4.1.2. Distracted Driving

Other researchers have looked at the effect of distracted driving. The problem of mobile phone usage and distracted driving has been noticed by the World Health Organization [110]. They noted that world-wide use of cell phones has increased by up to 11% in the past 5 to 10 years. Their data sugges<sup>t</sup> that cell phone usage increases the chance of a crash by a factor of four, and this is similar for hand-held phones and hands-free devices. Young et al. [111] noted that at the time, about one fourth of all crashes (trucks and personal vehicles combined) were due at least in part to distractions, particularly mobile phones and navigational systems. They reviewed much of the literature available at the time of their writing. Wilson and Stimpson [112] reviewed trends in distracted driving accidents and noted that deaths due to distracted driving had increased 28% from 2005 to 2008 when the rate was nearly 6000 deaths per year.

Olson et al. [113] studied distracted driving in 203 commercial drivers. The data involved 4452 critical events, such as crashes, near-crashes, and unintentional lane departures, along with 19,888 time periods that involved no special events. The found that 71% of all crashes and 46% of near crashes involved drivers who were engaged in tasks not related to driving. Overall, 60% of critical events occurred while the driver was performing non-driving tasks. Klauer et al. [114] conducted a study in which 42 young drivers (16.3 to 17.0 years of age) who had just received their driver's license and 109 experienced drivers were studied. Here the unit of measurement is the driver. Equipment, such as accelerometers and cameras, were used to detect distracted motion while driving. They found that distracting events like eating or cell phone dialing or texting led to an increased risk of accident, with odds ratios often exceeding 3.0.

In terms of safety optimization, the choice here is clear. Distracted driving, such as hands-on cell phone use and texting, should not be allowed. From a general public perspective, these have translated into driving laws in many countries as well as have been translated into company policies for many commercial transportation firms. In addition, there are several smart phone-based applications that disables texting while driving and/or encourage safe driving behavior. From a commercial driving perspective, there are wearable technologies (e.g., headsets embedded with sensors that are linked to a smartphone application) that are used by professional drivers that provide voice-alerts when their mirror-check rate deviates from a pre-set standard. This information is also shared with dispatchers to schedule rest-breaks as an intervention. While these smart-phone applications/technologies seem promising, there is not a large body of literature that examines the effectiveness of these interventions.

#### 4.1.3. Weather, Traffic Conditions, and Road Geometry

In Sections 4.1.1 and 4.1.2, we have discussed driver-related factors. In many cases, the crash likelihood and severity can be impacted by non-driver/external factors. Variables/features capturing weather (e.g., temperature, precipitation, wind speed, humidity, and visibility), traffic conditions (e.g., traffic flow, occupancy, density, and volume), and road geometry (e.g., elevation, curvature, road surface, and the number of lanes) represent the main external factors that impact the crash likelihood and severity [10,11,115]. Note that these factors should not be considered in isolation since their interactions are complex and can significantly change the crash likelihood. Thus, in this subsection, we highlight three relevant studies that have investigated the combined effect of such factors on crash risk.

Ahmed et al. [116] investigated the effect of the interaction between road geometric features, real-time weather parameters, and traffic data on crash likelihood. Using a Bayesian logistic regression framework, the authors developed two models for snowy and dry seasons. Based on their models

and case study, their results showed that in, both models, the main effects and at least one interaction term were significant. The authors showed that the crash risk during the snowy season was two times that of the dry season. Furthermore, the authors suggested that the crash risk likelihood may also be influenced by the interaction effects between the snowy, icy, or slushy road surface conditions with road segments involving steep grades.

In another study, Yu et al. [117] conducted their study on a 15-mile segmen<sup>t</sup> of the I-70 interstate in Colorado. The authors utilized: (a) 30 Remote Traffic Microwave Sensor (RTMS) sensors to extract real-time traffic data; (b) six weather stations for obtaining real-time weather data; and (c) the Roadway Characteristics Inventory (RCI) for obtaining descriptors of road geometry. Different scenarios were considered in the study based on the season and crash type. The results showed that the adverse weather condition combined with critical roadway conditions (e.g., steep slopes) can increase the crash likelihood significantly. Further, single vehicle (SV) and multiple vehicle (MV) models shared some common significant predictors such as precipitation and average speed. Furthermore, in the SV model, the significant variables were more related to weather conditions and vehicle speed. On the other hand, MV crashes were more affected by traffic-related variables.

Wang et al. [118] studied several of the factors that could lead to high risk traffic conditions. They considered traffic, weather, road geometry, and some behavioral aspects, such as trip generation and social demographics. These variables were taken as the characteristics of the region surrounding the crash, not the individuals involved in the crash. They used a case-control design with a 10:1 ratio of non-crashes to crashes. They used support vector machines (SVM) for variable selection and Bayesian logistic regression for inference. They found that the percentage of home-based work production, which includes commuters, was the only behavioral characteristic that had a significant effect on the risk of accident.

Xu et al. [115] developed crash prediction models at different levels of crash severity. Three levels of crash severity were considered: fatal/incapacitating injury crashes (KA), non-incapacitating/possible injury crashes (BC), and property-damage-only crashes (PDO). Results showed that under different crash severity levels, the effect of environmental variables is different. For example, in the all crashes model (KA, BC and PDO), adverse weather conditions would increase crash risk. However, under the injury crashes model (KA and BC), adverse weather conditions had the opposite effect which indicated that it could possibly reduce the likelihood that a crash would result in injuries and fatalities (possibly due to uncaptured changes in driver behaviors). Note also that the significant traffic-related variables are different in these two models which indicates that the interaction of the external variables would result in different level of crash risk and severity.
