the number

that rates you

Problem statement

The United States measures workplace injury primarily by asking employers to record it. The same recorded counts then determine the employer's workers' compensation insurance rate and their probability of an OSHA inspection, and in many firms determine a safety bonus. The NORA Musculoskeletal Health Council names the resulting circularity as a research priority: it calls for research to "[u]nderstand the effects of insurance rate setting and compliance inspection systems that are triggered by specific injury counts (OSHA recordable injuries and lost day injuries) that give employers a strong incentive to under-report," and asks for study of "alternate surveillance or rate setting practices based on safety factors (leading indicators such as exposure assessment and reduction practices) … that may give a more accurate estimate of current conditions and future risk." Attempts to size the error have not converged: matching state survey cases against workers' compensation records produced undercount estimates ranging from 20% to 70% depending on method and state. There is no independent measure of the national injury rate to check the reported one against.

Why this matters

Everything downstream is calibrated on this number. Inspection targeting, insurance pricing, the business case for an ergonomics program, the evaluation of whether a state's safe-patient-handling law worked, and the choice of which industries a research agency studies all take the reported rate as ground truth. The council states the practical consequence: underreporting "prevents an accurate assessment of costs and benefits of state-level injury prevention programs … making it difficult to assess the effectiveness of workplace MSD prevention programs or to target intervention efforts to populations with the highest burden and need." Musculoskeletal disorders — the largest category of disabling workplace injury — are the worst case, because there is "no standardized national reporting system for chronic work-related MSDs" at all, and because a gradual-onset condition has no incident moment to record. The undercount is also not random. It concentrates where the mechanisms bite hardest: workers treated as independent contractors or supplied by temporary agencies, whose injuries the council notes are underreported, and who are disproportionately placed in the highest-hazard industries — manufacturing, health care, construction, trucking and warehousing, and the wholesale and retail trades. That means the bias runs toward under-counting exactly the workers at greatest risk, and a rate that is wrong in a patterned way is worse than one that is merely noisy.

What’s been tried and why it hasn’t worked

Record linkage has been the main approach — matching the Survey of Occupational Injuries and Illnesses against state workers' compensation records — and it established that an undercount exists without pinning down its size, producing that 20–70% range across studies and states. Linkage inherits the problem it is trying to solve: a worker who never filed a comp claim because they feared lost wages, peer or employer pressure, or losing the job is missing from both sides of the match, so the method can only recover cases that at least one bureaucracy captured. The known suppression mechanisms are individually documented and collectively unfixed: employers returning injured workers on "light duty" to avoid recording lost days; bonus programs that pay for low recordable counts; treating workers as contractors rather than employees; complicated claim procedures; and worker fear of wage loss, reprisal or dismissal. Enforcement against recordkeeping violations addresses only the fraction that is falsification rather than incentive-shaped judgment — whether a case was "work-related," whether restricted duty counts — where the discretion is real and the pressure is one-directional. And the structural shift keeps outrunning the fix: the council notes surveillance is further hampered by the growth of contractors, temp agencies and professional employer organizations, so the fraction of the workforce whose injuries have an ambiguous owner keeps rising.

What would unlock progress

The reframing the council itself proposes is to stop trying to make the self-reported count honest and instead build a second, independently produced estimate that does not pass through the employer's recording decision — active surveillance (asking workers directly on a sampling frame), or leading indicators derived from observable exposure and control practices rather than from outcomes. If a leading-indicator index predicted future claims well enough, it could carry part of the rate-setting and targeting load, which would defuse the incentive at its source: a metric that no longer determines the insurance rate is a metric no one has a reason to manage. The adjacent precedent is crime statistics and public health, where victimization surveys and syndromic surveillance were built precisely because the administratively reported count was known to be shaped by the incentives of the reporting institution, and the two series are now read against each other.

Entry points for student teams

A team could build and pilot the worker-side active-surveillance instrument — a short, anonymous symptom-and-incident questionnaire administered on a defined sampling frame (one campus's facilities and dining staff, one union local, one hospital unit) — and compare its yield against the same population's OSHA 300 log, producing a measured local capture ratio. A second team could construct a candidate leading-indicator index from observable practices (lift-assist availability, job rotation, staffing ratios, near-miss reporting rate) and test its correlation with recorded outcomes across a set of employers. A third could work the mechanism side: an interview or vignette study of the specific recording decisions where discretion enters — work-relatedness, restricted duty, contractor status — mapping where the judgment actually bends. Relevant skills: survey methodology, occupational epidemiology, statistics, organizational behavior, policy analysis.