You are on page 1of 10

Transportation Research Record 1746 s Paper No.

01-2346

37

Simulation of Road Crashes by Use of Systems Dynamics


Arif Mehmood, Frank Saccomanno, and Bruce Hellinga
The framework of a systems dynamics model (SDM) is presented for simulating road crashes. For the purpose of illustration, the model is applied to two-vehicle rear-end crashes where both vehicles are assumed to be traveling in the same lane. The crash situation results from the introduction of some obstruction along the pathway, which causes drivers to adjust their speeds and separation distances. SDM extends the features of classical car-following theory in combination with crash avoidance models. Preliminary results are discussed. The SDM provides a number of useful insights into the complex process that gives rise to rear-end crashes for an assumed transportation scenario. These insights can serve to guide the development of effective safety countermeasures for reducing rear-end crashes. The potential relevance of the model for other crash situations is also discussed.

of transportation conditions. With these insights, a better evaluation of the effectiveness of alternative safety countermeasures is achieved.

CRASH SCENARIO CONSIDERED In this section, the basic features and underlying assumptions of a simple rear-end crash for SDM modeling are dened. Different pathways leading to this type of crash are described within the context of the simple scenario being considered. The authors decision to restrict this analysis to rear-end crashes should not reect a limitation of the approach itself. Rather, in this paper the authors have restricted the crash types being considered in order to illustrate the basic features of the approach. Application of SDM requires an explicit denition of the individual decisions and actions leading to some outcome of interest. Within the context of two-vehicle rear-end crashes, the behavior of the lead vehicle and the following vehicle to some extent can be described by classical car-following theory. Studies addressed herein (817) are among the considerable body of research that has been conducted to reect the car-following behavior of drivers for different traffic assumptions and conditions. However, by far the largest contribution to the car-following theory was made by the General Motors (GM) researchers in the late 1950s and early 1960s (8, 11, 12). In fact, most of the other car-following models are special cases of the GM model (e.g., 9, 10, 13). The GM model is a stimulus-response model of car-following. It assumes that the following vehicle driver responds by accelerating or decelerating based on a stimulus measure that is a function of the drivers own speed, speed relative to the lead vehicle, and relative separation distance from the lead vehicle. Despite the prevalence of the GM model (and its variants) in the research literature, the GM model exhibits several undesirable characteristics. First, when the lead and following vehicles are traveling at the same speed, the acceleration response dictated by the GM model is 0, regardless of the current separation distance. Second, the GM model assumes a symmetric behavior of the following vehicle driver. For example, if the following vehicle driver responds by accelerating at x ft/s2 when the relative speed is +y ft/s, then, under similar conditions of headway and speed the response to a relative speed of y ft/s will be x ft/s2. However, it has been observed that drivers act differently when the separation distance between vehicles is increasing or decreasing because drivers pay closer attention to spacing decreases than to spacing increases (18). Although classical car-following theory can explain some aspects of rear-end crash, the presence of these two limiting characteristics in the GM model led the authors to seek a more exible and encompassing approach such as SDM. SDM was also selected as the model of choice for road crash analysis since it can encompass a wider range

Systems dynamics models (SDMs) have been used in the past to describe and investigate complex interrelated systems or processes. These processes are characterized by many nonlinear causal relationships, which are often heuristic in nature and veried empirically. The basic theory of SDM was developed in the 1930s. SDM originated in engineering and rapidly spread into economic, social, and corporate applications. The theory was refined in the 1950s by Brown and Forrester (1) for application in the elds of industrial and management sciences. In this research an SDM was developed for the study of road crashes. To the authors best knowledge, this application of SDM to the analysis of road crashes has not been previously documented in the published literature, except for some recent work carried out by the ASTRA in Europe (2). Road crashes are caused by a complex interplay of factors and connecting pathways. The development of effective safety countermeasures for reducing these crashes requires a thorough understanding of the roles these factors play in different types of crashes. Road crashes have traditionally been modeled using statistical methods (37). Although these statistical models have proved to be valuable for identifying sections of roadways that may be unsafe (e.g., black spots), they are generally unable to assist in the evaluation of potential safety countermeasures. Specically, if the statistical model does not contain independent variables associated with the potential countermeasure, the safety benet of implementing that countermeasure cannot be determined by the model. The main purpose of this paper is to introduce an SDM framework and discuss its application to a simple rear-end crash involving two vehicles traveling in the same lane. SDM provides insights as to how rear-end crashes take place at a given location for a given set
Department of Civil Engineering, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada.

