Road Safety Engineering, Human Factors and Sustainable Planning

Learning objectives

Learn to characterize the mechanisms, assumptions and observational evidence connecting crash data, exposure and regression to the mean, bayesian safety screening and uncertainty, project appraisal and distributional impacts, integrated corridor planning and climate resilience. Demonstrate unit consistency, interpret safety and access alongside vehicle throughput, and compare model predictions against field observations before making recommendations.

How to study this unit

Work through all four lessons, reproduce each numerical illustration without copying the result, operate the linked investigations, then answer the unit assessment and prepare a short project memo. Recognize which assumptions are useful for learning and which prevent design decisions.

Lesson 8.1: Crash data, exposure and regression to the mean

Physical and planning meaning

Crash counts must be interpreted alongside exposure, severity, reporting bias and random variation. A site with unusually high crashes one year may naturally decline even without treatment. Before–after evaluation can therefore overstate improvements when a location is selected because of an exceptionally bad recent period.

Governing relationship

Illustrative crash rate per million vehicle-km = crashes × 10⁶/(AADT × 365 × length_km).

Interpreted worked case

For six reported crashes, AADT 12,000 and a 2 km segment observed for one year, the estimated rate is 6e6/(12000×365×2) ≈ 0.685 crashes per million vehicle-km. It says nothing about injury severity or statistical uncertainty.

Investigating an assumption

How would underreporting or different road-user composition affect comparisons of two apparently similar sites?

Important limitations

The calculation is descriptive; it is not a predictive crash model or causal measure of engineering countermeasure effectiveness.

Dimensional audit

Before computing, rewrite every quantity in one coherent system of units. A time rate needs a time denominator; a spatial density needs a road-length denominator; a percentage must become a decimal before entering an algebraic equation. Unit checking can reject many plausible-looking but incorrect answers without any specialist software.

Three distinct evidence-based investigations

Investigation 8.1.1

An agency is reviewing a low-lying corridor with intermittent drainage blockage after intense storms. The primary affected users include commuters, merchants and emergency responders. Before transferring the relationship Illustrative crash rate per million vehicle-km = crashes × 10⁶/(AADT × 365 × length_km). to the location, explain which inputs could be directly observed, which would need a proxy, and which are not meaningful without a local policy or design standard. Establish a baseline for flooded travel lanes and unreliable service; compare that baseline with an alternative centred on drainage maintenance plus a safe diversion strategy. Your written answer must separate improvements that follow from a formula from improvements that are only plausible hypotheses. Address conditions in which the intervention would fail.

Investigation 8.1.2

Use this worked reference as a challenge rather than as a template to memorize: For six reported crashes, AADT 12,000 and a 2 km segment observed for one year, the estimated rate is 6e6/(12000×365×2) ≈ 0.685 crashes per million vehicle-km. It says nothing about injury severity or statistical uncertainty. Replace one input with a plausible field observation, state why the replacement is plausible, and predict the direction of the result before computing. If the prediction differs from the calculation, investigate units and the sign convention first. Explain how uncertainty from measurement, behavior and a simplified model affects the credibility of a precise numerical answer.

Investigation 8.1.3

A project manager proposes to use a single number from this topic to approve an investment. Prepare a critique rooted in the following limitation: The calculation is descriptive; it is not a predictive crash model or causal measure of engineering countermeasure effectiveness. List two measurements that would improve the study, one externality that the calculation does not represent and one practical decision that could be deferred until the missing evidence is obtained. Avoid claiming the educational model has been professionally calibrated or validated on a real transportation network.

Explain the computation to a classmate

Without looking at the numerical illustration, reconstruct the logic of crash data, exposure and regression to the mean. Start with the field measurement and unit definition, define the reference condition, state the idealization in the equation (Illustrative crash rate per million vehicle-km = crashes × 10⁶/(AADT × 365 × length_km).), and then walk through the calculation so that a reader can independently check every value. Identify which result can be checked by a second technique. If observations contradict the simple model, explain how you would revise the model rather than discard the observations.

Field data and reporting protocol

For a short field exercise about crash data, exposure and regression to the mean, specify a time window, observation location, data dictionary and recording method. Protect individual privacy by avoiding unnecessary vehicle or personal identifiers. Use at least three quality checks: plausibility ranges for the raw input, missing-value counts, and a reconciliation of any aggregated totals. Present separate notes on systematic bias, day-to-day variability and uncertainty in the equation itself.

Decision boundary

The relationship in this lesson is appropriate for concept exploration and transparent back-of-the-envelope screening. A code-compliant transportation-system decision would additionally need jurisdiction-specific requirements, evidence of relevant site conditions and a documented professional review. Name the exact assumptions that would need replacement and identify stakeholders who might be excluded if the decision were made entirely from a vehicle-centred metric.

Lesson 8.2: Bayesian safety screening and uncertainty

Physical and planning meaning

Empirical Bayes methods combine observed crashes with information from comparable sites to temper regression-to-the-mean effects. A simple gamma–Poisson teaching model updates a prior distribution with observed counts, but a real safety performance function must adjust for traffic exposure and geometry.

Governing relationship

If λ ~ Gamma(α,β) in rate form and N ~ Poisson(λT), posterior mean is (α+N)/(β+T).

Interpreted worked case

With a prior α=4, β=2 site-years and three observed crashes in one site-year, the posterior mean is (4+3)/(2+1)=2.33 crashes/year. That differs from both the prior mean 2 and the observed count 3.

Investigating an assumption

Which comparison sites would be exchangeable with a busy urban roundabout? Why does a motorway average make a poor prior?

Important limitations

The illustrative conjugate update is not a substitute for an exposure-adjusted safety performance function or national guidance.

Alternative design comparison

Compare at least two interventions against the same baseline. Record whether each changes demand, available capacity, exposure, pavement condition, human behavior or monitoring quality. An alternative with the largest improvement in one performance indicator is not necessarily superior when its cost, emissions, access impacts and failure modes are considered.

