TRANSPORTATION RESEARCH RECORD 1172 93 Driver Deision Making at Traffi Signals RICHARD VAN DER HORST The driving task at signalized road intersetions is simplified substantially over that at nonsignalized ones. At the guid11ne level, the deision-making proess onsists mainly of the stopping/nonstopping deision at the moment that the signal hanges from green to yellow. Most red light running offenses appear to be related to this partiular moment. The results of a 1-year before-and-after behavioral study in the northern part of the Netherlands demonstrate that a yellow interval of appropriate length (4 s for SO km/hr and S s for 8 km/hr intersetions) uts the number of red light violations in half. Compared with fixed time ontrol, vehile-atuated signal ontrol leads to a 1-s shift in the probability of stopping with respet to the potential time to the stopline at the onset of yellow, lllustratlng the role of driver expetany in deision making. Through a omparison of these results with observed behavior at drawbridges and signalized railway grade rossings without gates, it is onluded that the absene of a separate yellow interval at railway grade rossings and drawbridges is disadvantageous. For all situations, a uniform stop signalization should be implemented. This signalization should onsist of a steady red signal preeded by a yellow interval that is properly designed to serve the driver adequately in normal driving as well as to meet minimal driver needs In deteriorated irumstanes. For task analysis purposes, the driving task is generally divided into three performane levels: ontrol, guidane, and navigation (1). In terms of the relationship with other traffi, the driving task at the guidane level for signalized road intersetions is onsiderably simplified ompared to that for nonsignalized intersetions. However, at signalized intersetions the road user has to deal with other tasks, suh as proessing information from traffi ontrol signals. One of the important behavioral aspets of the driver's task is the deision whether to stop or to proeed when the signal hanges from green to yellow and then to red. In an earlier study an optimal setting of the yellow interval was evaluated, and it was demonstrated that the type of signal ontrol system (fixed time versus vehile atuated) influenes driver expetany in this deision-making proess (2). The same kind of deision making is demanded when a driver approahes a drawbridge or a railway grade rossing at the onset of the red signal. The main differene in signalization in these ases, ompared with roadway intersetions, is the absene of a separate yellow interval. In this paper, by using the results of three studies dealing with driver deision making at road intersetions, drawbridges, and railway grade rossings, it TNO Institute for Pereption, P.O. Box 23, 3769 ZG Soesterberg, the Netherlands. will be argued that at least part of the problem of running red lights an be solved by applying simple measures. STOPPING OR NONSTOPPING The Traffi Rules and Traffi Signs Law in the Netherlands distinguishes among traffi ontrol signals at road intersetions, railway grade rossings, and drawbridges. In all these situations the required reation to the red signal (either steady or flashing) is quite unambiguous, namely, "stop." At road intersetions the red signal is always preeded by a yellow signal, whih, as the law states, means "Stop; but those drivers who are lose enough to the intersetion that they are not reasonably able to stop may proeed." The rationale for the yellow interval is, in priniple, to prevent vehiles that are within a given (short) distane of the intersetion from inevitably running through the red signal. At drawbridges the signal hanges from off to steady or flashing red. This hange is sometines preeded by prewaming flashing amber signals at a given distane. At railway grade rossings the signal hanges either from off (at railway rossings with gates) or from flashing white (at railway rossings without gates) to two signal heads alternately flashing red. The absene of a yellow phase at these stops is questionable, and the preise meaning of the flashing of a red signal an be questioned as well. When the signal hanges from green to yellow at a road intersetion, a deision is demanded. The riterion for this deision (the distane within whih it is not reasonable to stop) is not