SHARE Knowledge Hub

Loss Causation, Incident Models and Quantitative Data

Incident models help ask different questions about what failed; they do not prove a single root cause. Trace an event through immediate conditions, organisational decisions and failed barriers, then verify corrective action in the field. Use incident rates with a stated numerator, denominator, reporting scope and uncertainty; statutory reporting is separate from internal learning.

Updated 2026-10-02UK focus

Understanding Loss Causation, Incident Models and Quantitative Data

Incident models help ask different questions about what failed; they do not prove a single root cause. Trace an event through immediate conditions, organisational decisions and failed barriers, then verify corrective action in the field. Use incident rates with a stated numerator, denominator, reporting scope and uncertainty; statutory reporting is separate from internal learning.

Why this matters at work

A severity rate commonly divides lost workdays by exposure hours and multiplies by a stated scale. It reflects time loss, not the full severity of harm. Define whether calendar or scheduled workdays are used, how the day of injury is treated and how restricted work or fatalities are represented.

A count states how many defined events occurred but does not account for workforce size, hours, tasks or exposure. Counts remain useful for workload and absolute harm, especially rare fatalities, but they should not be used alone to compare units of different scale. Choose the denominator that represents opportunity for the event: workers or full-time equivalents for population risk; hours for time at risk; kilometres for driving; lifts, deliveries or maintenance interventions for task exposure; production units only when they closely track hazardous opportunity. If contractor incidents are included, contractor exposure must also be included; if hours exclude overtime or agency work, the rate is biased. State treatment of remote workers, visitors, public events and shared-service staff.

How organisations manage this in practice

Prefer elimination, substitution, engineered prevention and inherently safer design before administrative controls and personal protective equipment. For human failure, reduce opportunities and consequences through task, interface and system design rather than depending on perfect attention.

Scope must reach far enough to examine underlying management and organisational factors. Excluding design, procurement, contractor or senior decisions in advance can predetermine the result. Equally, tracing every historical weakness can delay urgent prevention without improving causal confidence.

Change analysis compares the event situation with a known safe or expected state. Examine what changed in people, task, equipment, material, software, procedure, environment, organisation, contractor, schedule or control status and when the change entered the system.

Combine credible event evidence, control checks and appropriately defined measures. Set an owner and intended control outcome, verify implementation and test effectiveness in the work. A falling injury rate does not prove that a high-consequence barrier is reliable. Bring unresolved uncertainties and overdue actions into management review and preserve the rationale for decisions about resources and priorities.

What does loss causation mean?

Loss causation explains how unwanted events develop from immediate failures, deeper conditions and management-system weaknesses.

A loss is not only an injury. It can be property damage, environmental release, production loss, occupational illness, business interruption, legal exposure, reputational harm. Loss-causation learning also includes near misses where the potential harm did not occur but a weak control was revealed. Loss causation asks how the pathway developed: what energy, substance, task or decision created harm; what barriers failed; why they failed; and what management system allowed those conditions to exist.

At Diploma level, the important skill is not naming models. It is using models to improve thinking. A domino model can help explain sequences, but it can oversimplify. A Swiss cheese model helps show barrier alignment, but it can become vague if the holes are not evidenced. A systems model helps challenge organisational factors, but can become too abstract if actions are not practical. Professional judgement means choosing the model that helps learning without forcing the evidence to fit the model.

Loss causation should also connect to quantitative data. Numbers can reveal patterns, but they can also hide reality. A falling accident count may reflect better controls, under-reporting or reduced workforce/exposure. A correctly exposure-normalised rate needs its own numerator, denominator and work-mix interpretation; fewer workers alone does not explain a lower rate. Near-miss data can show a strong reporting culture or a deteriorating control environment. A serious injury rate may be too unstable to interpret without context. Data is a prompt for inquiry, not a substitute for field verification.

