Quantitative Responses

Xiaoning Kang , Shyam Ranganathan , Lulu Kang , Julia Gohlke , Xinwei Deng This is my paper

Pith reviewed 2026-05-24 13:50 UTC · model grok-4.3

classification 📊 stat.ME

keywords Bayesian modelingjoint modelinglatent variablepreterm birthbirth weightauxiliary variableMCMCbirth records

0 comments

The pith

A Bayesian auxiliary variable model jointly analyzes preterm birth and birth weight by linking them with a latent variable.

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

The paper proposes a Bayesian method for jointly modeling qualitative responses like preterm birth and quantitative responses like birth weight when they are associated. Separate modeling ignores this link and loses predictive power, while the joint approach uses a latent variable to capture and measure the dependence. Posterior inference is done via MCMC, and simulations confirm better predictions for both response types. The model is applied to Virginia birth records data to study the mutual dependence between the two outcomes.

Core claim

The authors introduce a Bayesian auxiliary variable model that connects a probit model for the binary response with a linear regression for the continuous response through a shared latent variable, allowing joint estimation of parameters and assessment of association strength via the covariance structure in the latent space.

What carries the argument

The auxiliary latent variable that serves as the link between the qualitative and quantitative response models, enabling quantification of their dependency.

If this is right

Joint modeling leads to improved prediction capacity for both the qualitative and quantitative responses compared to separate models.
The strength of the dependency between preterm birth and birth weight can be directly assessed from the model parameters.
The MCMC algorithm provides efficient sampling from the joint posterior distributions.
Application to birth records reveals the mutual dependence in real data from Virginia Department of Health.

Where Pith is reading between the lines

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

If the single latent variable assumption holds, this approach could be adapted to other mixed-response datasets in health or social sciences.
Extending the model to include multiple latent variables might capture more complex dependencies not addressed here.
Policy applications could use the dependency measure to prioritize interventions that affect both birth outcomes.

Load-bearing premise

That a single latent variable structure is sufficient to capture the full association between the qualitative and quantitative responses without residual dependence.

What would settle it

Observing that the joint model's prediction errors on the birth records data are not smaller than those from independent models for preterm birth and birth weight would challenge the claim of improved prediction.

Figures

Figures reproduced from arXiv: 2008.06525 by Julia Gohlke, Lulu Kang, Shyam Ranganathan, Xiaoning Kang, Xinwei Deng.

**Figure 2.** Figure 2: Boxplots of RMSPE and mis-classification error for preterm birth data for each [PITH_FULL_IMAGE:figures/full_fig_p026_2.png] view at source ↗

**Figure 3.** Figure 3: Regression coefficient distributions for the explanatory variables (1 indicates the [PITH_FULL_IMAGE:figures/full_fig_p027_3.png] view at source ↗

read the original abstract

Many applications involve data with qualitative and quantitative responses. When there is an association between the two responses, a joint model will provide improved results than modeling them separately. In this paper, we propose a Bayesian method to jointly model such data. The joint model links the qualitative and quantitative responses and can assess their dependency strength via a latent variable. The posterior distributions of parameters are obtained through an efficient MCMC sampling algorithm. The simulation shows that the proposed method can improve the prediction capacity for both responses. We apply the proposed joint model to the birth records data acquired by the Virginia Department of Health and study the mutual dependence between preterm birth of infants and their birth weights.

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 a standard Bayesian latent-variable joint model for binary and continuous responses that shows the expected prediction gains in simulation and applies it to birth records, with no major internal problems.