very preise, and for that reason it is subjet to differenes in interpretation. Although the yellow phase is absent at railway grade rossings and drawbridges, the deisionmaking proess itself is not very different. Fators that may influene this proess involve the driver's motivation and attitude, the amount of preditability of the situation, the estimate of the onsequenes of not stopping (likelihood of running red and getting a fine or of getting involved in a onflit with interseting traffi or a ollision with the grade rossing gate), and the estimate of the onsequenes of stopping (disomfort, waiting time, likelihood of a read end ollision or that braking will end in a stop within the intersetion). Evaluating these fators also involves the driver's estimates of the required deeleration on the basis of the speed and distane to the stop line and the expetations of the duration of the yellow phase and the all-red period (or the time left until the gates lose or the train arrives). All of this disussion is based on the premise that the signals are implemented and positioned to be visible in due time for the approahing driver to make a deision. Also, the behavior of
94 other road users may influene the deision making. An example is the so-alled "mitshleppeffekt," in whih a nonstopping leading automobile influenes the following automobile to ontinue as well (3, 4). SIGNALIZED ROAD INTERSECTIONS So that the deision-making proesses at road intersetions, drawbridges, and railway grade rossings an be ompared, the results of the road intersetion study with respet to the timing of the yellow signal and the type of traffi ontrol strategy will be disussed briefly. More detailed information an be found elsewhere (2, 5). Duration of the Yellow Interval Extensive literature is available regarding the timing of the yellow interval. From an explorative study on the extent of the problem of red light running (6) and from a review of the literature on measures related to traffi signal ontrol (5), it was onluded that the optimum yellow timing was 4 s of yellow for 5 km/hr intersetions and 5 s for 8 km/hr intersetions. Compared with urrent values in the Netherlands, this means that the duration of the yellow interval should be extended by 1 s to serve the driver appropriately in normal driving onditions and to meet minimum driver needs in deteriorated irumstanes. With this hange in yellow timing the number of run red offenses would be redued onsiderably, simply beause 1-s prolongation of the yellow is not expeted to hange the drivers' behavior. Furthermore, more onsistent deision making (in the sense that "emergeny-type" braking and swerving are not needed anymore) might also be benefiial to traffi safety. On the other hand, the yellow time should not be longer than reommended beause the stopping driver has to be "rewarded" with red (onfirmation of appropriate behavior) and beause an overlong yellow time might lead to greater variability in the deision making, resulting, for example, in an inrease of the number of rear end ollisions. Evidene for the latter has been found in literature on the safety effets of flashing green (3, 7). Beause the system is better adapted to "normal" behavior, enforement measures might be muh more effetive. First, the group of off enders is expeted to be muh smaller, and seond, the beginning of red no longer funtions as a "neessary" extension of the yellow. A 1-s extension of the yellow was evaluated in a 1-year before-and-after study in both urban and rural loations. All loations were provided with vehile-atuated signal ontrol. In the ity of Leeuwarden, whih has 23 signalized intersetions, the yellow interval was hanged from 3 to 4 s. The hange was made without shortening the all-red time so that possible interfering effets would be prevented. The behavior of automobile drivers was observed at four seleted intersetions and was then analyzed quantitatively from time-lapse video reordings. The ameras reorded about 15, traffi signal yles, in whih 7, "deiding" vehiles (those nonstopping plus the automobiles that stopped first after the beginning of the yellow) were registered. The extension of the yellow interval at rural intersetions from 4 to 5 s was evaluated on a rural route to TRANSPORTATION RESEARCH RECORD 1172 Leeuwarden that had seven signalized intersetions. At two intersetions the behavior of