TermPlain-English meaningProfessional caution
Immediate causeThe direct unsafe act, condition, failure or event closest to the loss.Do not stop here; immediate causes often sit on deeper weaknesses.
Underlying causeThe organisational, supervision, competence, design or planning factor that made the immediate cause more likely.Needs evidence, not speculation.
Root causeA deeper system weakness that, if fixed, reduces recurrence risk across similar activities.Avoid claiming one root cause where several interacting causes exist.
Barrier failureA prevention, detection, control or mitigation measure did not work as intended.Define the barrier performance standard and evidence.
Latent conditionA hidden weakness built into the system before the event.Often linked to design, staffing, maintenance, procurement or leadership decisions.
Active failureA front-line action or omission close to the event.Understand performance influencing factors before blaming behaviour.

How should incident models be used?

Incident models are thinking tools; they should improve questions, not replace evidence.

A simple model is useful for communication, but a complex event often needs several lenses. For example, a fall from height may have an immediate cause such as missing edge protection, an underlying cause such as poor temporary works coordination, a human factor such as time pressure, a procurement issue around access equipment, and a leadership issue around programme pressure. One neat diagram may not capture that richness.

The best investigators use models to ask better questions: what was the sequence, which barriers were expected, what conditions degraded them, how did people interpret the situation, what data was available, what pressures existed, what previous weak signals were missed, and what management review should have detected the problem earlier?

ModelUseful forWhere it can fail
Domino modelExplaining a linear sequence and how removing one factor can prevent loss.Can imply one simple chain where real systems are more complex.
Swiss cheese modelShowing multiple barriers with weaknesses aligning.Can become a vague picture unless each barrier and hole is evidenced.
Bow-tie modelLinking threats, top event, prevention barriers, consequences and mitigation barriers.Can miss organisational context if barriers are treated as static boxes.
Systems modelExploring interactions between people, technology, organisation and environment.Can become too abstract if actions are not concrete.
Human factors modelUnderstanding performance influencing factors such as workload, design, competence and supervision.Can be misused to excuse reckless conduct or still blame the operator politely.
Barrier health modelTesting whether controls are available, effective and assured.Needs good evidence standards and ownership.

Which model should be used for which learning question?

A strong investigator chooses the model that best tests the evidence and explains the control failure.

Incident models are not badges of sophistication. They are lenses. If the question is sequence, a domino-style model may help. If the question is barrier weakness, a bow-tie or Swiss cheese lens may help. If the question is why normal work drifted over time, systems thinking may be more useful. Quantitative trend analysis then tests whether the event is isolated or part of a wider pattern.

For Diploma answers, the strongest response usually compares models, applies one to a scenario and then explains its limitations. A model should not force a simplistic answer or hide uncertainty.

Model or approachBest used forLimitations to challengeExample use
Heinrich / dominoExplaining a sequence of events and how removing one factor can interrupt loss.Can overstate linear causation and underplay organisational context.Simple injury sequence where a missing guard, unsafe access and supervision failure align.
Swiss cheeseShowing how multiple barriers have weaknesses that align.Can become vague unless each layer and hole is evidenced.Permit, isolation, supervision and alarm response all partially fail.
Bow-tie / barrier modelLinking threats, top event, consequences, prevention barriers and mitigation barriers.May look complete while barrier health, ownership and testing are weak.Loss of containment leading to fire, toxic release or environmental harm.
Root cause approachesSeparating immediate, underlying and system causes and building corrective actions.Can imply one root cause where there are interacting causes.Investigation action quality review after repeated similar events.
Systems thinkingUnderstanding interactions between people, equipment, procedures, leadership, procurement and work pressure.Can become too broad unless translated into practical controls.Normalisation of shortcuts across shifts or contractor interfaces.
Quantitative trend analysisTesting patterns across time, sites, teams, tasks or exposure groups.Can be distorted by under-reporting, small numbers and changing work mix.Near-miss increase in one plant area prompts barrier verification.

How can the same incident be analysed using two models?

Use two models to test the same conveyor event, then challenge each with evidence.

