Evaluation of Driver Behavior at Signalized Intersections

Transcription

1 10 Transportation Research Record 904 Evalation of Driver Behavior at Signalied Intersections ROBERT H. WORTMAN AND JUDSON S. MATTHIAS Time-lapse photography as sed to stdy driver behavior associated ith the traffic signal change interval at a total of six intersections in the Phoenix and Tcson metropolitan areas. In addition, nighttime stdies ere condcted at to of these intersections. An evalation of the time-lapse film permitted the determination of the approach speeds of the vehicles, the average deceleration rates of the stopping vehicles, the perception-reaction times of the drivers of the stopping vehicles, and the distance that the vehicle as from the intersection at the onset of the yello interval. The distance from the intersection as measred for both the stopping vehicles as ell as those that proceeded throgh the intersection. The reslts of the stdy indicated that the mean deceleration rates at the six sites ranged from 7.0 to 13.9 ft/s/s, and the mean vale for all observations as 11.6 ft/s/s. The observed mean perception-reaction time as approximately 1.3 s hile the 85th percentile times ranged from 1.5 to 2.1 s. Comparisons of intersections ith yello only verss yello pls all-red inter vals prodced mixed reslts in terms of differences in observed behavior. Even for intersections ith the same change interval design, there ere cases here the observed deceleration rates ere significantly different. The failre to properly recognie and nderstand driver behavior in signalied-intersection design can contribte to operational and safety problems. A revie of 1980 traffic accidents in Ariona reveals that approximately 3 percent of the reported accidents had disregard of a traffic signal listed as a contribting circmstance (1.). The recognition of driver behavior is particlarly critical in the design of the traffic signal change interval. The change interval has to recognied prposes. The first is to advise the motorist that the red interval is abot to connence, and the second is to allo vehicles that have legally entered the intersection sfficient time to clear the point of conflict prior to the release of opposing pedestrians or vehicles. The determination of the change interval considers driver perception-reaction time and vehicle deceleration rates; the se of nrealistic vales for these factors old potentially affect driver compliance, safety, and even intersect ion capacity. In the eqations sed in the determination of the dration of the change interval, a perception-reaction time of l s and a deceleration rate of 10 ft/s/s are sggested {. ). The latter vale has been decreased from 15 ft/s/s, hich as sed in practice for a nmber of years. The crrent Ariona Department of Transportation policy <1> indioates that 8 and 12 ft/s/s are the pper and loer limits for deceleration rates that are sed in establishing the change interval dration. This range provides some degree of latitde in applying engineering jdgment to the determination of the change intervali hoever, the policy frther states that 10 ft/s/s shold normally be sed. In recent years, several stdies have focsed on driver behavior dring the yello signal interval. For example, Williams (_!) reported that a stdy of an intersection in Ne Haven, Connectict, revealed that the average maximm deceleration rate for stopping vehicles as 9. 7 ft/s/s. Other stdies <iril have sggested that a reasonable deceleration rate old be in the magnitde of 10 ft/s/s. To previos stdies inclded field measrements of driver reaction time. In an early stdy of the change interval problem, Gais, Herman, and Maraddin (7) fond a mean reaction time of 1.14 s based on a -sample of 87 observations. A stdy in 1966 by Jenkins (8) contained a sample of 21 observations ot reaction time. An analysis or the data sample obtained by Jenkins reveals that the mean reaction time as approximately 1.4 s. The Transportation and Traffic Engineering Handbook (. ) states that excessively short or long yello intervals are ndesirable; ths, the connon practice is to se yello intervals of 3-5 s. If a longer dration is reqired to clear the intersection before the cross traffic enters, an all-red period can be sed in addition to the selected yello interval. The Ariona Department of Transportation policy <ll indicates that a yello interval of p to 6 s may be sed. If a longer change interval is reqired, then an all-red interval shold be sed. In Ariona, as in most states, it is legal for the vehicle to enter the intersection dring the yello signal indication. It is not necessary for the vehicle to have cleared the intersection prior to the onset of the red signal indication. The prpose of this stdy focsed on examining driver behavior that is related to the change interval. The intent as to docment driver behavior parameters and possible variations in behavior that exist in Ariona. More specifically, the stdy objectives focsed on the determination of the folloing items: 1. The actal deceleration rates and the range of deceleration rates that are sed by drivers, 2. Possible differences in driver behavior de to the intersection environment, and 3. The effect of the se of the all-red phase on driver behavior1 in addition, the stdy provided information abot the perception-reaction times of drivers as ell as measres of driver compliance in terms of the signal indications. It shold be noted that it as not feasible to condct before-and-after stdies of the effects of changes in the change interval design and dration at a particlar intersection. The stdy as limited, therefore, to examining behavior at existing intersections ith existing signal timing. Driver behavior associated ith the change interval is certainly a complex phenomenon in that there are nmeros facets to the overall problem. It as not the intent of this particlar stdy to inclde an evalation of the crrent practice sed in the determination of change intervals. METHOD OF STUDY Time-lapse photography as sed to record the daytime driver behavior associated ith the change interval at six intersections. To of the six intersections ere also observed dring the nighttime period. Vehicles approaching the intersection ere filmed for a fe seconds prior to the onset of the yello, dring the change interval, and ntil the vehicle either stopped or cleared the intersection. Given the onset of the yello signal indication, the stdy focsed on the first vehicle to stop and the last vehicle to pass throgh the intersection. Where mltiple approach lanes ere involved, this information as inclded for each lane. It as not possible to determine sch information sch as the age, sex, experience, and rote familiarity of the driver. The camera as located so that it as possible to record the intersection imd lhe signal indication ae ell as the operation of approaching vehicles ithin ft of the intersection. A sper 8-mm movie

