Crash Patterns, Collision Diagrams, and Diagnostic Hypotheses

Crash Patterns, Collision Diagrams, and Diagnostic Hypotheses

Verify the event information first

Diagnosis seeks contributing mechanisms behind a supported pattern. Begin by verifying crash locations, directions, dates, types, severity, and reporting consistency. A geocoding error can create a cluster at an intersection even when events occurred upstream. A narrative can disagree with the coded maneuver. Resolve or document those discrepancies before drawing a diagram or choosing a treatment.

Review the relevant vehicles and people, road conditions, time, weather, traffic control, and reported circumstances. An at-fault designation can inform the account but is not the complete safety explanation. Consider visibility, speed, task demand, vehicle characteristics, and post-crash consequences as well as the reported action.

Construct a collision diagram

A collision diagram schematically shows movements and impact patterns from historical records. It can use arrows for trajectories and consistent symbols for crash types, with labels for date, severity, or conditions. The diagram is a summary of evidence, not an exact reconstruction of every vehicle position.

Keep a legend, period, source, and location method. Avoid visually merging distinct events or omitting less convenient records. If an event's direction is uncertain, indicate that uncertainty rather than assigning a path that supports the favored explanation. Separate pedestrian and bicycle movements where relevant.

Patterns can suggest localized mechanisms. Repeated westbound turning conflicts with southbound through vehicles might justify checking the westbound driver's view toward northbound-origin traffic approaching southbound, speeds, gap choices, and control visibility. Do not assume a specific named quadrant without checking the actual coordinate layout and driver's position. The direction pattern narrows investigation; it does not prove one cause.

Construct a condition diagram

A condition diagram describes physical and operational features: lanes, control, crossings, driveways, markings, roadside objects, sight obstructions, grades, and relevant dimensions. It differs from the collision diagram, which describes events. A scaled drawing can support measurement, while a schematic should be labeled as such.

Overlaying or comparing the diagrams can connect a movement pattern with a physical feature. For example, events on one approach may correspond to vegetation or an unusual control arrangement. That connection is a hypothesis to test, not causal proof from a drawing. Check conditions during the crash period as well as the current inventory.

Disaggregate with a crash tree

A crash tree organizes counts into categories, such as segment versus intersection, departure versus other types, wet versus dry, or day versus night. Each split needs clear definitions and compatible totals. Use mutually exclusive categories where intended, and show unknown values rather than silently treating them as absence.

The tree can reveal where a broad total conceals a specific pattern. For example, many corridor events may be concentrated in wet-weather departures rather than intersection conflicts. The next step is to examine exposure, facility characteristics, and possible mechanisms. A tree does not establish statistical significance solely by subdividing the data.

Compare proportions appropriately

Suppose 26 of 38 departures are wet-weather events: about 68.4%. An illustrative comparable-network share of 18% suggests overrepresentation worth investigating. The comparison needs compatible definitions, weather exposure, facility conditions, period, and sample size. A higher proportion can occur because another category is lower, so also inspect counts and exposure.

Do not equate overrepresentation with proof of low friction. Drainage, tires, speed, curve recognition, reporting, or differing rain exposure can contribute. Formal inference requires an appropriate statistical approach and assumptions. The proportion comparison is a diagnostic starting point, not a complete causal test.

Build and test alternative explanations

For rear-end crashes near a signal, candidates might include an unexpected queue, limited visibility, access conflicts, timing, or approach speed. Specify what evidence would support each explanation. Queue observations, sight checks, narratives, and signal records may resolve different parts of the question.

For angle crashes, review gap selection, control recognition, sightlines, turning volumes, and speeds. For pedestrian events, include desire lines, turning movements, lighting, accessible routes, and transit. Avoid selecting a treatment from collision type alone; the same type can arise through different mechanisms.

Use multidisciplinary perspectives

Engineering and operations staff can review layout and control. Maintenance can identify recurring surface or vegetation conditions. Enforcement can explain observed behavior and report practices. Health and EMS can clarify injuries and response. Community members can describe experiences and missing movements. Select partners according to the problem rather than using a fixed team list for every site.

Record disagreement and evidence needs. A useful diagnosis explains why the team favors a mechanism and what remains uncertain. It should be understandable to decision-makers without requiring them to infer causation from a dense technical chart.

Build a defensible hypothesis

• Describe the type, severity, time, and movement pattern.
• Review narratives, sketches, exposure, and operating conditions.
• Test candidate contributors with field and other evidence.
• Record uncertainty and the next investigative step.

Produce a treatment-ready diagnostic statement

A strong statement combines the pattern, plausible mechanism, supporting evidence, affected users, and limits. For example: “The records and observations indicate recurrent late queue recognition on this approach; visibility and operating speed need further assessment before selecting a warning or operational treatment.” It does not say “rear-end crashes prove the signal is wrong.”

Use the FHWA diagnostic toolkit to connect tools and candidate measures. The output should support appropriate countermeasure selection and identify what the evaluation must later test.