Safety Performance Measurement, Audit and Analytics

Introduction and Context

Construction safety management requires more than compliance with procedures or counting injuries after they occur. It requires a reliable system for determining whether risk controls are present, effective, deteriorating, or producing unintended consequences. Safety performance measurement provides evidence for managerial decisions; audits examine whether systems and controls conform to requirements and work in practice; analytics converts safety data into patterns, explanations, forecasts, and priorities.

A mature measurement system combines leading indicators, which describe preventive activity and control effectiveness, with lagging indicators, which describe harmful outcomes. It also uses exposure-normalized measures, because raw event counts can misrepresent performance when workforce size, working hours, project duration, or production volume changes. Dashboards and trend analysis make information visible, while audits and benchmarking test the credibility and comparability of that information.

Measurement is not neutral. If managers reward low injury counts without checking reporting behaviour, workers may avoid reporting minor injuries, near misses, unsafe conditions, or occupational symptoms. Therefore, the purpose of measurement must be learning and risk reduction rather than the cosmetic achievement of favourable statistics.

Learning Objectives

After studying this chapter, the learner should be able to:

1. Distinguish leading, lagging, exposure, and contextual safety indicators.

2. Select indicators that reflect both safety activity and control effectiveness.

3. Calculate and interpret common exposure-normalized metrics.

4. Design a safety dashboard for project and organizational decisions.

5. Apply trend analysis, benchmarking, and audit methods critically.

6. Evaluate the financial and non-financial value of safety interventions.

7. Identify ethical risks, including under-reporting and data manipulation.

8. Use analytics and digital technologies responsibly in construction safety management.

1. The Measurement Problem in Construction Safety

Construction projects are temporary, heterogeneous, and production-oriented. Work locations change, subcontractors enter and leave, hazards vary by phase, and exposure is often intermittent. A single project injury rate may therefore be influenced by project type, hours worked, reporting culture, regulatory definitions, and chance. Measurement must be interpreted in relation to the work system.

A useful performance system answers four questions:

• What happened? Injuries, illnesses, property damage, releases, and high-potential events.
• What is being done? Inspections, training, corrective actions, planning, and observations.
• Are controls effective? Barriers, permits, exclusion zones, temporary works, lifting arrangements, and emergency systems.
• What should management do next? Allocate resources, redesign work, stop work, improve competence, or investigate systemic causes.

The last question is the most important. A metric has value only when it supports a sound decision.

2. Types of Safety Indicators

2.1 Lagging indicators

Lagging indicators record outcomes that have already occurred. Examples include fatalities, lost-time injuries, medically treated injuries, total recordable cases, occupational illnesses, high-potential incidents, equipment damage, and days away from work.

Their strengths are that they are relatively easy to understand, often required by regulators or clients, and useful for verifying whether harm has occurred. Their limitations are equally significant: serious events are statistically infrequent; results may be dominated by chance; and poor outcomes provide evidence after controls have failed. A low injury rate does not prove that risk is low.

Lagging indicators should be disaggregated by project, activity, contractor, hazard, body part, mechanism, severity, and potential severity. A minor actual outcome may arise from a high-potential event and should not be treated as evidence of a safe system.

2.2 Leading indicators

Leading indicators describe actions, conditions, or capabilities intended to prevent harm. Examples include:

• percentage of high-risk activities with approved task plans before work begins;
• percentage of critical controls verified as effective;
• percentage of corrective actions closed by their due dates;
• quality and frequency of supervisor field engagements;
• proportion of workers participating in meaningful pre-task planning;
• inspection completion and defect recurrence;
• percentage of safety-critical personnel with current competence;
• near-miss and hazard reports, interpreted with reporting quality and workforce size;
• emergency drill performance; and
• percentage of subcontractors passing assurance reviews.

A leading indicator is not automatically a good indicator. Counting toolbox talks may measure attendance rather than understanding. Counting observations may encourage superficial observations. The indicator should be linked to a credible control mechanism and should be capable of independent verification.