Three distinct evidence-based investigations

Investigation 8.2.1

An agency is reviewing a bridge rehabilitation requiring temporary lane closures. The primary affected users include construction teams and road users. Before transferring the relationship If λ ~ Gamma(α,β) in rate form and N ~ Poisson(λT), posterior mean is (α+N)/(β+T). to the location, explain which inputs could be directly observed, which would need a proxy, and which are not meaningful without a local policy or design standard. Establish a baseline for unexpected queues and disrupted access; compare that baseline with an alternative centred on staged works and active travel-time messaging. Your written answer must separate improvements that follow from a formula from improvements that are only plausible hypotheses. Address conditions in which the intervention would fail.

Investigation 8.2.2

Use this worked reference as a challenge rather than as a template to memorize: With a prior α=4, β=2 site-years and three observed crashes in one site-year, the posterior mean is (4+3)/(2+1)=2.33 crashes/year. That differs from both the prior mean 2 and the observed count 3. Replace one input with a plausible field observation, state why the replacement is plausible, and predict the direction of the result before computing. If the prediction differs from the calculation, investigate units and the sign convention first. Explain how uncertainty from measurement, behavior and a simplified model affects the credibility of a precise numerical answer.

Investigation 8.2.3

A project manager proposes to use a single number from this topic to approve an investment. Prepare a critique rooted in the following limitation: The illustrative conjugate update is not a substitute for an exposure-adjusted safety performance function or national guidance. List two measurements that would improve the study, one externality that the calculation does not represent and one practical decision that could be deferred until the missing evidence is obtained. Avoid claiming the educational model has been professionally calibrated or validated on a real transportation network.

Explain the computation to a classmate

Without looking at the numerical illustration, reconstruct the logic of bayesian safety screening and uncertainty. Start with the field measurement and unit definition, define the reference condition, state the idealization in the equation (If λ ~ Gamma(α,β) in rate form and N ~ Poisson(λT), posterior mean is (α+N)/(β+T).), and then walk through the calculation so that a reader can independently check every value. Identify which result can be checked by a second technique. If observations contradict the simple model, explain how you would revise the model rather than discard the observations.

Field data and reporting protocol

For a short field exercise about bayesian safety screening and uncertainty, specify a time window, observation location, data dictionary and recording method. Protect individual privacy by avoiding unnecessary vehicle or personal identifiers. Use at least three quality checks: plausibility ranges for the raw input, missing-value counts, and a reconciliation of any aggregated totals. Present separate notes on systematic bias, day-to-day variability and uncertainty in the equation itself.

Decision boundary

The relationship in this lesson is appropriate for concept exploration and transparent back-of-the-envelope screening. A code-compliant transportation-system decision would additionally need jurisdiction-specific requirements, evidence of relevant site conditions and a documented professional review. Name the exact assumptions that would need replacement and identify stakeholders who might be excluded if the decision were made entirely from a vehicle-centred metric.

Lesson 8.3: Project appraisal and distributional impacts

Physical and planning meaning

Transport appraisal weighs agency investment, travel time, safety, emissions, reliability and access. Discounting converts future monetary consequences into a common base-year value, but a project’s benefits and costs may fall on different groups. A large positive aggregate benefit–cost ratio can still conceal local displacement or reduced pedestrian safety.

Governing relationship

NPV = Σ (B_t−C_t)/(1+r)^t; BCR = PV(benefits)/PV(costs) when definitions match.

Interpreted worked case

A hypothetical project costs 10 million now and produces 1.6 million per year of net monetized benefit for eight years. At 5% discount, its benefit present value is about 10.34 million and NPV about 0.34 million, before residual value and distributional adjustments.

Investigating an assumption

Design a table showing which groups gain accessibility and which experience construction impacts, emissions or relocation.

Important limitations

Appraisal outputs depend on assumptions about value of time, induced demand, construction cost and the reliability of model forecasts.

Sensitivity and uncertainty

Vary one uncertain parameter while keeping others fixed, then vary a second parameter jointly. A steep response indicates that the result should be reported as a range rather than a single number. Do not equate a sensitivity exercise with a statistical confidence interval; uncertainty also includes model structure and measurement bias.

Three distinct evidence-based investigations

Investigation 8.3.1

An agency is reviewing an industrial connection used by large goods vehicles and irregular delivery convoys. The primary affected users include truck drivers and nearby residents. Before transferring the relationship NPV = Σ (B_t−C_t)/(1+r)^t; BCR = PV(benefits)/PV(costs) when definitions match. to the location, explain which inputs could be directly observed, which would need a proxy, and which are not meaningful without a local policy or design standard. Establish a baseline for reliable travel time, pavement damage and turning geometry; compare that baseline with an alternative centred on a freight-access management plan. Your written answer must separate improvements that follow from a formula from improvements that are only plausible hypotheses. Address conditions in which the intervention would fail.

Investigation 8.3.2

Use this worked reference as a challenge rather than as a template to memorize: A hypothetical project costs 10 million now and produces 1.6 million per year of net monetized benefit for eight years. At 5% discount, its benefit present value is about 10.34 million and NPV about 0.34 million, before residual value and distributional adjustments. Replace one input with a plausible field observation, state why the replacement is plausible, and predict the direction of the result before computing. If the prediction differs from the calculation, investigate units and the sign convention first. Explain how uncertainty from measurement, behavior and a simplified model affects the credibility of a precise numerical answer.

Investigation 8.3.3

A project manager proposes to use a single number from this topic to approve an investment. Prepare a critique rooted in the following limitation: Appraisal outputs depend on assumptions about value of time, induced demand, construction cost and the reliability of model forecasts. List two measurements that would improve the study, one externality that the calculation does not represent and one practical decision that could be deferred until the missing evidence is obtained. Avoid claiming the educational model has been professionally calibrated or validated on a real transportation network.

