# Methods: Italian library reading-group provision, 2025

Author: Alisson Bento. Secondary descriptive analysis of Istat Tables 3, 25 and 26, accessed October 30, 2026. Observation year 2025; fieldwork March–July 2026; release October 12, 2026.

## Universe and unit
The unit is a library whose predominant function is public reading, not a municipality, resident or member. Questionnaire p9 Q15=1 identifies that function; p11 Q20.1a is restricted to it. Table 3 gives 6,252 libraries (5,851 publicly owned, 401 privately owned). The wider census has 8,883 eligible libraries across all functions. Six ordinary size bands total 5,975 public-reading libraries; the separate metropolitan-capital category adds 277. The >50,000 ordinary band excludes that separate category.

## Variables and calculations
Table 25 supplies Yes, No and unknown/nonresponse percentages; Table 26 supplies rounded conditional means. Q20.1b asks Yes libraries for groups organized during 2025, not meetings. The mean footnote limits the denominator to item respondents; exact numerical-response counts are not published here.

Known-status sensitivity = 100 × Yes/(Yes + No). Missing-status lower scenario = Yes; upper scenario = 100 − No. These use rounded source percentages and are approximate, not confidence intervals. Independent rounding explains small deviations from Yes + Unknown. The six-band availability range is max(Yes) − min(Yes) = 32.7 percentage points; its max/min ratio is about 2.1315. The ratio of printed conditional means is 4/3, not an exact ratio of underlying means. Under explicitly assumed nearest-unit rounding, underlying means would allow a max/min ratio up to 4.5/2.5 = 1.8; nearest-tenth rounding leaves availability's endpoint ratio above 61.55/28.95 = 2.1261. Istat documents automatic rounding but does not specify tie handling.

No availability × mean product, total-group reconstruction, p-value, confidence interval or population-weighted estimate is calculated. The open-ended >3 group category is not replaced with 4. No attendance, resident access, demand, capacity, sales or causal inference is made.

## Reproduction
Run `python3 reproduce.py` beside `reading_groups_panel.csv`. The dependency-free script uses retained source inputs in the analytical CSV to reproduce derived sensitivities and metrics, writing `recomputed_sensitivities.csv` and `recomputed_metrics.json`. It does not independently re-extract the agency workbook. To audit extraction, download the official workbook linked in sources.json and compare Table 3 column B size rows 38–44 and total row 60, Table 25 cells recorded in the CSV, and Table 26 cells recorded there. The frozen official workbook SHA-256 is supplied in rights.md. Download a fresh source only if you intend to check for revisions; a changed source hash may produce changed results.

## Uncertainty
The census began with 9,963 units; 265 were unresolved and 8,883 were eligible and open. Its 97.3% resolution rate includes resolved ineligible units and is not a reading-group item response rate. Frame omissions, reporting errors and item missingness remain. Provincial administrative integration can create missing information, notably in Bolzano. Exact group-number response coverage remains unverified; any combined intensity extension requires aligned Yes counts, numerical-response counts and group totals.