2.3 Control-performance indicators

Control-performance indicators occupy the space between activity and outcome. They ask whether a barrier actually works. For example, a project may report that lifting plans were prepared, but a stronger measure verifies whether exclusion zones were maintained, lifting accessories were certified, loads were within capacity, and supervisors corrected deviations.

For high-risk construction work, control indicators should focus on the critical controls whose failure could lead to a fatal or catastrophic event. A practical hierarchy is:

1. Identify the credible major-harm scenarios.

2. Specify the controls that prevent or mitigate each scenario.

3. Define observable standards for each control.

4. Assign an accountable owner.

5. Verify presence and effectiveness in the field.

6. Escalate repeated or critical failures.

2.4 Contextual and capacity indicators

Contextual indicators explain the conditions under which performance occurs. They may include overtime, labour turnover, schedule compression, percentage of inexperienced workers, subcontracting intensity, weather exposure, design changes, production interruptions, and workforce availability. Capacity indicators include safety staffing, competence resources, maintenance resources, and time allocated to planning.

These indicators are not necessarily measures of safety itself. They are explanatory variables that help managers avoid simplistic conclusions.

3. Exposure-Normalized Metrics

Raw counts are unsuitable for comparing projects with different exposure. A common normalization is the incident rate per 200,000 hours:

\[

R_{200,000}=\frac{N\times 200,000}{H}

\]

where \(N\) is the number of defined cases during the period and \(H\) is total hours worked. The constant 200,000 represents 100 full-time workers working 2,000 hours per year; it is a conventional comparison base, not a law of safety. Some jurisdictions or organizations use 100,000, 1,000,000, or another base. The base must always be stated.

For a rate per one million hours:

\[

R_{1,000,000}=\frac{N\times 1,000,000}{H}

\]

The numerator definition must be stable. A rate should not combine first-aid cases in one project with recordable cases in another. Hours worked should be measured consistently and should not include unworked paid hours unless the adopted standard explicitly requires it.

A frequency rate generally relates event numbers to exposure, whereas a severity rate relates consequences to exposure. One possible severity rate is:

\[

S_{200,000}=\frac{D\times 200,000}{H}

\]

where \(D\) is the number of days away from work or restricted-work days, according to the selected definition. Days must be counted under a clearly stated rule, such as calendar days or scheduled workdays, and the rule must remain consistent.

An organization may also calculate a lost-time injury frequency rate:

\[

LTIFR=\frac{L\times 1,000,000}{H}

\]

where \(L\) is the number of lost-time injuries and \(H\) is hours worked. This rate is meaningful only when the definitions of lost-time injury and hours are consistent across the comparison.

Worked Example 1: Comparing exposure-normalized rates

Project A records 3 recordable cases in 480,000 hours. Project B records 2 cases in 160,000 hours. Using a base of 200,000 hours:

\[

R_A=\frac{3\times200,000}{480,000}=1.25

\]

\[

R_B=\frac{2\times200,000}{160,000}=2.50

\]

Project B has fewer cases but twice the rate. The conclusion is not that Project B is intrinsically unsafe: confidence intervals, case mix, reporting culture, work type, and statistical variation must also be considered. The calculation only establishes the normalized point estimates.

Rates, proportions, and denominators

A completion percentage is calculated as:

\[

P=\frac{n}{N}\times100

\]

where \(n\) is the number meeting a criterion and \(N\) is the eligible population. For example, 45 of 50 critical-control verifications completed on time gives 90%. The denominator must be defined: planned inspections, eligible inspections, or inspections actually attempted produce different results.

A metric based on workers, shifts, tasks, hours, lifts, or permits answers different questions. Hours are appropriate for broad injury comparisons, while the number of lifts may be more meaningful for crane-control verification. The denominator should represent the exposure or opportunity relevant to the event.

Statistical caution

Rare events produce unstable rates. If a project records no fatalities, that does not establish zero fatality risk. When comparing rates, managers should examine the number of events, exposure size, confidence intervals where appropriate, and whether the populations and definitions are comparable. Small differences should not be overinterpreted. Control failures and high-potential events may be more decision-relevant than small changes in a low injury rate.