38

Paper No. 01-2346

Transportation Research Record 1746

FIGURE 1

Initial features of rear-end crash considered (in equilibrium).

of crash scenarios, many of which cannot be adequately described by car-following theory (e.g., sideswipes and single-vehicle lossof-control). To illustrate SDM, the components, decisions, and actions in rearend crashes are dened. These are not intended to be a comprehensive description of precrash decisions and actions but rather to provide a representative set of realistic denitions that can be used to illustrate the SDM process. Figure 1 illustrates the initial features of a simple same-lane rearend crash. The lead vehicle encounters some obstruction or target along its path (same lane of travel for all vehicles involved), causing its driver to react in order to avoid a crash with the target. For both drivers two actions are possible: (a) a reduction in speed or (b) a change in direction of travel or lane change. For this application it has been assumed that vehicles continue to travel in the same lane, such that only a speed reduction is possible. The target obstruction can be either stationary (e.g., debris on the road) or moving (e.g., a slow-moving vehicle). In the case of a moving target, the speed of the target must be lower than the speed of the lead vehicle for a rear-end crash to occur. The driver of the following vehicle observes the stream of traffic (both the lead vehicle and the target) to adjust his or her speed in order to avoid a crash with the lead vehicle. Microlevel simulation models (e.g., NETSIM, INTRAS, FRESIM, CARSIM) assume that the following vehicle driver observes only the lead vehicle in determining his or her speed. In reality it has been observed that driver responses are often determined on the basis of the behavior of several downstream vehicles, not just the single vehicle immediately downstream. The required safe speed for the following vehicle is established to maintain a safe separation distance from the lead vehicle. As the current separation distances between the lead vehicle and the target and between the following vehicle and the lead vehicle are reduced, the driver of the following vehicle increases his or her reliance on the

current separation distance to the target. This relationship is established as a function of the current separation distances between the vehicles and required safe separation distance for the following vehicle. The current separation distance between the lead vehicle and following vehicle and lead vehicle and the target is established for different time intervals, based on their initial speeds, required safe speeds, and acceleration/deceleration rates. The potential for a rearend crash between either the lead vehicle and the target or the following vehicle and the lead vehicle is a probabilistic function of the current separation distance at time t, the minimum required safe stopping distance at time t, current speed of the vehicle at time t, and required safe speed at time t.

SDM MODEL STRUCTURE In this section an SDM model that replicates individual driver behavior for all drivers involved for the simple rear-end crash scenario dened above is presented. The model is built using system dynamics methodology and the ithink software platform. Figure 2 shows the overall structure of the SDM model being discussed in this paper. The model consists of ve main sectors: the target, the lead vehicle, the following vehicle, the separation distance, and the crash potential. Each sector performs certain functions and interacts with the other sectors (through feedback links) to describe the rear-end crash process. The arrows linking the sectors represent their interaction. The details of each sector including relationships that link them together are described below. The developed model is a microscopic simulation model, which describes individual driver behavior. This model is formulated such that the speed of the target is set exogenously by the user, whereas the speed and separation distances of the lead and following vehi-

FIGURE 2

Overall structure of SDM model.

Mehmood et al.

Paper No. 01-2346

39

cles are determined endogenously subject to the specied rules and conditions in the SDM model. As noted previously, it is assumed that vehicles are moving in the same lane and that applying the brakes is the only response option available to drivers to avoid rear-end crashes.

Target Sector In Figure 3, the target obstruction sector is an exogenous sector whose function is to describe the target in terms of its required or safe speed and its safe separation distance with respect to the lead vehicle. In this sector the stock variable (VT) represents the speed of the target at time t. The rate of change in the speed of the target (RT) during the time interval dt is determined by the initial and desired speed of the target for an assumed interval of time over which this change takes place. Any change in the speed of the target will change its relative position with respect to the lead vehicle. The separation distance between the target and the lead vehicle is calculated in the Separation Distance Sector.

