Safety Training Retention and the Forgetting Curve: What the Research Actually Shows

Safety training retention is the quiet problem behind every green training matrix. A worker sits through a hazard briefing, passes the check, signs the sheet — and then the clock starts. Memory decays with disuse, and the certificate on file freezes the moment of learning at a date that is now weeks or months in the past. This article walks through what the peer-reviewed literature actually says about how fast safety knowledge fades, why the way training is delivered changes how much sticks, and what the loss is worth in injuries and dollars. Every figure below is the researchers' finding, linked to its original paper — not a vendor promise.

The forgetting curve: retention starts falling almost immediately

The foundational work is Hermann Ebbinghaus's forgetting curve from 1885, and for more than a century it went essentially unreplicated at scale. In 2015, Jaap Murre and Joeri Dros published a full replication in PLOS ONE — "Replication and Analysis of Ebbinghaus' Forgetting Curve" — that reproduced the original curve's shape closely.

Their measure is savings: how much faster you relearn material than you learned it the first time, which captures how much is still retained. In the replication, savings fell steeply and early. Relearning savings sat near half within roughly an hour of first study (about 0.50), dropped to roughly a third by one day, and kept sliding over the following days and weeks. The plain-language summary the industry usually quotes — that around half of newly learned material can fade within days without reinforcement — is a fair reading of that curve, provided you remember it describes a savings measure of memory strength, not a literal count of forgotten facts.

The operational point survives all the caveats: a single training event is not a durable state. It is a spike that begins decaying the moment it ends. A class last spring says very little about what is actually in a worker's head on the morning of today's job. That is the gap between attendance and understanding, extended across time.

Storage strength versus retrieval strength

Cognitive science gives the decay a mechanism. A memory has two independent strengths: storage strength, which, once built, barely fades, and retrieval strength — the ability to actually call the knowledge up when you need it — which declines with disuse. A worker can have genuinely learned a hazard control and still be unable to retrieve it at the panel or the trench at the moment it matters. This is why retention is not a character question about attentive versus careless workers. It is the default behaviour of human memory, and it is why reinforcement, not repetition of the original event, is the lever that moves it.

Delivery method changes how much is retained

If retention decays no matter what, the next question is whether some training builds more of it to begin with. The strongest evidence here is Burke and colleagues' 2006 meta-analysis in the American Journal of Public Health, which pooled 95 quasi-experimental studies covering 20,991 workers.

Burke's team sorted training into three tiers by how much active participation it demanded: least engaging (lectures, pamphlets, videos), moderately engaging (programmed instruction, feedback), and most engaging (hands-on practice, behavioural modelling, dialogue). Knowledge-acquisition effect sizes rose cleanly across the tiers — from 0.55 for the least-engaging methods, to 0.74 for the moderately engaging, to 1.46 for the most engaging. That is a nearly threefold difference between passively watching and actively doing, and the same pattern showed up in reductions in accidents, injuries, and illnesses. The methods that make a worker produce the answer, rather than recognize it, build more knowledge and prevent more harm.

That finding reframes the most common safety-training format. A worker watching a video and signing a sheet is operating in Burke's weakest tier — the one that both retains least and decays from a lower peak. It is also why the question do toolbox talks work has a nuanced answer: a talk that is genuinely a dialogue lands very differently from one that is a monologue with a signature at the end.

What retention is worth: injuries and cost

Two more studies put a number on the stakes of safety systems that actually work in practice, rather than only on paper.

Levine, Toffel, and Johnson published a randomized study in Science in 2012 — "Randomized government safety inspections reduce worker injuries with no detectable job loss." Because Cal/OSHA inspections were assigned at random, the design isolates cause. In the four years after an inspection, treated firms had 9.4% fewer injury claims and 26% lower workers' compensation costs than closely matched uninspected firms — with no detectable loss of employment, sales, or survival. The paper is about the value of effective oversight, not about training retention specifically; what it establishes is the size of the prize when a safety system genuinely changes conditions rather than just documenting them.

Closer to home, a University of British Columbia evaluation led by Dr. Chris McLeod, using Infrastructure Health and Safety Association (IHSA) and WSIB data, examined Ontario construction firms that earned COR (Certificate of Recognition). Participating in COR was associated with a 28% lower lost-time injury rate relative to comparable non-certified firms (incidence rate ratio 0.72; 95% confidence interval 0.64–0.84). COR certification, in other words, was associated with a lower lost-time injury rate in that evaluation — a useful anchor when you are weighing what a genuinely implemented safety program is worth, and one reason COR and the WSIB Health and Safety Excellence Program both reward evidence of implementation over evidence of intent.

The countermeasure is spacing, not more training days

The forgetting curve has a well-established remedy, and it is not re-running the annual course. It is spaced reinforcement: brief, repeated retrieval of the same material at intervals, which flattens the decay curve and rebuilds retrieval strength each time. The mechanism is the opposite of a one-day marathon. A worker who is asked to recall and apply a hazard control on the day it is actually relevant — and again when that hazard resurfaces weeks later — holds it far better than one who was told it once in a classroom and never revisited it.