4. Indicator Design and the Prevention of Perverse Incentives

A balanced indicator set should include:

• outcome measures, such as injuries and occupational illness;
• preventive measures, such as planning and verification;
• control measures, such as barrier effectiveness;
• learning measures, such as quality of investigations and recurrence prevention;
• contextual measures, such as overtime and labour turnover; and
• data-quality measures, such as reporting timeliness and audit agreement.

Indicators should satisfy several design tests:

1. Relevance: Is the indicator connected to a material hazard or decision?

2. Validity: Does it measure the intended construct?

3. Reliability: Would different competent assessors obtain similar results?

4. Sensitivity: Can it detect meaningful deterioration or improvement?

5. Controllability: Can the responsible manager influence it?

6. Timeliness: Is information available while intervention remains possible?

7. Resistance to gaming: Can the measure be improved without improving safety?

8. Comparability: Are definitions and denominators stable?

Avoiding under-reporting

A low count of near misses may mean few near misses, poor recognition, fear of blame, or administrative barriers. To avoid rewarding silence:

• never use low reporting alone as evidence of good performance;
• reward useful hazard identification and effective corrective action, not merely report volume;
• protect confidentiality and prohibit retaliation;
• investigate changes in reporting patterns;
• compare reporting with field observations, audits, worker interviews, and exposure;
• track reporting timeliness, quality, and closure rather than raw counts only;
• distinguish actual severity from potential severity; and
• include worker perceptions of psychological safety.

A mature system may initially show more reports after trust improves. That increase can represent improved visibility rather than deteriorating conditions.

5. Dashboards and Trend Analysis

A dashboard is a decision interface, not a decorative collection of gauges. It should identify the audience, decision horizon, risks, thresholds, trends, owners, and required actions. A project manager may need weekly critical-control status, while a corporate director may need monthly trends across projects.

A useful dashboard contains:

• a small number of strategic indicators;
• clear definitions and data dates;
• actual values, targets, and tolerance limits;
• trend lines rather than isolated values;
• segmentation by project, contractor, work type, and hazard;
• overdue corrective actions and escalation status;
• high-potential events and critical-control failures;
• data-quality warnings; and
• a stated management action for each red condition.

Colour coding should not replace explanation. A green result may be misleading if data completeness is poor or if the metric encourages non-reporting.

Trend analysis

Trend analysis may use moving averages to reduce random short-term fluctuation. A simple three-period moving average is:

\[

MA_t=\frac{x_t+x_{t-1}+x_{t-2}}{3}

\]

where \(x_t\) is the value in the current period. This is descriptive and should not be treated as a forecast law. It is unsuitable when the underlying definition, exposure, or project phase changes substantially.

Managers should examine:

• direction and persistence of change;
• changes in exposure and workforce composition;
• seasonality and project phase;
• step changes after interventions;
• concentration by contractor or activity;
• recurrence of similar events; and
• divergence between leading and lagging indicators.

For example, declining injury rates combined with declining near-miss reporting, fewer inspections, and increased overtime is not a reassuring trend. It is a signal requiring investigation.

Control charts and statistical process-control methods can be useful when data are sufficiently frequent and the process is reasonably stable. They should not be applied mechanically to rare-event data or to mixtures of different project populations. A special-cause signal indicates a need to investigate, not automatic proof of causation.

6. Safety Audits and Assurance

An audit is a systematic, documented, and objective process for obtaining evidence and evaluating it against defined criteria. An inspection generally focuses on physical conditions and immediate compliance; an audit examines the management system, implementation, evidence, and effectiveness. An investigation seeks to understand an event and prevent recurrence. These activities are complementary, not interchangeable.

Audit planning

An effective audit begins by defining:

• scope, objectives, and criteria;
• project phase and activities;
• legal, contractual, organizational, and technical requirements;
• sampling strategy;
• auditor competence and independence;
• evidence sources;
• grading and escalation rules; and
• reporting and follow-up arrangements.