STOCK FLOW DIAGRAM OF REAR-END CRASH SDM Figure 3 illustrates the stock ow diagram for the SDM model being considered in this study. Stocks represent accumulations of all transfers of information (ows) describing the process. The basic features of stock ow diagrams of this nature have been discussed elsewhere (19, 20). In Figure 3 the rectangles represent stocks, the valve symbols represent ows, and the circles represent intermediate computations. The circles with the symbol ~ represent cause and effect relationships between two variables. All variables in Figure 3 are dened in Table 1, along with basic inputs used in their determination. Each sector of the stock flow diagram is discussed separately below.

Lead Vehicle Sector In this sector the behavior of the lead vehicle driver is simulated. The process describing this sector is similar to the following vehicle sector as described in the next section. The only difference between the lead and following vehicle sector is that the driver of the following vehicle considers both the separation distance between the target and

FIGURE 3

Stock Flow diagram of SDM model.

40

Paper No. 01-2346

Transportation Research Record 1746

TABLE 1

Variable Descriptions and Inputs

(continued) the lead vehicle, and the separation distance between the lead and the following vehicle when adjusting his or her speed. While adjusting his or her speed, the lead vehicle driver considers only the separation distance between the target and itself. In the present model, it is assumed that the driver maintains a speed equal to or less than the lead vehicle speed and also maintains a separation distance equal to or greater than the required safe separation distance. This required safe separation distance is the minimum distance required by the lead vehicle driver or the following vehicle driver to safely reduce his or her speed or to stop prior to crashing. In Figure 3, the stock variable V2 represents the current speed of the following vehicle. This speed is initially assumed to be 80 ft/s (24.4 m/s) or 55 mph (128.7 km/h). The equation to determine the current speed of the lead vehicle at simulation time t is V 2 (t ) = V 2 (t dt ) + AD 2 (t ) dt (1)

Following Vehicle Sector In this sector the behavior of the following vehicle driver is simulated. First it is assumed that the driver behaves normally, that is, can observe the target at the appropriate time and reduces his or her speed in order to avoid a rear-end crash. For this initial application it is assumed that weather, pavement, and vehicle conditions are also normal. These assumptions will be relaxed later in order to better appreciate the full spectrum of conditions leading to rear-end and other types of crashes.

The variables of the model as used in these equations and as illustrated in Figure 3 are dened in Table 1.

Mehmood et al.

Paper No. 01-2346

41

TABLE 1 (continued)

Variable Descriptions and Inputs

The variable dt represents the simulation interval, over which all calculations of the model are carried out. For this application a value for dt of 0.25 s was assumed. In Equation 1, the acceleration/deceleration rate of the following vehicle speed (AD2) at time t is based on the current speed of the following vehicle (V2) at time t, required safe speed of the following vehicle (RV2) at time t, and the perception-reaction time of the following vehicle driver (T2). It is assumed for this initial application that T2 is 2.5 s. The equation used to determine AD2 is AD 2 (t ) = [ RV 2 (t ) V 2 (t )] T 2 (2)

RV 2 (t ) = V 1 (t ) ES 2 (t )

(3)

In Equation 3 the factor ES2 is obtained from a cause and effect relationship of the variables Ratio2 and ESDR. This function is illustrated in Figure 4. The x-axis in this gure represents the product of variables Ratio2 and ESDR. The y-axis represents the assumed values of ES2. The variable Ratio2 is dened as Ratio 2 = S 2 (t ) RS 2 (t ) ( 4)

The required safe speed of the following vehicle (RV2) with respect to the lead vehicle at time t depends on (a) the speed of the lead vehicle (V1) at time t and (b) the effect of separation distance ratio on the required safe speed of the following vehicle (ES2) at time t, such that

In Figure 4, a value less than 1.0 reects a reduced margin of safety which implies an exponentially increasing effect on the part of the driver to adjust his or her speed, and vice versa for values greater than 1.0. The variable ESDR is a cause and effect relationship of variables S1 and S2 illustrated in Figure 5. S1 is the current separation distance

