pith. sign in

arxiv: 2504.01148 · v5 · pith:FAT5V5AXnew · submitted 2025-04-01 · 📊 stat.AP

Methodological insights in Bayesian Age-Period-Cohort analysis: an application to the case of Puerto Rico's fertility decline

Pith reviewed 2026-05-22 22:33 UTC · model grok-4.3

classification 📊 stat.AP
keywords Bayesian APC modelfertility declinePuerto Ricocohort effectsage-period-cohort analysisdemographic modelingtotal fertility ratePoisson likelihood
0
0 comments X

The pith

Bayesian APC model shows cohort effects predominate in Puerto Rico fertility decline

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

The paper develops and applies a Bayesian Age-Period-Cohort model to Puerto Rico fertility rates from 1948 to 2022 in order to separate how much of the decline comes from age patterns, time periods, or birth cohorts. A sympathetic reader would care because the results indicate that recent birth cohorts carry most of the explanatory weight, with groups born in 1963-1967 and later displaying markedly lower fertility than earlier ones. The analysis also finds no evidence that women are simply shifting births to older ages, which differs from patterns reported elsewhere. This is presented as the first APC application to fertility data for Puerto Rico and the first to conclude that cohort effects outweigh period effects in this setting.

Core claim

Both the Bayesian model and a frequentist counterpart attribute greater importance to cohort effects than to period effects when accounting for fertility changes; birth cohorts from 1963-1967 onward exhibit notably low fertility rates, and the data show no postponement of births to later maternal ages.

What carries the argument

The Bayesian Age-Period-Cohort model with Poisson likelihood, random-walk-of-order-2 autoregressive priors on the age, period, and cohort effects, and Scaled Beta2 priors on the precision parameters, used to produce stable estimates that separate the three components despite their linear dependence.

If this is right

  • Public policies addressing fertility would need to target factors specific to recent birth cohorts rather than broad period shocks.
  • Fertility rates are expected to stay low because the low-fertility cohorts will continue to dominate reproductive ages.
  • Interventions focused on reversing birth postponement are not supported by the Puerto Rico data.
  • The agreement between Bayesian and frequentist results increases in the cohort-dominant interpretation for this population.

Where Pith is reading between the lines

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

  • Generational shifts in values, economic conditions, or migration patterns specific to those cohorts may be the main drivers worth investigating next.
  • The same modeling approach could be tested on other low-fertility Caribbean or Latin American populations to check whether cohort dominance is regional.
  • Population projections for Puerto Rico could incorporate these cohort-specific fertility schedules to improve forecasts of future age structure and labor supply.

Load-bearing premise

The chosen random-walk priors on the effects and Scaled Beta2 priors on precisions resolve the age-period-cohort identifiability problem on this dataset without biasing the results toward cohort effects.

What would settle it

Re-estimating the same data under alternative priors or with a frequentist APC method that instead assigns greater weight to period effects would falsify the claim of cohort predominance.

read the original abstract

Age-Period-Cohort (APC) models are of special importance in Demography and Epidemiology for analyzing panel data according to three different factors: biological (age), technological (period) and cultural (cohort). The main goal of APC modeling is to separate the explanation of both period and cohort effects to the phenomenon. The objective of this paper is to develop a Bayesian Age-Period-Cohort framework that can model a wide range of demographic and epidemiological phenomena and improve upon existing statistical methodologies. The APC framework consists of addressing three main challenges: (1) the identification problem of all APC models, usually managed by imposing constraints on effect groups, (2) considering expert knowledge in the model definition, and (3) efficient solution of computational issues. By allowing full parameter uncertainty, use of robust priors, and an efficient computational implementation, a Bayesian methodology manages these concerns. Bayesian models also produce results that allow intuitive implementation and support theoretical knowledge. Our original methodology consists of the use of (i) a Scaled Beta2 prior distribution for the scale parameters, (ii) imposing different period and cohort constraints and comparing them,(iii) user-friendly implementation that can be easily adapted to the event, and (iv) various model comparison criteria that leads to reasonable interpretation of APC effects. We examine the dramatic collapse of fertility in Puerto Rico, an application that is difficult to model due to the accelerated changes and has interesting demographic implications that challenge the predominance of period effects in lowest-low fertility countries, emphasizing the cohort (cultural) momentum. The scope of the methodology introduced here is wide, including applications to obesity or smoking studies, for example.

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 develops a Bayesian Age-Period-Cohort (APC) model with Poisson likelihood, RW(2) autoregressive priors on the age, period, and cohort effects, and Scaled Beta2 priors on the precision parameters to analyze fertility rates in Puerto Rico from 1948-2022. It concludes that cohort effects predominate over period effects in explaining the fertility decline, with birth cohorts from 1963-1967 onward exhibiting notably low fertility rates, and reports no evidence of birth postponement (contrary to patterns in other lowest-low fertility countries). The analysis is compared to frequentist methods, with claimed agreement between the two, and is presented as the first APC study of Puerto Rican fertility data.