Evidence should include document review, field observation, interviews, records, photographs where permitted, and verification of corrective action. Auditors should sample enough work to support a reasonable conclusion and should record limitations.

Audit questions

A mature audit asks more than whether a document exists:

• Was the hazard identified accurately?
• Was the control selected according to the hierarchy of controls?
• Was the control implemented at the workface?
• Did workers understand and use it?
• Was it maintained under changing conditions?
• Did supervision verify it?
• Were previous failures corrected and prevented from recurring?

Grading and closure

Findings should distinguish nonconformity, observation, opportunity for improvement, and good practice. Critical findings require immediate risk control and escalation. Corrective action should address the underlying cause, identify an owner and due date, and be verified for effectiveness. Closing an action because a document was uploaded is not the same as demonstrating risk reduction.

Audit independence is a governance issue. Internal auditors may understand operations but face conflicts of interest; external auditors may provide independence but lack contextual knowledge. A risk-based combination is often appropriate.

7. Benchmarking and Comparative Interpretation

Benchmarking compares performance with a reference group, internal standard, regulatory expectation, or historical baseline. It can identify unusual results and stimulate learning, but it can also encourage superficial competition.

Valid benchmarking requires comparability of:

• case definitions;
• exposure denominators;
• project types and risk profiles;
• workforce and subcontractor arrangements;
• reporting culture;
• data period and maturity; and
• inclusion or exclusion rules.

Internal benchmarking can compare similar bridge, building, tunnelling, or infrastructure projects. External benchmarking should disclose definitions and limitations. A project with a higher reported rate may have better reporting and more transparent governance than a project with a lower rate.

Benchmarking should therefore compare practices and control effectiveness, not just injury rankings. Questions such as “How does the best-performing project verify lifting controls?” are often more useful than “Which project has the lowest rate?”

8. Cost–Benefit Analysis and Managerial Decisions

Safety investments should be evaluated using both economic and ethical reasoning. Preventing harm cannot be reduced entirely to monetary savings, especially where life, health, dignity, and legal duties are involved. Nevertheless, economic analysis helps compare options and communicate resource requirements.

A simple expected-loss estimate is:

\[

E(L)=\sum_{i=1}^{n}p_iC_i

\]

where \(p_i\) is the estimated probability of outcome \(i\), and \(C_i\) is its consequence cost. The estimate may include medical costs, delay, repair, investigation, legal cost, insurance effects, and productivity loss. Intangible and human consequences should be identified separately rather than hidden in a weak monetary estimate.

For an intervention with benefit \(B\) and cost \(C\), a simple benefit–cost ratio is:

\[

BCR=\frac{B}{C}

\]

A ratio greater than one suggests quantified benefits exceed quantified costs under the assumptions used. It does not establish that an intervention is optional where law or ethical duty requires it. Nor does it justify using uncertain probabilities to discount severe catastrophic risks.

For projects with multi-year cash flows, present value may be calculated as:

\[

PV=\frac{F_t}{(1+r)^t}

\]

where \(F_t\) is a future cash flow at time \(t\), and \(r\) is the selected discount rate per period. Discounting is sensitive to assumptions and may undervalue long-term health effects; therefore, sensitivity analysis and non-monetary criteria are necessary.

Worked Example 2: Intervention appraisal

A fall-prevention redesign costs $80,000. Management estimates an avoided direct and operational loss of $140,000 over the evaluation period, excluding the human value of preventing injury. The simple benefit–cost ratio is:

\[

BCR=\frac{140,000}{80,000}=1.75

\]

The quantified benefit exceeds cost. A responsible decision still tests the probability estimate, considers legal requirements, examines uncertainty, and asks whether the redesign reduces exposure more effectively than alternative controls. The result is decision support, not permission to accept preventable harm.

