You are on page 1of 7

Accident Analysis and Prevention 35 (2003) 427433

A comparison of headway and time to collision as safety indicators


Katja Vogel
Swedish National Road and Transport Research Institute, S-581 95 Linkping, Sweden Received 13 September 2001; received in revised form 21 January 2002; accepted 13 February 2002

Abstract The two safety indicators headway and time to collision (TTC) are discussed and compared with respect to their usefulness in determining the safety of different trafc situations, like different locations in a junction. Over a 6-day-period trafc ow measures were taken in a four-way junction with stop signs on the minor road. It was found that for vehicles in a car following situation headway and TTC are independent of each other. The percentage of small headways is relatively constant across different locations in the junction, while the percentage of small TTC values varies between different locations. It is recommended to use headway for enforcement purposes, because small headways generate potentially dangerous situations. TTC, on the other hand, should be used when a certain trafc environment is to be evaluated in terms of safety, because it indicates the actual occurrences of dangerous situations. 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Safety assessment; Safety indicator; Time to collision; Headway

1. Introduction Time headway (H) is one of the indicators that is used to estimate the criticality of a certain trafc situation. It has been dened as the elapsed time between the front of the lead vehicle passing a point on the roadway and the front of the following vehicle passing the same point (Evans, 1991, p. 313). In some countries, this indicator is also used by the authorities to impose nes for close following. Another widely used safety indicator is time to collision (TTC), a concept introduced by Hayward (1972). It indicates the time span left before two vehicles collide, if nobody takes evasive action. In this study, these two indicators will be compared, both theoretically and in relation to empirical data. 1.1. Time headway Time headway is measured by taking the time that passes between two vehicles reaching the same location (see Eq. (1)). H = ti ti1 (1)

with ti denoting the time at which the vehicle i passes a certain location and ti1 the time at which the vehicle ahead of vehicle i passes the same location.

Tel.: +46-13-20-43-06; fax: +46-13-14-14-36. E-mail address: katja.vogel@vti.se (K. Vogel).

Different countries have slightly different rules with regard to the legal or recommended safety distance. In the US, e.g. several driver training programs (Michael et al., 2000) state that it is impossible to follow a vehicle safely with a headway of less than 2 s. In Germany, the recommended minimum distance is half the speedometer, which means, a car traveling at 80 km/h should keep a distance of at least 40 m. This rule translates to a recommended time headway of 1.8 s. Fines are imposed when the time headway is smaller than 0.9 s. In Sweden the National Road Administration recommends a time headway of 3 s in rural areas, and the police use a time headway of 1 s as orientation for imposing nes. Researchers investigated whether any connection between preferred time headway, accident involvement, and driver characteristics existed, but the results are not consistent. Evans and Wasielewski (1982), e.g. claimed that drivers who keep longer time headways tend to have a history of fewer accidents and violations. On the other hand, the same authors stated 1 year later (Evans and Wasielewski, 1983), that no reliable relation between preferred time headway and accident involvement could be detected. Van Winsum and Heino (1996) investigated in a simulator study whether a closer following distance was connected to more expertise in accurately estimating TTC, but the relationship they found was not signicant. Michael et al. (2000) found that a substantial percentage of drivers in several urban locations did not observe the 2 s rule, but compliance increased moderately when hand-held signs urged drivers to heed the rule.

0001-4575/02/$ see front matter 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 0 0 1 - 4 5 7 5 ( 0 2 ) 0 0 0 2 2 - 2

428

K. Vogel / Accident Analysis and Prevention 35 (2003) 427433

However, no direct relation to the occurrence of trafc conicts was provided. 1.2. Time to collision TTC is computed according to Eq. (2). TTCi = Xi1 (t) Xi (t) li Xi (t) Xi1 (t) Xi (t) > Xi1 (t) (2)