A maintenance worker is injured clearing a blocked conveyor after an unexpected restart. HSE HSG245 prompts investigation of immediate, underlying and root causes, including plant, procedures, people and organisation. A “failed to isolate” conclusion is only a starting fact: inspect the isolation point, stored-energy route, permit, production pressure, competence and supervisor decision.

The Swiss-cheese view asks how several defences were weak together. A bow-tie view names the top event as unexpected energy at the work point, tests preventive barriers such as lock-off/try-out and mitigation such as exclusion and emergency stop, then checks who owned proof. Neither model should be used to force the evidence into a preferred story.

EvidenceLayered-defence questionBarrier questionAction and proof
Unclear isolation labelWhy did identification defence fail?Could the wrong point be locked?Re-label, walk down and test with workers.
Stored energyWhich engineering/procedure layer was absent?Was zero energy verified before access?Add try-out and release check; observe live task.
Production pressureWhich management layer favoured speed?Who may stop the job?Resource downtime and review stop decisions.
Restart injuryHow did aligned holes reach the worker?Did mitigation act in time?Record injury, redesign barriers and verify close-out.

How should quantitative safety data be interpreted?

A safety rate is interpretable only with its definition, denominator, scope and uncertainty.

HSE incident-investigation guidance supports learning from events, while HSE RIDDOR definitions separates legal reporting from all internal events. If a site records 4 injuries over 400,000 work hours, its stated rate is 4 ÷ 400,000 × 100,000 = 1.0 per 100,000 hours. Compare only with the same injury definition, worker/contractor inclusion, hours and period. Four events are too few to treat a small rate change as proof that risk changed.

A higher near-miss count may reflect better reporting. Pair rates with exposure, serious-potential events, barrier tests, worker interviews and overdue corrective actions. A low lost-time count cannot prove that isolation or guarding is sound.

MetricRequired definitionMisleading inferenceCheck
Injury rateCount, hours, period, staff/contractor scope and multiplier.Lower rate proves safer site.Same scope and uncertainty; field controls.
Near-miss countReporting route and severity-potential classification.More reports prove more danger.Reporting culture and serious-potential themes.
Action closureVerified effective controls, not just completed forms.Closed action proves barrier works.Observe task and test recurrence.
RIDDOR totalCorrect GB/NI statutory reportability.Only RIDDOR events matter.Keep internal learning beyond threshold.

How can the hierarchy of controls turn incident-causation findings into stronger prevention measures?

A worked example shows how models and data change learning quality.

The table gives practical examples, evidence expectations and common decision points so the section can be used for learning, supervision, audit and review.

Investigation elementWeak answerStronger Diploma-level answer
EventWorker slipped because they were not looking.Worker slipped while carrying a box on a contaminated walkway during a rushed delivery change.
Immediate causeUnsafe behaviour.Loss of footing on wet surface while line of sight was restricted by the load.
Underlying causesWorker should take care.No spill ownership, poor lighting, delivery route not assessed, supervisor accepted shortcut.
Human factorsCarelessness.Time pressure, obstructed view, normalised shortcut and unclear stop authority.
DataOnly one accident this year.Two previous near misses and repeated housekeeping actions at the same location.
ActionRetrain worker.Redesign route, improve spill response, review delivery planning, verify housekeeping and monitor recurrence.

Can fewer injuries still mean a higher local frequency rate?

Yes. A count changes with the amount of work observed as well as with risk and reporting. Define the event numerator, workforce scope, time period and exposure denominator before calculating a rate. Keep the count visible alongside the rate so a reader can reproduce the comparison. Related guidance: HSE: Measures of workplace injury — definitions and formulae.

The following invented two-year example uses the same employee population and local recordable-injury definition in both years. Year 1 has 8 recorded injuries and 800,000 employee hours; Year 2 has 6 and 400,000 hours. For this teaching example only, choose a multiplier of 200,000 hours: frequency = recorded injuries ÷ corresponding hours × 200,000. This is a locally declared convention, not a UK statutory reporting threshold or an HSE national benchmark. Related guidance: HSE: Measures of workplace injury — definitions and formulae.