9. Integrated Analytics and Artificial Intelligence

Safety analytics can integrate incident records, inspection findings, corrective actions, worker feedback, equipment data, weather, schedules, and exposure. Descriptive analytics explains what happened; diagnostic analytics explores why; predictive analytics estimates future risk; prescriptive analytics evaluates possible interventions.

Machine-learning models may identify patterns in text reports, images, location data, or equipment signals. However, prediction is not causation. A model may learn that a particular contractor, shift, or project is associated with events because of confounding factors or reporting differences. It may reproduce historical bias and may create privacy or surveillance concerns.

Responsible use requires:

• a legitimate purpose and data minimization;
• documented data provenance and quality checks;
• validation on new projects and changing conditions;
• human review for consequential decisions;
• explainability appropriate to the decision;
• protection of personal and sensitive information;
• monitoring for bias and model drift; and
• a prohibition on using algorithms to blame workers for systemic failures.

Computer vision may flag missing personal protective equipment, but it cannot reliably determine whether the selected control is adequate or whether the work method is safe. Digital tools augment competent judgment; they do not replace engineering control, supervision, worker participation, or legal accountability.

10. Illustrative Case: MetroLink Station Construction

Background

MetroLink is an illustrative case involving a large urban rail-station project with excavation, lifting, temporary works, concrete operations, and multiple subcontractors. The project had 1.2 million hours worked during its first 18 months. Management reported two lost-time injuries and described performance as better than the corporate target.

Problem

A quarterly review showed that near-miss reports fell by 46% in one quarter, while overtime increased, three experienced supervisors left, and subcontractor turnover rose. The dashboard showed green injury indicators but did not display critical-control verification. Worker interviews suggested that reports were sometimes treated as evidence of poor individual performance.

Analysis

The safety team normalized the injury data and examined exposure, project phase, and reporting processes. It found that the apparent improvement was accompanied by incomplete inspection records and delayed corrective-action closure. A field audit identified inconsistent edge protection at temporary work areas, weak segregation around deliveries, and lifting-plan deviations. No catastrophic event had occurred, but several failures had high potential.

The team avoided concluding that the injury rate proved deterioration or that the reporting decline proved improvement. Instead, it assessed the measurement system itself as unreliable.

Approach and solution

Management introduced a balanced dashboard containing:

• lost-time and recordable cases per one million hours;
• high-potential events by hazard;
• percentage of critical controls verified in the field;
• overdue high-risk corrective actions;
• quality-assessed worker hazard reports;
• overtime and supervisor coverage;
• subcontractor assurance results; and
• data-completeness status.

Reports were separated from disciplinary processes except where there was deliberate misconduct. Workers received feedback on action taken. Critical-control failures triggered immediate intervention. A monthly assurance review sampled records against actual work and required closure verification.

Outcome

During the next two quarters, hazard and near-miss reporting increased, while the proportion of reports receiving timely feedback improved. The project recorded no fatal or high-potential harm in the period, but management did not claim that the increase in reporting alone proved improved safety. Field verification showed improved edge protection and lifting-zone control, and recurring findings declined. The case demonstrates that better transparency can make performance appear worse before it becomes more reliable.

Learning points

1. Outcome metrics must be read with exposure and context.

2. Reporting volume is meaningful only when reporting quality and trust are assessed.

3. Critical-control verification can be more actionable than low-frequency injury counts.

4. Audit findings should test effectiveness at the workface.

5. A dashboard should reveal uncertainty and data limitations.

6. Safety performance improvement requires both technical controls and fair governance.

11. Challenges, Ethics, and Governance

Safety data may be incomplete, delayed, inconsistent, or strategically reported. Common problems include changing definitions, missing subcontractor hours, duplicate events, retrospective record alteration, denominator errors, and failure to record occupational health conditions. Governance should establish data ownership, version control, access rights, retention rules, verification responsibilities, and escalation thresholds.

