Site-Specific and Systemic Network Screening

Site-Specific and Systemic Network Screening

Establish the focus and population

Network screening identifies locations or groups for potential safety investment. Start by specifying the outcome, road population, period, and purpose. Fatal and serious injury crashes, pedestrian conflicts, or rural departures can require different measures and data. Group comparable facilities before interpreting rankings.

FHWA screening guidance describes establishing a focus, identifying facilities, selecting measures and a method, then screening and evaluating results. These decisions should be documented so the ranking can be reviewed and repeated.

Site-specific screening

Site-specific approaches assess particular intersections, segments, or corridors. Measures can include frequency, rates, severity-weighted scores, predicted frequency, EB-expected frequency, or excess expected performance relative to a suitable peer baseline. Each uses different inputs and emphasizes different aspects of harm or unusual performance.

Frequency is easy to understand and reflects burden, but tends to emphasize sites with much exposure. Rates normalize a selected exposure while becoming unstable at small denominators and potentially misadjusting nonlinear relationships. Severity weighting represents consequences but can be dominated by rare events. Model-based measures require suitable models, exposure, definitions, and expertise.

EB can reduce RTM bias from unusually high or low observations, but does not reveal the physical cause or guarantee treatable benefit. A positive excess measure supports further diagnosis. A site with low or negative excess can still warrant a systemic treatment or an accessibility improvement. The measure should support the objective rather than become a universal rule for denying action.

Systemic screening

Systemic analysis identifies risk features associated with a targeted severe outcome across a network. It can be valuable where serious events are dispersed and unlikely to recur at the same location in a short period. Examples include curve, roadside, crossing, or speed characteristics associated with the selected crash type.

Use crash information to identify relevant outcomes and associated features, inventory the features across the eligible network, assess candidate risk profiles, and select suitable treatments and priorities. The absence of a recent crash cluster at one site does not establish absence of risk. Conversely, one feature does not guarantee a crash or automatically justify every proposed treatment.

Systemic concerns a network risk mechanism. Systematic can describe a consistent procedure or deployment practice. They are not identical terms. A systematic review can be applied to either a site-specific or systemic analysis. FHWA systemic approach

Compare data and resource needs

A site-frequency screen needs reliable locations and event definitions. Rate comparisons also need exposure and compatible facility groups. An EB screen needs appropriate SPFs and model parameters. Systemic analysis needs a usable inventory of relevant features and evidence connecting them with the targeted harm. No method removes the need for quality checks.

Resource requirements include staff skills, tools, inventory maintenance, data management, field verification, and interpretation. Choose a defensible method the agency can sustain. A sophisticated model with unsupported inputs can be less useful than a transparent simpler screen followed by diagnosis and a plan to improve data.

Segment corridors and avoid boundary artifacts

Segment definitions should reflect meaningful changes in geometry, operation, traffic, or context. A very long segment can dilute a local pattern; very short segments can create sparse counts and unstable results. Document how intersection crashes, ramps, and overlapping features are assigned.

A sliding window moves a defined length along a corridor in specified increments, calculating the measure for each position. It can reveal patterns obscured by fixed boundaries. Overlapping windows contain shared crashes and are not independent observations; avoid counting each flagged window as a separate distinct problem or benefit opportunity.

Peer grouping and consistent assignment matter as much as software output. Review whether a flagged location reflects a real pattern, a boundary choice, duplicate records, or geocoding error. Field and narrative review remain necessary after the ranking.

Work a site-specific versus systemic choice

A city has repeated severe angle crashes at several signalized intersections with detailed data. A site-based screen can identify where performance and burden warrant diagnosis. A county has severe departures scattered across many curves, with few repeated events at any single curve. A systemic risk-feature approach can identify groups for proactive treatment.

Both agencies can use the complementary method. The city can review risk profiles beyond its current HIN, and the county can investigate a pronounced cluster. Do not declare one approach universally superior or limit systemic work to rural roads. The distribution of the targeted outcome and data needs guide the choice.

Choose the screening question

• Site-specific: where does performance warrant focused diagnosis?
• Systemic: where are applicable risk features present across the network?
• Measures: what data, units, uncertainty, and resources support the choice?

Interpret and communicate the list

Screening produces a starting point for investigation and program development. Report the measure, period, units, population, assumptions, and uncertainty. Show whether rankings change under a different reasonable measure or severity focus. Explain that ranks do not establish blame or an exact number of preventable crashes.

For a nontechnical audience, distinguish the burden question from the unusual-performance question. “This location has many serious events” differs from “This location performs above comparable-site expectations.” Both can matter, but they lead to different analyses. The next step is to verify the pattern, diagnose plausible mechanisms, and assess feasible treatments with applicable evidence.