2 Transportation Research Record camera ith a oom lens and a 16-mm movie camera ere sed for data collection. Becase of the age of the 16-mm eqipment, it proved to be less reliable and more expensive to operate than the sper 8-mm niti ths, it as sed only on a limited basis. Also, higher-speed film for the night observations as readily available for the sper 8-mm camera i therefore, that camera as sed for all of the nighttime stdies. The cameras ere operated at a frame speed of 18 frames/s to increase the accracy of the time measrements. The day filming as accomplished by sing a Kodak Kodachrome color film. This made it possible to easily distingish the signal indications as ell as hen brake lights ere illminated on a vehicle. For the night filming, a Kodak Ektachrome color film as sed. Althogh it as not possible to see the entire vehicle at night, the lights on the vehicle and the traffic signals ere recorded on the film. If available, the intersection as filmed from a nearby bilding or strctre. If sch a facility as not available at a particlar location, the Ariona Department of Transportation frnished a trck ith an elevating platform that old extend to a height of approximately 30 ft. The camera as placed on a tripod, hich reslted in a total height of abot 35 ft. The trck as parked in a parking lot or vacant area approximately ft from the intersection. Where the trck as employed for filming, it as located abot ft from the edge of the roaday. There as concern that the presence of the trck might inflence the behavior of the drivers. Based on observations by the stdy teams at the sites, there as no evidence that the drivers ere cogniant of the filming activities. Distances from the intersection ere noted by sing reference points on the roaday. Generally, strips of tape ere placed at 50-ft intervals along the edge of the approach. In some cases, the dashed-lane striping as measred and sed as the reference for distance from the intersection. Site Selection The folloing six intersections in the Phoenix and Tcson metropolitan areas ere selected for stdy: 1. Phoenix metropolitan area: University Drive and Rral Road (located in Tempe), Sothern Avene and Mcclintock Drive (located in Tempe), and US-60 and Greenfield Road (located in Mesa) i and 2. Tcson metropolitan area: First Avene and Roger Road, Sixth Street and Campbell Avene, and Broaday Bolevard and Colmbs Bolevard. Only one approach at each of the intersections as stdied, and the stdy approach as located on the first street that is shon in the list of intersections. Table 1 smmaries the characteristics of each of the intersection approaches that ere observed. Major difficlties ere encontered in the selection of intersections that ere to be sed for stdy observations. These difficlties served to severely limit the choice of intersections that cold even be considered. Even ith the option of placing the camera in a nearby bilding or on a trck ith an elevating platform, it as freqently difficlt to find a sitable bilding or an acceptable place to locate the trck. Also, if the traffic signal at an intersection is operated as part of an areaide system, approaching vehicles freqently may not be in proximity to the intersection at the onset of the yello interval. The same may be tre for signalied intersections in otlying areas that have traffic-actated controllers. When sch conditions exist, the probability of obtaining a data sample is relatively small, and the time reqired for data collection is greatly increased. Jrisdictions are relctant to temporarily isolate a signal from the control system becase of liability qestions. Filming of Intersection Approaches Becase of the arrangements that ere necessary for access to bildings or to se a trck ith a platform, filming activities ere accomplished in periods of several hors dration. For example, a stdy team old freqently film an approach for a 6- to 8-h period. For the night observations, filming as ndertaken dring the months of December and Janaryi ths, data collection generally began abot 6:00 p.m. and as terminated abot 11:00 p.m. It took abot h of filming at each site to obtain a sample sie of approximately 100 stopping vehicles. Even ith this filming period, there ere some intersection approaches that yielded less than 100 samples hen the film as analyed. It might seem that this dration of filming is somehat excessive for sch a sample siei hoever, there are to explanations. First, cycle lengths ere sometimes as high as 120 s. Second, there ere nmeros cases for hich there ere no vehicles in the proximity of the approach at the onset of the yello interval. Becase of the dration of filming activities, it as not possible to focs on particlar periods of time dring the day. The observations that ere made, therefore, represent a cross section of the traffic conditions at an intersection. Data Redction By sing the distance reference points that ere Table 1. Characteristics of stdy sites. Site Chnnge Estimated lnlcrvalb ADT" (s) Approach Configration Left-Trn Signaliation University Drive* and Rral Road Sothern Avene* and McClintock Drive US-60* and Greenfield Road First Avene* and Roger Road Sixth Street* and Campbell Avene Broaday Bolevard* and Colmbs Bolevard S 800 SY SY SY+3AR 3Y+2AR To lanes pls exclsive left-trn lane Three lanes pls exclsive left-trn lane Three lanes pls exclsive left-trn Jane To lanes pls exclsive left-trn lane 3Y + 2AR To lanes pls exclsive right- and lefttrn lanes 3.6Y+ 2AR Three lanes pls exclsive right- and left-trn lanes Exclsive left-trn phase Exclsive left-trn phase Exclsive left-trn phase Trns permitted on a permissive basis dring throgh movement Trns permitted on a permissive basis dring throgh movement except in peak hors hen lefttrn lane is sed as a reversible lane Exclsive left-trn phase pls permissive left trns dring throgh movement Note: ADT =average daily traffic, and denotes street on hich the observed approach as located. ADT is for the street on hic h lh e c1 b se rv d approach as located. Dci nqtes design and draholi o f c h i..nga lntfj rv1. For example, SY+ 3AR indkates 5 s of yello pls 3 s of ah-rd time.

