Which factors contribute most to the severity of road traffic accidents in the UK

between 2005-14?

Abstract:
In this study, we investigate the contributing factors to the severity of road traffic accidents
in the UK between 2005 and 2014. Given the intricate dynamics of road safety,
understanding these contributing elements is essential for developing strategies to reduce
the frequency and severity of the accidents. This research delves into the multifarious
factors such as the spatial, temporal, and environmental influences. Our objective is to
dissect the various aspects that come into play during traffic incidents, such as speed limits,
road conditions and weather conditions. We aim to bridge a critical knowledge gap by
presenting an exhaustive analysis of the determinants of accident severity, which could
inform policymaking and lead to improved road safety measures. This comprehensive
approach seeks to shed light on the complex interplay of factors that contribute to the
severity of traffic accidents, with the ultimate goal of enhancing public health and safety.

Literature Review:
The meticulous exploration of weather-related road accidents in England and Wales, as
elucidated in Edwards' spatial analysis, lays the fundamental groundwork for understanding
the influences on road traffic accidents (Edwards, 1996). The study delineates a clear
positive correlation between the incidence of weather hazards—such as rain and high
winds—and the proportion of accidents, underscoring the pertinence of weather as a
contributing factor. This relationship persists even when accounting for regional disparities in
road network quality and demographics, suggesting that weather conditions significantly
sway the fatality of accidents. However, Edwards' study also emphasises the complexity of
accident causation, acknowledging that weather is but one of a myriad factor that may
precipitate a road traffic incident. The analysis indicates that while the majority of accidents
occur during non-hazardous weather conditions, adverse weather can exacerbate existing
risks. The research further posits that habitual exposure to certain weather conditions may
enable drivers to adapt and mitigate the associated risks through a learning process.
Notably, the study calls for expanded research to dissect the interplay of weather with other
critical parameters like vehicle type, road category, driver demographics, and driving
behaviours (af Wåhlberg et al., 2015). It intimates the potential for more nuanced insights
into accident causation by examining these factors in conjunction, particularly given that
accidents in fog and snow, though highly seasonal, form a minimal part of the overall
accident statistics. Building upon Edwards' seminal work, subsequent research must
endeavour to explicate the intricate dynamics of these contributory elements. In my
investigation I draw strength upon the utility of spatial mapping in correlating accident
severity with urban planning features. Furthermore, by integrating clustering methods I aim
to uncover spatial-temporal patterns that the single-method studies may overlook.

1
Methodology:

The statistical data was extracted from the DfT website which has information about the
road traffic statistics collected over the last 23 years, across 45,865 manual count points
(Department for Transport, 2022).
Sub-sampling: To address the dataset's size, a sub-sampling method was employed for some
analysis, reducing it from over a million rows to around 200,000 occurrences, aiming for
approximately 22,000 samples for each year. This approach allowed for a more manageable
dataset size while retaining representative samples.
Data Wrangling: Data pre-processing involved tasks such as creating dummy variables for
categorical data and handling missing values. Certain columns were undummified post-
clustering for improved interpretability.
Chi-Squared Testing: Chi-squared tests were employed to assess the representativeness of
the subset obtained through the sampling strategy. These tests provided insights into
whether the proportions of samples from different year-month groups were significantly
different.
Visualisation and Spatial Mapping: Visualisations, including scatter plots and density maps,
were instrumental in identifying accident concentrations. Spatial information was mapped
onto a UK map, revealing accident hotspots in cities and regions, particularly in urban
environments and during daylight hours away from junctions.
Clustering with DBSCAN: DBSCAN, a density-based clustering algorithm that is well-suited
for identifying spatial clusters was utilised to uncover patterns within the data. By
configuring DBSCAN with an epsilon of 0.8 and a minimum sample size of 8, clusters were
formed for each year. This enabled the identification of common characteristics among road
traffic accidents during specific time periods. Spatial correlations were explored through hot
spot analysis, involving the calculation of spatial weights and Moran's I statistics. This
analysis helped identify significant clusters and spots, categorizing areas as having either a
"Low" or "High" incidence of accidents.

Large datasets are computationally challenging which is made manageable vis-à-vis the
above. In summary, these analytical techniques are most appropriate for road traffic
accident data because they address the challenges posed by the dataset's size, complexity,
and spatial nature.

Presentation and Analysis of Data

Figure 1 reveals a significant variability in road numbers and classes. This suggests a possible
correlation between diverse road types and accident risks, highlighting the need for tailored
road safety measures across different road environments. As shown in Figure 2, with 30mph
as the most prevalent speed limit where accidents occur, it's likely that urban areas, typically
subjected to this speed limit, are more prone to accidents. This finding challenges the focus
of road safety initiatives, steering them towards urban settings.

