The reliability of food safety inspection disclosures has long been questioned. Providing meaningful disclosures to consumers based on public data is a core challenge for environmental health. We present new results, based on retrospective data and a novel randomized peer review trial, to improve the reliability of the restaurant food safety grading system that was implemented in Seattle and King County, Washington, in 2017. To address critiques of inspections as driven by inspector differences and merely “snapshots-in-time,” we used data from multiple inspections and adjusted scores within each inspector area to isolate relative performance. We studied the geographic distribution, predictability, and reliability of this adjusted grading system.
Our findings are threefold. First, critical violations are much more reliably cited than non-critical violations, providing much more reliable inputs for grading. Second, and contrary to common belief, little evidence exists that repeat violations or the time trend of scores are associated with increased risk, hence providing no useful information for grading. Third, the adjusted system, based on quantile rankings of average critical points in up to four inspections, is much more robust to widespread inspector differences in stringency. Compared with conventional systems, this system reduces unwarranted geographic and intertemporal differences and improves predictive power. We provide easy-to-use, open-source statistical software for jurisdictions to implement this approach.