3 12 Transportation Research Record 904 Table 2. Stdy site speed characteristics. Approach Speed Last Vehicle Throgh the Intersection Posted Speed Limit Sample Mean Speed Standard Intersection Approach (mph) Sie (mph) Deviation University Drive Sothern Avene Day Night US First Avene Sixth Street Broad ay Bolevard Day Night All approaches First Vehicle to Stop 85th 85th Percentile Sample Mean Speed Standard Percentile (mph) Sie (mph) Deviation (mph) Table 3. Distance from intersection at beginning of yello interval. Last Vehicle Throgh Intersection First Vehicle to Stop Mean Mean Intersection Distance Standard Distance Standard Approach (ft) Deviation (ft) Deviation University Drive Sothern Avene Day l Night US First Avene Sixth Street Broad ay Bolevard Day Night All approaches established on each of the stdy approaches, a grid as developed so that the location of a vehicle cold be determined hen the film as projected on a screen. The grid indicated the distance from the intersection. In all cases, this distance as measred from the crossalk. For each of the vehicles that ere the first to stop after the beginning of the yello interval, the folloing information as extracted from the film record: 1. Distance from the intersection at the beginning of the yello interval, 2. Location of the vehicle hen the brakes ere applied (as indicated by the brake lights), 3. Location of the vehicle hen it stopped, 4. Time reqired for the vehicle to stop, 5. Perception-reaction time (determined as the time beteen the beginning of the yello interval and the application of the brakes), and 6. Type of vehicle if other than a passenger car or light trck. Based on this information, the approach speed and the average deceleration rate as compted for each stopping vehicle. In addition, the behavior of the last vehicle to pass throgh the intersection after the beginning of the yello as determined by making the folloing observations: 1. Location of the vehicle at the beginning of the yello interval, 2. Time elapsed from the onset of the yello interval ntil the vehicle entered the intersection, 3. Type of vehicle if other than a passenger car or light trck, and 4. If the vehicle entered the intersection on the red signal indication. For the vehicles that did not stop, the approach speed as also compted. In this case, the determination of speed as accomplished by sing the elapsed time and the distance traveled. RESULTS Table 2 indicates the observed speed characteristics at the stdy sites. At all six stdy locations, the mean approach speeds of the last vehicle throgh the intersection ere higher than the first vehicle to stop. It shold be recognied that the measrement of the speed of the first vehicle as made at a point here some deceleration may have occrred. Also, speed measrements for the throgh vehicles ere made before the onset of the yello interval, and some of those vehicles may have been accelerating. It as not possible, hoever, to determine if these vehicles ere accelerating from the films that ere made. Althogh the posted speed limits ranged from 30 to 50 mph, there as mch less variation in the vehicle approach speeds. The difference beteen the high and lo mean speed as less than 6 mph for the last vehicle throgh the intersection and less than 8 mph for the first vehicle to stop. This indicates that there as not that mch diversity in the vehicle operations hen comparing the sites. With respect to the comparison of the intersection approaches here the day and night stdies ere condcted, there as no significant difference in the approach speeds for the first vehicle to stop. The approach speeds ere significantly higher at night for the throgh vehicles in the case of the Sothern Avene site and ere significantly loer at night at the Broaday Bolevard site. It might be expected that the nighttime approach speeds old consistently be higher becase of the loer volmes. Table 3 smmaries the mean distance from the intersection at the beginning of the yello interval for the to grops of vehicles at each site. As old be expected, the distance from the intersection as considerably less for the last vehicle throgh the intersection. In fact, the mean distance differences ranged from approximately 100 to 150 ft for the stdy approaches. Figre 1 depicts the cmlative freqency distribtions of the distance from the intersection for all sites. A smmary of the observed perception-reaction

4 Transportation Research Record Figre 1. Cmlative freqency 100 r r ,, :::::;;;iii ::=o;_., --, distribtion of vehicle location at onset of yello interval. 80 LAST VEHICLE THROUGH "' a: >- ::> r;r a:. "' ::>... :'5 ::> :.: ::> QL-... :;..-= J' L ' ' DISTANCE FROM INTERSECTION (FEET) times is given in Table 4. As indicated previosly, the perception-reaction time as determined by measring the time beteen the onset of the yello interval and the application of the brakes. These times are for the stopping vehicles only. Generally, the stdy team as nable to determine the perception-reaction time for the drivers ho chose to proceed throgh the intersection. There ere a fe cases dring the night stdies here it as possible to detect that drivers began to apply the brakes bt then proceeded throgh the intersection. In these cases, a brief flickering of the brakelights as recorded on the film. The cmlative freqency distribtion crve for all approaches is shon in Figre 2. It as theoried that the perception-reaction time is related to sch factors as the location of the vehicle at the beginning of the yello interval, the approach speed, or even possibly the deceleration rate that as sed by the driver. These hypotheses ere tested by sing correlation and regression analyses and compting the coefficient of determination (r 2 ). When these variables ere analyed in terms of the relation ith the length of the perception-reaction time, the folloing r 2 - vales ere obtained: Variable Distance from intersection Approach speed Deceleration rate r These reslts old serve to indicate that there as little relation beteen the perception-reaction time and these variables. Table 5 smmaries the deceleration rates that ere observed at the stdy locations, and the cmlative freqency distribtion of the deceleration rates is shon in Figre 3. It shold be emphasied that these observed vales reflect driver behavior nder the set of conditions experienced at the stdy sites. Frther analysis of the variation in the deceleration rates ith respect to the stdy sites is discssed later. The