read the letter

The main thing here is a Bayesian auxiliary variable construction that links a qualitative response like preterm birth to a quantitative one like birth weight through a latent variable, then samples the joint posterior with MCMC. The simulation checks that joint modeling beats separate fits when dependence exists, and the authors run the model on Virginia birth records to look at that dependence in practice. That is the core of the paper. It is new only in the narrow sense of spelling out this particular auxiliary-variable setup and MCMC scheme for this data combination; the underlying idea of latent linking for mixed types is already established. The work does well by keeping the focus on prediction improvement and by including both a controlled simulation and a real health-data example, which makes the practical case clearer than a methods-only paper would. The evidence is not circular because the simulation uses independent data generation and the birth records are external. The central claim holds up under the stated assumptions. The soft spots are modest and proportional. The model rests on one latent variable being enough to capture the association; if residual dependence remains, the joint posterior could be off, though the simulation tests the case where the assumption is true. The abstract gives little on prior choices or convergence checks, so a reader would want those details to judge robustness, but nothing in the description points to an identifiability failure or a simulation design flaw that would invalidate the reported gains. This paper is for applied statisticians or biostatisticians who already work with mixed responses in health records and want a ready Bayesian joint option. A reader who needs exactly this setup or who is comparing joint versus separate models in similar data would get concrete value from the simulation numbers and the data example. I would bring it to a reading group as maybe. I would not cite it in my own work in the next year. It deserves peer review because the method is coherent, the evidence is independent, and the application is relevant; referees can usefully check the technical implementation and any sensitivity to the latent-variable assumption.

Referee Report

1 major / 2 minor

Summary. The paper proposes a Bayesian auxiliary variable model for jointly modeling a binary qualitative response and a continuous quantitative response. A latent variable links the two responses to capture and quantify their dependence; posterior inference uses an efficient MCMC algorithm. Simulations demonstrate improved predictive performance for both responses relative to separate modeling, and the method is applied to Virginia birth records data to study the dependence between preterm birth and birth weight.

Significance. If the latent-variable construction is correctly specified and the MCMC mixes adequately, the framework supplies a coherent joint posterior for mixed responses that can improve prediction when dependence is present. The simulation design and birth-records application constitute independent checks rather than tautological fits, which strengthens the practical claim.

major comments (1)

[Model specification (likely §2)] The central modeling assumption—that a single latent variable suffices to capture all dependence between the binary and continuous responses without residual association—is load-bearing for the prediction-improvement claim, yet the manuscript provides no formal test (e.g., posterior predictive check for residual correlation) or sensitivity analysis to this assumption.

minor comments (2)

[Abstract and §4] The abstract states that the simulation shows improved prediction but supplies no numerical metrics (e.g., MSE, AUC, or coverage) or details on prior choices and convergence diagnostics; these should be added to the main text or supplementary material for reproducibility.
[§2] Notation for the auxiliary variable and the link functions between the latent variable and the two response types should be made fully explicit with a single equation block rather than scattered definitions.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their thoughtful review and constructive feedback on our manuscript. We address the single major comment below and outline the revisions we will make.

read point-by-point responses

Referee: The central modeling assumption—that a single latent variable suffices to capture all dependence between the binary and continuous responses without residual association—is load-bearing for the prediction-improvement claim, yet the manuscript provides no formal test (e.g., posterior predictive check for residual correlation) or sensitivity analysis to this assumption.

Authors: We agree that the assumption of a single latent variable fully capturing the dependence (i.e., conditional independence of the responses given the latent) is central to the model and to the reported gains in predictive performance. The auxiliary-variable construction is deliberately specified in this way to induce the observed association through the shared latent, consistent with standard joint modeling approaches for mixed responses. However, the manuscript indeed does not include a formal posterior predictive check for residual correlation or a sensitivity analysis to this modeling choice. In the revised version we will add (i) a posterior predictive check that compares the observed pairwise association (e.g., tetrachoric or polyserial correlation) against the posterior predictive distribution under the fitted model, and (ii) a brief sensitivity analysis that refits the model after introducing an additional direct residual correlation parameter and reports the resulting change in predictive metrics. These additions will directly address the concern and strengthen the justification for the reported improvements. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents a standard Bayesian latent-variable data-augmentation construction for jointly modeling binary and continuous responses. The abstract and summary describe the model linking responses via a latent variable, with posterior sampling via MCMC, and validation via simulation and birth-records application. No load-bearing step reduces by construction to a fitted parameter or self-citation chain; the simulation design tests improvement when dependence exists and is independent of the model equations themselves. The derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