Year 1 is 8 ÷ 800,000 × 200,000 = 2.0 per 200,000 hours. Year 2 is 6 ÷ 400,000 × 200,000 = 3.0. The injury count falls 25%, while the calculated rate rises 50% because hours halve. The arithmetic disproves the simple claim that the lower count alone demonstrates improvement; it does not prove a particular cause or a statistically reliable risk increase. Related guidance: HSE: Measures of workplace injury — definitions and formulae.

Before interpreting the difference, check whether event coding, late reports, contractor inclusion, overtime recording, work mix and reporting incentives changed. If contractor injuries enter the numerator while their hours are missing, the comparison is not coherent; correct the scope or report the gap. Small counts are volatile. Seek suitable statistical advice before claiming a meaningful trend or ranking sites. Related guidance: HSE: Measures of workplace injury — definitions and formulae.

HSE national LFS injury incidence rates use people and their reference-period definitions, usually expressed per 100,000 workers. They must not be compared directly with this local hours-based frequency as if the units, populations and case definitions matched. Use national figures only with their source methods, coverage, uncertainty and revision status. Related guidance: HSE: Measures of workplace injury — definitions and formulae.

The decision is to examine controls and reporting alongside the data. Sample the tasks behind the injuries, check failed barriers and review action effectiveness. The local comparator linked below reproduces the declared arithmetic and deliberately makes no compliance, significance or causal judgement. Related guidance: HSE: Measures of workplace injury — definitions and formulae.

Can fewer injuries still mean a higher local frequency rate
PeriodRecorded injuriesCorresponding employee hoursReproducible calculationLocal rate per 200,000 hours
Year 18800,0008 ÷ 800,000 × 200,0002.0 Related guidance: HSE: Measures of workplace injury — definitions and formulae; HSE HSG245 investigation steps.
Year 26400,0006 ÷ 400,000 × 200,0003.0 Related guidance: HSE: Measures of workplace injury — definitions and formulae; HSE HSG245 investigation steps.
Change−25% count−50% hours(3.0 − 2.0) ÷ 2.0 × 100+50% rate — interpretation needs evidence Related guidance: HSE: Measures of workplace injury — definitions and formulae; HSE HSG245 investigation steps.

Guidance for these decisions: HSE: Measures of workplace injury — definitions and formulae; HSE HSG245 investigation steps. Apply the legal route for the work location; illustrative scenarios are SHARE teaching examples.

What data quality questions should professionals ask?

Data is only useful if it is trustworthy, comparable and linked to decisions.

Performance data can mislead when reporting is weak, samples are too narrow or indicators reward the wrong behaviour. The traps below help readers challenge the number before accepting the conclusion.

QuestionWhy it mattersExample action
Are definitions consistent?Sites may classify injuries or near misses differently.Use a standard reporting taxonomy and audit samples.
Is under-reporting likely?Low numbers may hide fear, fatigue or contractor exclusion.Compare worker feedback, inspections and first-aid records.
Is the denominator right?Rates need hours, headcount, shifts or exposure that match the risk.Use task exposure where possible for high-risk work.
Are contractors included?Excluding contractors hides interface risk.Define scope and report contractor data transparently.
Are actions effective?Closed actions may not reduce risk.Verify controls in the field after close-out.
Are rare severe events visible?High-consequence low-frequency risks may not show in trend data.Use critical control assurance and scenario review.

How should an investigator test competing explanations?

Keep facts, inferences and unanswered questions distinct. A witness account is evidence to evaluate in context, not a complete mechanism by itself. A model helps organise questions; it cannot turn a plausible story into an established cause. Related guidance: HSE HSG245 investigation steps.

In an illustrative conveyor-restart event, a worker says the equipment had been isolated while the shift log says production restarted at 14:10. Secure people and evidence, then establish the event sequence from available records, equipment state, accounts and competent technical examination. Do not energise equipment or recreate exposure to make an investigation easier. Related guidance: HSE HSG245 investigation steps.