tablation of the vehicles entering the intersection on the red signal indication is given i'n Table 4. Perception-reaction times. Intersection Mean Time Standard 85th Percentile Approach (s) Deviation Time (s) University Drive Sothern Avene Day Night US First Avene Sixth Street Broad ay Bolevard Day Night All approaches l Table 6. Generally, the intersections ith the shorter yello interval and the all-red phase reslted in a higher percentage of vehicles entering on the red indication. This old be expected de to the shorter yello dration. The First Avene site had an extremely high percentage of vehicles entering on the red signal compared ith the other locations. Althogh this intersection is sitated in a more otlying area than the other to stdy locations in Tcson, there is no clear explanation for this percentage being so mch higher than the other sites. When comparing the variation beteen the day and night conditions, the percentage of entering vehicles as drastically redced dring the night conditions. This may be partly explained by the fact that there ere less vehicle qees dring the nighttime period. The deceleration rate at the Broaday Bolevard site also decreased dring the nighttime period. COMPARATIVE ANALYSIS OF STUDY SITES The second major portion of the analysis effort dealt it:h the comparison of observed behavior at the different stdy sites. Basically, these comparisons ere intended to test the inflence of the driving environment and signal timing practices. The mean and standard deviation for each of the

5 14 Transportation Research Record 904 Figre 2. Cmlative freqency distribtion of observed perceptionreaction times. ;:::- f:'i "' <l. >- "'. > >-- :5 >: PERCEPT I ON-REACT I ON TI ME (SECONDS) Table 5. Observed deceleration rate. Mean Change Approach Mean Rate Standard 85th Percentile Intersection Approach Interval 8 Speed (mph) (ft/s/s) Deviation Rate (ft/s/s) University Drive 5Y I 1.5 Sothern Avene Day 5Y Night 5Y US-60 5Y+ 3AR First Avene 3Y + 2AR Sixth Street 3Y + 2AR Broaday Bolevard Day 3.6Y + 2AR Night 3.6Y+ 2AR All approaches e Denotes design and dration of change interval. For example, SY + 3AR indicates 5 s of yello pls 3 s of all-red time. Figre 3. Cmlative freqency distribtion of observed deceleration rates. 100 r----t----t-----,-----,-----t-:::;;;;;;;-----., ;:::- "' 60 >- O' "'. 40 > >-- <[ -' >: o DECELERATION RATE ( fps 2 )

6 Transportation Research Record Table 6. Percentage of last vehicles throgh intersection on red indication. Sample Sie of Last Vehicle Percentage Change Throgh Entering on Intersection Approach Interval" Intersection Red Indication University Drive Sothern Avene Day Night US AR First Avene 3Y+ 2AR Sixth Street 3Y + 2AR Broaday Bolevard Day 3.6Y+ 2AR Night AR a Denotes design and dration of change interval. For example, SY+ 3AR indkates S s of yello pls 3 s of all red time. Table 8. Comparison of Tcson area sites (day observations). Sites Compared First Sixth Street Avene and First and Broaday Avene and Broaday Item Bolevard Sixth Street Bolevard First vehicle to stop Approach speed s s s Perception-reaction time N s s Distance from intersection s s s Deceleration rate N s N Last vehicle throgh intersection Approach speed N s s Distance from intersection N s s Note: N = dhference is not significant, and S =significant difference. Table 7. Comparison of Phoenix area sites (day observations). Sites Compared US-60 US-60 Sothern and and Avene and University Sothern University Item Drive Avene Drive First vehicle to stop Approach speed N N N Perception-reaction time N N N Distance from intersection N N N Deceleration rate s N s Last vehicle throgh intersection Approach speed s s N Distance from intersection N N N Table 9. Comparison of day and night behavior. Site Sothern Broaday Item Avene Bolevard First vehicle to stop Approach speed N N Perception-reaction time N N Distance from intersection N N Deceleration rate N s Last vehicle throgh intersection Approach speed s N Distance from intersection N N Note: N = difference is not significant, and S = signjficant difference. Note: N =difference is not significant, and S =significant difference, measred parameters ere compted. Where data from intersections ere combi ned for a particlar analysis, t hese vales ere compted for the combined data. Potential differences in behavior ere analyed by examining differences in the means for specific grops or pairs. In all cases, the 95 percent confidence level as sed for the prpose of assessing statistical significance. Table 7 gives a smmary of the reslts for a comparison of the sites in the Phoenix metropolitan area. For this analysis, a particlar site as compared ith each of the other sites in that metropolitan area. Note that there as a statistical ly significant difference in the deceleration rate at the University Drive site even thogh there as no significant difference in the other parameters for the first vehicle t o s top. For the last vehicle throgh the i ntersection after the beginning of the yello interval, t he re e r e t o cases here the differences in the approach speeds ere significant. A similar comparison of the sites in Tcson yielded a somehat different set of reslts (Table 8). Foe this grop of intersections, the comparison revealed significant differences in a nmber of measres of behavior. Table 9 gives the reslts of the comparison of the day and night stdies at the to sites selected for that prpose. In this case, the observed day and night behavior at each of the to sites ere compared. Again, there ere some differences beteen the to stdy sites. As as previosly indicated, the observed deceleration rates ere significantly loer at night for the Broaday Bolevard location. The Sothern Avene site revealed no difference in the day and night comparison, except that the mean approach speed for the last vehicle throgh the intersection as significantly higher at night. All three of the stdy locations in Tcson had change intervals ith the all-red phase: ths, an analysis of the inflence of the yello only verss the yello pls all-red intervals reqired a comparison of the Phoenix and Tcson area sites. For this analysis, the Sothern Avene and the University Drive stdy approaches in the Phoenix area ere sed. The data for these intersections ere compared ith the information for the three Tcson intersections. The reslts of this analysis are given in the table belo (note, N = difference is not significant, and S = significant difference): First vehicle to stop Approach speed Perception-reaction time Distance from intersection Deceleration rate Last vehicle throgh intersection Approach speed Distance from intersection Difference s N s s s N The mean approach speeds for both the stopping and throgh vehicles ere significantly higher in Tcson. In contrast, hoever, the distance from the intersection at the beginning of the yello interval as significantly less and the deceleration rate as higher in Tcson. CONCLUSIONS The reslts of this stdy reflect the behavior of drivers given the existing intersections and traffic signal timing. It is not possible to anticipate hat the reslting driver behavior old be ith a different set of conditions. Based on the observa-