2
Figure 1: Heatmap of Co-efficient of Variation Figure 2: Pie Chart of Speed Limits

Impact of existing highways infrastructure

Figure 3 shows "non-junction pedestrian crossing", with the count exceeding 1 million when
compared with "zebra crossing" which is indicative of the safety measures associated with
the measure. In figure 4, the ordinal “5” represents dry conditions being the most frequent
for road traffic accidents. The relationship between this and behaviour of drivers in normal
conditions can be explored to assess the casual nature of this link.

Figure 3: Pedestrian Crossing Physical Facilities Figure 4: Road Surface Conditions

Effects of environmental factors

Figure 5 illustrates the prevalence of road traffic accidents in seemingly “fine without high
winds” conditions hinting there are other more revealing factors. Figure 6 also contrary to
the prevailing assumption, shows that most accidents occur during daylight hours with
streetlights present (Raynham et al., 2019). This suggests a possible underestimation of risks
during these seemingly safe conditions. This could inform future road safety campaigns and
driver education programs.

3
Figure 5: Weather Conditions Figure 6: Light Conditions

Comparing number of vehicles and casualties

The box plots for traffic accidents show that most incidents involve a low number of vehicles
and result in a small number of casualties, with the median values indicating that the most
common occurrence is one vehicle involved and few casualties. Notably, there are outliers in
both plots, which denote that while uncommon, there are significant instances where the
number of vehicles or casualties is substantially higher than the average accident.

Figure 7: Boxplots for number of vehicles and the number of casualties

Spatial Analysis

The data analysis reveals distinct patterns in road accidents across the United Kingdom.
Major cities such as London, Liverpool, the Midlands, Newcastle, Middlesbrough, Glasgow,
Edinburgh, and Cardiff are focal points for accidents, suggesting that urban areas witness a
higher incidence. Specifically, the central and southern regions of the UK exhibit
concentrated accident rates, reflective of increased population density and traffic flow.
Notably, urban centres display more pronounced data clusters, aligning with the correlation
between population density and accidents. Furthermore, a noteworthy observation is the
significant decline in accident, signalling potential improvements in road safety measures
and regulations.

4
 Figure 8: Scatter plot for year 2005 (left) and 2014 (right)

Clustering Analysis:
1. Cluster Dominance Shifts: The summary data clearly illustrates that the most
prominent clusters change from year to year. This dynamic pattern suggests that
accident patterns are not static but evolve over time.
2. Junction Control Influence: The choice of junction control appears to have a
significant impact on cluster formation. Some years exhibit a prevalence of "Give way
or uncontrolled" junctions, while others favour "No Junction." This indicates that the
type of junction control is a factor in cluster differentiation.
3. Road Class and Speed Limits: The summary data emphasizes that road class and
speed limits play crucial roles in cluster differentiation. While certain clusters share
similar road class characteristics, disparities in speed limits contribute to the
formation of distinct clusters. Cluster 1.0 commonly features 30mph zones as
expected from the data.
4. Urban or Rural Areas: The data consistently points to accidents occurring in both
urban and rural areas, affirming that accidents are not confined to a single
environment. However, this variable does not seem to be the primary driver of
cluster differentiation.
5. Number of Vehicles and Casualties: Some clusters consistently involve fewer vehicles
and fewer casualties across multiple years. For instance, Cluster 1.0 often represents
cases with reduced vehicles and casualties, indicating a recurring pattern of less
severe accidents (i.e., less fatal) in those clusters.
6. Weather Conditions and Darkness Presence: Weather conditions and the presence
of darkness are not prominent factors in cluster formation. These variables
consistently show values of 0 across the years, suggesting that they may have limited
influence on accident clustering.
7. Yearly Variations: Each year exhibits unique cluster compositions. This underscores
the dynamic nature of accident clustering over time. The variations strongly suggest
that external factors such as changes in traffic regulations, alterations in road
infrastructure, or fluctuations in weather conditions significantly influence accident
patterns.

5
Discussion:

The evaluation of road traffic accidents through clustering and spatial mapping has yielded
noteworthy insights into their underlying factors. The clustering algorithm, DBSCAN,
segregated the data into discernible patterns, revealing commonalities in accident
characteristics. This method’s strength lies in its ability to handle large datasets and identify
outliers, crucial for understanding accident dynamics. Significantly, clusters often revealed
accidents with recurring themes, such as the number of vehicles involved and the road
conditions at the time of the accident. These results align with studies that highlight the
prevalence of such factors in influencing accident severity (Masuri et al., 2012). The spatial
mapping provided a geographical perspective, uncovering hotspots in urban centres like
London and Birmingham (Wang et al., 2009). This spatial concentration of accidents mirrors
findings from urban studies that pinpoint high-density areas as critical points for road safety
interventions (Hashimoto et al., 2016). Moreover, the hot spot analysis further stratified the
risk areas, identifying locations with a notably high frequency of vehicles involved in
accidents. Such areas can be the focal point for deploying targeted road safety measures.
The implications suggest that certain urban areas and road conditions are more prone to
severe accidents, supporting the notion that road safety efforts should be highly
contextualized.

