Everyone defines one. No one draws them together.
Search the literature and you find each concept explained on its own island. KiwiRAP pairs collective and personal risk. Austroads documents the Star Rating Score. A separate Austroads manual covers the Infrastructure Risk Rating. FHWA explains the systemic approach. The closest thing to a unified treatment is a dense government prose report — no single figure places all of them on shared axes.
That gap matters, because the confusion is practical: people read “collective risk” and “systemic risk” as the same data processed two ways, and they assume the Infrastructure Risk Rating sits on top of the AusRAP star rating. Neither is true. Below is the corrected map.
Scope: this page covers road & network risk
It stays deliberately at the level of roads and corridors — how crash risk is measured along a road and across a network. It does not cover vehicle-level safety, which is a complementary lens: how well a vehicle protects its own occupants (crashworthiness) and the harm it can impose on others (aggressivity), as captured in the Used Car Safety Ratings developed at MUARC. Road risk, vehicle risk and road-user behaviour sit at different levels of the Safe System and work together. With thanks to Emeritus Professor Max Cameron for the prompt.
Two lanes, one destination
Every metric on this page exists to prevent fatal and serious injury (FSI). They split into a reactive lane that scores crashes that already happened, and a proactive lane that scores the risk built into the road itself. Both, in practice, rank corridors — and the systemic method, shown beneath, draws on both.
*Shown FSI-based: Collective and Personal Risk here use fatal-and-serious-injury (FSI) crashes. Under the FSI-Equivalent methodology you also incorporate minor-injury crashes — each weighted by a severity index — to estimate the expected FSI burden.
What feeds each one
Score the crashes that happened
Built entirely from historic crash records. It can only see danger after it has occurred.
- Collective risk — crash density: fatal-and-serious-injury crashes per km per year (or, under the FSI-Equivalent method, all injury crashes weighted by a severity index). Ranked into corridors.
- Personal (individual) risk — crash rate: FSI per vehicle-kilometre travelled. The exposure-adjusted, traveller’s-eye companion.
Strength: pinpoints proven trouble. Blind spot: a dangerous road with no crash yet stays invisible.
Score the risk built into the road
Built from a coded survey of the road’s own attributes — no crash history required. It can flag a dangerous road before anyone is hurt.
- Star Rating Score (SRS) → 1–5 stars. The detailed iRAP/AusRAP model, ~78 attributes per 100 m.
- Infrastructure Risk Rating (IRR). A lighter NZ-origin model, ~9 attributes, its own log-product equation — a sibling of SRS, not built on top of it.
- Systemic screen. The infrastructure-screen phase of the broader systemic method — rank candidate sites by SRS/IRR, after crashes are diagnosed by pattern.
Strength: finds latent risk early. This is where “systemic” lives.
Collective risk vs systemic risk
They are not the same dataset processed two ways. They sit in different lanes and answer different questions — yet both rank corridors.
| Collective risk | Systemic risk | |
|---|---|---|
| Data source | Historic crashes (FSI + minor injury) | Crash patterns (movement & place, type) + road attributes |
| Stance | Reactive — scores what happened | Proactive — diagnose risk factors, screen network-wide |
| Weighting | Severity Index per record, summed (FSI-Equivalent variant) | Risk-factor patterns; SRS / IRR screen |
| What it surfaces | Corridors where trauma has concentrated | Candidate corridors sharing severe-crash risk factors — some with little crash history yet |
| Output | FSI per km per year (or Estimated FSI-Equivalent), ranked into corridors | A prioritised screen: candidates ranked by SRS / IRR |
The punchline a crash-based view misses: on a real network safety plan, the worst-ranked corridor on the systemic page can carry zero FSI crashes — flagged purely because its star rating score says the road is built for harm. Collective risk would never surface it.
Actual FSI vs FSI-Equivalent
Collective Risk is a crash-density measure — the annual fatal-and-serious-injury (FSI) burden on a corridor, divided by its length. But “FSI burden” hides a choice most analysts make without noticing: there are three different numerators, and they answer different questions. Never write a bare “FSI” — always name which one.
| Numerator | What it counts | Collective Risk |
|---|---|---|
| Actual FSI crashes | crashes with at least one fatal or serious injury | FSI crashes ÷ (N yrs × km) |
| Actual FSI casualties | fatalities + serious injuries (people) | FSI casualties ÷ (N yrs × km) |
| Estimated FSI-Equivalent | expected FSI inferred from all injury crashes via severity indices | Σ e_i ÷ (N yrs × km) |
Actual FSI is observed trauma — use it for transparent historical reporting, public communication and evaluation. Estimated FSI-Equivalent is a prediction, not a count: it draws on the far larger pool of all injury crashes and replaces the randomness of whether a crash happened to be fatal/serious with an average severity for similar crashes. That makes it often more stable on sparse corridors and better for prioritisation — but only after calibration/validation, and it must always be labelled Estimated. Crash-density (FSI crashes per km per year) is the iRAP/KiwiRAP risk-mapping tradition; casualty-density better reflects trauma where one crash injures many (bus, pedestrian group, motorcycle pillion).
