Readability of U.S. food handler training materials: a natural language processing analysis of worker study guides and the federal Food Code

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Introduction In many U.S. jurisdictions a food handler card requires studying a training document and passing a written examination; one state legally requires that the document be furnished to every applicant. Whether these documents can be read by the workers required to use them is unknown. The common assumption that certification is required in all fifty states is also inaccurate: among the jurisdictions whose rules we verified, some mandate training but delegate its content to private vendors, so no worker-facing manual is publicly available. Methods A natural language processing pipeline (pdfplumber, spaCy, textstat) was applied to 20 documents retrieved under a documented protocol: 14 worker-facing study guides across 11 jurisdictions in eight states, the 2022 FDA Food Code, and five Spanish-language counterparts, excluded from analysis. These jurisdictions contain roughly 26.7 million residents, about 8 percent of the U.S. population. The document was the unit of analysis. All 417 sentences of 80 tokens or more were audited, and results are reported under three extraction rule sets. Results One document stood apart. Washington’s Food Worker Manual met the recommended sixth-grade Flesch–Kincaid target under all three extraction rule sets, at a mean sentence length of 10.6 words, while addressing the same ten content domains as the other guides. The remaining 13 of 14 guides exceeded that target under every rule set (guide mean 10.33, 95% CI 9.15–11.11), and at most one met the Gunning Fog target of 8 (mean 13.08, 95% CI 11.85–13.91). The Food Code was more demanding than every guide (Flesch–Kincaid 15.30; Gunning Fog 19.02), with mean sentence lengths of 38.87 words against 25.00. Discussion The Washington result is consistent with the cognitive load account that motivated the measures and indicates that lower reading levels are attainable in this domain, giving agencies a working model rather than an exhortation. Where materials are delegated to private vendors, their accessibility cannot be audited. Two states have adopted alternatives to the written examination under the Americans with Disabilities Act, differing in whether they preserve the holder’s occupational range. Readability formulas measure textual difficulty, not comprehension; direct testing with workers is the necessary next step.