# Swiss private-purpose reading distribution: methods Scope: national permanent residents aged 15+ in private households. Books read privately in the twelve months preceding interview, including ebooks and excluding comics. Observations are separate survey waves, not a panel. Total/all-purpose reading ends in 2019 and is not used. Source: BFS workbook 36216339, Bücher_Privat, national rows 99 (2014), 63 (2019), 5 (2024). Columns B,D,F,H,J hold any,1–3,4–12,13+,none; adjacent columns hold published marginal 95% CI half-widths. Unknown/no-answer are omitted by BFS. Keep residual 100−any−none distinct from none. No missing value is replaced by zero. Data version: 2025-11-14. Define R=population percentage any books, H=population percentage 13+ books, Q=100H/R (frequent percentage among reported readers). H=RQ/100. For each interval, prevalence-associated component=change(R)×mean(Q)/100; composition-associated component=change(Q)×mean(R)/100. Components sum to change(H). Periods: 2014→2019, 2019→2024, 2014→2024. This is a symmetric arithmetic allocation of published margins, not effects, explanations or people moving categories. 2024 methodological break: primarily CAWI with CATI alternative replaced CATI plus supplemental CAWI/PAPI; shortened August–October fieldwork, changed questionnaire presentation and nonresponse weighting including income information. No behavioral attribution of observed changes. Marginal source CIs do not supply covariance for ratio/component intervals; none are invented. Reproduction (Python 3 standard library only): python reproduce.py --input selected-source-inputs.csv --output output Produces selected-inputs-and-ratios.csv and decomposition.csv, retaining Decimal computational precision. Input contains three bounded published national rows, not microdata. No network or private path needed. Original raw sources are optional and not redistributed; URLs/checksums are in sources.json and source-hashes.json. Narrative rounds for readability; CSV retains source precision and 40-significant-digit arithmetic (not additional measurement accuracy). Independent direct-XLSX-XML computation agrees within 1.8e-15 percentage points. One-decimal inputs and ±0.05 endpoint rounding checks retain the larger composition-associated component for the decade and 2019→2024. Rounding bounds are not CIs. Source-informed calculation specification was fixed after inspecting the published inputs and before derived computation; it was not outcome-blind preregistration. BFS already charts category prevalence; additions are conditional shares, residual accounting and the explicit decomposition.