7 16 Transportation Research Record 904 tions and analyses that ere made a part of this stdy, the folloing conclsions can be dran: 1. The observed mean deceleration rates chosen by drivers at the varios stdy sites ranged from 7.0 to 13.9 ft/s/s. The mean deceleration rate for all observations as 11.6 ft/s/s. 2. The day and night comparison of driver behavior revealed mixed reslts in that the mean deceleration rate as not significantly different dring the nighttime period at o ne site1 hoever, there as a decrease in t he mean rate from to 9. 7 ft/s/s at the Broaday Bolevard location. 3. With respect to the perception-reaction times of drivers, the observed mean time for each of the sites r anged f rom l. 16 to J..55 s. The mean for all approaches as l. 30 s. The 85th percentile vale for the t i me measrement rang.ed from l. 5 t o 2.1 s. Althogh the nighttime stdies revealed mean times that ere slightly less at both sites, these differences ere not statistically significant. Factors sch as approach speed, distance from the intersection at the beginning of the yello interval, and the deceleration rate that is s ed by the driver had little or no inflence on the perceptionreaction time. 4. The comparison of stdy sites in each of the metropolitan areas also provi ded mixed reslts. In the Phoenix area, the mean decel eration rate at the University Drive location as significantly less than at the other to sites even thogh most of the other measres shoed no significant differences. Similar comparisons for the Tcson sites revealed differences in a nmber of the measred parameters. S. In terms of the comparison of the yello only verss the yello pls all-red change intervals, there ere no significant differences in the perception-reaction t i mes. This as tre hen the itersection ith an all-red inter val in the Phoenix area as compared ith the other sites that had a yello only interval. It as also tre hen the inte rsectio.ns in Tcson ere compared ith t hose in Phoenix. The analysis of the deceleration rates for the to types of change intervals did not yield consistent reslts. Althogh the deceleration rates for the Tcson sites ere higher than those observed in the Phoenix area, there as a variation in the significance of the differences beteen intersections ith the same type of change interval. In revieing the comparisons of the observed behavior, the se of some degree of cation shold be exercised. For example, small differences in deceleration rates may be statistically significant bt may have no appreciable affect on the evalation and soltion of traffic problems. In addition, the project staff spent considerable time observing the operations of these intersections dring the corse of the stdy. Althogh the observed decelera t i on rates and the perceptionreaction times varied from the vales recommended by crrent practice, it shold be recognied that the intersections appeared to be operating reasonably ell ith respect to the change interval. ACKNOWLEDGMENT This p ape r i s based on research ndertaken as part of a Highay Planni ng and Research Pi:ogram project entitled, An Evalation o f Dr i ver Behavior at Si g nalied Intersections. The research as condcted by the Ar iona Transportation and Traffic I nstitte, College of Engineering, University of Ariona, and the Col l ege of Engineering, Ariona Stat e University, in cooperation ith the Ariona Department of Transportation and the Federal Highay Administration (FHWA), U.S. Departme nt of Transportation. The contents of t his paper reflect or vies, and e are responsible for the facts and the accracy of the dat a presented herein. The c ontents do not necessar i ly reflect t he official v ies or policies of the Ar hona Department o f Transportation or FHWA. This paper does not cons t itte a standa rd, specif i cation, or reglation. Trade or manfactre rs' names that may appear herein are cited only bec ase they are considered essential to t he obj ectives of the paper. The U.S. gove rnment a nd the State of Ariona do not endorse prodcts or manfactrers. Discssion Peter s. Parsonson The most controversial isse in the preparation of the second edition of the Institte of Transportation Engineers (ITE) Transportation and Traffic Engineering Handbook (.2_) as the calclation of the yello and all-red intervals. The final draft of the handbook, pblished in 1982, called for the se of a perception-reaction time of l s. The paper by Wortman and Matthias does not inclde ithin its scope an evalation of this crrent practice; bt it is important that the discssions of their paper inclde a comparison of their findings ith the ITE handbook recommendations. Wortman and Matthias observed more than 800 stopping drivers and fond an average perceptionreaction time of 1.3 s. ITE ses l.o, so e need to consider hether the ITE procedre gives yello times that are 0.3 s too short. There are to reasons hy this paper does not represent a challe nge to the c rrently accepted vale of 1.0 s for pe rception-reaction time. First, the perception-reaction times reported by Wortman and Ma t t hias are f o r the first ve hicle to stop after the yello begins. For all e kno, many of these vehicles ere so far from the intersection hen the yello came on that it as ot of the qestion to contine throgh the intersection. These drivers cold react at their leisre and brake comfortably to a stop. They t ell s nothi ng abot ho fast they old react hen t here is a r e a l decision to be made. Second, it is necessary to consider perceptionreaction time and deceleration rate jointly, as these to variables are tied together as determinants of stopping distance and reqired yello time. The Wortman and Matthias data for perceptionreaction time and deceleration rate, taken together, prodce calclated yello times comparable ith those derived from the ITE gidelines. For example, consider an approach ith a speed of SO mph. The ITE vales of 1 s for perceptionreacti on time and 10 ft/s/s for deceleration rate yield a yello t i me o f 1.7 s. The c a lclation, hen sing t he Wortman and Matthias vales o f 1. 3 s a nd 11.6 ft/s / s, g i ves 4.5 s, hich i s j st a cople o f tenths of a second l ess than the I TE reslt. That difference makes sense becase ITE assmes a et r oad, hile the Wortman and Matthias d ata c ame from dry roads. The average d r i vers in t he Wo r t man and Matthias stdy may be alloi ng themselves a little extra time to react becase they kno they can easily make p for it by braking a little more heavily on a dry road. At 35 mph, the to sets of vales give an identical 3.5 s. It is ell knon (!Q.l that the minimm yello time c an be compted by dividing the stopping dis-