The FSI-Equivalent formula
where e_i = the expected FSI value (severity index) for crash i
Each crash’s severity index e_i is driven by crash characteristics — typically urban/rural & speed environment · midblock vs intersection · intersection control · road-user group (vehicle, motorcycle, pedestrian, cyclist) · crash/movement type · and a speed-limit scaling factor where supported. These specific factors are model-specific (for example NZ’s DSI-equivalents) — not a universal table.
Personal (individual) risk — the exposure-normalised companion
Collective Risk rises with traffic volume: a busy arterial scores high simply because more people are exposed. Personal Risk divides the burden by exposure, answering “how risky is this road per trip?”
= FSI per 100 million vehicle-kilometres travelled (VKT)
A high-volume arterial can carry high collective but moderate personal risk; a quiet rural road can carry low collective but high personal risk. Report them side by side — one finds where the most trauma sits, the other finds where each driver is most exposed.
What “systemic risk” actually means
Systemic risk is the term most often misread — and it is not a single metric, nor the same thing as an AusRAP star rating. It is a broader, agency-defined safety method for finding and treating risk proactively. Two of its phases are the ones a dashboard usually shows:
Diagnose reads crash records, but groups them by type and context — not by location — to find the recurring focus crash types and the road features behind them. Screen then finds the candidate sites across the network sharing those features, using the Star Rating Score and IRR, and prioritises them for treatment without requiring each site to have its own crash history. The Star Rating Score and IRR are an infrastructure screen used inside the method — not the method itself, which continues on into countermeasure selection, prioritisation, delivery and evaluation.
This is why a Network Safety Plan can show what looks like crash analysis under “systemic risk.” A Systemic Risk — Movement & Place view is the diagnosis input (crashes grouped by pattern); the Systemic Risk — Corridor / Intersection views are the infrastructure screen (candidates ranked by Star Rating Score / IRR). They are two faces of one method — and they read as genuinely “systemic” only when the screen is tied back to the diagnosed focus crash types, rather than presented as a bare star-rating output. FHWA’s systemic approach is consistent with this diagnose-then-screen logic — proactive and risk-factor based, not waiting for crashes to cluster — though it doesn’t prescribe these specific Australian tools.
The contrast with collective risk stays the same: it is the unit of analysis. Collective risk ranks one corridor by its own crash density; systemic groups crashes by type and pattern across the network and screens by risk factor. Crash history still informs the diagnosis — systemic isn’t crash-history-free — it simply doesn’t wait for each individual site to accumulate crashes.
Siblings, not a stack
A frequent error is to picture a tower: AusRAP at the base, IRR layered on it, Star Rating on top. The evidence says otherwise. One road-attribute survey feeds two independent models. The only genuine “on top of” is the Star Rating banding sitting on the Star Rating Score.
One boundary to keep clear: the Star Rating is AusRAP’s own infrastructure methodology — collective and personal risk are a separate, complementary Crash Risk Mapping protocol (the iRAP / KiwiRAP tradition), used alongside star ratings, not part of them.
The Star Rating Score bands into stars on the iRAP scale (vehicle-occupant thresholds shown):
Bicyclists and pedestrians use different (higher) bands, because their scores come from different crash-type equations. SRS is computed separately for each of four road-user groups: vehicle occupant, motorcyclist, pedestrian, bicyclist.
The Infrastructure Risk Rating is a parallel model
IRR was developed by Waka Kotahi (NZ) in 2016, inspired by the iRAP star-rating approach but with its own governing equation — IRR = log₁₀(R_SRS × R_SA × R_CW × ((R_LRH+R_RRH)/2) × R_LU × R_ID × R_AS × R_TV) — its own ~9 coded attributes, its own tool, and its own five risk bands (not stars). A trap to avoid: the R_SRS term inside that equation is a road-stereotype risk factor — it is not the AusRAP Star Rating Score. The names collide; the quantities are different. IRR shares a lineage with SRS; it is not derived from it.
A third predictive approach: crash prediction models
Alongside the AusRAP/iRAP Star Rating and the Infrastructure Risk Rating, road safety also uses crash prediction models — Safety Performance Functions (SPFs) that estimate how many crashes a given site type should expect from its traffic volume and road features. Paired with the Empirical Bayes method (which tempers that estimate with the site’s own crash record) and Crash Modification Factors (CMFs, which quantify how much a treatment changes risk), these are the most accurate of the predictive methods — and the most data-hungry.
