The unit of analysis changes the story

A study published on 2 September 2026 challenges broad claims that US food recalls are steadily rising and that undeclared allergens dominate every recall dataset. Its main analysis covers 24,761 conventional human-food product recall records in the FDA system for fiscal years 2013–2025. Microbiological hazards were the largest category in 12 of those 13 years.

The first distinction is what the researchers counted. A single company’s recall action can cover several products or pack sizes, each appearing as a separate record. The 24,761 product records in this analysis represented 6,674 recall events. Neither total measures the number of people made ill, and neither is a reliable measure of the physical quantity of food recalled.

Microbiological hazards led the FDA categories

Researchers recoded the reasons given in the records using 70 ordered keyword rules. The five main categories were microbiological, allergen, chemical, physical and other quality issues. Microbiological hazards accounted for 11,443 product records, or 46.2%; allergens accounted for 6,717, or 27.1%. The microbiological category included pathogens as well as mold and infestation.

The team checked the coding through an internal audit and a stratified manual review of 300 records, reporting 95.7% overall precision in that review. Some 7.7% of the full analysis remained uncategorised. The authors report that assigning every uncategorised record to allergens would still leave the microbiological category ahead in the cumulative count.

No simple upward trend

Average annual product records were lower in fiscal 2018–2025 than in 2013–2017: 1,593 versus 2,403. The series peaked around 2015–2016 and reached its lowest count in 2021. That pattern matters because a rebound from a low year can look like a sustained increase when the comparison window is short.

The statistical evidence also needs precision. The Mann–Kendall test did not establish a significant monotonic trend across the 13 annual observations. Count models that allowed for the data’s extra variation indicated a modest decline. Those results do not justify treating each year as part of a smooth downward line, but they also do not support the simple claim of persistent growth.

FDA and FSIS are not interchangeable datasets

A separate comparison with USDA’s Food Safety and Inspection Service produced a different hazard mix. Allergens led the FSIS event counts, whereas microbiological hazards led the FDA product records. These agencies cover different product categories and oversight systems; even the reported counting units differ. Their totals should not be placed side by side as if they measured identical activity.

The FSIS archive available to the researchers ended on 2 March 2025, leaving its final fiscal year incomplete. The CDC outbreak data used for a further comparison extended only through 2023. These uneven endpoints limit any claim about the most recent years or about a direct relationship between recall activity and illness.

What a recall dashboard should explain

The study also identifies inconsistent quantity units and repeated event-level quantities across product rows. Summing such fields can count the same recalled stock more than once. A useful comparison therefore needs a stated time window, agency, counting unit and hazard definition before an increase or a leading cause can be interpreted.

The analysis is not proof that food has become universally safer, nor a causal test of a particular law. Detection, reporting, classification and product traceability all shape recall records. For businesses tracking safety performance, the paper’s practical value is to make those measurement choices visible while keeping allergen control and microbiological control as distinct, continuing priorities.

SOURCES & TRANSPARENCYHow we work ↗
food recallsFDAallergen management