Human behaviour is often blamed for serious road crashes, but its impact is rarely uniform. New research suggests that risky driving can have very different consequences depending on weather, road gradient, traffic conditions and vehicle type.

Published in Accident Analysis & Prevention, the study analysed crash reports from Chuxiong Prefecture in China’s Yunnan Province between 2017 and 2022.

Rather than looking only for statistical associations, the researchers used a counterfactual machine-learning approach to estimate how specific human factors may affect crash severity under different conditions.

Moving Beyond Simple Correlations

Traditional road-safety studies often compare crashes involving behaviours such as distraction, fatigue or unsafe following distances with other crashes. The problem is that weather, road geometry, vehicle type and traffic volume can influence both driver behaviour and the eventual outcome.

The research team addressed this by developing a dynamic weighted double machine learning framework. The method combined LightGBM and XGBoost models and adjusted for multiple potential confounding factors.

Crashes attributed mainly to non-human causes, including mechanical failures or unavoidable environmental events, were used as the control group. This allowed the researchers to estimate how crash severity might differ when particular human behaviours were present, while accounting for other conditions.

Five Human Factors Were Examined

The analysis focused on five risky-driving factors and assessed how their estimated effects changed across road and traffic conditions.

One important result concerned inadequate safety distance. The researchers found that following too closely often appeared alongside other dangerous circumstances, increasing the likelihood of property-damage-only and injury crashes. More than half of cases involving inadequate spacing were associated with a higher likelihood of these outcomes, although the behaviour played a smaller role in fatal crashes.

This suggests that unsafe following distance may often amplify existing danger rather than determine the most severe outcome on its own.

Weather Changes Some Risks

Reversing behaviour showed particularly strong sensitivity to weather conditions. Its estimated effect on crash severity varied depending on the surrounding environment, highlighting why the same manoeuvre may carry different levels of risk in rain, fog or other difficult conditions.

Distracted or fatigued driving and travelling in the wrong lane were also influenced by wider traffic and road characteristics. Daily truck traffic and road gradient affected the estimated relationship between these behaviours and crash severity.

Mountain roads create a particularly demanding environment because sharp curves, steep gradients and reduced visibility can leave drivers with less time and space to correct mistakes.

Heavy Vehicles Add Complexity

The study also found an interaction between road gradient and vehicle type in crashes involving overloaded vehicles.

Overloading on steep sections may therefore carry different consequences from the same behaviour on flatter roads. The researchers argue that enforcement against overloading and reversing should be strengthened on steep sections carrying large volumes of truck traffic.

They also suggest restricting heavy trucks to right-hand lanes on particularly high-risk motorway sections.

Why The Findings Matter

The main message is that “human error” is too broad a category to explain crash severity on its own. A dangerous behaviour can become more or less serious depending on the road, weather, surrounding traffic and vehicle involved.

For road authorities, that means safety measures may work better when targeted at specific combinations of risk rather than applied uniformly across an entire motorway network.

The modelling approach also has limitations. Counterfactual machine learning can strengthen causal analysis, but its conclusions still depend on the quality of crash reports, how causes are classified and whether important confounding factors have been measured.

For drivers, the practical lesson is simpler: risky behaviours become especially dangerous when several unfavourable conditions coincide. Maintaining adequate distance, avoiding distraction and respecting vehicle and lane restrictions may matter even more on steep, busy or weather-affected mountain roads.