Comparative Fertility

Analysis and visualization codebase behind:

Pattaro, S., Vanderbloemen, L. and Minton, J. (2020). Visualizing fertility trends for 45 countries using composite lattice plots. Demographic Research 42(23): 689–712. doi: 10.4054/DemRes.2020.42.23

…and its in-progress 2026 update (see UPDATE_WORKPLAN.md). Companion interactive app: https://datascapes.shinyapps.io/cumulative_fertility_app/.

Following the link from the 2020 paper? The repository exactly as it stood at publication is preserved at the tag demres-2020 — everything after that tag is the 2026 update. The published pipeline is also kept unmodified in the current tree (see Repo structure below).

Replicating the analyses: what to download, from where

Raw source data are not redistributed in this repo (HFD/HFC terms); the directories below are git-ignored. To rebuild everything from source:

1. Human Fertility Database (HFD) — https://www.humanfertility.org

   • Register (free) and log in.
   • Go to Data → Zipped Data Files and download the "All types of HFD data" bundle (HFD.zip, ~37 MB).
   • Extract into data/hfd_2026/. The pipeline needs asfrRR.txt (period age-specific fertility rates, Lexis squares); the bundle's birthsRR.txt/exposRR.txt support the alternative rates-from-counts route, and cpfrRR.txt/ccfrVH.txt are useful cross-checks on our own cumulative calculations.

2. Human Fertility Collection (HFC) — https://www.fertilitydata.org

   • From Zipped data files, download "ASFR and CPFR, standardized age scale → All birth orders combined, females" (HFC_ASFRstand_TOT.zip).
   • Optionally also "By birth order, females" (HFC_ASFRstand_BO.zip) — unused by the current pipeline, held for the birth-order extension (below).
   • Extract into data/hfc_2026/.

3. Run the 2026 pipeline (R; tested on R 4.1 with readr/dplyr/tidyr/ lattice/viridis — no latticeExtra required):

   Rscript scripts/ingest_2026.R    # raw files -> data/data_combined_and_standardised_2026.csv
   Rscript scripts/figures_2026.R   # published figure set on the new panel
   Rscript scripts/wall_2026.R      # Phase 3: fertility-ceiling / compression analysis

To reproduce the published 2020 figures no download is needed: the frozen 2016 combined dataset (data/data_combined_and_standardised.csv) is committed, and scripts/two_contour_version_of_figures.R is the script that produced the figures in the paper (kept byte-for-byte as published; the 2026 scripts never modify it).

Reproduction vs. replication of the 2020 findings. The published results remain reproducible from this repository (frozen data + frozen scripts, and the demres-2020 tag preserves the full 2020-era state including the manuscript history under manuscript/). They are, however, no longer exactly replicable from fresh source downloads: both the HFD and HFC revise historical series, and the revisions between the 2016 and 2026 releases are documented in reports/validation_2026.md (largest for Ireland, Spain and Denmark). This is inherent to living databases — and part of the case for updating. The frozen CSV in this repo is the authoritative record of the paper's inputs; anything not marked _2026 is 2020-era.

The paper

The paper introduces a refinement of the Lexis surface for fertility: level plots (colour = age-specific fertility rate, ASFR) composed with contour lines marking cumulative pseudo-cohort fertility (CPCFR) milestones — 2.05 (replacement) and 1.50 — so that period, age, cohort and cumulative-milestone information are readable in a single panel, small-multiplied across 45 countries. It appeared in Demographic Research's Special Issue on Demographic Data Visualization.

Headline empirical result (2020): once a country's cohorts fall below replacement (CPCFR < 2.05) they tend not to return; Norway and the USA were the two exceptions. The 2026 update reverses the headline: with data extended from 2014–15 to 2023–25, both exceptions have collapsed (Norway TFR 1.98 in 2009 → 1.40 in 2023; USA → ~1.6 in 2024), and the updated Norway panel shows the 2.05 contour returning to the surface for cohorts born 1958–68 and then escaping vertically — permanently, on current data — for cohorts born after ~1972.

The app

cumulative_fertility_app/ is a Shiny app (deployed at the shinyapps.io link above) that lets users build their own fertility Lexis surfaces per country: level-plot shading, milestone contours, period gridlines, linked tooltips/cross-sections, and a 3D surface view. Tooling note: the static composites use lattice/latticeExtra (produce_composite_lattice()), while the interactive surfaces are plotly — the app never used ggplot2. It still runs on the 2016 data; refreshing its data and redeploying is a pending step of the 2026 update (Phase 4).