Abstract-only review limits visibility into exact parameter counts and assumptions; the model relies on standard Bayesian priors and a latent-variable linking assumption whose details are not provided.

free parameters (1)

prior hyperparameters
Bayesian models require specification of prior distributions whose parameters are chosen or tuned and not derived from the data.

axioms (1)

domain assumption A latent variable adequately captures the dependence between qualitative and quantitative responses
The joint model is built around this linking mechanism as stated in the abstract.

invented entities (1)

latent auxiliary variable no independent evidence
purpose: To link the two response types and quantify their mutual dependence
Introduced as the core modeling device; no independent evidence outside the model is described.

pith-pipeline@v0.9.0 · 5644 in / 1189 out tokens · 25992 ms · 2026-05-24T13:50:43.405659+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We introduce a latent variable of U to facilitate this task. Assume the binary response follows the Bernoulli distribution Z=1 if U≥0 ... with U|β1,x∼N(x′β1,1) ... Y|β2,σ²,x∼N(x′β2,σ²). To link ... bivariate normal ... Σ=[1,ρσ;ρσ,σ²]
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

The joint model links the qualitative and quantitative responses and can assess their dependency strength via a latent variable.

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

3 extracted references · 3 canonical work pages

[1]

Bayesian optimal blocking of factorial designs,

Ai, M., Kang, L., and Joseph, V. R. (2009), “Bayesian optimal blocking of factorial designs,” Journal of Statistical Planning and Inference , 139(9), 3319–3328. Catalano, P. J., and Ryan, L. M. (1992), “Bivariate latent variable models for clustered 20 discrete and continuous outcomes,” Journal of the American Statistical Association , 87(419), 651–658. C...

work page 2009
[2]

QQ Models: Joint Modeling for Quantitative and Qualitative Quality Responses in Manufacturing Systems,

Deng, X., and Jin, R. (2015), “QQ Models: Joint Modeling for Quantitative and Qualitative Quality Responses in Manufacturing Systems,” Technometrics, 57(3), 320–331. Dunson, D. B. (2000), “Bayesian latent variable models for clustered mixed outcomes,” Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 62(2), 355–

work page 2015
[3]

Dynamic latent trait models for multidimensional longitudinal data,

Dunson, D. B. (2003), “Dynamic latent trait models for multidimensional longitudinal data,” Journal of the American Statistical Association , 98(463), 555–563. Fitzmaurice, G. M., and Laird, N. M. (1995), “Regression models for a bivariate discrete and continuous outcome with clustering,” Journal of the American statistical Association , 90(431), 845–852....

work page 2003

[1] [1]

Bayesian optimal blocking of factorial designs,

Ai, M., Kang, L., and Joseph, V. R. (2009), “Bayesian optimal blocking of factorial designs,” Journal of Statistical Planning and Inference , 139(9), 3319–3328. Catalano, P. J., and Ryan, L. M. (1992), “Bivariate latent variable models for clustered 20 discrete and continuous outcomes,” Journal of the American Statistical Association , 87(419), 651–658. C...

work page 2009

[2] [2]

QQ Models: Joint Modeling for Quantitative and Qualitative Quality Responses in Manufacturing Systems,

Deng, X., and Jin, R. (2015), “QQ Models: Joint Modeling for Quantitative and Qualitative Quality Responses in Manufacturing Systems,” Technometrics, 57(3), 320–331. Dunson, D. B. (2000), “Bayesian latent variable models for clustered mixed outcomes,” Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 62(2), 355–

work page 2015

[3] [3]

Dynamic latent trait models for multidimensional longitudinal data,

Dunson, D. B. (2003), “Dynamic latent trait models for multidimensional longitudinal data,” Journal of the American Statistical Association , 98(463), 555–563. Fitzmaurice, G. M., and Laird, N. M. (1995), “Regression models for a bivariate discrete and continuous outcome with clustering,” Journal of the American statistical Association , 90(431), 845–852....

work page 2003