Using Near-Miss Clusters to Profile Safety Risk

Kent County Council (KCC) wanted to use near-miss data to proactively identify road safety risk areas on their road network. Specifically, they were interested in harsh braking and harsh swerving events, focusing on the most 'extreme' events - near-misses with high g-force ratings at high speeds.

Context for the Programme

KCC's annual Crash Remedial Programme (CRM) has 3 work streams:

Cluster Site Programme - uses 3 years of collision data to identify clusters
Junction Programme - junction-specific programme that uses 10 years of collision data
Route Programme - segmented by road section on longer routes using 5 years of collision data.

The current challenge for KCC is that the basis for all 3 of these streams is reactive analysis based on collision datasets; a crash has to occur in order for data to be collected about road safety risk at that location. the benefit of including near-misses into the CRM was it provided a proactive and risk-focused approach.

Applying Connected Vehicle Data

The first stage of the programme involved a Compass Data Science dashboard of near-misses across the County area. The dashboard identified locations where there was a minimum of 3 harsh braking and/or harsh swerving events within a 62 metre radius. This radius size was used because it is the average size of Kent's collision cluster events.

KCC wanted to prioritise these clusters based on the highest average speeds. They also wanted to identify loss of control and high-speed (>25mph) events that were likely to result in fatal or serious injuries and head-on collisions above certain survivability thresholds and focus on these locations. To do this, the County identified each trip that made up the braking and/or steering event in all the near-miss clusters. They then took the average speed of the vehicles registering these near-miss events to establish an average speed value for each near-miss cluster.

Compass also provided a normalisation value to determine what proportion of vehicles travelling through that cluster area experienced a harsh braking or swerving event. This helped the County to prioritise a list of sites by allowing KCC to filter by those with higher rates of near-misses per vehicle travelling.

Findings

KCC shortlisted 16 sites. They then did a sweep through sites they were already aware of and then focused their attention on new sites that they didn't know about or weren't already being investigated. They cross-compared these sites with their Highway Improvement Plans to see if any of these sites had notable community concerns. Each site was then categorised based on its risk level and whether it had a clear cause.

For the 16 sites:

6 were identified as high risk, having a clear cause that could be mitigated with remedial action
4 were identified as high risk, but unclear cause, requiring further data collection to identify underlying risk and cause.
3 were identified as having a clear cause, but the mitigation was not justified for the level of risk
3 were identified as having an unclear cause and unclear risk level, requiring more data.

Site Example - Shoreham Rd

Shoreham Rd is a single-carriage way, rural road that as national speed limits. The site has a steep kink that passes under a bridge on a blind corner.

The near-misses at this site were clustered on the corner where the bridge was, for vehicles in both directions. There were quite high g-forces and hard left and right braking. This location also had a considerable collision history, with 8 collisions between 2014 - 2023: 5 slights, 3 serious, and 1 fatal. When the fatality occured, there was a review and improvements made at this location

Looking at the 2024 data for the same site, KCC could see that the number of events and the severity of those events had reduced, indicating that interventions were having an effect. That being said, the iRAP assessment for this location was considered quite poor, mainly due to a combination of speeds, run-off space, the corner bend, and the bridge.

Site Example - A226 Rochester Rd

This site was identified in the top-left quadrant of the risk and cause matrix: relatively high risk but without a clear cause. This location was a 50mph zone, single carriage way with a long island in the middle. It's a residential junction and some property access points. The site has never appeared in KCCs previous work because it has a historically low number of collisions

The near-miss data displayed a few specific patterns:

All of the near-misses were heavy vehicles
A majority happened around Summer months
They are all braking-related

Following more investigation, it wasn't clear what caused the near-misses. KCC thought it could be due to a number of criteria, such as a right turn junction, the properties, drivers mistaking the turning lane as an overtaking lane, a speed camera warning sign and a blind bend. There are ongoing observations and data analysis at this site to try and determine the cause/s of the near-misses.

Lessons Learned

There were 6 key lessons:

It was a valuable process to better understand the data and identify risk
It has helped shift thinking away from outcomes (crashes) to cause (risks)
Running the analysis for a longer period (>9 months) with a higher tolerance threshold
Ground truth or cross-referring to other datasets is essential
Maintenance and engagement with maintenance is critical such as worn signage or line markings.
Thresholds for inclusion on review can be flexed based on existing works programmes