about 3, deiding vehiles was observed for about 4,5 traffi signal yles. Details about method, proedure, and quantitative analysis may be found elsewhere (2, 8). One year after the timing was hanged, the number of runred offenses at urban intersetions appears to have been halved, from 1.1 to.5 perent of the total number of vehiles and from 13.4 to 6.7 perent of the number of deiding vehiles. At rural intersetions the same redution was ahieved. The proportion of nonstopping automobiles with a stop line-passing time (TPS) greater than or equal to a given time t after the onset of yellow (Figure 1) shows a small shift of about.13 s for the urban intersetions (Dmax = 6., Kolmogorov-Smimov test, one-tailed, p <.1). Results for the "between" period (6 months after the extension of the yellow) differ by this same amount from the "before" period, hut the helween period does not differ signifiantly from the 1 yr after period. At rural intersetions, no signifiant differene was found between the before and 1 yr after periods.., D "'Cit :' a.- g. -.. "'.'.'.:. - en Cl. :... -:, C> 1 BO 6 : ' - g Al 4 2 t yellow 2 4 -before nl3bo o-o one year ofter n: 1494 5 6 7 Time ofter yellow,1(se) FIGURE I Proportion of the nonstopplng ars with TPS t after the onset of the yellow, for before and 1 yr after the extension of yellow at urban Intersetions. By inluding the information on the stopping drivers, the probability of stopping with respet to the potential time to the stop line (TIS) was alulated, taking into aount the individual approah speed of eah vehile and assuming that vehiles will ontinue with a onstant speed. A log-linear model fit resulted in a shift of.17 s between before and after situation. This (relatively small) hange in driver behavior appears to be present for the after 6 months period as well and does not hange thereafter. From this field study it is onluded that a 1-s extension of the yellow time (from 3 to 4 s in urban areas and from 4 to 5 s in rural areas) uts the total number of run-red offenses in half, although a small adaptation effet was also observed. However, after 6 months, no further hange in driver behavior was found. The results are ompletely in agreement with the expetations of a simple model for alulating the required yellow time. Type of Traffi Signal Control It might be expeted that different types of traffi signal ontrol will differ in terms of exposure to potential offenses and will also perhaps differ in the proess used by the drivers to deide
van der Horst whether to stop or to proeed. If a vehile-atuated ontrol strategy is used, the green phase, in priniple, will be extended as long as there are vehiles in a given detetion area. When this ontrol strategy operates well, it will result in muh fewer potential red runners than does fixed time yle ontrol, in whih the times seleted for ending the green are independent of the traffi at the moment. Zegeer and Deen (9) found a large redution in the number of run red offenses when a green extension system was applied. The redution was due mainly to the differene in exposure, but what about the differenes in the disipline at the red light at the moment that the deision is required? The expetations of automobile drivers approahing the traffi light might differ for different types of ontrol, resulting in different deision behavior in general, regardless of the duration of the yellow. In the field study on the effets of yellow timing that was desribed in the previous setion, all the traffi signal ontrols were vehile atuated. In Figure 2, the probability of stopping at intersetions with vehile-atuated ontrol is ompared with data on fixed time ontrol, available from the literature (112). All data sets are based on field studies, exept those of Mahalel et al. (1). They onduted a laboratory experiment on the deision-making behavior of automobile drivers at traffi lights. A remarkable 1-s shift exists between the probability of stopping for a fixed yle time ontrol and that of stopping for a vehile-atuated ontrol. Evidently, a vehile-atuated ontrol leads to a shift in the riterion that drivers use. The drivers deide to proeed in an earlier stage of the approah proess, when ompared with the deision made for fixed-time ontrol. In the field studies, it was observed that the harateristis of the deision proess itself are not different, as is indiated by the similarity of the slopes of the urves at a.5 probability of stopping. The slope of the laboratory experiment at this point tends to be different, indiating a somewhat deviant behavior in more artifiial irumstanes (1). "' a. 2 1.. 