Explain the computation to a classmate

Without looking at the numerical illustration, reconstruct the logic of project appraisal and distributional impacts. Start with the field measurement and unit definition, define the reference condition, state the idealization in the equation (NPV = Σ (B_t−C_t)/(1+r)^t; BCR = PV(benefits)/PV(costs) when definitions match.), and then walk through the calculation so that a reader can independently check every value. Identify which result can be checked by a second technique. If observations contradict the simple model, explain how you would revise the model rather than discard the observations.

Field data and reporting protocol

For a short field exercise about project appraisal and distributional impacts, specify a time window, observation location, data dictionary and recording method. Protect individual privacy by avoiding unnecessary vehicle or personal identifiers. Use at least three quality checks: plausibility ranges for the raw input, missing-value counts, and a reconciliation of any aggregated totals. Present separate notes on systematic bias, day-to-day variability and uncertainty in the equation itself.

Decision boundary

The relationship in this lesson is appropriate for concept exploration and transparent back-of-the-envelope screening. A code-compliant transportation-system decision would additionally need jurisdiction-specific requirements, evidence of relevant site conditions and a documented professional review. Name the exact assumptions that would need replacement and identify stakeholders who might be excluded if the decision were made entirely from a vehicle-centred metric.

Lesson 8.4: Integrated corridor planning and climate resilience

Physical and planning meaning

A transport corridor links traffic operations, road structure, drainage, emissions, ecology, land use and access. The cheapest initial option can be fragile in extreme rain or incompatible with an expanding walking network. Scenario analysis tests alternatives against several plausible futures rather than selecting a single forecast as certain.

Governing relationship

Expected annual loss = Σ probability_i × consequence_i for a well-defined set of mutually exclusive scenarios.

Interpreted worked case

Suppose an annual flood probability is 5% and estimated direct repair loss is 20 million cost units if the event occurs. The simple expected annual direct loss is one million before indirect disruption, recovery time and correlated hazards.

Investigating an assumption

Choose two resilient alternatives and explain how they perform under both normal demand growth and a more extreme precipitation scenario.

Important limitations

Expected values do not capture tail risk, equity, emergency access or the long life of infrastructure without deeper analysis.

Measurement protocol

Specify which observations distinguish the proposed engineering relationship from a coincidental correlation. Record station location, observation interval, vehicles or users included, calibration checks and missing data. A correct equation evaluated on a biased dataset can create a less defensible conclusion than a rough estimate accompanied by an explicit uncertainty interval.

Three distinct evidence-based investigations

Investigation 8.4.1

An agency is reviewing a busy intersection surrounded by stores, loading areas and narrow footpaths. The primary affected users include local businesses, cyclists and walkers. Before transferring the relationship Expected annual loss = Σ probability_i × consequence_i for a well-defined set of mutually exclusive scenarios. to the location, explain which inputs could be directly observed, which would need a proxy, and which are not meaningful without a local policy or design standard. Establish a baseline for conflicting curb use and vehicle queues; compare that baseline with an alternative centred on reassignment of curb space with inclusive access. Your written answer must separate improvements that follow from a formula from improvements that are only plausible hypotheses. Address conditions in which the intervention would fail.

Investigation 8.4.2

Use this worked reference as a challenge rather than as a template to memorize: Suppose an annual flood probability is 5% and estimated direct repair loss is 20 million cost units if the event occurs. The simple expected annual direct loss is one million before indirect disruption, recovery time and correlated hazards. Replace one input with a plausible field observation, state why the replacement is plausible, and predict the direction of the result before computing. If the prediction differs from the calculation, investigate units and the sign convention first. Explain how uncertainty from measurement, behavior and a simplified model affects the credibility of a precise numerical answer.

Investigation 8.4.3

A project manager proposes to use a single number from this topic to approve an investment. Prepare a critique rooted in the following limitation: Expected values do not capture tail risk, equity, emergency access or the long life of infrastructure without deeper analysis. List two measurements that would improve the study, one externality that the calculation does not represent and one practical decision that could be deferred until the missing evidence is obtained. Avoid claiming the educational model has been professionally calibrated or validated on a real transportation network.

Explain the computation to a classmate

Without looking at the numerical illustration, reconstruct the logic of integrated corridor planning and climate resilience. Start with the field measurement and unit definition, define the reference condition, state the idealization in the equation (Expected annual loss = Σ probability_i × consequence_i for a well-defined set of mutually exclusive scenarios.), and then walk through the calculation so that a reader can independently check every value. Identify which result can be checked by a second technique. If observations contradict the simple model, explain how you would revise the model rather than discard the observations.

Field data and reporting protocol

For a short field exercise about integrated corridor planning and climate resilience, specify a time window, observation location, data dictionary and recording method. Protect individual privacy by avoiding unnecessary vehicle or personal identifiers. Use at least three quality checks: plausibility ranges for the raw input, missing-value counts, and a reconciliation of any aggregated totals. Present separate notes on systematic bias, day-to-day variability and uncertainty in the equation itself.

Decision boundary

The relationship in this lesson is appropriate for concept exploration and transparent back-of-the-envelope screening. A code-compliant transportation-system decision would additionally need jurisdiction-specific requirements, evidence of relevant site conditions and a documented professional review. Name the exact assumptions that would need replacement and identify stakeholders who might be excluded if the decision were made entirely from a vehicle-centred metric.

Unit synthesis exercise

Connect crash data, exposure and regression to the mean with integrated corridor planning and climate resilience: imagine a corridor proposal that improves the first measure but worsens the last. Write a one-page recommendation showing the trade-off, a minimum dataset for testing both effects, and the circumstances that would reverse your recommendation.

Reading the laboratory outputs

Every chart is calculated for hypothetical parameters. Change one parameter at a time to understand the model. Check a point by hand. Export a CSV, note the axes, then identify at least one physically relevant effect omitted by the model. No laboratory in this book is approved for design or real-time operational control.