with Xi denoting the speed of vehicle i, Xi the position of vehicle i, li the length of vehicle i and i1 the vehicle ahead of vehicle i. The TTC is the time that is left until a collision occurs if both vehicles continue on the same course and at the same speed. Therefore it is the time that is needed to cover the distance between the lead and the following vehicle with the relative speed between the lead and the following vehicle. Hayward (1972) rst introduced this concept, and it is discussed extensively in Hydn (1987), for example. For calculation of TTC to be possible, the vehicles in question have to be on collision course, but not necessarily in a car-following situation. TTC in car-following situations is only dened when the speed of the following vehicle is higher than the speed of the lead vehicle. In research TTC has often been used as a safety indicator for certain maneuvers by determining the minimum TTC measured during the maneuver (e.g. van Winsum and Heino, 1996; Hirst and Graham, 1997; Janssen and Nilsson, 1991). According to Svensson (1998), TTC is an indicator for a trafc conict and is, thus, inversely related to accident risk (smaller TTC values indicate higher accident risks and vice versa). In a recent article, Minderhoud and Bovy (2001) suggest a method, which allows using TTC to compare the safety of different drivers, road environments, or situations in general. The basic idea is to sample TTC values over time, and to examine how often a certain driver undershoots a given lower safety limit, or how often this limit is breached on a particular road stretch or under particular conditions. In the literature different opinions can be found as to which value should be used as safety limitsuggestions range from 1.5 s in urban areas (Svensson, 1998) to 5 s (Maretzke and Jacob, 1992). Minderhouds and Bovys (2001) method will be applied on empirical data in this study, and different threshold values will be compared. 1.3. Relation and comparison A comparison of the two equations shows that more variables have to be known to determine TTC than to determine H. The relationship between these two measures is presented in Eq. (3). Xi TTC = H, i Xi1 X where li1 H =H = gap Xi (3)

with Xi1 denoting the speed of the lead vehicle, Xi the speed of the following vehicle and li1 the length of the lead vehicle. To obtain TTC, the speed of both the involved vehicles has to be known in addition to the time gap. An interesting difference between the two measures exists with respect to trafc safety. It could be formulated such that time headway is a step further away from a crash than TTC. This claim is based on the reasoning described later. Let us consider a vehicle in the following mode. Such a vehicle can have a relatively small headway, but a large or even undened TTC value. This situation occurs, when Xi1 is equal to or larger than Xi (cf. Eq. (3)). The situation becomes critical only when something in the constellation changes, like when the lead vehicle brakes such that Xi1 i . Thus, under stable circumstances, becomes smaller than X a small time headway can be maintained over extended periods of time without resulting in an immediately dangerous situation. If, on the other hand, the TTC value of the following vehicle is small, something has to change in the constellation if a crash is to be avoided. In a car-following situation, the average relative speed between following and lead vehicle cannot be larger than 0 if a collision is to be avoided in the long run. To summarize, vehicles with small time headways can (and frequently do) have large or undened TTC values, while small TTC values (in car following) are impossible for vehicles with long time headways. In a car-following situation, TTC can, in fact, never be smaller than H , because the term Xi /(Xi Xi1 ) (cf. Eq. (3)) can never be smaller than 1. This is due to the fact that Xi1 is never negative, which would mean that the lead vehicle was reversing. In the special case of a stopped lead vehicle (Xi1 = 0), the actual time gap (H ) equals TTC. As mentioned earlier, in a car-following situation TTC can never be smaller than the time gap between the lead and the following vehicle (H ). Thus, if the two values are to be compared, it seems reasonable to exclude those cases that are not safety critical with respect to any of the two measures (Table 1). In order to determine the threshold between safety critical small and safe long headways, the existing literature on the concept of free and following vehicles was consulted. A free vehicle is by denition not in interaction with any vehicle ahead of it. For this reason, the analyses in the present study were limited to following vehicles. The denition for a free vehicle that was adopted here is based on an empirical analysis by Vogel (2002), which shows that
Table 1 Relationship between TTC, headway and safety Headway Small TTC Small Large Danger imminent Potential danger Large Impossible Safe