8.. 6 -. 4 :.:; &l. 2 n.,.- -: _. v..,,._....... '1/ _. "' I/ Fixed - lime ;; " //..... liol-..._ ' It. 4 - -(II)... I o-(12),:/ Veh i le - oluoled... /....... / /J I --- (13)... "/,/ -(Bl 2 3 4 6 7 Time lo stop line, TTS l se.) FIGURE 2 Probability of stopping as a funtion of the time to stop line (TTS) for fixed time (1-12) versus vehile-atuated (8, 13) signal ontrol systems. Given this dependeny of behavior on the type of traffi signal ontrol, the hypothesis an be formulated that automobile drivers who are austomed to vehile-atuated ontrol will adapt their behavior and expet to see a response when approahing the traffi signal during green. Contrary to the fixed time ontrol situation, the drivers expet that the green interval will ontinue. Therefore, when the green phase ends 8 (that is, the maximum green extension period has been reahed), their expetany is violated, with onsequenes for their deision-making proesses (1). It is expeted that similar effets will our in other situations. For example, in a string of interonneted signalized intersetions (progressive signal ontrol systems), the experiene at previous intersetions auses the expetany that the driver will have or get the green interval at the next one as well. To onlude, in spite of a negative effet on the deisionmaking behavior of automobile drivers per se, a vehileatuated traffi ontrol system redues the number of run-red offenses substantially in omparison with fixed time ontrol. This redution is mainly aused by a redution in exposure, however. RED SIGNALS AT DRAWBRIDGES In the Netherlands, drawbridges are ommon. The operation of these bridges, espeially on roads with heavy traffi flows, is sometimes problemati. On a freeway, for example, bridgekeepers sometimes have problems onduting the opening proedure beause drivers persist in proeeding. The proedure is automati one started by the bridgekeeper and onsists of prewarning flashing signals (ombined with a sign showing a drawbridge) loated 9, 6, and 3 m before the bridge (the flashing starts suessively about 3 s before the red signal), the onset of alternating flashing red lights above the stop line on the ramps of the bridge, the losing of the gates (8 s after red), and finally the opening of the bridge. The bridgekeeper an interrupt this proedure as needed if the desending gate threatens to hit a proeeding automobile. Aording to omplaints by the bridgekeepers, 1 out of 5 or 1 losures has to be interrupted (14). Why don't the drivers stop? In this ase, it is not a matter of poor visibility of the signals or the gates. One assumption is that motorists, beause they are familiar with this system and have visual ontat with the bridgekeeper, know that the bridgekeeper is able to interrupt the proedure. The interation between motorist and bridgekeeper and their mutual antiipation results in a nonstopping motorist. However, hiding the bridgekeeper behind one-diretional urtains was not an effetive ountermeasure (14). From video reordings, the behavior of the nonstopping automobiles was analyzed in more detail. Reordings of 82 losures with 348 nonstopping automobiles after the onset of the red signal were analyzed Figure 3 gives the proportion of the nonstopping automobiles with a TPS greater than or equal to a given time t after the onset of red. It an be seen that 12 perent of the nonstopping automobiles pass the stop line after the gates begin to lose (8 s after the onset of red). If the behavior at this drawbridge is ompared with that at signalized rural intersetions with similar approah speeds, it is obvious that motorists make the deision to proeed muh earlier at the drawbridge. On the basis of a speed of 1 km/hr, a reation time of 1 s, and an average deleration rate of 3 mjs2, a learane interval of 5.6 s would be suffiient. Yet 3 perent of the nonstopping automobiles pass the stop line later than this, ompared to 6 perent for rural intersetions. Even with a speed of 14 km/hr, drivers would be able to stop before the stop line at a time distane of 7.5 s. Yet 14 perent of the 95