8 Transportation Research Record tance by the approach speed, as follos: Y =stopping distance/speed= [vt + (v 2 /2a)] /v = t + (v/2a) (I) Figre 4. Certain relation involving yello time, travel time, and approach speed. If stopping distance is measred directly by observing the behavior of traffic at the onset of the yello interval, the reqired yello time can be calclated ithot the need to make assmptions for t and a. From 50 mph, for example, Zegeer (!!) fond that 90 percent of the drivers ill decide to stop if they are 350 ft from the intersection hen the yello begins. A yello time long enogh for 90 percent of the drivers is as follos: Y =stopping distance/speed= 350/(50 x 1.47) = 4.76 s (2) This minimm yello is entirely empirical and is independent of assmptions for t and a. It can be sed to calclate combinations of t and a that ill satisfy it, as follos: 4.76 = t + ((50 x l.47)/2a] = t + (36.75/a) (3) A table of combinations that satisfy this relation is given belo:! a B This table shos that provision of a yello of 4.76 sallos the driver stopping from 350 ft a ide range of responses. A qick-reacting driver can decelerate comfortably to a stop, hile one sloer to react mst compensate by braking more forceflly. For example, a driver ho reacts in 1-s flat can decelerate comfortably at 9.B ft/s/s, hile the driver ho reqires 1.3 s mst brake more heavily at 10.6 ft/s/s. The point is that it is the combination of t and a that conts. The combination reported by Wortman and Matthias is in harmony ith the combination sggested by ITE. The small difference probably can be acconted for as dry road verss et road. Figre 4 shos Zegeer's observed distances for 50 and 90 percent stopping, converted to travel time to the intersection by dividing by the approach speed. The data are for 2100 drivers responding to the yello interval in Kentcky. The pper crve means that 90 percent of these drivers stopped if the yello came on hen they ere abot 4. B s of travel time from the intersection. The crve is qite flati vales of jst nder 5 s ere observed over the entire range of speeds stdied. At 50 mph, 4.B s is a yello time long enogh for 90 percent of the Kentcky drivers. These motorists seem to be indicating that they ant a constant yello of almost 5 s, regardless of the approach speed. Figre 4 also shos a crve for yello time calclated by sing the ITE vales of t = 1 and a = 10, as follos: Y= t + (v/2a) = 1+[v/(2x10)] = 1 + (v/20) (4) Consider a driver approaching at 35 mph. Sppose he or she is shon a yello calclated by sing t = 1 and a = 10, hich yields 3.6 s. No sppose the vehicle is 4 s from the intersection hen the yello comes on. There is a 70 percent chance that the driver ill decide to stop (obtained by interpolating beteen Zegeer' s 50 and 90 percent crves) There is a 30 percent chance that he or she ill c,j U) ' :iii j:: 3: 0..J..J 6 s 4 6 c,j U) 5 ' :iii j:: O 1 -i.o..j, : g 3 I 3 2 ' r--.---' APPROACH SPEED, MPH decide to clear, in hich case the driver ill then rn the red becase of the deficient yello. The driver seems to ant 4.B s--he or she needs 4.0 s--bt the driver gets only 3.6 s, so he or she may ell rn the red. The Zegeer data sggest that yellos calclated by the ITE gidelines, or by the Wortman and Matthias data, are too short for speeds less than 50 mph if e are trying to meet the needs of the 90th percentile driver. The eight plotted points on Figre 4 cannot be explained ithot frther discssion of entering on the red becase of yello time deficiency. The First Avene site as observed to have an extremely high percentage of vehicles entering on the red signal compared ith other locations. The reason for this seems clearly to be that this site as mch more deficient in yello time than any of the others. Table 10 shos the deviations of the observed yello times from those calclated by sing t = 1 s and a = 10 ft/s/s for the eight sites. Figre 5 shos the percentage entering on the red plotted against these deviations expressed as a srpls or deficiency in yello time. The figre shos that the First Avene site can be nderstood if the percentage entering on the red increases exponentially (or logarithmically) ith increasing deficiency of yello. The figre also sggests that, at ero deficiency (obtained by sing t = 1 and a = 10 for or calclations), e can expect abot 3 percent of the vehicles to enter on the red. This might be considered an acceptable level, inasmch as it is a common rle of thmb in the positive-gidance area that an operational deficiency exists if traffic conflicts or erratic manevers are exhibited by more than 3 percent of the approaching vehicles. The final comment is an explanation of the eight plotted points on Figre 4. These points represent the first vehicles to stop at the eight sites reported by Wortman and Matthias. Specifically, the ordinate of each point is the mean travel time to the intersection for the first vehicles to stop. The ordinates are not the yello times at these intersections. The points plotted as triangles are the for sites shon in Table 10 to have a srpls of yello time, and the circles are for locations deficient in yello time. All of these drivers decided to stop becase they correctly jdged that their travel times to the intersection ere too great to allo entry on the yello. The average ordinate of these points, hich represents 100 percent stopping, lies close to Zegeer's 90 percent crve and indicates good agreement. It is most interesting to note that the sites deficient in yel-