This is where day-to-day safety practice and the retention research finally meet. The problem has never been that supervisors do not know reinforcement works. It is that checking understanding, per worker, on the right hazard, on the right day, did not scale. A competent person cannot sit with fifteen workers every morning and still run the job, so the industry settled for the artifact that does scale — the signature — and quietly accepted the decay.

How Dashpot applies the science

Dashpot Safety is built around exactly this mechanic. Prompted by the daily Job Safety Plan a crew already fills out, it runs a short, per-worker exchange tied to that day's hazards: a roughly three-minute matched video, then a roughly two-minute guided dialogue — the link arrives by SMS, the exchange runs in the phone's browser — in which the worker explains the hazard and its control back in their own words — about five minutes in total, on the worker's own phone, no app. Because it runs every day against the actual hazards on site, it is spaced reinforcement by construction, and the same hazard resurfacing weeks later becomes a natural second retrieval rather than a staged re-test.

The output is a dated, per-worker, per-hazard record of demonstrated comprehension — the kind of contemporaneous evidence a validator, an auditor, or a court can read, and the strongest exhibit in a WSIB evidence story. Dashpot supplies the records half only — but the mechanics inside it are not neutral. Spaced retrieval, active recall, a worker producing the explanation rather than watching one: in Burke's meta-analysis, that most-engaging tier showed not only the largest knowledge gains but the greatest reductions in accidents, injuries, and illnesses. Dashpot itself does not promise fewer injuries — the injury evidence above measures training methods and safety programs, not any single product — but it is deliberately built on the methods that carried the strongest results. It complements legislated training rather than replacing it.

Frequently asked questions

How quickly does safety training retention decline?

Retention starts falling almost immediately after a single training event. In Murre and Dros's 2015 PLOS ONE replication of the Ebbinghaus forgetting curve, relearning savings — a measure of retained memory strength — dropped to roughly half within about an hour of first study and continued declining over the following days, reaching about a third by one day. The exact numbers depend on the material and the person, but the shape is consistent: a single class produces a spike that decays steadily unless it is reinforced.

Does the way training is delivered affect how much workers retain?

Yes, substantially. Burke and colleagues' 2006 meta-analysis in the American Journal of Public Health, covering 95 studies and 20,991 workers, found knowledge-acquisition effect sizes rising from 0.55 for the least-engaging methods (lectures, pamphlets, videos) to 1.46 for the most-engaging (hands-on practice and behavioural modelling) — a nearly threefold difference — with larger reductions in accidents and injuries for the more engaging methods. Training that requires a worker to actively produce and apply the material retains better than training they passively watch.

What is the best way to improve safety training retention?

The countermeasure the literature supports is spaced reinforcement: brief, repeated retrieval of the same safety material at intervals, rather than a single long session. Spacing rebuilds retrieval strength each time and flattens the forgetting curve. In practice that means revisiting a hazard control when it is actually relevant to the day's work, and again when that hazard recurs, so the knowledge is exercised close to the moment it is needed rather than left to decay after an annual course.

Does better safety training and implementation actually reduce injuries?

The evidence points that way, though the studies measure programs and oversight rather than any single tool. Levine, Toffel, and Johnson's randomized 2012 Science study found 9.4% fewer injury claims and 26% lower workers' compensation costs in the four years after a Cal/OSHA inspection, versus matched uninspected firms. A University of British Columbia evaluation found that earning COR certification was associated with a 28% lower lost-time injury rate versus comparable non-certified firms. Both show what genuinely implemented safety systems are worth.

Does a training certificate prove current comprehension?

No. A training certificate proves a worker completed a course — and, where the course included a test, that they passed an assessment on the date it was issued. Neither reliably establishes what the worker understands today. The forgetting curve makes any dated certificate a weakening signal as time passes: it records a training event, not present retrieval strength — the ability to call the control up on today's job. That is why a per-worker, per-hazard record of recent, demonstrated understanding is a stronger answer to the question of whether a crew currently grasps the hazards in front of them.

---

Dashpot Safety complements legislated certification training such as Working at Heights and WHMIS — it never replaces it. Nothing here is legal advice.

Sources: Murre & Dros 2015, "Replication and Analysis of Ebbinghaus' Forgetting Curve," PLOS ONE 10(7):e0120644; Burke et al. 2006, "Relative Effectiveness of Worker Safety and Health Training Methods," American Journal of Public Health 96(2):315–324; Levine, Toffel & Johnson 2012, "Randomized government safety inspections reduce worker injuries with no detectable job loss," Science 336(6083):907–911; McLeod et al. / UBC evaluation of the IHSA COR program, IHSA impact-evaluation report and IHSA magazine summary. Verified July 2026.