TRANSPORTATION RESEARCH RECORD 1172 96 u 1.,,it' I a.-: 8 -':'-.. o.u.. - 6 g Al -IJ).. Drawbridge Rural intersetions Urban inlersetion1 Railway grade roninq 4... -.;.s:. -- Cl.. e 2......_,.. 2 4 6 8 1 12 14 Time after ional hange, t (se l FIGURE 3 Proportion of nonstopplng automobiles with TPS t after the onset of the yellow (for road Intersetions) or red (for drawbridges and railway rossings) signal. nonstopping automobiles pass later. It appears that drivers are not willing to stop, rather than not able to stop. Given our knowledge of motorist behavior at signalized intersetions, however, the funtioning of the drawbridge signalization itself is questionable. About 3 s before the red signal is ativated, the prewarning signal at 9 m starts flashing. Even when the prewarning signal is pereived, the unertainty about the exat time that the red will appear is great beause of the long interval (3 s) that passes with no omplementary message to the motorist. The signaling system is in the same state at the end of the yle, when the gates have been opened again. At the end of the yle the warning lights are used as a warning for the queue of standing ars. In itself, this measure is effetive to an extent, but it devalues the main funtion of these lights. Approahing drivers might be unertain about whether the beginning or the end of the yle is indiated. At a speed of 1 km/hr, motorists who have just missed the warning signal will reah the stop line after 32.5 s. At that moment, the red signal is on for 2.5 s. Logially, it annot be expeted that those motorists will deide to stop at the onset of the red signal. They will proeed and inevitably run the red. Thus the meaning of the red signal is rather ambiguous. Following drivers see those automobiles in front running a red light. Beause their situation is only marginally different (often heavy traffi flows with small headways), they are easily tempted to proeed as well (the previously mentioned mitshleppeffekt). A separate yellow signal, preeding the red signal and ombined with a muh shorter warning time, might prevent a lot of these situations. RED SIGNALS AT RAILWAY GRADE CROSSINGS Reently, a study was onduted on the deision-making behavior of motorists who were approahing railway grade rossings that had signal ontrol and were without gates (15). At this type of railway rossing, the yellow interval is absent: The signal hanges diretly from the safe situation (flashing white) to the unsafe situation (alternating flashing red). This is another situation in whih road users may not able to stop and therefore neessarily proeed through red. The behavior of 66 motorists at the onset of the red signal was analyzed at two railway rossings. In ontrast with the behavior at the drawbridge, motorists at this type of railway rossing are more willing to stop than those at signalized urban intersetions with omparable approah speeds (Figure 3). The probability of stopping at a given time distane appears to be higher at railway grade rossings, even in omparison with the data on fixed time ontrol. At railway grade rossings the lak of a yellow phase seems to work in the opposite diretion; that is, the obediene to the red light is better than at signalized intersetions. However, a detailed analysis of the stopping vehiles indiated that some of the stops were realized only with a high deeleration rate (> 4 rn/s2), and a few of them ended very lose to the first railway trak. Suh "emergeny" stops an easily lead to a stop _on the trak itself or to a rear end ollision by a following automobile. Beause of these risks, this kind of behavior is not desirable. It an be prevented by a separately signalized learane interval with a yellow indiation. CONCLUSIONS AND RECOMMENDATIONS At signalized road intersetions, an extension of the yellow time by 1 s resulted in the number of run red offenses being ut in half. In the literature, evidene was found that 4 s of yellow for 5 km/hr zones and 5 s for 8 km/hr zones are optimum values. Longer times are not desirable beause of expeted seondary effets. The omparison between behavior at vehile-atuated and fixed time ontrol systems indiates that motorists use expetations about the funtioning of the signals and at aordingly. The relatively small number of run red offenses at vehileatuated signals is mainly due to a low exposure of deiding vehiles. Disipline at the red light per se appears to be less than for fixed time signals, probably beause of a