Markov switching multinomial logit model: An application to accident-injury severities

School of Civil Engineering, 550 Stadium Mall Drive, Purdue University, West Lafayette, IN 47907, United States
Accident Analysis & Prevention (Impact Factor: 1.87). 07/2009; 41(4):829-838. DOI: 10.1016/j.aap.2009.04.006
Source: PubMed


In this study, two-state Markov switching multinomial logit models are proposed for statistical modeling of accident-injury severities. These models assume Markov switching over time between two unobserved states of roadway safety as a means of accounting for potential unobserved heterogeneity. The states are distinct in the sense that in different states accident-severity outcomes are generated by separate multinomial logit processes. To demonstrate the applicability of the approach, two-state Markov switching multinomial logit models are estimated for severity outcomes of accidents occurring on Indiana roads over a four-year time period. Bayesian inference methods and Markov Chain Monte Carlo (MCMC) simulations are used for model estimation. The estimated Markov switching models result in a superior statistical fit relative to the standard (single-state) multinomial logit models for a number of roadway classes and accident types. It is found that the more frequent state of roadway safety is correlated with better weather conditions and that the less frequent state is correlated with adverse weather conditions.

Download full-text


Available from: Fred L. Mannering, Jul 25, 2014
  • Source
    • "Yu et al. (2015) estimated the influence of weather conditions on mountain freeway crash potential using a correlated random parameter tobit model. Malyshkina and Mannering (2009) modeled unobserved heterogeneity by assuming that the variance between two unobserved roadway safety statuses follows a Markov switching pattern on injury severity. But the disadvantage of random parameters models is that it may not be able to capture the heterogeneity across different data groups, and therefore result in biased estimations. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Traffic crashes occurring on rural roadways induce more severe injuries and fatalities than those in urban areas, especially when there are trucks involved. Truck drivers are found to suffer higher potential of crash injuries compared with other occupational labors. Besides, unobserved heterogeneity in crash data analysis is a critical issue that needs to be carefully addressed. In this study, a hierarchical Bayesian random intercept model decomposing cross-level interaction effects as unobserved heterogeneity is developed to examine the posterior probabilities of truck driver injuries in rural truck-involved crashes. The interaction effects contributing to truck driver injury outcomes are investigated based on two-year rural truck-involved crashes in New Mexico from 2010 to 2011. The analysis results indicate that the cross-level interaction effects play an important role in predicting truck driver injury severities, and the proposed model produces comparable performance with the traditional random intercept model and the mixed logit model even after penalization by high model complexity. It is revealed that factors including road grade, number of vehicles involved in a crash, maximum vehicle damage in a crash, vehicle actions, driver age, seatbelt use, and driver under alcohol or drug influence, as well as a portion of their cross-level interaction effects with other variables are significantly associated with truck driver incapacitating injuries and fatalities. These findings are helpful to understand the respective or joint impacts of these attributes on truck driver injury patterns in rural truck-involved crashes.
    Accident Analysis & Prevention 12/2015; 85:186-198. DOI:10.1016/j.aap.2015.09.005 · 1.87 Impact Factor
  • Source
    • "For example, Malyshkina et al. (2009) estimated a Markov switching model of crash frequency which indicated that the frequency of crashes fundamentally shifted between states, over time, due to unobserved factors. Subsequent work has also shown that the severity of crashes is not stable overtime (Malyshkina and Mannering, 2009; Xiong et al., 2014), with unobserved heterogeneity again suggesting Markov transitioning between multiple states. While the empirical findings from these studies suggest time-varying heterogeneity over relatively short time periods, 1 they do not explicitly address the issue of possible long-term shifts in the effect that specific explanatory variables may have on the frequency and/or severity of crashes. "
    [Show abstract] [Hide abstract]