K. Vogel / Accident Analysis and Prevention 35 (2003) 427433

429

vehicles with a time headway of more than 6 s choose their traveling speed independent of the vehicle ahead. The choice of this threshold value is supported by the fact that no author considers a TTC that is larger than 6 s to be dangerous. This means that even a stopped lead vehicle will not force the follower to perform dangerous avoidance maneuvers if the headway is larger than 6 s (and the following driver can see the vehicle ahead). Only for speeds higher than 130 km/h a larger time headway is needed to stop a vehicle safely behind a stopped vehicle. An additional reason for limiting the analysis to following vehicles only is the possibility to compare different situations with each other. It allows, e.g. comparing the safety level for situations with different trafc densities, either across locations, or in the same location at different times. If it were of interest whether the safety level is inuenced by time of day (daytime versus night-time), a simple comparison of the percentage of small TTCs across all vehicles would mostly reect the lower trafc volume during night-time. This might lead to the assumption that safety increases during the night. If, on the other hand, only those vehicles are considered that actually are in a car-following situation, a relative increase of small TTC values might be found during the night, which could for instance be explained with driver fatigue and extended reaction times. The last reason for restricting the analysis to following vehicles is of methodological nature. A consequence of the measurement technique used in this study is that TTC values become less reliable the longer the time headway for the vehicle in question. This is explained in more detail in Section 3. By excluding vehicles with a long headway from

the analysis, those unreliable TTC values are excluded as well.

2. Method The location of the study was a four-way junction in a mid-sized town in Sweden. The junction is one of the most accident-prone locations in town. Stop signs were placed on the subordinate road and right-of-way signs were placed on the main road. The posted speed limit was 50 km/h on all arms of the junction. All arms had one lane in each direction, except for one arm on the main road, which had a separate lane for left-turning vehicles (Fig. 1). The junction was located in the outskirts of the town, the two streets forming it were major roads leading to residential areas. The continuation of the main road after measurement site Main1 (indicated in Fig. 1) led out of town, the continuation after measurement site Main5 led to the city center. Lampposts were installed at the junction on all four arms. For 6 days in spring 2000, trafc ow point measurements were taken 24 h a day at seven locations around the junction. The weather during the measurement week was either sunny or cloudy, but there was no precipitation. The measurement devices (described in detail in Anund, 1992) record the speed of each passing vehicle, its direction, the time interval between two passing vehicles, and the axle distance of each vehicle, from which the vehicle type can be deduced to a certain extent (Srensen, 1996). The locations of the measurement devices are indicated in Fig. 1. The measuring sites Main1 and Main4, as well as

Fig. 1. Schematic overview of the junction with indication of the location of the measurement sites.

430

K. Vogel / Accident Analysis and Prevention 35 (2003) 427433

Sub1 and Sub2 were located at 115 m distance from the center of the junction, the sites Main2 and Main3 were located at 17.5 m distance from the center of the junction. Main5 was located on the main road, approximately 600 m away from Main4. There was no junction in the vicinity, except for one small and not very frequently used by-road leading to a block of houses, at a distance of about 20 m from Sub2. On site Main3 (direction out of town), left-turning vehicles could be registered separately due to the divided lane. Measurement sites were dened to be corresponding, when they were installed on equivalent locations with respect to the junction. In this sense, the locations Sub1 (direction towards junction) and Sub2 (direction towards junction) are corresponding, as well as the locations Main2 (direction away from junction) and Main3 (direction away from junction).

following vehicle was recorded at the measurement site, and ti1 being the time when the lead vehicle was recorded at the measurement site. 3.1. Trafc density Neither the subordinate nor the main road operated at capacity level any time. For both roads the number of vehicles per hour was very similar across weekdays, the pattern looked different for Saturday and Sunday, though, as there were no rush hour peaks. The increase in trafc volume during the morning and the afternoon rush hours on weekdays was more pronounced on the main road. The percentage of following vehicles lay on average slightly above 30% on the main road and somewhat below 30% on the subordinate roads. The percentage of following vehicles increased with increasing trafc volume, but at any time of day there were at least 50% free vehicles on both roads. 3.2. Relationship between time to collision and headway