differene in expetany. Although the willingness of motorists to stop at drawbridges seems to be poor, at least some of the offenses an be explained by operational defiienies of the signal system itself. The lak of a separate indiation for the learane interval is a major one. By introduing a yellow interval, only one interpretation of the red signal is possible, namely "stop." At railway grade rossings the absene of the yellow interval appears to have an opposite effet. The sudden onset of the red signal sometimes auses a kind of a pani reation, with a very abrupt stop. This is espeially the ase with motorists who are so lose to the railway trak L.'iat proeeding would have been more appropriate. Although the obediene to the red light is better at grade rossings than at signalized intersetions, large deviations from normal behavior our. A separate yellow signal would therefore be advantageous at railway grade rossings as well. Beause the same behavior is demanded from the road user in all three examples given, one uniform signalization for stopping is reommended: a steady red signal preeded by a yellow interval that is properly designed to serve the driver adequately in normal driving as well as to meet minimal driver needs in deteriorated irumstanes. Speifially, 4 s of yellow for 5 km/hr zones and 5 s for 8 km/hr zones are reommended. ACKNOWLEDGMENTS This researh was supported by the Transportation and Traffi Engineering Division and the Road Safety Diretorate, both of the Duth Ministry of Transport.
va11 der Horst REFERENCES 1. G. J. Alexander and H. Lunenfeld. Dri.ver Expeta11y in Highway.Design and Traffi Operations. Report FHWA-T-86-1. F.1-JWA, U.S. Department of Transportalion, May 1986. 2. A. R. A. van der Horst and A. Wilmink. Drivers' Deision Making at Signalised Intersetions: An Oplimisation of the Yellow liming. Traffi Engineering and ConJrol, Vol. 27, No. 12, Deember 1986, pp. 615-622 3. H. Knoflaher. Der Einfluss des Griinblinkens auf die Leistungsfiihigkeit und Siherheit Lihtsignalgeregelter Strassenkreuzungen. Strassenforshung, Vol. 8, Bundesanslllll flir Bauten und Tehnil, Vienna, 1973. 4. W. Kokelke. Reaktions- und Bremsvrhalten an Lihtsignalanlagn. Strassenverkehrstehnik, Vol. 1, 198, pp. 11-16. 5. A. R. A. van dr Horst and J. GodLhelp. De roodlihtdisipline van bestuurders van motorvoertuigen in. relatie lo het beiiindige11 van de groenfase: een literatuurstudie. Report lzf 1982 C-11. TNO Instirnte for Pereption, Soesterberg, the Netherlands, 1982. 6. K. J. Poll and A. R. A. van der Horst. De roodlihldisipline van bestuurders van motorvoertuigen: een verkennend ondenoek. Report IZF 1981 C-12. TNO lnstirute for Pereplion, Soesterbrg, the Netherlands, 1981. 7. D. Mahall and. M. Zaidel. Safely Evaluation of a F1ashing Green Light in a Traffi Signal. Traffi Engineering and ConJrol, Vol. 26. No. 2, February 1985, pp. 79-81. 8. A. R. A. van der Horst, W. ten Broeke, and P. J. Bakker. De roodlihtdisipline van bestuurders van mo/orvoerluigen onder invloed van een aanpassing van de geeltijd. Report IZF 1985 C-11. TNO ln.stitule for Pereption, Soes1erbrg, the Netherlands, 1985. 9. C. V. Zegeer and R. C. Deen. Green Extension Systems at Hlgh Speed Intersetion. l!'e Journal, Vol. 48, No. 11, November 1978, pp. 19-24. 1. D. Mahalel, D. M. Zaidel, and T. Klein. Driver's Deision Proess on Tennination of the Green Light. Aident Analysis and Prevention, Vo.I. 17, No. 5, 1985, pp. 373-38. 11. W. L. Williams. Driver Behavior During the Yellow Interval. In Transportation Researh Reord 644, TRB, National Researh Counil, Washington, D.C., 1977, pp. 75-78. 12. F. R. Hulsher. The Problem of Stopping Drivers After the Termination of the Green Signal at Traffi Lights. Traffi Engineering and Conirol, Vol. 25, No. 3, Marh 1984, pp. 11--116. 13. Y. Sheffi and H. Mahmassani A Model of Driver Behavior at High-Speed Signalized Intersetions. Transportation Siene, Vol. 15, No. 1, February 1981, pp. 5--61. 14. W. H. Janssen. Voorlopige resultaten van het ondenoek op de vehjbrug. Internal memo. TNO Institute for Pereption, Soesterberg, the Netherlands, 1985. 15. B. Tenlink, A. R. A. van der Horst, P. J. Bakker, and W: ten Broeke. Gedrag bij AKJ spoorwegovergangen. Report lzf 1987 C-25. TNO Institute for Pereption, Soesterbrg, the Netherlands, 1987. Publiation of this paper sponsored by Committee on Vehile User Charateristis. 97