Ethically, organizations must protect workers from retaliation, respect confidentiality, and ensure that metrics do not shift responsibility for system design onto individuals. Worker representatives should participate in indicator selection and interpretation. A just culture distinguishes human error, at-risk behaviour influenced by system conditions, and reckless conduct; it does not treat every adverse outcome as individual failure.

Regulatory requirements vary by jurisdiction. Organizations must follow applicable occupational safety and health reporting rules, privacy and employment law, contractual requirements, and recognized management-system standards. External reporting should be accurate, transparent about definitions, and not selectively presented to create a misleading impression.

12. Best-Practice Implementation Framework

A practical implementation sequence is:

1. Establish the hazard and risk profile, including major-harm scenarios.

2. Define a measurement dictionary containing numerator, denominator, scope, owner, frequency, and data source.

3. Select a balanced set of outcome, leading, control, context, and data-quality indicators.

4. Test each indicator for validity, usefulness, and gaming risk.

5. Build reliable data capture and independent quality checks.

6. Create dashboards matched to operational, project, and corporate decisions.

7. Conduct risk-based audits and verify corrective-action effectiveness.

8. Review trends with exposure, phase, contractor, and reporting-culture information.

9. Benchmark comparable populations and disclose limitations.

10. Evaluate interventions using economic, legal, ethical, and risk-reduction criteria.

11. Review the measurement system itself at planned intervals.

12. Retire indicators that no longer support decisions or create harmful incentives.

The central principle is that measurement should make risk more visible and control more effective. It should not become a competition to produce attractive numbers.

Chapter Summary

Safety performance measurement combines lagging outcomes, leading activities, critical-control verification, contextual conditions, and data-quality measures. Exposure-normalized rates improve comparison but do not eliminate uncertainty or differences in project risk. Dashboards should support decisions through definitions, trends, segmentation, escalation, and transparent limitations. Audits test conformity and effectiveness through objective evidence, while benchmarking is useful only when populations and definitions are comparable. Cost–benefit analysis can support resource decisions but cannot replace legal, ethical, and professional duties. Analytics and AI can improve visibility, provided that data quality, privacy, bias, explainability, and human accountability are addressed. The most mature systems resist under-reporting by rewarding learning and effective controls rather than low event counts.

Key Terms

Leading indicator: A measure of preventive activity, capability, or control performance.

Lagging indicator: A measure of harm or loss that has already occurred.

Critical control: A barrier whose failure could result in fatal or catastrophic harm.

Exposure: The amount of work or opportunity for an event, commonly represented by hours worked.

Exposure-normalized rate: An event count related to a defined exposure base.

Frequency rate: The number of events relative to a stated exposure denominator.

Severity rate: Consequence days or other severity measure relative to exposure.

Dashboard: A visual decision interface presenting selected performance information.

Trend analysis: Examination of direction, persistence, variation, and change over time.

Moving average: An average of observations over a specified rolling number of periods.

Audit: A systematic, documented, objective evaluation against defined criteria.

Inspection: A focused examination of conditions, equipment, or work practices.

Assurance: Confidence, supported by evidence, that controls and systems are effective.

Benchmarking: Comparison with a reference group, standard, or baseline.

Under-reporting: Failure to record or communicate events, hazards, or symptoms that should be reported.

Data quality: Fitness of data for its intended use, including accuracy, completeness, consistency, and timeliness.

High-potential event: An event that could reasonably have produced a severe outcome, regardless of actual harm.

Benefit–cost ratio: Quantified benefits divided by quantified costs under stated assumptions.

Psychological safety: A climate in which people can raise concerns without fear of humiliation or retaliation.

Model drift: Decline in the validity of an analytical model as conditions or data change.

Review Questions

Short Answer Questions

1. Distinguish between a leading indicator and a lagging indicator, giving one construction example of each.

2. Why are raw injury counts inadequate for comparing projects?

3. State two limitations of using near-miss frequency as a safety measure.

4. What is the difference between an inspection and an audit?

5. Why should critical-control verification be included in a safety dashboard?

Long Answer Questions

1. Explain how a balanced safety measurement system can combine outcome, preventive, control, contextual, and data-quality indicators.