42

Paper No. 01-2346

Transportation Research Record 1746

FIGURE 4 Effect of separation distance ratio on required safe speed of the following vehicle.

between the lead vehicle and the target at time t, and S2 is the current separation distance between the following and the lead vehicle at time t. In Figure 5 the x-axis represents the ratios of S2 and S1, and the y-axis represents the assumed values of ESDR. The simulation starts at equilibrium where the initial value of Ratio2 and ESDR is set equal to 1.0. This is because the current separation distance for the following vehicle (S2) at initial speed [assumed value of 80 ft/s (24.4 m/s)] is set equal to the required safe separation distance (RS2). This assumes that the driver of the following vehicle does not have to change his or her speed. However, when the speed of the lead vehicle changes, the variables Ratio2 and ESDR will change, and the following vehicle driver will adjust his speed to maintain a required safe separation distance with the lead vehicle. Presumably, other factors such as pavement conditions, pavement friction, road geometry, and traffic conditions can affect the required safe speed for the following vehicle. For the initial analysis in this paper, it was assumed that ideal conditions and these other relationships were not incorporated into the model. The nonlinear relationship in Figure 4 and other nonlinear relationships used in the SDM road crash model represent the cause and effect relationship. These are initially based on an intuitive or heuristic understanding of these relationships as opposed to having been calibrated on the basis of empirical evidence. As noted by Legasto et al., the heuristic nature of cause and effect relationships such as that illustrated in Figure 4 is not a source of weakness of SDM but rather a source of strength (21). Through these types of functions, SDM is able to represent essential phenomena that might otherwise be omitted from lack of sound empirical evidence. Note that this does

not mean that these cause and effect relationships should be purely subjective in nature but rather that they should be initially dened and incorporated into the process, and later investigated in more depth using whatever empirical evidence is available. Any change in the current speed of either the following or the lead vehicle will have the effect of changing the current separation distance between the following and the lead vehicle (S2). If the speed of the following vehicle increases relative to the lead vehicle, the separation distance (S2) will decrease. In the model, the required safe separation distance (RS2) at time t between the following and the lead vehicle is expressed as follows (22, 23). RS 2 (t ) = V 2 (t ) TP 2 (5)

TP2 is preferred time headway of the following vehicle driver in seconds. van Winsum has reported that there are substantial differences in the value of preferred headway between drivers (22). For example, drivers who are less skilled generally choose to drive at a larger time headway. In this paper it is assumed that TP2 = 2 s. The initial value of the current separation distance (S2) for an initial following vehicle speed of 80 ft/s (24.4 m/s) is calculated using Equation 5 as 160 ft (48.8 m/s) (S2 = 80 2 = 160). However, as the speed of the following vehicle changes, so do its required safe separation distance and current separation distance. Changes in the value of required safe separation distance will have the effect of changing the following vehicle separation distance ratio (Ratio2).

Separation Distance Sector As shown in Figure 3, in this sector there are two stocks: one determines the current separation distance between the target and the lead vehicle (S1), and the other determines the current separation distance between the lead vehicle and the following vehicle (S2). It is assumed that the change in the current separation distances between these vehicles depends on their relative speeds and the simulation time interval. Presumably, other factors can affect this relationship, such as the vehicle weight, pavement conditions, pavement friction, and grade. For the initial analysis in this paper, it was assumed that ideal conditions and these other relationships were not incorporated into the model.

FIGURE 5

Effect of current separation distance ratio on Ratio2.

Mehmood et al.

Paper No. 01-2346

43

Crash Potential Sector Two rear-end crash scenarios are considered in this sector: (a) between the lead vehicle and the target, and (b) between the lead vehicle and the following vehicle. For both scenarios the potential for a crash depends on three basic factors: current speed, difference in speed, and difference in separation distance. These factors yield a crash potential between the lead vehicle and the target, and the lead vehicle and the following vehicle. In this paper, the authors have initially dened this crash potential in terms of a simple normative measure of the relative speeds and respective separation distances of the vehicles. This relationship will likely be the subject of future work and empirical verication. Initially, the following vehicle crash potential (CP2) has been expressed as CP 2 = ( EV 2 D Eratio 2 EV 2) (6)

FIGURE 7 Effect of following vehicle separation distance ratio on crash potential of the following vehicle.