Extended safety and appraisal case: a high-crash corridor

A regional road agency notices that a 2-kilometre corridor has six reported crashes over the past year. Daily traffic volume is estimated at 12,000 vehicles. A descriptive rate is 6 × 10^6/(12,000 × 365 × 2) ≈ 0.685 crashes per million vehicle-kilometres. Such a rate can help organize exposure information, but comparisons require consistent injury reporting, segment definitions, traffic measurement and time windows. The agency should not rank a pedestrian-dense town-centre road against a limited-access motorway as if they are exchangeable sites with identical risk exposure.

The agency proposes to fund a guardrail because the corridor experienced unusually many crashes in the selection year. There is a selection-bias problem: choosing sites at their observed maximum tends to create apparent improvements when subsequent observations move toward ordinary conditions. Regression to the mean is not a defense for doing nothing; it is a reason to evaluate interventions properly. If an illustrative gamma prior for a yearly crash rate has mean two and effective prior exposure two site-years, observing three crashes in one new year yields posterior mean (4+3)/(2+1)=2.33 crashes/year. This toy model illustrates shrinkage but does not account for variable traffic exposure or roadway characteristics. An actual empirical Bayes analysis needs an appropriate safety performance function and reference-site data.

Collect the circumstances of reported crashes, injury severity, turning movements, speed distribution, road-user types, lighting, road-edge conditions and relevant exposure. A higher share of serious crashes may justify action even when a simple total-count comparison does not rank the site highest. Assess the validity of police reports and hospital-linked records under local privacy and consent requirements; reporting rates can differ by road-user type. Build a collision diagram rather than relying only on a coloured dot map. Identify hazardous mechanisms that could be altered by engineering action.

Compare interventions targeting the identified risk: speed management, a protected crossing, better junction sight lines, shoulder improvement, forgiving roadside design or enforcement in conjunction with engineering changes. Each has a different causal pathway. A treatment does not have a credible expected crash reduction merely because its name sounds safety-related. Research or locally validated crash modification factors are required for formal benefit estimates. Treat site-specific applicability, implementation fidelity and maintenance as part of the evaluation.

For a broader project, suppose an investment costs ten million cost units at the start, while an estimate of monetized benefit is 1.6 million per year for eight years. At discount rate 5%, the present value of that constant annuity is 1.6 × [1−(1.05)^−8]/0.05 ≈ 10.34 million. The resulting NPV is around 0.34 million before omitted items. The difference is small relative to plausible uncertainty in construction cost or forecast benefits. Sensitivity analysis and distributional impacts should therefore be central, not an afterthought.

A credible corridor decision also tests flood resilience, freight routing, emergency access, school crossings and induced demand. Present a balanced dashboard that separates economic value, severe-injury risk, accessibility impacts and who pays. State how recommendations would change if crash underreporting is large, if a new school increases pedestrian exposure, or if a flood-prone bridge experiences more frequent closures. The purpose of the exercise is to improve decision quality under uncertainty, not to turn a simple calculator output into a claim of proven safety.

Unit 8: Road Safety, Sustainability and Planning

Book 14 · Transportation Engineering · Civil Engineering

Learning objectives

Learn to characterize the mechanisms, assumptions and observational evidence connecting crash data, exposure and regression to the mean, bayesian safety screening and uncertainty, project appraisal and distributional impacts, integrated corridor planning and climate resilience. Demonstrate unit consistency, interpret safety and access alongside vehicle throughput, and compare model predictions against field observations before making recommendations.

How to study this unit

Work through all four lessons, reproduce each numerical illustration without copying the result, operate the linked investigations, then answer the unit assessment and prepare a short project memo. Recognize which assumptions are useful for learning and which prevent design decisions.

Lesson 8.1: Crash data, exposure and regression to the mean

Physical and planning meaning

Crash counts must be interpreted alongside exposure, severity, reporting bias and random variation. A site with unusually high crashes one year may naturally decline even without treatment. Before–after evaluation can therefore overstate improvements when a location is selected because of an exceptionally bad recent period.

Governing relationship

Illustrative crash rate per million vehicle-km = crashes × 10⁶/(AADT × 365 × length_km).

Interpreted worked case

For six reported crashes, AADT 12,000 and a 2 km segment observed for one year, the estimated rate is 6e6/(12000×365×2) ≈ 0.685 crashes per million vehicle-km. It says nothing about injury severity or statistical uncertainty.

Investigating an assumption

How would underreporting or different road-user composition affect comparisons of two apparently similar sites?

Important limitations

The calculation is descriptive; it is not a predictive crash model or causal measure of engineering countermeasure effectiveness.

Dimensional audit

Before computing, rewrite every quantity in one coherent system of units. A time rate needs a time denominator; a spatial density needs a road-length denominator; a percentage must become a decimal before entering an algebraic equation. Unit checking can reject many plausible-looking but incorrect answers without any specialist software.

Three distinct evidence-based investigations

Investigation 8.1.1

An agency is reviewing a low-lying corridor with intermittent drainage blockage after intense storms. The primary affected users include commuters, merchants and emergency responders. Before transferring the relationship Illustrative crash rate per million vehicle-km = crashes × 10⁶/(AADT × 365 × length_km). to the location, explain which inputs could be directly observed, which would need a proxy, and which are not meaningful without a local policy or design standard. Establish a baseline for flooded travel lanes and unreliable service; compare that baseline with an alternative centred on drainage maintenance plus a safe diversion strategy. Your written answer must separate improvements that follow from a formula from improvements that are only plausible hypotheses. Address conditions in which the intervention would fail.

Investigation 8.1.2

Use this worked reference as a challenge rather than as a template to memorize: For six reported crashes, AADT 12,000 and a 2 km segment observed for one year, the estimated rate is 6e6/(12000×365×2) ≈ 0.685 crashes per million vehicle-km. It says nothing about injury severity or statistical uncertainty. Replace one input with a plausible field observation, state why the replacement is plausible, and predict the direction of the result before computing. If the prediction differs from the calculation, investigate units and the sign convention first. Explain how uncertainty from measurement, behavior and a simplified model affects the credibility of a precise numerical answer.