9 18 Transportation Research Record 904 Table 10. Deviations of yello time from calclated standard. Percentage 85th Percentile Calclated Srpls(+) or Entering on Approach Speed 8 Standard Yello Actal Yello Deficiency (-) 0 Intersection Approacn Red (mph) Timeb (s) Time (s) (s) University Drive o.8 Sothern Drive Day Night US First Avene Sixth Street l Broad ay Bolevard Day Night LO 43.l AvOrn p.ri oflasl tjhl c l a 1hro g h tha inl e('.se:ct lon D-nd nc st vctlic:lt.l io.s101,. bderlvi=.tj by sln, 1he eq.:lora I + ( '/lii 1. 64,.411). '"' h r e IA = ) O f t/'fi/!j. nd g ==approach gradient. c DeriYci.1 by s\jlra.o:lintt m: lnl Yello timl\ fro m 1--"lltclll li:d :!l l:m'1nnl 'tcllo time. Figre 5. Percent entering on red verss yello-time srpls or deficiency. c a: , Cl 20 a: 0 DAY FIRST NIGHT 7 10 I- 6 a: A-a.-...e l SURPLUS C+l OR DEFICIENCY C-l IN YELLOW TIME, SEC. lo all plotted loer than those ith a srpls. That is, drivers approaching yello-deficient signals ere illing to stop from locations closer to the intersection, hich reqired qicker reaction and/or heavier braking. The data sggest that the drivers are aare of the lengths of the yello times at these intersections and do factor this into their decision hether to stop. This finding conflicts ith the 1979 report by Stimpson (12), hich indicated that the length of the yello interval does not affect driver behavior. Discssion Jack A. Btler The ork of Wortman and Matthias has prodced data that confirms the findings of other researchers and reinforces crrent trends in calclating vehicle change intervals. The conclsions of the athors, hoever, do not flly explain the data and fail to anser qestions, stated and implied, in their report. This discssion ill expand their analysis and propose frther explanations of the data. The first goal of the sbject stdy as the determination of actal deceleration rates applied in stopping a vehicle at the monitored signalcontrolled intersections. By employing field stdy methods to srvey nsspecting drivers, the possibility of driver behavior modification as removed. Given that it has been shon (13) that acceptable deceleration rates fall ithin the range of 8-12 ft/s/s (15th to 85th percentiles), the observed vales generally fall ithin the pper half of the acceptable range. The to deviating means can be explained by the approach geometry. Althogh not i nclded in the report, the University Drive site incldes an at-grade railroad crossing ithin the stopping distance. Wortman disclosed in a conversation that they had observed drivers applying a higher deceleration rate on the pstream side of the crossing. The reported mean of 7.0 ft/s/s is an average that may not accrately indicate a rate sed at a signalied intersection. The pper limit on the range of observed rates of ft/s/s at Sixth Street incldes the effect of gravity on sloing the vehicle. Parsonson and Santiago (14) pblished a report that details an easy method to evalate the effect of roaday slope on vehicle deceleration. Removing the deceleration de to gravity on the 2 percent phill grade prodces a rate of 12.6 ft/s/s. It is this vale that shold be sed in discssing driver behavior, as it is the force perceived by the motorist. All observations ere made on dry pavement. Becase it might be expected that et pavement conditions old loer the stdy vales for deceleration, Wortman and Matthias' findings tend to reinforce the se of 10 ft/s/s as a mean deceleration rate in yello interval calclations. The second goal of the stdy as the collection of nighttime data and the evalation of varios intersection environments. The data do not indicate any differences among these factors except as noted above. The third goal, evalation of all-red intervals, as not flly accomplished. The measre of effectiveness selected--vehicles entering on red--is indicative of yello interval performance. It has been shon (5) that the length of the yello interval does not-effect driver behavior. Indeed, it is the yello interval that mst conform to driver behavior. A too-short yello ill natrally prodce a large nmber of red-aspect violations. The sbject stdy provides some confirmation of that principle. Data ere not presented that allo the comparison of the all-red timing and intersection geometry. Frther ork is needed to accomplish this third goal. It is possible to calclate more reasonable yello intervals by applying the folloing formla (14) : y = t + [v/(2a g)] here t v a reaction time {s), 85th percentile approach speed (ft/s), deceleration rate (ft/s/s), and (5)

10 Transportation Research Record g = grade as percent divided by 100. When this is done, a clear relation emerges beteen yello time deficiency and red-aspect violations. Table 11 lists the vales relevant to this discssion Tables 12 and 13 illstrate yello and allred times over a variety of speeds and intersection idths. The 85th percentile speeds ere estimated by adding one standard deviation to the 50th percentile speed. The means for the last car throgh and the first to stop ere combined, as ere the standard deviations. This techniqe attempts to balance the effect of acceleration of vehicles that did not stop ith the deceleration that occrs dr- Table 11. Reported and derived data. Sothern Avene University Item Drive Day Night US-60 Yello interval (s) All-red interval (s) th percentile speed (ft/s) Calclated yello (s) Actal mins calclated yello (s) Last throgh vehicles entering on red(%) Mean stopping distance (ft) Stopping distance implied by calclated yello (ft) Distance traveled at 85th percentile speed dring actal yello (ft) Mean reaction time (s) Mean rate of deceleration (ft/s/s) Deceleration rate implied by crrent yello for 85th percentile speed (ft/s/s) First Avene Broaday Bolevard Sixth Street Day Night I.I 12.6' Vale has been adjsted to remove effect of approach grade. Table 12. Time (in secondsl of yello interval. Yello Intervals by Grade of Approach 85th Uphill(%) Percentile Speed I Level Donhill(%) -I I Note: Yello interval table vales derived from the formla: y = t + [v/(2a g)) here y =yello interval (s), t =reaction time (set at 1.0 s), v = 85th percentile approach speed (Ft/s), a= decejeration rate (set at 10 feet/s/s), and g =grade of approach over the braking distance (percent divided by 100). Table 13. Time (in secondsl of all red interval. 85th All-Red Intervals by Width of Approach (ft) Percentile Speed l." I.I I l Note: All-red interval table vales derived from the formla: r = (+l)/v here r =all-red interval (s), =idth of intersection (ft), I= length of vehicle (set et 20 ft), and v = speed of vehicle (ft/s).