EV2D is the effect of the difference in speed between the current and required safe speed of the following vehicle on the following vehicle crash potential. Eratio2 is the effect of current and required safe separation distance ratio on the lead vehicle crash potential, and EV2 is the effect of current speed of the following vehicle on the following vehicle crash potential. In Equation 6 the factors EV2D, Eratio2, and EV2 are determined by using the cause and effect relationship illustrated in Figures 6, 7, and 8. Again these relationships have been dened heuristically and need to be veried. The x-axis in Figure 6 represents the ratio between the current speed of the following vehicle and required safe speed of the following vehicle. The y-axis represents the assumed values of EV2D. To obtain the shape of the curve in Figure 6, it was initially assumed that the values of EV2D increase exponentially with the increase in ratio between current and required safe speed of the following vehicle. The obtained values of EV2D are used in Equation 6 to estimate the crash potential for the following vehicle. The second factor affecting the following vehicle crash potential is Eratio2. This relationship is illustrated in Figure 7. The x-axis in this gure represents the values of following vehicle separation distance ratio (Ratio2). The y-axis is assumed values of Eratio2. To obtain the shape of curve in Figure 7, an exponential relationship was assumed between Eratio2 and Ratio2. The values of Eratio2 are used in Equation 6 to estimate the CP2.

The third factor affecting the following vehicle crash potential is EV2, illustrated in Figure 8. The x-axis in this gure represents the normalized values of current speed of the following vehicle. The y-axis is assumed values of EV2. To obtain the shape of the curve in Figure 8 it was assumed that EV2 increases exponentially with increases in the current speed of the following vehicle. A normalized value of current speed of the following vehicle from 0 to 1.0 reects an increased margin of safety, and a value greater than 1.0 reects a decreased margin of safety. The values of EV2 are used in Equation 6 to estimate the crash potential of the following vehicle (CP2). The crash potential for the lead vehicle (CP1) is established in a similar way. In this expression the current speed of the lead vehicle, and the speed difference and separation distance between the lead vehicle and the target are used to estimate this potential.

PRELIMINARY SIMULATION RESULTS The simulation run starts where all variables are parameterized to reect equilibrium in all stocks. Initially, it was assumed that the speeds of the target, lead, and following vehicles are the same at 80 ft/s (24.4 m/s). Furthermore, the current separation distances between the target and the lead vehicle and between the lead and following vehicle are assumed initially to be equal to their required safe separation distance of 160 ft (48.8 m). The initial simulation clock time is set at time t = 1.0 s (the default setting in the ithink software). It is now assumed that the speed of the target is decreased exogenously with a deceleration rate of 16 ft/s2 (4.9 m/s2) beginning at

FIGURE 6 Effect of ratio between the current speed and required safe speed of following vehicle on the crash potential of following vehicle.

FIGURE 8 Effect of current speed of the following vehicle on the following vehicle crash potential.

44

Paper No. 01-2346

Transportation Research Record 1746

FIGURE 9

Speed profiles of target, lead vehicle, and following vehicle (1 ft

0.3 m).

time t = 10.0 s. This reduction in the speed of the target could be in response to the onset of congestion or the presence of some debris along the path of the target. Figure 9 illustrates the speed prole of the target, lead vehicle, and following vehicle as obtained from the SDM for a 50-s simulation interval. Figure 10 illustrates the separation distance between the lead vehicle and the target, and the lead vehicle and the following vehicle over this same simulation interval. This is related directly to the crash potential for the lead and following vehicles as per Equation 6. This potential is plotted separately for both the lead and the following vehicle in Figure 11 for the 50-s simulation interval. As shown in Figure 9, the reduction in the speed of the target decreases the current separation distance between the target and the lead vehicle. The lead vehicle driver perceives the reduction in the target speed and the desired separation distance between the target and lead vehicles and reduces his speed accordingly to match that of the target. This is done to maintain the desired separation distance between the target and the lead vehicle. The reduction in the speed of the lead vehicle decreases the current separation distance from the following vehicle. The driver of