The 2026 update

Tracked in UPDATE_WORKPLAN.md (the living plan, decision log, and handoff document). Status at 2026-07-11: data refresh (Phase 1), regenerated core figures (Phase 2), and the fertility-ceiling analysis (Phase 3) are done; app refresh and debate-facing outputs (Phase 4) and a Quarto/BibTeX manuscript modernization are planned. Key new artifacts:

• reports/validation_2026.md — new-vs-old build comparison (all discrepancies explained: HFD back-revisions, one code-mapping fix that restores France post-2008, source switches for Korea)
• figures/figures_2026/ — the published figure set on the new panel, plus ceiling-overlay and reachability figures
• reports/wall_analysis_2026.md — the "wall" headline numbers: the effective ceiling of the fertility lifecourse (highest age with ASFR ≥ 1/200) fell from ~45 (1950s) to ~41 (1980s–90s) and has crept back only to ~43 (2020s), while cumulative fertility above age 40 remains ~0.03–0.07 children — so rising ages at first birth compress replacement into a narrowing corridor ending in the early 40s. 43 of 45 countries have post-war cohorts observed past age 44 that never reached 2.05.

Repo structure

├── UPDATE_WORKPLAN.md            living plan + decision log — start here
├── reports/                      machine-generated review documents
├── data/
│   ├── hfd_2026/, hfc_2026/      raw 2026 downloads (git-ignored; see above)
│   ├── hfc/                      2016-era HFC remnants + code_definitions.csv (shared lookup)
│   ├── data_combined_and_standardised.csv       2016 build — FROZEN (published paper)
│   ├── data_combined_and_standardised_2026.csv  2026 build (+ hfd/hfc/interp source column)
│   └── derived_2026/             analysis tables (ceiling, late-mass, reachability)
├── scripts/
│   ├── (2016–2020 scripts)       FROZEN — reproduce the published paper
│   ├── ingest_2026.R             Phase 1: raw -> combined CSV
│   ├── functions_2026.R          shared data prep + composite plot function
│   ├── figures_2026.R            Phase 2: published figure set, new data
│   └── wall_2026.R               Phase 3: ceiling / compression / reachability
├── figures/
│   ├── (2016–2020 directories)   FROZEN
│   └── figures_2026/             all update figures
├── manuscript/                   published paper history, incl. peer-review rounds — FROZEN
├── cumulative_fertility_app/     Shiny app (Phase 4: data refresh + redeploy)
└── poster/, presentation/, stl/, support/, script.R, …   2016–19 working archive

Architecture rule: the published layer is never edited; every update artifact carries a _2026 suffix or lives in a _2026/reports directory. Data flows one way: raw (ignored) → ingest → combined CSV → shared prep (functions_2026.R) → figures + derived tables + human-readable reports.

Has anyone used the method? (uptake since 2020)

Honestly: not much. Semantic Scholar (July 2026) records 2 citations, 0 influential: the special issue's own editorial (2021), and — the one direct methodological descendant — "Visualising Age-Specific Fertility Patterns in the Former German Democratic Republic (GDR) and in East Germany, 1956–2024" (Comparative Population Studies, 2026), which applies Lexis-surface visualization to the East German case. (Google Scholar typically indexes a few more than Semantic Scholar; worth a manual check. Whether the CPS paper is independent of the original authors has not been verified.) The paper also appears in the HFD's official list of publications using HFD/HFC data.

Fertility research and discussion since 2020

The update lands in a very different conversation from the one the paper entered. Since publication: record-low fertility in Norway (1.40, 2023) and the Nordics generally, a record-low US TFR (~1.6, 2024), South Korea's descent to ~0.72, a COVID-era fluctuation now visible in the HFD's monthly STFF series, and a lively public debate about accelerating decline — including the "connectivity/smartphone" hypothesis (Burn-Murdoch's FT columns, Alice Evans's essays) and its skeptics (a 4G-rollout study finding detectable effects only among teens). Academic anchors include Kearney, Levine & Pardue's "puzzle of falling US birth rates," Hellstrand/Nisén/Myrskylä and Comolli et al. on the Nordic decline, Yoo & Sobotka on tempo in ultra-low-fertility Korea, and Beaujouan on latest-late fertility. The update's stance (see workplan): the composite plots contribute stylized facts any explanation must match — which ages, which cohorts, what timing, tempo or quantum — rather than adjudicating between co-timed candidate causes.

Birth order: the unexecuted extension

No written plans for a birth-order analysis exist anywhere in the repo or the manuscript history — but the data has now been downloaded for it twice (HFC_ASFRstand_BO in 2016, again in 2026, plus the HFD bundle's *bo.txt files), which records an intention if not a plan. It is the natural next analytical step: parity-specific composite surfaces would separate the postponement of first births (where the compression against the fertility ceiling binds hardest) from progression to second and higher-order births (where Zeman et al.'s 2018 decomposition locates much of the recent cohort decline). The Phase 3 machinery (wall_2026.R) generalizes to parity-specific ASFRs without structural change.