Significance. If the attribution of fertility changes primarily to cohort effects proves robust, the work would offer useful demographic insight into Puerto Rico's distinctive low-fertility, high-migration, aging population context and could inform targeted policy. The explicit Bayesian formulation with RW(2) and Scaled Beta2 priors, together with the side-by-side frequentist comparison, constitutes a methodological strength for applied demography.

major comments (2)
  1. [Model specification and identifiability discussion] The central claim that cohort effects predominate rests on the posterior decomposition of the APC effects. The model places independent RW(2) priors on age, period, and cohort together with Scaled Beta2 priors on their precisions, yet the manuscript provides no demonstration that this structure resolves the linear dependence (cohort = period − age) in a manner invariant to reparameterization, sum-to-zero placement, or hyperprior choice. Without such evidence the ranking of effect importance cannot be confirmed as data-driven rather than prior-driven on this short, sparse series.
  2. [Results and comparison to frequentist model] The abstract states agreement between Bayesian and frequentist fits and attributes more importance to cohort effects, but the results section does not report quantitative measures of relative importance (e.g., posterior variance shares, deviance explained by each component, or cross-validation metrics) nor sensitivity checks under alternative constraints. This omission is load-bearing for the headline conclusion that cohort effects explain the decline more than period effects.
minor comments (2)
  1. [Abstract] The abstract contains a duplicated sentence: 'Birth cohorts born in 1963-1967 onward have notably low fertility rates.' appears twice consecutively.
  2. [Methods] Notation for the Scaled Beta2 priors and the precise form of the RW(2) precision parameters should be stated explicitly with equations to allow replication.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight important areas for strengthening the manuscript. We respond to each major comment below and commit to revisions that address the concerns raised.

read point-by-point responses
  1. Referee: The central claim that cohort effects predominate rests on the posterior decomposition of the APC effects. The model places independent RW(2) priors on age, period, and cohort together with Scaled Beta2 priors on their precisions, yet the manuscript provides no demonstration that this structure resolves the linear dependence (cohort = period − age) in a manner invariant to reparameterization, sum-to-zero placement, or hyperprior choice. Without such evidence the ranking of effect importance cannot be confirmed as data-driven rather than prior-driven on this short, sparse series.

    Authors: We agree that an explicit demonstration of robustness to identifiability constraints is needed. Our model employs standard sum-to-zero constraints and RW(2) priors as in the Bayesian APC literature, but we did not report sensitivity to alternative placements or hyperpriors. In the revised manuscript we will add a dedicated subsection with sensitivity analyses under varied constraints and Scaled Beta2 hyperparameter values, confirming that the cohort predominance conclusion is preserved. revision: yes

  2. Referee: The abstract states agreement between Bayesian and frequentist fits and attributes more importance to cohort effects, but the results section does not report quantitative measures of relative importance (e.g., posterior variance shares, deviance explained by each component, or cross-validation metrics) nor sensitivity checks under alternative constraints. This omission is load-bearing for the headline conclusion that cohort effects explain the decline more than period effects.

    Authors: The referee correctly identifies that the results section relies primarily on visual comparisons rather than quantitative decomposition. We will revise the results to report posterior variance shares for each APC component, the proportion of deviance explained by age, period, and cohort, and any feasible cross-validation metrics. These additions, together with the sensitivity checks noted above, will be included to substantiate the claim of cohort predominance. revision: yes

Circularity Check

0 steps flagged

No circularity: posterior attribution of cohort importance follows from data fit under standard priors

full rationale

The paper fits a Bayesian APC Poisson model with RW(2) priors on age/period/cohort effects and Scaled Beta2 priors on precisions to Puerto Rico fertility counts (1948-2022). The central claim that cohort effects predominate is obtained by inspecting posterior means and variances after MCMC sampling; this quantity is not shown by any equation in the manuscript to equal a function of the data or priors by algebraic identity. No self-citation is invoked to justify the identifiability constraint, no fitted parameter is relabeled as a prediction, and no ansatz is smuggled via prior work. The derivation chain therefore remains self-contained against the observed rates.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The model rests on standard Poisson count likelihood and RW(2) autoregressive structure for the APC effects; these are domain-standard assumptions rather than new inventions. The Scaled Beta2 prior on precision parameters is a modeling choice whose justification is not detailed in the abstract.

free parameters (1)
  • precision parameters for RW(2) effects
    Scaled Beta2 priors are placed on the precision hyperparameters of the age, period, and cohort random walks; these are estimated from the data.
axioms (2)
  • domain assumption Poisson likelihood for age-specific birth counts
    Standard for fertility count data; invoked implicitly by the Bayesian APC setup.
  • domain assumption RW(2) autoregressive priors on age, period, and cohort effects
    Common smoothing device in Bayesian APC models to handle the linear dependence among the three factors.

pith-pipeline@v0.9.0 · 5804 in / 1562 out tokens · 41003 ms · 2026-05-22T22:33:13.839113+00:00 · methodology

discussion (0)

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