Investigation 8.1.3

A project manager proposes to use a single number from this topic to approve an investment. Prepare a critique rooted in the following limitation: The calculation is descriptive; it is not a predictive crash model or causal measure of engineering countermeasure effectiveness. List two measurements that would improve the study, one externality that the calculation does not represent and one practical decision that could be deferred until the missing evidence is obtained. Avoid claiming the educational model has been professionally calibrated or validated on a real transportation network.

Explain the computation to a classmate

Without looking at the numerical illustration, reconstruct the logic of crash data, exposure and regression to the mean. Start with the field measurement and unit definition, define the reference condition, state the idealization in the equation (Illustrative crash rate per million vehicle-km = crashes × 10⁶/(AADT × 365 × length_km).), and then walk through the calculation so that a reader can independently check every value. Identify which result can be checked by a second technique. If observations contradict the simple model, explain how you would revise the model rather than discard the observations.

Field data and reporting protocol

For a short field exercise about crash data, exposure and regression to the mean, specify a time window, observation location, data dictionary and recording method. Protect individual privacy by avoiding unnecessary vehicle or personal identifiers. Use at least three quality checks: plausibility ranges for the raw input, missing-value counts, and a reconciliation of any aggregated totals. Present separate notes on systematic bias, day-to-day variability and uncertainty in the equation itself.

Decision boundary

The relationship in this lesson is appropriate for concept exploration and transparent back-of-the-envelope screening. A code-compliant transportation-system decision would additionally need jurisdiction-specific requirements, evidence of relevant site conditions and a documented professional review. Name the exact assumptions that would need replacement and identify stakeholders who might be excluded if the decision were made entirely from a vehicle-centred metric.

Lesson 8.2: Bayesian safety screening and uncertainty

Physical and planning meaning

Empirical Bayes methods combine observed crashes with information from comparable sites to temper regression-to-the-mean effects. A simple gamma–Poisson teaching model updates a prior distribution with observed counts, but a real safety performance function must adjust for traffic exposure and geometry.

Governing relationship

If λ ~ Gamma(α,β) in rate form and N ~ Poisson(λT), posterior mean is (α+N)/(β+T).

Interpreted worked case

With a prior α=4, β=2 site-years and three observed crashes in one site-year, the posterior mean is (4+3)/(2+1)=2.33 crashes/year. That differs from both the prior mean 2 and the observed count 3.

Investigating an assumption

Which comparison sites would be exchangeable with a busy urban roundabout? Why does a motorway average make a poor prior?

Important limitations

The illustrative conjugate update is not a substitute for an exposure-adjusted safety performance function or national guidance.

Alternative design comparison

Compare at least two interventions against the same baseline. Record whether each changes demand, available capacity, exposure, pavement condition, human behavior or monitoring quality. An alternative with the largest improvement in one performance indicator is not necessarily superior when its cost, emissions, access impacts and failure modes are considered.

Three distinct evidence-based investigations

Investigation 8.2.1

An agency is reviewing a bridge rehabilitation requiring temporary lane closures. The primary affected users include construction teams and road users. Before transferring the relationship If λ ~ Gamma(α,β) in rate form and N ~ Poisson(λT), posterior mean is (α+N)/(β+T). to the location, explain which inputs could be directly observed, which would need a proxy, and which are not meaningful without a local policy or design standard. Establish a baseline for unexpected queues and disrupted access; compare that baseline with an alternative centred on staged works and active travel-time messaging. Your written answer must separate improvements that follow from a formula from improvements that are only plausible hypotheses. Address conditions in which the intervention would fail.

Investigation 8.2.2

Use this worked reference as a challenge rather than as a template to memorize: With a prior α=4, β=2 site-years and three observed crashes in one site-year, the posterior mean is (4+3)/(2+1)=2.33 crashes/year. That differs from both the prior mean 2 and the observed count 3. Replace one input with a plausible field observation, state why the replacement is plausible, and predict the direction of the result before computing. If the prediction differs from the calculation, investigate units and the sign convention first. Explain how uncertainty from measurement, behavior and a simplified model affects the credibility of a precise numerical answer.

Investigation 8.2.3

A project manager proposes to use a single number from this topic to approve an investment. Prepare a critique rooted in the following limitation: The illustrative conjugate update is not a substitute for an exposure-adjusted safety performance function or national guidance. List two measurements that would improve the study, one externality that the calculation does not represent and one practical decision that could be deferred until the missing evidence is obtained. Avoid claiming the educational model has been professionally calibrated or validated on a real transportation network.

Explain the computation to a classmate

Without looking at the numerical illustration, reconstruct the logic of bayesian safety screening and uncertainty. Start with the field measurement and unit definition, define the reference condition, state the idealization in the equation (If λ ~ Gamma(α,β) in rate form and N ~ Poisson(λT), posterior mean is (α+N)/(β+T).), and then walk through the calculation so that a reader can independently check every value. Identify which result can be checked by a second technique. If observations contradict the simple model, explain how you would revise the model rather than discard the observations.

Field data and reporting protocol

For a short field exercise about bayesian safety screening and uncertainty, specify a time window, observation location, data dictionary and recording method. Protect individual privacy by avoiding unnecessary vehicle or personal identifiers. Use at least three quality checks: plausibility ranges for the raw input, missing-value counts, and a reconciliation of any aggregated totals. Present separate notes on systematic bias, day-to-day variability and uncertainty in the equation itself.

Decision boundary

The relationship in this lesson is appropriate for concept exploration and transparent back-of-the-envelope screening. A code-compliant transportation-system decision would additionally need jurisdiction-specific requirements, evidence of relevant site conditions and a documented professional review. Name the exact assumptions that would need replacement and identify stakeholders who might be excluded if the decision were made entirely from a vehicle-centred metric.