Comparing these results with existing literature reveals both consistencies and deviations.
Similar to previous research, this study confirms that certain road types and weather
conditions are significant contributors to accident severity. However, by leveraging clustering
and spatial analysis, this study offers a more granular understanding of these factors. It
moves beyond the broader strokes of statistical correlations, presenting a spatial-temporal
picture that can be pivotal for real-world applications. In conclusion, the analytical journey
underscores the intricate interplay of various factors contributing to road traffic accidents.
This integrated approach not only enriches the current literature but also furnishes practical
insights for road safety strategies, emphasising the need for localised and data-driven
interventions.

6
Conclusion and Recommendations

The analysis of road traffic accidents from 2005 to 2014 reveals several notable trends and
characteristics. Firstly, specific types of road surface conditions and pedestrian facilities
emerge as significantly more prevalent in these incidents. Over the years, there has been a
discernible decrease in the total number of accidents, indicating potential improvements in
road safety or effective traffic management. When examining the distribution of accidents
throughout the week, the data shows a fairly uniform spread across different days, albeit
with minor variations. In terms of speed zones, the majority of accidents occur in areas
designated as 30 mph zones (Mountain et al., 2005). This is followed by a lesser but still
notable number of incidents in 60 mph and 40 mph zones. The light condition during most
accidents is typically daylight with street lighting available, suggesting that visibility factors
play a less significant role in these occurrences. Finally, the prevailing weather condition
during the majority of accidents is fine weather without high winds, challenging the
assumption that adverse weather is a primary contributor to road accidents.

The research findings in traffic safety have significant implications for policy and practice. To
reduce accidents, it is crucial to prioritise safer road design by widening roads and installing
safety barriers. Speed control measures, such as lowering speed limits and using speed
cameras, are also essential. Education and awareness programmes targeting all road users
can promote safe driving habits and deter reckless behaviour. Strict enforcement of existing
road safety laws, including seatbelt requirements and penalties for impaired driving, must
be a top priority. Additionally, investing in research for innovative solutions, like advancing
driver assistance technology and enhancing vehicle safety features, is necessary to further
enhance road safety efforts. In summary, a comprehensive road safety policy should
encompass safer road design, speed control, education, enforcement, and research to create
a safer and more secure road environment.

Strengths:
1. Comprehensive Data Analysis: The study extensively uses data from the UK
government, providing a robust foundation for understanding the factors influencing
the severity of road traffic accidents.
2. Use of Multivariate Analysis: The application of multivariate statistical analysis
enables the examination of intricate relationships among variables, enhancing the
depth and reliability of the findings.
Limitations:
1. Variable Selection: The study assesses the determinants of accident severity using a
limited number of variables, excluding potentially influential factors such driver
behaviour, gender, and age.
2. Statistical Techniques: While the study employs commonly used statistical
techniques, such as clustering, multinomial logistic regression could be attempted.
For this study, the effort to evaluate qualitative features would not suit encoding.

7
References:

Edwards, J.B. (1996) ‘Weather-related Road accidents in England and Wales: A spatial
analysis’, Journal of Transport Geography, 4(3), pp. 201–212. doi:10.1016/0966-
6923(96)00006-3.

Hassan, Bhasin. (2023). QM: Quantitative Methods Repository. [GitHub].

[Link]

Department for Transport (2022) Road traffic statistics, Map Road traffic statistics - Road
traffic statistics. Available at: [Link]
4.340/basemap-regions-countpoints (Accessed: 2 January 2024).

af Wåhlberg, A.E., Barraclough, P. and Freeman, J. (2015) ‘The driver behaviour

questionnaire as accident predictor; a methodological re-meta-analysis’, Journal of
Safety Research, 55, pp. 185–212. doi:10.1016/[Link].2015.08.003.

Raynham, P. et al. (2019) ‘The role of lighting in Road Traffic Collisions’, Lighting Research
& Technology, 52(4), pp. 485–494. doi:10.1177/1477153519870857.

Wang, C., Quddus, M.A. and Ison, S.G. (2009) ‘Impact of traffic congestion on road
accidents: A spatial analysis of the M25 motorway in England’, Accident Analysis
& Prevention, 41(4), pp. 798–808. doi:10.1016/[Link].2009.04.002.

Hashimoto, S. et al. (2016) ‘Development and application of traffic accident density

estimation models using kernel density estimation’, Journal of Traffic and
Transportation Engineering (English Edition), 3(3), pp. 262–270.
doi:10.1016/[Link].2016.01.005.

Mountain, L.J., Hirst, W.M. and Maher, M.J. (2005) ‘Are speed enforcement cameras more
effective than other speed management measures?’, Accident Analysis &
Prevention, 37(4), pp. 742–754. doi:10.1016/[Link].2005.03.017.