Empirical Evaluation of Drivers’ Behavior Response

1 In this paper, we focus on the accident messages displayed on freeway changeable message signs 2 (CMS), and study their effect on drivers’ route choice behavior. Previous studies on the effect of 3 CMS messages have shown mixed results, due to differences in event types and modeling choices. 4 Therefore, the purpose of this paper is to ascertain the real effect of CMS accident messages, 5 and also to compare two commonly used statistical models. We calculate the proportion of total 6 flow heading to off-ramps or freeway interchanges (called “diversion rate” hereafter) at diverging 7 locations, and use the change in diversion rate as the indicator for behavior change. We first draw 8 insights from two case studies, and find that the effect of CMS accident messages on the diversion 9 rate is minor and the effect of visible congestion is dominant. We then compare two commonly 10 used statistical analyses, accounting for the effect of visible congestion. The correlation analysis 11 compares the diversion rate with and without CMS accident messages, while the causality analysis 12 compares the diversion rate right before and after CMS accident messages are turned on or off. 13 With empirical data from three study sites, we use the causality analysis to show that the real effect 14 of CMS accident messages on diversion rate is insignificant. However, the correlation analysis 15 shows positive correlation between CMS accident messages and diversion rate, indicating that this 16 analysis cannot be used to draw causal inferences and that other factors have played a role in 17 changing the diversion rate. 18 INTRODUCTION 1 In this paper, we evaluate the effect of accident messages displayed on freeway changeable mes2 sage signs (CMS) on drivers’ route choice behavior. (Changeable message signs are also known 3 as variable message signs or dynamic message signs.) 4 CMS is an important component of the Advanced Traveler Information System (ATIS), 5 whose purpose is to generate and disseminate traveler information to enable informed travel de6 cisions. Compared with other traveler information sources like TV, radio, on-board navigation 7 devices, and smartphone apps, CMS has two distinct features. First, CMS only targets en route 8 drivers traveling through certain locations, mainly for them to change their routes (and sometimes 9 travel modes, like park-and-ride) but not their departure time. Second, the CMS message is visible 10 to all drivers within its range. The rest of the information sources either have a limited penetration 11 rate, or may not be active at all times, or both. 12 CMS messages are mostly about travel time and accident information. Sometimes, CMS 13 will broadcast roadwork information, weather-related cautions, etc. In this paper, we will focus 14 on accident messages instead of travel time or roadwork, for two reasons. First, accidents are 15 not foreseeable, and therefore accident messages are more likely to be valuable to travelers and 16 motivate their behavior change. Second, travel time does not show where the problem is even 17 when its value is higher than normal. Comparatively, accident messages are more accurate, with 18 information on the location and description of the accidents, and therefore are more likely to induce 19 behavior change. 20 The CMS system has been in use for a long time, dating back to at least the 1960s. The 21 system is expensive: Typical installation cost for each freeway CMS is around $200,000, excluding 22 the cost for operation and maintenance. In California alone, there are about 771 such signs on the 23 freeway, which cost at least $150 million for installation. 24 However, we have very limited understanding of how effective the CMS system is. This 25 statement is true for ATIS generally. The effectiveness of the system is largely anecdotal. Accord26 ing to a recently published NCHRP report (27), only 30 percent of the agencies reported having 27 evaluation data that demonstrate the benefits of providing information to the traveling public, and 28 only 40 percent have an ongoing program for evaluating the provision of traveler information. 29 The goal of our research is to evaluate the effectiveness of CMS messages and to under30 stand how travelers react to this information. The main challenge in answering this question is 31 to distinguish the effect of CMS from the effect of other sources of information such as visible 32 congestion and radio. 33 This paper is organized as follows. In the literature review section, we describe the previous 34 work and the gap in modeling the effect of CMS on travelers. After describing our experimental 35 design, data, and study sites, we start by drawing insights from two case studies. We then perform 36 two types of statistical analyses: The first analysis is simple and focuses on correlation only; the 37 second analysis borrows ideas from regression discontinuity and determines the causal effect of 38 the CMS message. At the end, we conclude the paper and provide directions for future research. 39 LITERATURE REVIEW 40 We will start with a brief description of the literature on the network effect of traveler information. 41 The majority of the literature on the effect of CMS uses either the survey method or the empirical 42 method, which are described below in detail. 43