Lesson 8.3: Project appraisal and distributional impacts

Physical and planning meaning

Transport appraisal weighs agency investment, travel time, safety, emissions, reliability and access. Discounting converts future monetary consequences into a common base-year value, but a project’s benefits and costs may fall on different groups. A large positive aggregate benefit–cost ratio can still conceal local displacement or reduced pedestrian safety.

Governing relationship

NPV = Σ (B_t−C_t)/(1+r)^t; BCR = PV(benefits)/PV(costs) when definitions match.

Interpreted worked case

A hypothetical project costs 10 million now and produces 1.6 million per year of net monetized benefit for eight years. At 5% discount, its benefit present value is about 10.34 million and NPV about 0.34 million, before residual value and distributional adjustments.

Investigating an assumption

Design a table showing which groups gain accessibility and which experience construction impacts, emissions or relocation.

Important limitations

Appraisal outputs depend on assumptions about value of time, induced demand, construction cost and the reliability of model forecasts.

Sensitivity and uncertainty

Vary one uncertain parameter while keeping others fixed, then vary a second parameter jointly. A steep response indicates that the result should be reported as a range rather than a single number. Do not equate a sensitivity exercise with a statistical confidence interval; uncertainty also includes model structure and measurement bias.

Three distinct evidence-based investigations

Investigation 8.3.1

An agency is reviewing an industrial connection used by large goods vehicles and irregular delivery convoys. The primary affected users include truck drivers and nearby residents. Before transferring the relationship NPV = Σ (B_t−C_t)/(1+r)^t; BCR = PV(benefits)/PV(costs) when definitions match. to the location, explain which inputs could be directly observed, which would need a proxy, and which are not meaningful without a local policy or design standard. Establish a baseline for reliable travel time, pavement damage and turning geometry; compare that baseline with an alternative centred on a freight-access management plan. Your written answer must separate improvements that follow from a formula from improvements that are only plausible hypotheses. Address conditions in which the intervention would fail.

Investigation 8.3.2

Use this worked reference as a challenge rather than as a template to memorize: A hypothetical project costs 10 million now and produces 1.6 million per year of net monetized benefit for eight years. At 5% discount, its benefit present value is about 10.34 million and NPV about 0.34 million, before residual value and distributional adjustments. Replace one input with a plausible field observation, state why the replacement is plausible, and predict the direction of the result before computing. If the prediction differs from the calculation, investigate units and the sign convention first. Explain how uncertainty from measurement, behavior and a simplified model affects the credibility of a precise numerical answer.

Investigation 8.3.3

A project manager proposes to use a single number from this topic to approve an investment. Prepare a critique rooted in the following limitation: Appraisal outputs depend on assumptions about value of time, induced demand, construction cost and the reliability of model forecasts. List two measurements that would improve the study, one externality that the calculation does not represent and one practical decision that could be deferred until the missing evidence is obtained. Avoid claiming the educational model has been professionally calibrated or validated on a real transportation network.

Explain the computation to a classmate

Without looking at the numerical illustration, reconstruct the logic of project appraisal and distributional impacts. Start with the field measurement and unit definition, define the reference condition, state the idealization in the equation (NPV = Σ (B_t−C_t)/(1+r)^t; BCR = PV(benefits)/PV(costs) when definitions match.), and then walk through the calculation so that a reader can independently check every value. Identify which result can be checked by a second technique. If observations contradict the simple model, explain how you would revise the model rather than discard the observations.

Field data and reporting protocol

For a short field exercise about project appraisal and distributional impacts, specify a time window, observation location, data dictionary and recording method. Protect individual privacy by avoiding unnecessary vehicle or personal identifiers. Use at least three quality checks: plausibility ranges for the raw input, missing-value counts, and a reconciliation of any aggregated totals. Present separate notes on systematic bias, day-to-day variability and uncertainty in the equation itself.

Decision boundary

The relationship in this lesson is appropriate for concept exploration and transparent back-of-the-envelope screening. A code-compliant transportation-system decision would additionally need jurisdiction-specific requirements, evidence of relevant site conditions and a documented professional review. Name the exact assumptions that would need replacement and identify stakeholders who might be excluded if the decision were made entirely from a vehicle-centred metric.

Lesson 8.4: Integrated corridor planning and climate resilience

Physical and planning meaning

A transport corridor links traffic operations, road structure, drainage, emissions, ecology, land use and access. The cheapest initial option can be fragile in extreme rain or incompatible with an expanding walking network. Scenario analysis tests alternatives against several plausible futures rather than selecting a single forecast as certain.

Governing relationship

Expected annual loss = Σ probability_i × consequence_i for a well-defined set of mutually exclusive scenarios.

Interpreted worked case

Suppose an annual flood probability is 5% and estimated direct repair loss is 20 million cost units if the event occurs. The simple expected annual direct loss is one million before indirect disruption, recovery time and correlated hazards.

Investigating an assumption

Choose two resilient alternatives and explain how they perform under both normal demand growth and a more extreme precipitation scenario.

Important limitations

Expected values do not capture tail risk, equity, emergency access or the long life of infrastructure without deeper analysis.

Measurement protocol

Specify which observations distinguish the proposed engineering relationship from a coincidental correlation. Record station location, observation interval, vehicles or users included, calibration checks and missing data. A correct equation evaluated on a biased dataset can create a less defensible conclusion than a rough estimate accompanied by an explicit uncertainty interval.

Three distinct evidence-based investigations

Investigation 8.4.1

An agency is reviewing a busy intersection surrounded by stores, loading areas and narrow footpaths. The primary affected users include local businesses, cyclists and walkers. Before transferring the relationship Expected annual loss = Σ probability_i × consequence_i for a well-defined set of mutually exclusive scenarios. to the location, explain which inputs could be directly observed, which would need a proxy, and which are not meaningful without a local policy or design standard. Establish a baseline for conflicting curb use and vehicle queues; compare that baseline with an alternative centred on reassignment of curb space with inclusive access. Your written answer must separate improvements that follow from a formula from improvements that are only plausible hypotheses. Address conditions in which the intervention would fail.