    ABSTRACT: This study explores the temporal stability of factors affecting driver-injury severities in single-vehicle crashes. Using data for single-vehicle crashes in Chicago, Illinois from a nine-year period from January 1, 2004 to December 31, 2012, separate annual models of driver-injury severities (with possible outcomes of severe injury, minor injury, and no injury) were estimated using a mixed logit model to capture potential unobserved heterogeneity. Likelihood ratio tests were conducted to examine the overall stability of model estimates across time periods and marginal effects of each explanatory variable were also considered to investigate the temporal stability of the effect of individual parameter estimates on injury-severity probabilities. A wide range of variables potentially affecting injury severities was considered including driver-contributing factors, location and time of day, crash-specific factors, driver attributes, roadway characteristics, environmental conditions, and vehicle characteristics. The results indicated that, although data from different years share some common features, the model specifications and estimated parameters are not temporally stable. In addition, complex temporal stability behaviors were observed for individual parameter estimates such as driver gender, apparent physical condition of driver, type of vehicle, vehicle occupancy, road surface, weather, and light conditions. It is speculated that this temporal instability could be a function of the urban nature of the data, possible variations in police-reporting of crash determinants over time, the impact of continuing improvements in vehicle safety features and drivers’ response to them, and/or the effects of macroeconomic instability that was present over the time period considered in this study. Although the source of temporal stability is not clearly known, the general subject of temporal instability warrants substantial attention in future research. The possible presence of temporal instability in injury-severity models can have significant consequences in highway-safety practice where accurate forecasting of the impacts of alternative safety countermeasures is sought.
  • Source
    • "Xie et al. (2012) compared the effectiveness of latent class logit models and MNL models in the assessment of driver injury severities in rural single-vehicle crashes and verified the applicability of them as an alternative of each other in traffic safety analyses. Other discrete choice models, such as mixed logit models, ordered Probit, Markov switching MNL, ordered logit, sequential logit and nested logit models, were also employed to analyze crash injury severities to address different methodological issues associated with the specific datasets (Abdel-Aty and Abdelwahab, 2004; Altwaijri et al., 2011; Anastasopoulos and Mannering, 2011; Das et al., 2008; Golob et al., 2008; Gray et al., 2008; Haleem and Abdel-Aty, 2010; Holdridge et al., 2005; Hu and Donnell, 2011; Malyshkina and Mannering, 2009, 2010; Moore et al., 2011; Park et al., 2010; Savolainen and Mannering, 2007; Wang and Abdel-Aty, 2008; Xie et al., 2007) Savolainen et al. (2011) provided a comprehensive review to summarize these statistical modeling approaches for crash injury severity analyses and shed light on their application restrictions with various data issues. However, most of these discrete choice models are established based on certain assumptions . "
    [Show abstract] [Hide abstract]
    ABSTRACT: Rear-end crash is one of the most common types of traffic crashes in the U.S. A good understanding of its characteristics and contributing factors is of practical importance. Previously, both multinomial Logit models and Bayesian network methods have been used in crash modeling and analysis, respectively, although each of them has its own application restrictions and limitations. In this study, a hybrid approach is developed to combine multinomial logit models and Bayesian network methods for comprehensively analyzing driver injury severities in rear-end crashes based on state-wide crash data collected in New Mexico from 2010 to 2011. A multinomial logit model is developed to investigate and identify significant contributing factors for rear-end crash driver injury severities classified into three categories: no injury, injury, and fatality. Then, the identified significant factors are utilized to establish a Bayesian network to explicitly formulate statistical associations between injury severity outcomes and explanatory attributes, including driver behavior, demographic features, vehicle factors, geometric and environmental characteristics, etc. The test results demonstrate that the proposed hybrid approach performs reasonably well. The Bayesian network reference analyses indicate that the factors including truck-involvement, inferior lighting conditions, windy weather conditions, the number of vehicles involved, etc. could significantly increase driver injury severities in rear-end crashes. The developed methodology and estimation results provide insights for developing effective countermeasures to reduce rear-end crash injury severities and improve traffic system safety performance. Copyright © 2015 Elsevier Ltd. All rights reserved.
    Accident Analysis & Prevention 07/2015; 80. DOI:10.1016/j.aap.2015.03.036 · 1.87 Impact Factor
Show more