the following vehicle also reacts in a similar fashion to reduce his or her speed to match that of the lead vehicle. This is done to maintain the required safe separation distance. In Figure 9, the target, lead, and following vehicle speeds are adjusted accordingly until a new equilibrium speed prole is established. This occurs when the vehicle and target speeds are equal to each other. For a stationary target, all vehicles will be stopped. In Figure 10, the current separation distance between the target and the lead vehicle is reduced as a result of a reduction in the speed of the target. Here the lead vehicle driver tries to match the target speed. When the lead vehicle speed and target speed become equal, the current separation distance between the target and the lead vehicle stabilizes at its new equilibrium value. This is equal to the required safe separation distance for the given lead vehicle speed. Similarly, the separation distance between the lead vehicle and the following vehicle declines and stabilizes at a point where their speeds are equal. This is shown in Figure 10. As shown in Figure 11, at simulation clock time t = 10.0 s the crash potential for the lead vehicle increases sharply. This sharp increase is due to the increase in the speed difference and decrease

FIGURE 10 Current separation distance profiles of lead and following vehicles (1 ft 0.3 m).

Mehmood et al.

Paper No. 01-2346

45

FIGURE 11

Crash potential profiles of lead and following vehicles.

in separation distance between the target and lead vehicle. When the lead vehicle speed is reduced to match the speed of the target and the current separation distance between target and lead vehicle matches the required safe separation distance, the crash potential is reduced. This stabilizes in the long run to a value less than the initial value of crash potential because the speed of the lead vehicle is reduced. At this point, uniform speed proles and a safe separation distance between the lead vehicle and the target are obtained. The crash potential prole for the following vehicle is also illustrated in Figure 11. As expected, the highest potential for a rear-end crash occurs when the vehicle speeds are highest and their separation distance is lowest.

USE OF SDM The use of the SDM model is to explain the causes of an undesirable behavior and identify the policy variables capable of eliminating the undesirable behavior. The policy variables are those parameters in

the model that change the behavior of the model in response to the change in the parameter values. For example, in the case of road crashes, the desirable behavior would be a reduction in accident potential over the simulated interval under a range of transportation conditions. To illustrate how SDM can be used to evaluate safety countermeasures, consider the introduction of countermeasures that primarily impact driver perception reaction times (e.g., on-board vision enhancement systems or improved roadway lighting). Figure 12 illustrates the rear-end crash potential of the following vehicle for three driver perception/reaction times: 1.5, 2.5, and 3.5 s. The 2.5 s represents normal perception/reaction time of the following vehicle driver. The 1.5 s input reects possible improvement associated with the countermeasure, and 3.5 s represents retarded perception/reaction time associated with possible drug or alcohol use or driver fatigue. As expected, an improvement (i.e., reduction) in the crash potential associated with the introduction of the countermeasure is noted. Also as expected, an increase in the perception/reaction time increases the crash potential for the following vehicle.

FIGURE 12 Change in crash potential of the following vehicle by changing perception-reaction time of the following vehicle driver.