2. Discuss the methodological and ethical risks of ranking construction projects using injury rates alone.

3. Describe a risk-based audit process from planning through corrective-action verification.

4. Explain how dashboards and trend analysis can support different levels of construction management.

5. Critically evaluate the use of artificial intelligence in construction safety analytics.

Analytical / Application Questions

1. Project X records four recordable cases in 320,000 hours, while Project Y records three cases in 180,000 hours. Calculate each rate per 200,000 hours and explain why the results should not be interpreted without contextual information.

2. A project’s injury rate declines by 20%, near-miss reports decline by 50%, overtime rises by 30%, and critical-control verification falls from 92% to 74%. Analyse at least three plausible interpretations and recommend management actions.

3. A fall-prevention intervention costs $120,000 and is estimated to avoid $210,000 in quantified losses. Calculate the simple benefit–cost ratio and identify at least four non-financial or uncertainty-related issues that should influence the decision.

Multiple Choice Questions with Answers

Question 1

Which is the strongest example of a critical-control indicator?

A. Number of toolbox talks delivered

B. Percentage of verified excavation edge-protection controls that meet the required standard

C. Number of workers employed

D. Total project expenditure

Correct Answer: B. Percentage of verified excavation edge-protection controls that meet the required standard

Question 2

Project A has 3 cases in 480,000 hours. What is its rate per 200,000 hours?

A. 0.625

B. 1.25

C. 2.50

D. 6.25

Correct Answer: B. 1.25

Question 3

Why can a declining near-miss count be misleading?

A. Near misses are always unrelated to risk

B. It may reflect reduced reporting or reduced psychological safety

C. Near misses cannot be counted

D. It proves that controls are effective

Correct Answer: B. It may reflect reduced reporting or reduced psychological safety

Question 4

In the formula R = N × B / H, what does H represent?

A. The hazard rating

B. The number of high-potential events

C. Total hours worked or another defined exposure denominator

D. The monetary cost of harm

Correct Answer: C. Total hours worked or another defined exposure denominator

Question 5

What primarily distinguishes an audit from an inspection?

A. An audit examines evidence against defined criteria and system effectiveness

B. An audit is always shorter

C. An inspection never involves field observation

D. An inspection is legally more important in every jurisdiction

Correct Answer: A. An audit examines evidence against defined criteria and system effectiveness

Question 6

A benefit–cost ratio of 1.75 means that, under the stated assumptions:

A. The intervention costs 1.75 times its benefits

B. The quantified benefits are 1.75 times the quantified costs

C. The probability of injury is 1.75%

D. The project has 1.75 injuries per worker

Correct Answer: B. The quantified benefits are 1.75 times the quantified costs

Question 7

Which practice best reduces the risk of rewarding under-reporting?

A. Rewarding projects solely for zero injuries

B. Removing near-miss reporting from dashboards

C. Assessing report quality, feedback, control effectiveness, and reporting culture

D. Disciplining every worker associated with a report

Correct Answer: C. Assessing report quality, feedback, control effectiveness, and reporting culture

Question 8

What is a principal limitation of benchmarking injury rates across projects?

A. Rates never use denominators

B. Projects may differ in definitions, exposure, hazards, and reporting culture

C. Benchmarking cannot use historical data

D. All projects have identical risk profiles

Correct Answer: B. Projects may differ in definitions, exposure, hazards, and reporting culture

Question 9

A three-period moving average is calculated by:

A. Dividing the current value by the previous value

B. Adding three selected observations and dividing by three

C. Multiplying all observations by exposure hours

D. Subtracting the oldest observation from the newest

Correct Answer: B. Adding three selected observations and dividing by three

Question 10

Which statement about AI safety prediction is most accurate?

A. Prediction automatically establishes causation

B. AI eliminates the need for competent safety professionals

C. Models require validation, bias monitoring, and human accountability

D. Historical data are always unbiased

Correct Answer: C. Models require validation, bias monitoring, and human accountability