Multivariate time series analysis of traffic congestion measures in urban areas as they relate to socioeconomic indicators

Abstract Traffic congestion has significant adverse implications for the environment and economy. Many state and local transportation agencies have implemented traffic congestion management practices to alleviate the negative implications of urban traffic. One of the major drawbacks of traffic congestion management practices is that they do not account for socio-demographic and economic factors, which have a significant impact on traffic congestion. Understanding the influence of these factors is very crucial because they can help to communicate the system's performance management and target setting. Only a few studies analyzed the relationship between traffic conditions (e.g., traffic demand and vehicular traveling speed) with a limited number of socio-economic factors. Moreover, most of the existing models ignore the temporal and spatial autocorrelations of traffic congestion, which may significantly limit their reliability and effectiveness. This study is developed with the purpose of identifying the most relevant external factors that affect traffic congestion performance measures. To conduct the research, we have used three urban congestion performance measures collected from 51 metropolitan areas across the U.S. over a four-year period, 2013–2016: travel time index, planning time index, and congested hours. We have used multivariate time series models to account for the complex inter-relationships among the performance measures and socioeconomic factors to identify the most influential factors affecting system performance. We have finally developed predictive models to estimate the traffic congestion measures using these factors. The results of rigorous modeling show that the factors influencing the traffic congestion measures are monthly average daily traffic (MADT), the number of employed, rental vacancy rate, building permits, fuel price index, and Economic Conditions Index (ECI). The prediction models indicated that the effects of these factors are statistically significant and could be used to forecast future trends in three performance measures accurately.