# Reading plays and attending theatre: portable source reproduction Original secondary analysis by Alisson Bento. Source: US Census Bureau and National Endowment for the Arts, July 2022 Survey of Public Participation in the Arts. Target population: US civilian noninstitutionalized household adults 18+; annual recall before interviews 16–25 July 2022. Play reading is not book-only or necessarily leisure-only. Attendance does not identify the same play, title, sequence, ticket purchase or motivation. The public bundle contains 20 estimates, four missing-counterpart ranges, methods and source fingerprints. It contains no person-level source records. Musical and nonmusical comparisons have distinct complete-case bases and overlapping audiences; do not add their margins. Percentages and the paired conditional difference use different declared units; all standard errors are percentage points. Python 3.10 or later, standard library only, is sufficient. Obtain the exact official jul22pub.csv and jul22rep.csv separately through the URLs in source-fingerprints.json. Keep their original complete bytes. Then run: `python -I -S reproduce.py --sources SOURCE_DIRECTORY --out OUTPUT_DIRECTORY` The source directory must contain those two files. Both SHA-256 fingerprints are checked before extraction. The recipe has no network calls, secret, account, private parent-directory input or supplied-results dependency. It validates unique person keys, complete positive-adult replicate matching and agreement of the full-sample weight. It calculates every ratio under 160 replicates and retains their covariance in the inverse-conditional difference. Original four-implied-decimal weight scale and Core 1 expansion cancel in the ratios. No population totals are estimated. The regenerated estimates.csv, missingness-bounds.csv and results.json reproduce the supplied portable outputs byte-for-byte in the documented standard-library implementation. The source NumPy calculation and independent raw-data kernel also agree numerically within recorded floating-point tolerance; public CSV integer-count formatting follows this portable recipe. Explicit LF output is used on every platform. Exact source fingerprints identify this frozen release; if an official source changes, stop rather than silently substitute it. Approximate 95% BRR/Fay 0.5 sampling confidence intervals are separate from missing-counterpart extreme-assignment bounds. The latter are not CIs or probabilities of missing behaviour. Retain the wide intervals and conditioning bases when quoting 10.66% or 20.41%. The joint 39 unweighted respondents do not support a precise narrative about reader motivation or conversion. No 2026 current-rate or causal claim is made. NEA Reading at Risk, 2004, Table 3 already supplies an inverse broad literary/performing-arts audience comparison. This reference credits that precedent and offers a narrower 2022 same-respondent play/nonmusical case with paired uncertainty and missing-response sensitivity. It does not claim the conditional-denominator method is new or compare 2002 and 2022 as a trend. Source reports, images, photographs and logos are not redistributed. Original analysis and charts: CC BY 4.0. Original code: MIT. Retain source credit and interpretation boundaries; see reuse.json and CODE-LICENSE.txt. No official endorsement is implied.