Investigation 8.4.2

Use this worked reference as a challenge rather than as a template to memorize: Suppose an annual flood probability is 5% and estimated direct repair loss is 20 million cost units if the event occurs. The simple expected annual direct loss is one million before indirect disruption, recovery time and correlated hazards. Replace one input with a plausible field observation, state why the replacement is plausible, and predict the direction of the result before computing. If the prediction differs from the calculation, investigate units and the sign convention first. Explain how uncertainty from measurement, behavior and a simplified model affects the credibility of a precise numerical answer.

Investigation 8.4.3

A project manager proposes to use a single number from this topic to approve an investment. Prepare a critique rooted in the following limitation: Expected values do not capture tail risk, equity, emergency access or the long life of infrastructure without deeper analysis. List two measurements that would improve the study, one externality that the calculation does not represent and one practical decision that could be deferred until the missing evidence is obtained. Avoid claiming the educational model has been professionally calibrated or validated on a real transportation network.

Explain the computation to a classmate

Without looking at the numerical illustration, reconstruct the logic of integrated corridor planning and climate resilience. Start with the field measurement and unit definition, define the reference condition, state the idealization in the equation (Expected annual loss = Σ probability_i × consequence_i for a well-defined set of mutually exclusive scenarios.), and then walk through the calculation so that a reader can independently check every value. Identify which result can be checked by a second technique. If observations contradict the simple model, explain how you would revise the model rather than discard the observations.

Field data and reporting protocol

For a short field exercise about integrated corridor planning and climate resilience, specify a time window, observation location, data dictionary and recording method. Protect individual privacy by avoiding unnecessary vehicle or personal identifiers. Use at least three quality checks: plausibility ranges for the raw input, missing-value counts, and a reconciliation of any aggregated totals. Present separate notes on systematic bias, day-to-day variability and uncertainty in the equation itself.

Decision boundary

The relationship in this lesson is appropriate for concept exploration and transparent back-of-the-envelope screening. A code-compliant transportation-system decision would additionally need jurisdiction-specific requirements, evidence of relevant site conditions and a documented professional review. Name the exact assumptions that would need replacement and identify stakeholders who might be excluded if the decision were made entirely from a vehicle-centred metric.

Unit synthesis exercise

Connect crash data, exposure and regression to the mean with integrated corridor planning and climate resilience: imagine a corridor proposal that improves the first measure but worsens the last. Write a one-page recommendation showing the trade-off, a minimum dataset for testing both effects, and the circumstances that would reverse your recommendation.

Reading the laboratory outputs

Every chart is calculated for hypothetical parameters. Change one parameter at a time to understand the model. Check a point by hand. Export a CSV, note the axes, then identify at least one physically relevant effect omitted by the model. No laboratory in this book is approved for design or real-time operational control.

Extended safety and appraisal case: a high-crash corridor

A regional road agency notices that a 2-kilometre corridor has six reported crashes over the past year. Daily traffic volume is estimated at 12,000 vehicles. A descriptive rate is 6 × 10^6/(12,000 × 365 × 2) ≈ 0.685 crashes per million vehicle-kilometres. Such a rate can help organize exposure information, but comparisons require consistent injury reporting, segment definitions, traffic measurement and time windows. The agency should not rank a pedestrian-dense town-centre road against a limited-access motorway as if they are exchangeable sites with identical risk exposure.

The agency proposes to fund a guardrail because the corridor experienced unusually many crashes in the selection year. There is a selection-bias problem: choosing sites at their observed maximum tends to create apparent improvements when subsequent observations move toward ordinary conditions. Regression to the mean is not a defense for doing nothing; it is a reason to evaluate interventions properly. If an illustrative gamma prior for a yearly crash rate has mean two and effective prior exposure two site-years, observing three crashes in one new year yields posterior mean (4+3)/(2+1)=2.33 crashes/year. This toy model illustrates shrinkage but does not account for variable traffic exposure or roadway characteristics. An actual empirical Bayes analysis needs an appropriate safety performance function and reference-site data.

Collect the circumstances of reported crashes, injury severity, turning movements, speed distribution, road-user types, lighting, road-edge conditions and relevant exposure. A higher share of serious crashes may justify action even when a simple total-count comparison does not rank the site highest. Assess the validity of police reports and hospital-linked records under local privacy and consent requirements; reporting rates can differ by road-user type. Build a collision diagram rather than relying only on a coloured dot map. Identify hazardous mechanisms that could be altered by engineering action.

Compare interventions targeting the identified risk: speed management, a protected crossing, better junction sight lines, shoulder improvement, forgiving roadside design or enforcement in conjunction with engineering changes. Each has a different causal pathway. A treatment does not have a credible expected crash reduction merely because its name sounds safety-related. Research or locally validated crash modification factors are required for formal benefit estimates. Treat site-specific applicability, implementation fidelity and maintenance as part of the evaluation.

For a broader project, suppose an investment costs ten million cost units at the start, while an estimate of monetized benefit is 1.6 million per year for eight years. At discount rate 5%, the present value of that constant annuity is 1.6 × [1−(1.05)^−8]/0.05 ≈ 10.34 million. The resulting NPV is around 0.34 million before omitted items. The difference is small relative to plausible uncertainty in construction cost or forecast benefits. Sensitivity analysis and distributional impacts should therefore be central, not an afterthought.

A credible corridor decision also tests flood resilience, freight routing, emergency access, school crossings and induced demand. Present a balanced dashboard that separates economic value, severe-injury risk, accessibility impacts and who pays. State how recommendations would change if crash underreporting is large, if a new school increases pedestrian exposure, or if a flood-prone bridge experiences more frequent closures. The purpose of the exercise is to improve decision quality under uncertainty, not to turn a simple calculator output into a claim of proven safety.

