# Brazil online purchasing and reading-category purchases, 2024

Author: Alisson Bento

## Question and units
Compare subgroup percentage-point differences in H2 online purchasing among Internet users with H3 books/magazines/newspapers purchases among online purchasers. Ages 10+ in permanent private households; Internet use in previous three months; buying or ordering in previous twelve months, including payment outside the Internet. H3 derives from questionnaire H5 item I. Income is total monthly household income in minimum-wage bands, not individual or equivalized income.

## Vintage
The 2025 collection report explicitly documents a reduced commerce module with H2 only, confirmed in the questionnaire. The latest verified matched pair is 2024; do not splice 2025 participation with 2024 categories. Fieldwork March–August 2024 uses respondent-relative preceding-twelve-month periods.

## Extraction
Source archive: https://cetic.br/media/microdados/849/tic_domicilios_2024_individuos_xlsx_v1.1.zip
SHA-256: 5bca9a2f5a3e5fa2a9fa55b0c04a477472f3a77282089b468145d9b5d8ecbbf7

H2 affirmative column C; H3 selected category column I. estimates.csv preserves 34 selected source cells from the proportions workbook and matching margins from tic_domicilios_2024_individuos_tabela_margem_de_erro_v1.1.xlsx. Matching margins use the same worksheet, row and column as the proportion. weighted_numerator comes from the same position in the total workbook; it is a population estimate, not sample size. Eight H2/H3 worksheets were re-extracted read-only from the four official workbooks and matched the prior extraction after trimming empty trailing cells.

## Calculations and uncertainty
Contrasts subtract category B from A within each indicator. No ratio, joint-share product, causal decomposition, time change or significance claim is calculated. Source 95% margins are in percentage points; estimate ± margin endpoints are bounded to [0,100] for display only. Separate margins do not yield covariance for a difference or difference-of-gaps. Survey uses a stratified three-stage cluster design, selection weights, nonresponse adjustment and PNAD Contínua calibration; individual uncertainty uses 200 replicates and 1.96 times design-based standard error.

## Missingness and precision
Dashes denote item nonresponse in the source, and rounded zeros are not necessarily no responses. Selected cells are numeric. No numeric small-cell suppression threshold was established in reviewed documentation. Unknown income, no income and nonresponse remain separate; margins can be very wide. No imputation, pooling or subgroup ranking. Full precision is retained for audit; prose can round to one decimal. Category is books OR magazines OR newspapers, not books-only, ebooks, genres, expenditure or reading frequency.

## Reproduce
Run python reproduce.py beside estimates.csv to verify and regenerate descriptive-contrasts.csv. For independent original-source checking, download the official archive directly from the producer and run python reproduce.py --archive PATH_TO_ZIP; this optional step needs openpyxl. The script verifies the frozen archive hash and all 34 proportions and margins. No network requests or respondent records are required by the script.

## Rights
See rights.md for dataset-specific CC BY 4.0 attribution and modifications. sources.json records source URLs and archival hashes.