3. Results Between 20,000 and 40,000 vehicles were registered at each location, depending on trafc volume. For each vehicle the driving speed, driving direction and the passing time were recorded. Each driving direction was analyzed separately for each site. Only vehicles with time headways of at most 6 s were considered in the analysis. Time headway and TTC were calculated as follows. Time headway was calculated precisely as in Eq. (1). It has to be noted that time headway usually is dened as the elapsed time between the front of the lead vehicles and the front of the following vehicles reaching the same location. In the present study, the vehicles were recorded as soon as the front axle passed the measurement site. Thus, the elapsed time between the two front axles reaching the same location is taken as approximation of time headway. For the calculation of TTC another approximation had to be made. According to Eq. (2), the numerator should be equal to the distance headway minus the length of the lead vehicle, which is the distance gap between the two vehicles. As distance headway was not directly available from the data in this study, it was derived from time headway and traveling speed of the following vehicle, based on the assumption that the vehicles traveled at a constant speed during the measurement period. Vehicle length was approximated by adding 1.80 m to the axle distance of the lead vehicle. 1.80 m is equal to the average difference between vehicle length and axle distance for passenger cars. It was felt that this approximation was acceptable, because 95% of all recorded vehicles were passenger cars. However, no vehicle types were excluded from the analysis. The denominator should be the relative speed between the two vehicles at measurement time ti . It was approximated by the difference between the speed of the following vehicle at measurement time ti and the speed of the lead vehicle at measurement time ti1 , with ti being the time when the

For each site and each direction correlations between time headway and TTC were calculated for cases with valid TTC values (speed of following vehicle larger than speed of lead vehicle). At each site, the correlation was substantial if all vehicles were considered (average correlation r = 0.423; S.D. = 0.169), but close to 0 when only following vehicles were considered (average correlation r = 0.077; S.D. = 0.047). If only following vehicles are considered, TTC and H can be regarded as practically independent of each other and can be investigated separately. 3.3. Headway For each measurement site and each direction, the percentage of measured time headways that were below 1, respectively 2 s, given all following vehicles, was calculated. The results are presented in Fig. 2. The length of the black eld shows the percentage of time headways below 1 s, and the length of the gray eld shows the percentage of time headways between 1 and 2 s. The sum of the lengths of both shows the percentage of time headways below 2 s (percentage scale in lower left-hand corner), always given all following vehicles at the site in question. As can be seen in Fig. 2, the percentage of vehicles that drive with a time headway below 2 s is relatively evenly distributed on the main road (30.3% on average, S.D. = 5.99). The percentage of following vehicles traveling with a time headway below 1 s is on average 1.5% with a S.D. of 0.95. There are no big differences between the direction towards the junction and the direction away from the junction. On the subordinate road, on the other hand, more vehicles drive with a time headway below 2 s when they are on their way towards the junction. In general, the percentage of small headways is relatively similar on the corresponding sites. A

K. Vogel / Accident Analysis and Prevention 35 (2003) 427433

431

Fig. 2. Percentage of vehicles with time headways smaller than 2 and 1 s, respectively, given all following vehicles, for each measurement site and each direction. The percentage scale is indicated in the lower left-hand corner.

comparison of the sites close to the junction (both directions on Main2 and Main3) with the sites further away from the junction (both directions on all other sites) does not show any noteworthy differences with respect to percentage of small time headway values.

3.4. Time to collision TTC values are presented in a similar fashion in Fig. 3. The length of the black rectangle indicates the percentage of vehicles that had a TTC value of less than 1 s at this

Fig. 3. Percentage of vehicles with TTC values smaller than 5, 4, 3, 2 and 1 s, given all following vehicles, for each measurement site and each direction. The percentage scale is indicated in the lower left-hand corner.

432

K. Vogel / Accident Analysis and Prevention 35 (2003) 427433

measurement site, based on all following vehicles. The length of the dark gray eld indicates the percentage of vehicles with TTC values between 1 and 2 s, and so on. Large differences with respect to the percentage of small TTC values can be found between the different measurement sites. Generally, the sites closer to the junction have a higher percentage of small TTC values, especially values below 1 s occur almost only close to the junction. Visual inspection shows that most corresponding measurement sites have relatively similar percentages of small TTC values. The corresponding measurement sites Main3 (towards the junction, split into two lanes) and Main2 (towards the junction) are an exception. Main2 has a much larger percentage of small TTC values than Main3. Moreover, the percentage of small TTC values on Main2 is considerably larger than at any other site. 3.5. Comment on the use of inferential statistics Due to the following reasons, no inferential tests were performed to examine whether any of the observed differences were statistically signicant or not: (a) the number of measurement sites was small (15), (b) the percentage values for the different time boundaries (<1, 2 s, etc.) were not normally distributed and (c) variances were not homogeneous for any of the corresponding groups.