Consider at least two explanations: the intended isolation was not effective, or an effective isolation was subsequently removed before hand-back. Record what evidence would distinguish them, such as the actual isolation arrangement, authorisation/handover records, relevant control logs and consistent timing. Also test whether another energy source or misunderstood equipment boundary exists. Neither hypothesis is a finding until adequately supported. Related guidance: HSE HSG245 investigation steps.

Interview people fairly about what they saw, knew, expected and could do at the time. Compare accounts and technical evidence without treating disagreement as proof of dishonesty. A finding of operator error needs a description of the action and the system conditions that allowed harm, including design, instructions, authority, resources and supervision where evidence supports them. Related guidance: HSE HSG245 investigation steps.

Turn each supported control weakness into a specific change and an effectiveness test. If hand-back authority was unclear, test the revised process with an incoming shift and an unresolved defect; if a technical isolation weakness is found, competent engineering verification is required. A retraining attendance sheet cannot demonstrate either recovery. Record remaining uncertainty and who resolves it before the affected work restarts. Related guidance: HSE HSG245 investigation steps.

Guidance for these decisions: HSE HSG245 investigation steps. Apply the legal route for the work location; illustrative scenarios are SHARE teaching examples.

What evidence shows loss causation is being used well?

Evidence should show learning has changed risk controls, not only that a report was written.

Sample a real event: preserve evidence, test immediate and organisational causes, identify barrier owners, assign changes and observe the next comparable task. For data, audit numerator, denominator, contractor scope and serious-potential classification before presenting a trend.

EvidenceWhat it should showWeakness to challenge
Investigation reportEvidence, timeline, causes, barriers, human factors and proportionate actions.Report stops at worker behaviour.
Action trackerOwners, due dates, hierarchy of control, verification and close-out.Actions closed without checking effectiveness.
Data dashboardLeading, lagging, severity, near-miss quality and critical control indicators.Vanity metrics such as inspections completed only.
Learning reviewThemes reviewed by management and fed into risk registers.Learning shared but risk assessment unchanged.
Board or senior reviewMaterial risks, weak signals and assurance gaps escalated.Senior review only sees low accident rate.
Worker consultationFront-line experience tests whether actions are practical.Actions designed without the people doing the work.

How should controls be owned and reviewed?

Assign a competent owner, check that task controls work, involve workers and act when evidence changes.

For Loss causation, incident models and quantitative data, record the actual hazard and task, the chosen control, the person responsible, the evidence that it works and the point at which work must stop or be reassessed. HSE management guidance for Great Britain describes planning, worker consultation, checking and action. HSENI guidance for Northern Ireland describes competent help, risk control and review. These management steps do not prove that a particular task control is suitable; use the topic-specific evidence and the applicable GB or NI law elsewhere on this page.

Where ISO 45001 is adopted, connect incident findings, corrective-action ownership and effectiveness checks to management review. A named model or certificate cannot establish causation or make a declining incident count reliable. Decisions still need traceable evidence, comparable data and demonstrated risk reduction; no exact standard-clause conformity is asserted here. HSE: ISO 45001 management standard.

How should incident and data competence be assessed across roles?

Assess evidence handling, causal reasoning, operating decisions and data reproduction separately. An investigator may need technical specialists for a control mechanism, while a data owner needs the source records and definitions to reproduce a figure. No one should imply certainty outside their competence. Related guidance: HSE HSG245 investigation steps.

Use the conveyor hypotheses and two-year rate example as linked exercises. Ask the participant to show the evidence used, a competing explanation, a control decision and an appropriate limitation. The assessor should be competent in the specific investigation, operational or analytical task and should challenge a conclusion that outruns the evidence. Related guidance: HSE HSG245 investigation steps.

Record scope, observed work, assessor, limits and corrective learning. Reassess after a new event type, reporting-system/definition change, weak investigation, inconsistent data or failed action test. The existing investigator/supervisor/data-owner training routes remain useful; these examples extend them to workers and governance roles. Related guidance: HSE HSG245 investigation steps.

