Safety Data Types, Sources, and Users

Safety Data Types, Sources, and Users

Begin with the analytical question

A safety analyst needs more than a crash total. To screen a network, diagnose a pattern, select a treatment, or evaluate a program, identify the outcome and explanatory information needed. A crossing investigation may need pedestrian movements, vehicle speeds, signal operation, visibility, narratives, and injury outcomes. A statewide plan needs consistent patterns across owners and populations. A research model needs comparable observations and exposure.

Safety data users include roadway owners, engineers, planners, enforcement, licensing agencies, health and EMS professionals, researchers, policymakers, and communities. Their needs differ. An engineer may need precise locations; a health analyst may need person-level injuries; a program manager may need implementation and behavioral measures. Agree on definitions before combining their results.

NHTSA identifies six state traffic-records systems: crash, driver, vehicle, roadway, citation/adjudication, and injury surveillance. Traffic-volume data are essential exposure information but do not replace citation/adjudication in that six-system list. NHTSA traffic records

Police crash records

State or provincial crash systems typically record event location, time, road and weather conditions, vehicles, persons, maneuvers, injury status, and contributing circumstances. Coded elements support aggregation; narratives and sketches can reveal details that the codes omit. The reporting agency and jurisdiction determine the available variables and reporting thresholds.

Separate crash-level, vehicle-level, and person-level information. One crash can contain several vehicles and people. Counting every person record as a separate crash inflates frequency. Likewise, the most severe injury in a crash does not describe every occupant. Use stable identifiers and appropriate joins to prevent duplication.

Police records are not a complete census of every collision. Minor events may be unreported, reporting thresholds vary, locations can be imprecise, and investigators may have limited information about injury or distraction. These limitations do not make the records useless; they determine which comparisons and conclusions are defensible.

National data and research systems

The U.S. Fatality Analysis Reporting System (FARS) is a census of qualifying fatal motor-vehicle traffic crashes. The event must involve a motor vehicle in transport on a trafficway customarily open to the public and a death within 30 days from crash injuries. It does not count every injury crash, and immediate on-scene death is not required. NHTSA FARS

The Crash Report Sampling System (CRSS) uses a sample of police-reported crashes for national estimates. A sampled estimate needs weights and uncertainty; it is not a direct count of all crashes at a local intersection. Choose a system according to its purpose rather than assuming any national database can answer every local question.

FHWA's Highway Safety Information System (HSIS) provides research data from participating jurisdictions with linkable crash, roadway, and traffic information. Its value is structured research and analysis, not universal coverage of every U.S. road. Review each dataset's years, variables, and availability before using it. FHWA HSIS

For Canada, Transport Canada's National Collision Database (NCDB) compiles police-reported motor-vehicle collisions on public roads. Published tables and open extracts can differ in available variables and scope. U.S. and Canadian classifications, units, and reporting practices need reconciliation before comparison. Transport Canada statistics and data

Roadway and traffic information

Roadway inventory describes segments, intersections, and interchange or ramp features. Typical analytical information includes facility type, lanes, median, control, geometry, roadside characteristics, speed environment, and access. Traffic data can include annual daily volumes, turning movements, speeds, pedestrian and bicycle counts, and heavy-vehicle proportions.

The Model Inventory of Roadway Elements (MIRE) supplies standardized roadway-element definitions. MIRE Fundamental Data Elements (FDE) are a specified subset for HSIP safety analysis. Required subsets differ among non-local paved, local paved, and unpaved roads. Do not call every useful inventory element a mandatory FDE or assume the same set is required for every road. FHWA MIRE

Keep inventory and exposure time-consistent. Current lane configuration may differ from the configuration when a crash occurred. A volume observed after a major development cannot automatically represent earlier years. Versioning helps prevent a model or diagnosis from mixing incompatible conditions.

Health, enforcement, and implementation information

EMS, hospital discharge, trauma, and vital records can clarify injuries and response, but serve different populations and purposes. Clinical diagnosis differs from roadside suspected injury classification. Linkage requires appropriate authority, privacy controls, and validation; an apparent match is not automatically exact.

Citation and adjudication data describe enforcement and legal processing. More citations may reflect more enforcement rather than more violations. Driver and vehicle records can describe licensure, registration, or relevant characteristics, subject to access limits. Interpret those administrative measures within their collection purpose.

Municipal maintenance, complaints, project records, and program logs help describe conditions and implementation. They can reveal a failed light, changing vegetation, or delayed treatment maintenance. Complaints are selective and should supplement rather than replace systematic analysis.

Match the source with its purpose

• Crash records describe events, vehicles, and people.
• Inventory and traffic records describe conditions and exposure.
• Health records clarify injury and care outcomes.
• Implementation records describe the intervention actually delivered.

Choose and document the combination

For a suspected nighttime crossing problem, combine crash narratives and coded time with lighting inventory, field visibility, speeds, pedestrian exposure, and transit activity. If injuries seem underrepresented, consider validated health linkage. Specify what each source adds and what it cannot establish.

Document source, definition, time coverage, units, location method, missingness, revisions, and access restrictions. Then explain the uncertainty in the result. The best dataset is not the one with the most rows; it is a suitable combination of information for the question and the people who must make the decision.