Reference frameworks include the US Highway Safety Manual (HSM) and the New Zealand Crash Estimation Compendium (CEC). In practice they are reserved for detailed option assessment and economic appraisal on busy routes, while IRR (a short variable list) suits whole-network screening and AusRAP/iRAP suits strategic routes. Added with thanks to Dr Shane Turner, who highlighted it on this page.
Ranking a corridor systemically
At corridor level, a systemic risk view is typically sorted in descending order of an aggregated Star Rating Score — worst-built corridor at the top — with an Infrastructure Risk Rating shown alongside as a second infrastructure measure. Crash counts appear for context, but they do not drive the ranking. That is the whole point of the proactive lane: it orders the network by how the road is built, not by where people have already been hurt — the mirror image of collective risk, which ranks the same corridors by severity-weighted crash history.
One nuance worth keeping straight: the systemic approach begins with crash analysis too — grouping crashes by movement & place and crash type to find the recurring risk factors — before screening the network by infrastructure score. What makes it “systemic” rather than “collective” is the unit of analysis: by crash-type pattern across the network, not by a single location’s crash density.
Building a corridor model — five steps
The metric is only half the job; a defensible ranking follows a repeatable method. The thresholds below are common screening defaults, not mandates — sense-check them against your own network.
- Build clean corridor geometry. Match each crash to a road segment, group segments into homogeneous sections, then into named corridors. Aggregate sections shorter than ≈500 m — short lengths inflate per-km rates.
- Choose your numerator, explicitly. Keep three separate fields — Actual FSI crashes, Actual FSI casualties, Estimated FSI-Equivalent — and never collapse them into one vague “FSI” column.
- Use enough crash history. Around 5 years is the usual compromise: long enough to dampen random noise, short enough to still reflect current conditions. Use more years on low-crash networks, but flag that the road may have changed.
- Apply the FSI-Equivalent value crash-by-crash. Assign each injury crash its own severity index, then sum across the corridor — don’t apply a single average to the corridor as a whole.
- Rank on two axes, not one. Combine absolute burden (total FSI-Equivalent) with rate (FSI-Equivalent per km per year). A short hotspot and a long high-burden corridor are both real priorities; a per-km rate alone hides one of them.
Prioritisation logic
A strong corridor priority usually satisfies at least two of: high FSI-Equivalent per km per year · high total FSI-Equivalent · a real actual-FSI history · high Personal Risk · a poor infrastructure score (AusRAP SRS / IRR / Safe System) · a clear, treatable crash pattern. The last one carries weight — a corridor with high FSI-Equivalent but no treatable pattern is usually a weak investment case, though systemic treatments, speed management, or strong SRS/IRR risk can still justify action where crash history is sparse.
Traps to avoid
- Don’t import severity indices uncalibrated. FSI-equivalents reflect a jurisdiction’s own crash coding and severity reporting; there is no universally transferable production table without harmonised definitions and local calibration.
- Don’t let short corridors dominate. Apply a minimum length (≈500 m) or aggregate short sections before computing per-km rates.
- Don’t use Collective Risk alone. It finds where the most trauma sits, but it is driven by exposure — always pair it with Personal Risk and a proactive infrastructure measure (SRS / IRR / Safe System).
References
- Austroads — AusRAP Road Safety Star Rating (SRS equation, attributes, road-user groups).
- iRAP — Methodology Fact Sheet #7: Star Rating bands (SRS-to-star thresholds).
- Austroads — Infrastructure Risk Rating Manual for Australian Roads (AP-R587A-19).
- Queensland TMR — Infrastructure Risk Rating (IRR) Manual (model equation, eight road/roadside attributes, risk bands — the freely available companion to AP-R587A-19).
- KiwiRAP — Measures of Risk (collective vs personal risk formulas).
- Office of Road Safety (AU) — Local Government Network Risk Assessment Frameworks.
- FHWA — Why a systemic approach (reactive vs proactive).
- Austroads — AusRAP FAQs.
- iRAP — Methodology fact sheets index.
- iRAP — Crash Risk Mapping (crash risk per km and per km travelled).
- Waka Kotahi NZ — Crash Estimation Compendium & Calculating DSI equivalents (severity indices).
- RACQ — AusRAP and road safety (crash risk mapping: collective vs individual risk).
- AASHTO — Highway Safety Manual (HSM) — Safety Performance Functions, the Empirical Bayes method and Crash Modification Factors (the safety-science predictive approach).
- Monash University Accident Research Centre (MUARC) — Used Car Safety Ratings — vehicle-level crashworthiness & aggressivity (scope reference).