discussed independently of each other in the following paragraphs. 4.2. Headway At all sites on the main street the percentage of following vehicles that drove with a headway of under 2 s was relatively similar at around 30%. This is an indication that drivers tend to choose a rather constant headway across situations; they do not use longer headways in more complex situations. Instead, they reduce their speed when approaching the junction (Vogel, 2001). On the subordinate road, more small headways were measured on the way towards the junction than away from it. This can be explained by the fact that approaching a junction is a very predictable situation. The stop signs or the already waiting vehicles make clear that every driver has to reduce speed, hence, drivers expect to see the brake lights on the vehicle ahead upon closing in on the junction. They are not surprised and, therefore, do not brake hastily to increase the headway, instead they continue as usual and might even accept a temporarily smaller headway before they also brake for the stop sign. When heading away from the junction, on the other hand, most drivers have only just nished accelerating and have not yet settled at their preferred headway. 4.3. Time to collision

4. Discussion At rst the drawbacks of the approach will be discussed, to enable the reader to judge the interpretation of the results. The fact that the actual vehicle length was unknown, but derived from axle length, is considered to be a minor problem, because this approximation of the vehicle length is deemed to be sufcient. Slightly more problematic is the fact that the speed of the vehicles was not measured at the same time, but in the same place, one after the other. Thus, at the time when the speed of the following vehicle is measured, the speed of the lead vehicle can have changed. As mentioned earlier, the problem was addressed by restricting the analysis to vehicles with a headway of less than 6 s, but even in this time interval speed changes are not unlikely in urban areas. 4.1. Relationship between time to collision and headway For all vehicles the correlation between H and TTC was signicant. This is not surprising, because TTC can never be smaller than H, as explained earlier. As a result, across the wide range of measured headways, vehicles with larger headways tend to have larger TTC values. If the headway range is restricted to 06 s, however, the correlation disappears. This implies that headway and TTC are independent of each other for following vehicles and therefore they yield information that is not redundant. Due to this fact, they are

Compared with headway, TTC values vary more across different locations around the junction. Generally, on the sites further away from the junction fewer small TTC values were found than on the sites closer to the junction. For the former sites, dangerous TTC values occurred with a frequency of around 0.10.3%. Less than 0.1% of the TTC values lay below 2 s. The sites close to the junction have a higher percentage of small TTC values, the smallest values occur mostly for the direction towards the junction. Especially on site Main2 the percentage of small TTC values is high; on average about every 20th of all the following vehicles has a TTC value of under 5 s. Given that approximately 30% of all 2800 vehicles that travel daily towards the junction are following, about 42 occurrences of dangerous TTC values can be expected daily in the location Main2 towards the junction. It is noticeable that the percentage of dangerously small TTC values is much lower on the corresponding measurement site Main3 towards the junction. All other corresponding sites have very similar percentages of small TTC values. Interestingly, only the corresponding sites Main2 and Main3 towards the junction have a quite different road designon Main3 there is a separate lane for left-turning vehicles, and it is thinkable that this is the reason for the much lower percentage of dangerous TTC values on this site. It is easily imaginable that the separate lane for left-turning vehicles has a large safety effect, because they usually have to wait longer in the junction and have to yield to oncoming

K. Vogel / Accident Analysis and Prevention 35 (2003) 427433

433

trafc. A left-turning lane releases both the pressure on the left-turning vehicles that arises when a queue develops behind them, and the potential anger of those who have to wait behind a blocked left-turner. If the number of left-turners is approximately equally high at site Main2 as on Main3, adding a left-turning lane could lead to an increase in trafc safety. It seems less likely that driving direction per se has an effect (going into town vs. leaving town), because no such effect can be observed on the other corresponding sites. Additionally, the junction was located well away from both the city center and the city limits.