How should incident and data competence be assessed across roles
Role and assessorObserved decisionLimits and failure responseRecord and reassessment
Worker/witness — competent interview/task leadGive a factual account, distinguish seen from assumed and identify unresolved work restrictions.Do not pressure the person to diagnose root cause or repeat an unsafe event.Relevant factual account and briefing; review changed reporting route or task. Related guidance: HSE HSG245 investigation steps; HSE: Measures of workplace injury — definitions and formulae.
Supervisor — competent operational assessorSecure people/evidence and maintain a restriction pending competent recovery.Reporting or an investigation in progress cannot itself release impaired equipment.State, communications and recovery evidence; review event or control/hand-back change. Related guidance: HSE HSG245 investigation steps; HSE: Measures of workplace injury — definitions and formulae.
Investigator — competent investigation reviewerBuild chronology, compare hypotheses and link supported causes to tested action.Record uncertainty and technical scope; no unsafe recreation or blame-only conclusion.Evidence log, causal challenge and action test; reassess new event or poor reasoning. Related guidance: HSE HSG245 investigation steps; HSE: Measures of workplace injury — definitions and formulae.
Data owner/analyst — independent competent analystReproduce 2.0/3.0 rates and +50% change; test matched scope and units.Do not compare local hours-rate with national person-rate or infer significance from arithmetic.Dictionary, data lineage, calculation and caveats; review definition/system change. Related guidance: HSE HSG245 investigation steps; HSE: Measures of workplace injury — definitions and formulae.
Manager/director — competent assurance reviewerResource recovery and challenge lower counts with exposure, reporting and control evidence.A green metric or closed ticket cannot establish that controls improved.Reasoned decision and effectiveness sample; review repeat event, weak data or action failure. Related guidance: HSE HSG245 investigation steps; HSE: Measures of workplace injury — definitions and formulae.

Guidance for these decisions: HSE HSG245 investigation steps; HSE: Measures of workplace injury — definitions and formulae. Apply the legal route for the work location; illustrative scenarios are SHARE teaching examples.

What should Diploma learners and investigators practise?

An investigator demonstrates evidence-led causal analysis and tests action effectiveness.

RoleDemonstrationAssessorRetained recordReassessment trigger
InvestigatorBuild event chronology, compare layered-defence and barrier questions, distinguish fact from hypothesis.Competent investigation lead observes case exercise and checks field verification.Case file, evidence log, challenge notes and action test.New event type, weak causal analysis or failed action.
SupervisorShow isolation/try-out at the conveyor and preserve evidence after a restart.Safety lead observes a task simulation and checks stop authority.Task observation and authorisation record.Plant change, near miss or isolation failure.
Data ownerReproduce numerator, hours, contractor scope and rate calculation.Independent analyst checks source data and trend caveats.Data dictionary and signed sample check.Definition, system or reporting-rule change.

In an observed conveyor-restart exercise, the investigator first secures people and evidence, draws the event sequence, distinguishes the immediate unsafe state from the failed isolation and supervision controls, then proposes a tested action with an owner and due date. A supervisor or competent investigator assesses the reasoning against the evidence and records the action test, not just attendance at an investigation course. Repeat the exercise when equipment, isolation methods or reporting rules change. HSE HSG245 investigation steps support the information, analysis, control and action sequence; a model name alone does not prove a cause.

Which acronyms and terms are useful for loss causation, incident models and quantitative data?

These terms distinguish a control or legal decision from a label.

TermMeaningUse here
RIDDORReporting of Injuries, Diseases and Dangerous Occurrences RegulationsSeparate GB 2013 and NI 1997 instruments.
Top eventLoss of control at centre of a bow-tieUnexpected energy at conveyor work point.
NumeratorEvent count in a rateNeeds a clear case definition.
DenominatorExposure base in a rateHours or workforce must be stated.
ERICPDEliminate, Reduce, Isolate, Control, Personal protective equipment, DisciplineBarrier selection and sustained checks.