Engineering Projects

Book 14 — Engineering projects

Every project uses a simulated or explicitly permissioned dataset. No student project is a professional design approval.

Project 1: Traffic Stream Theory

Develop a defensible 30-minute classified count and compare vehicle flow with estimated passenger flow; explain which number better captures accessibility.

Required deliverables: an engineering memo (scope, assumptions, alternative options and limitations); a clearly labeled table of data; two charts with captions and units; one uncertainty assessment; and a recommendation that states the evidence required before implementation.

Quality rubric: 25% model definition and units; 25% evidence and reproducibility; 20% alternatives and externalities; 20% uncertainty and risk; 10% communication and transparent limitations.

Project 2: Traffic Data and Roadway Performance

Construct a travel-time and bottleneck field-study plan with route selection, data-quality checks and a short recommendation for recurrent congestion.

Required deliverables: an engineering memo (scope, assumptions, alternative options and limitations); a clearly labeled table of data; two charts with captions and units; one uncertainty assessment; and a recommendation that states the evidence required before implementation.

Quality rubric: 25% model definition and units; 25% evidence and reproducibility; 20% alternatives and externalities; 20% uncertainty and risk; 10% communication and transparent limitations.

Project 3: Geometric Design and Human Factors

Prepare geometric alternatives for a road segment with pedestrian crossings, crossfall and limited land; document stopping-distance assumptions and access impacts.

Required deliverables: an engineering memo (scope, assumptions, alternative options and limitations); a clearly labeled table of data; two charts with captions and units; one uncertainty assessment; and a recommendation that states the evidence required before implementation.

Quality rubric: 25% model definition and units; 25% evidence and reproducibility; 20% alternatives and externalities; 20% uncertainty and risk; 10% communication and transparent limitations.

Project 4: Intersections and Traffic Control

Compare two signal plans by effective green, queue persistence and pedestrian clearance constraints; do not issue actual timings.

Required deliverables: an engineering memo (scope, assumptions, alternative options and limitations); a clearly labeled table of data; two charts with captions and units; one uncertainty assessment; and a recommendation that states the evidence required before implementation.

Quality rubric: 25% model definition and units; 25% evidence and reproducibility; 20% alternatives and externalities; 20% uncertainty and risk; 10% communication and transparent limitations.

Project 5: Pavements and Material Performance

Create an inspection and lifecycle plan for a flexible and rigid pavement option using hypothetical materials, drainage and truck proportions.

Required deliverables: an engineering memo (scope, assumptions, alternative options and limitations); a clearly labeled table of data; two charts with captions and units; one uncertainty assessment; and a recommendation that states the evidence required before implementation.

Quality rubric: 25% model definition and units; 25% evidence and reproducibility; 20% alternatives and externalities; 20% uncertainty and risk; 10% communication and transparent limitations.

Project 6: Roadway Drainage and Infrastructure Systems

Evaluate a roadway crossing under more intense storms and debris blockage; explain how rational runoff and Manning conveyance differ from complete hydraulic design.

Required deliverables: an engineering memo (scope, assumptions, alternative options and limitations); a clearly labeled table of data; two charts with captions and units; one uncertainty assessment; and a recommendation that states the evidence required before implementation.

Quality rubric: 25% model definition and units; 25% evidence and reproducibility; 20% alternatives and externalities; 20% uncertainty and risk; 10% communication and transparent limitations.

Project 7: Transit, Walking and Cycling

Redesign a transit stop and nearby pedestrian crossing to improve person mobility while addressing mobility access and cycling conflict.

Required deliverables: an engineering memo (scope, assumptions, alternative options and limitations); a clearly labeled table of data; two charts with captions and units; one uncertainty assessment; and a recommendation that states the evidence required before implementation.

Quality rubric: 25% model definition and units; 25% evidence and reproducibility; 20% alternatives and externalities; 20% uncertainty and risk; 10% communication and transparent limitations.

Project 8: Road Safety, Sustainability and Planning

Prepare a corridor appraisal using safety, exposure, climate resilience, user costs, induced travel and distributional concerns, with a transparent scenario table.

Required deliverables: an engineering memo (scope, assumptions, alternative options and limitations); a clearly labeled table of data; two charts with captions and units; one uncertainty assessment; and a recommendation that states the evidence required before implementation.

Quality rubric: 25% model definition and units; 25% evidence and reproducibility; 20% alternatives and externalities; 20% uncertainty and risk; 10% communication and transparent limitations.

Technical references

Teaching references and standards

The source list below helps students locate authoritative resources. The book does not reproduce or certify compliance with any copyrighted design code. FHWA guidance applies in US contexts; other jurisdictions require their own adopted standards and professional review.

• FHWA Manual on Uniform Traffic Control Devices, 11th Edition with Revision 1 (current as of November 2026): https://mutcd.fhwa.dot.gov/kno_11th_Editionr1.htm
• FHWA Traffic Signal Timing Manual — Chapter 3: https://ops.fhwa.dot.gov/publications/fhwahop08024/chapter3.htm
• FHWA Traffic Signal Timing Manual — Chapter 6: https://ops.fhwa.dot.gov/publications/fhwahop08024/chapter6.htm
• FHWA Traffic Flow Theory: https://www.fhwa.dot.gov/publications/research/operations/tft/
• FHWA Highway Safety Manual resources: https://highways.dot.gov/safety/data-analysis-tools/highway-safety-manual
• FHWA Pavement and Materials: https://www.fhwa.dot.gov/pavement/
• FHWA Hydraulic Engineering: https://www.fhwa.dot.gov/engineering/hydraulics/
• NCHRP Transportation Research Board: https://www.trb.org/
• United States Access Board — Accessibility Guidelines: https://www.access-board.gov/