Acknowledgements I would like to thank VINNOVA, Sweden and the Swedish National Road Administration for providing the nancial support that made this study possible. I also thank Albert Kircher for helpful comments on the manuscript. References
Anund, A., 1992. Beteendeeffekter av olika fretrdesregleringar (notat T 133). Linkping: VTI. Evans, L., 1991. Trafc Safety and the Driver. Van Nostrand Reinhold, New York. Evans, L., Wasielewski, P., 1982. Do accident-involved drivers exhibit riskier everyday driving behavior? Acc. Anal. Prev. 14, 5764. Evans, L., Wasielewski, P., 1983. Risky driving related to driver and vehicle characteristics. Acc. Anal. Prev. 15, 121136. Hayward, J.C., 1972. Near miss determination through use of a scale of danger (trafc records 384). Highway Research Board, Washington, DC. Hirst, S., Graham, R., 1997. The format and presentation of collision warnings. In: Noy, I.Y. (Ed.), Ergonomics and Safety of Intelligent Driver Interfaces. Lawrence Erlbaum, Mahwah, NJ. Hydn, C., 1987. The Development of a Method for Trafc Safety Evaluation: The Swedish Trafc Conicts Technique, Doctoral Dissertation. Lund University, Lund, Sweden. Janssen, W.H., Nilsson, L., 1991. An experimental evaluation of in-vehicle collision avoidance systems. In: Proceedings of the Paper Presentation at the 24th ISATA International Symposium on Automotive Technology and Automation. Maretzke, J., Jacob, U., 1992. Distance warning and control as a means of increasing road safety and ease of operation. In: Proceedings of the Paper Presentation at the FISITA92: Safety, the Vehicle and the Road. XXIV FISITA Congress, London. Michael, P.G., Leeming, F.C., Dwyer, W.O., 2000. Headway on urban streets: observational data and an intervention to decrease tailgating. Transport. Res. Part F 3 (2), 5564. Minderhoud, M.M., Bovy, P.H.L., 2001. Extended time-to-collision measures for road trafc safety assessment. Acc. Anal. Prev. 33, 8997. Svensson, ., 1998. A method for analysing the trafc process in a safety perspective, Doctoral Dissertation. University of Lund, Lund, Sweden. Srensen, G., 1996. System fr bestmning av fordonskoder (meddelande 762). Linkping: VTI. van Winsum, W., Heino, A., 1996. Choice of time-headway in car-following and the role of time-to-collision information in braking. Ergonomics 39 (4), 579592. Vogel, K., 2001. Day- and night-time differences in speed in an urban intersection. In: Proceedings of the Paper Presentation at the Vision in Vehicles 9, Brisbane, Australia. Vogel, K., 2002. What characterizes a free vehicle in an urban area? Transport. Res. Part F 5 (1), 1529.

5. Conclusion It was found that headway and TTC are independent of each other for following vehicles. Due to the fact that TTC values cannot be smaller than headway values, a short headway can be interpreted as potential danger, because only vehicles that travel with short headways have the possibility to produce small TTC values. A small TTC value, on the other hand, represents actual danger, because an accident can only be avoided by changing the situation actively. The two values are suitable for different purposes, because they provide different information. It is recommended that authorities use headway as criterion for tailgating, because it is easy to measure, it is easily understandable and interpretable, and most important of all, it is directed against potential danger, which effectively prevents dangerous TTC values from occurring at all. TTC values, on the other hand, should be used if the actual safety of a situation has to be evaluated. A particular road design or driver can be evaluated with respect to safety by examining the actual percentage of dangerously small TTC values within a given time frame. Similarly, it is possible to evaluate the safety of in-car systems like intelligent transport systems (ITSs) by comparing the same driver with and without the system. Trafc environments can be compared with respect to safety, and the same environment can be analyzed at different times of day. Possible re-design of a road stretch or recommendations for ITS devices can be based on empirical grounds. The method could prove to be very useful within the eld of trafc simulation, as long as the trafc model is based on accurate distributions of speed and headway.

You might also like