Exploring weather-related factors affecting the duration of multiple congestion levels caused by traffic incidents using a multivariate joint frailty survival model

• A multivariate joint survival model is developed in this study. • The findings highlight the temporal correlations among different congestion levels. • How the transitions between varying congestion states influence the likelihood and timing of severe congestion events. • Key risk factors are identified and corresponding countermeasures are proposed. Hazard-based duration models have gained popularity in predicting traffic incident durations. However, most studies analyze congestion duration as a whole, overlooking the varying levels of congestion (minor, moderate, and severe), which may be interrelated and influenced by different factors. This study proposes multivariate joint survival analysis models to examine the relationships across these congestion levels using traffic incident data from New York State (2017–2019), treating minor and moderate congestion as recurrent events, with severe congestion as a terminal event. By incorporating a frailty term, unobserved heterogeneity among road segments is accounted for. The results show that real-time weather factors, such as temperature, wind speed, visibility, and precipitation (rain/snowfall), exhibit varying effects on the duration of different congestion levels, with these effects fluctuating over time. For example, in 2017–2019, low temperatures increase the duration of minor congestion by 40.88 %, 26.66 %, and 52.69 %, respectively. Conversely, for severe congestion, low temperatures also show stable temporal effects but reduce congestion duration by 70.81 %, 60.07 %, and 70.81 %, respectively. Rainy weather increases the duration of moderate congestion by 54.10 %, 31.94 %, and 54.10 %, respectively, while snowy weather reduces it by 41.38 %, 37.19 %, and 27.48 %. More importantly, a significant correlation is found between minor or moderate congestion, which are recurrent events, and severe congestion, the terminal event. Furthermore, a positive correlation between minor and moderate congestion suggests that unobserved factors jointly influence the duration of both. The study confirms the superiority of the proposed joint model for analyzing traffic incident duration and provides practical insights for transportation policymakers to massively ease congestion more effectively.