46

Paper No. 01-2346

Transportation Research Record 1746

CONCLUSIONS An SDM model of simple two-vehicle rear-end crashes has been developed to demonstrate the application of SDM to road crashes in general. This model replicates the behavior of two individual drivers traveling in the same lane and how they adjust their speeds to avoid a rear-end crash. The rear-end crash scenario began with the introduction of some obstruction (moving or stationary) in the path of the lead vehicle. This causes the drivers of the lead and the following vehicle to adjust their speeds and separation distance accordingly. This paper has demonstrated that the SDM approach provides a useful platform for describing and simulating road crashes. The SDM approach offers the necessary exibility and scope for application to other important crash situations. SDM is able to represent essential phenomena that might otherwise be omitted for lack of sound empirical evidence. It is recognized that further calibration and validation of the model components are required; however, this paper has shown that SDM is useful for evaluating changes in crash potential from the introduction of alternate safety countermeasures. REFERENCES
1. Assessment of Transport Strategies. Design and Specications of a System Dynamics Model, Deliverable D2. Commission of the European Communities, Brussels, Belgium, 1998. 2. Assessment of Transport Strategies. The ASTRA System Dynamics Model Platform (ASP), Deliverable D3. Commission of the European Communities, Brussels, Belgium, 1999. 3. Hauer, E. On the Estimation of the Expected Number of Accidents. Accident Analysis and Prevention Vol. 18, No. 1, 1986. 4. Persaud, B. Black Spot Identication and Treatment Evaluation. The Research and Development Branch, Ontario Ministry of Transportation, 1990. 5. Miaou, S.-P., and H. Lum. Statistical Evaluation of the Effects of Highway Geometric Design on Truck Accident Involvements. In Transportation Research Record 1407, TRB, National Research Council, Washington, D.C., 1993, pp. 1123. 6. Saccomanno, F. F., and C. Buyco. Generalized Loglinear Models of Truck Accident Rates. In Transportation Research Record 1172, TRB, National Research Council, Washington, D.C., 1988, pp. 2331. 7. Jovanis, P., and H. Chang. Disaggregate Model of Highway Accident Occurrence Using Survival Theory. Accident Analysis and Prevention, Vol. 21, No. 5, 1989, pp. 445458.

8. Chandler, R. E., R. Herman, and E. W. Montrol. Traffic Dynamics: Studies in Car-Following. Operation Research, Vol. 6, No. 2, 1958, pp. 165184. 9. Forbes, T. W., H. J. Zagorski, E. L. Holshouser, and W. A. Deterline. Measurement of Driver Reactions to Tunnel Conditions. HRB Proc., Vol. 37, 1958, pp. 6066. 10. Forbes, T. W. Human Factor Considerations in Traffic Flow Theory. In Highway Research Record 15, HRB, National Research Council, Washington, D.C, 1963, pp. 6066. 11. Gazis, D. C., R. Herman, and R. W. Rothery. Nonlinear Follow-theLeader Models of Traffic Flow. Operation Research, Vol. 9, No. 4, 1961, pp. 545567. 12. Herman, R., and R. B. Potts. Single Lane Traffic Theory and Experiment. In Proceedings of the Symposium on the Theory of Traffic Flow, Warren, Mich., Elsevier Publishing, Amsterdam, Netherlands, 1961, pp. 120146. 13. May, A. D., and H. E. M. Keller. Non-Integer Car-Following Models. In Highway Research Record 199, HRB, National Research Council, Washington, D.C., 1967, pp. 1932. 14. Gipps, P. G. A Behavioural Car-Following Model for Computer Simulation. In Transportation Research B, Vol. 15, No. 2, 1981, pp. 105111. 15. van Aerde, M. Single Regime Speed-Flow-Density Relationship for Congested and Uncongested Highways. Presented at the 74th Annual Meeting of the Transportation Research Board, Washington, D.C., 1995. 16. Hogema, J. H. Modeling Motorway Driving Behavior. Presented at the 77th Annual Meeting of the Transportation Research Board, Washington, D.C., 1998. 17. Zhang, Y., L. E. Owen, and J. E. Clark. Multiregime Approach for Microscopic Traffic Simulation. In Transportation Research Record 1644, TRB, National Research Council, Washington, D.C., 1998, pp. 103115. 18. Leutzbach, W. Introduction to the Theory of Traffic Flow. SpringerVerlag, New York, 1988. 19. Forrester, J. W. Industrial Dynamics, MIT Press, Cambridge, Mass, 1961. 20. ithink Technical Documentation. High Performance Systems, Inc., Hanover, N.H., 1994. 21. System Dynamics: Studies in the Management Sciences, Vol. 14 (A. A. Legasto, J. W. Forrester, and M. J. Lyneis, eds.). North-Holland Publishing, New York, 1980. 22. van Winsum, W. The Human Element in Car-Following Models. In Transportation Research F, Vol. 2, 1999, pp. 207211. 23. Aycin, M. F., and R. F. Benekohal. Comparison of Car-Following Models for Simulation. In Transportation Research Record: Journal of the Transportation Research Board, No. 1678, TRB, National Research Council, Washington, D.C., 1999, pp. 116127.
Publication of this paper sponsored by Committee on Safety Data, Analysis, and Evaluation.

You might also like