NCCI released a new Insights report, Exploring Workers Compensation Trends by Industry and State, the week of July 20, 2026, and it complicates the headline that everyone repeats about workers’ compensation: that lost-time claim frequency keeps falling. It does, on average. NCCI’s 2026 State of the Line put the 2025 decline at 2%. But the Insights report shows the average is hiding real divergence underneath, and for a self-insured employer the average is not the number that matters.
Two findings stand out. First, since 2021, most industries have posted average annual frequency declines of at least 5%, but three did not: Health Care, Office, and Private Education were flat in 2023 to 2024 (preliminary). Second, at the state level, Washington, DC and Nevada are the only two jurisdictions with rising overall frequency since 2021. NCCI ties both to Leisure and Hospitality. Hawaii, DC, Nevada, and Florida each carry more than 10% of their market in Leisure and Hospitality, but only DC and Nevada saw frequency rise in both that class and overall.
Who it affects
Self-insured employers, single-parent captives, and public entity pools whose payroll concentrates in one of the flat or rising segments: hospital and health systems, universities and private schools, and hospitality-heavy employers in DC or Nevada. A diversified national manufacturer benchmarking to the aggregate decline is roughly right. A regional hospital system, a university, or a casino operator is not; its exposure sits precisely where the decline has stalled or reversed.
The reserve mechanism: expected claim count, not severity
This is a frequency and class-mix issue that feeds the expected claim ratio, not a severity story. When you set an expected loss pick for a recent, green accident year, you lean on an a priori frequency assumption because the year has little developed data. The Bornhuetter-Ferguson method makes that assumption explicit: the less mature the year, the more weight the expected piece carries. If your expected claim count trends off a national “frequency down 5% a year” figure while your actual book sits in a flat Health Care segment or a rising Nevada hospitality one, you are granting yourself a frequency credit you did not earn. The result is a systematically low IBNR pick that only unwinds as the young years develop, one accident year at a time.
Construction is the cautionary counter-case. It logged the largest frequency drop, about 7% in 2023 to 2024, yet severity rose about 13% over the same window and drove most of the industry’s cost growth. Frequency relief and severity pressure can point in opposite directions inside the same class, so a frequency assumption alone never settles the pure premium.
What this means for your next review
Ask whether your expected claim counts are benchmarked to a national frequency trend or to your specific industry and state. If your actuary applies one blended decline across the whole program, request an industry-and-state overlay for any segment above roughly 15% of payroll. The correction is usually small in any single year and compounds quietly across several, which is exactly the kind of drift that surfaces as adverse development two or three years later. See also our workers’ compensation IBNR guide, the companion note on severity outrunning frequency relief, and the frequency read on recreational marijuana states.