11 20 Transportation Research Record 904 ing the driver reaction time. The vales for t and a are 1.0 s and 10 ft/s/s, respectively. One also finds a close agreement beteen the predicted stopping distance and that that as actally observed. The longer reaction times ere offset by higher deceleration rates. The only exception is First Avene, here the observed mean is 15 percent less than expected. This is also the approach ith the highest red-aspect violation rate ( 30 percent) and the greatest yello time deficiency. It may be conclded that many drivers chose not to apply the high rate df deceleration necessary on this approach to keep from entering on red. A similar, bt less severe, yello time deficiency and red-aspect violation relation exists at the Sixth Street and Broaday Bolevard sites. The loer violation rate observed at night for Broaday Bolevard is directly related to the loer approach speed at night. The yello deficiency can be frther refined by determining the deceleration rates implied by the existing signal timing at the 85th percentile approach speed. These vales all fall ithin the acceptable range except for those locations that experience the highest red-aspect violation rates. At the First Avene site, a deceleration rate of 16.3 ft/s/s old be necessary given a 1.0-s reaction time. The.higher-speed drivers have obviosly indicated their dislike of sch a high rate by entering on red.,, Taken together, the data sho that drivers reject deceleration rates significantly greater than 12 ft/s/s and that there is an evident relation beteen the length of the yello and red-aspect violation. Implied in this is the observed absence of drivers ho adjst their behavior significantly to conform to yello times. It may be frther stated that driver action as not affected by the presence of an all-red interval, hich gives spport to the findings of Benioff and others (15) that an increase in red-aspect violation does not accompany the se of sch intervals. Beyond that, it is not possible to examine the given all-red times, since the critical factor of intersection idth as omitted. Absent as ell as a discssion of right-angle and other accidents that old provide insight on the effectiveness of allred intervals. The final goal of the sbject stdy as the recording of driver reaction times. The athors admit in the paper that they had no ay of determining the reaction time of drivers ho chose not to stop. It may be frther sspected that they had no ay of determining the tre reaction time; i.e., did the drivers react in a leisrely manner. In spite of those shortcomings, the data seem to refte the contention of the American Association of State Highay Officials (AASHO) (16) that the 95th percentile alert reaction time isl. O s. The reslts also place an pper limit on the range of driver reaction times. As Parsonson shos in his discssion, for any given speed there is a range of reaction times and deceleration rates that prodce the same stopping distance. Becase the varios sites had similar approach speeds, it is not possible to select the correct combination. Frther research is needed to resolve the isse, inclding the qestion of the applicability of observed vales to design calclations. Committee 4A-16 (Use and Timing of Vehicle Change Intervals) of ITE ill soon begin field experiments to evalate the effectiveness of a range of reaction times and deceleration rates in calclatinq the yello interval length. rhe ork of Wortman, Matthias, and their team members comes at a time hen a decision mst be made on selecting a realistic change interval calclation methodology. In that context, their report provides considerable gidance. REFERENCES 1. Ariona Traffic Accident Smmary Ariona Department of Transportation, Phoenix, L. Rach. Traffic Signals. In Transportation and Traffic Engineering Handbook, 2nd ed., Prentice-Hall, Engleood Cliffs, NJ, Traffic Engineering Policies, Gides, and Procedres. Ariona Department of Transportation, Phoenix, Policy PGP-4-4B-3-0, Oct W.L. Williams. Driver Behavior Dring the Yello Signal Interval. TRB, Transportation Research Record 644, 1977, pp W.A. Stimpson, P.L. Zador, and P.J. Tarnoff. The Inflence of the Time Dration of Yello Signals on Driver Response. ITE Jornal, Nov. 1980, pp P.S. Parsonson and A. Santiago. Traffic-Signal Change Period Mst be Improved. Pblic Works, Sept. 1981, pp D. Gais, R. Herman, and A. Maraddin. The Problem of the Amber Signal Light in Traffic Flo. Operations Research, Vol. 8, No. 1, Jan.-Feb R.S. Jenkins. A Stdy of Selection of Yello Clearance Intervals for Traffic Signals. Michigan Department of State Highays and Transportation, Lansing, Rept. TSD-TR , Feb ITE. Transportation and Traffic Engineering Handbook, 2nd ed. Prentice-Hall, Engleood Cliffs, NJ, H.H. Bissell and D.L. Warren. The Yello Signal Is Not a Clearance Interval. ITE Jornal, Vol. 51, No. 2, Feb. 1981, pp C.V. Zegeer. Effectiveness of Green-Extension Systems at High-Speed Intersections. Division of Research, Brea of Highays, Kentcky Department of Transportation, Lexington, Res. Rept. 472, May W.A. Stimpson and others; A.M. Voorhees and Associates, Inc. The Inflence of the Time Dration of Yello Traffic Signals on Driver Response. Insrance Institte for Highay Safety, Washington, DC, Jan P. Olson and R. Rothery. Deceleration Levels and Clearance Times Associated ith Amber Phase of Traffic Signals. Traffic Engineering, April 1972, pp and P. Parsonson and A. Santiago. Design Standards for Timing and Traffic Signal Clearance Period Mst Be Improved to Avoid Liability. Compendim of Technical Papers, ITE Annal Meeting, 15. Ag. 1980, pp B. Benioff, F. Dock, and c. Carson. A Stdy of Clearanoe Intervals, Flashing Operation, and Left-Trn Phasing at Traffic Signals. FHWA, Rept. FHWh-RD-78-77, May A Policy on Design Standards for Stopping Sight Distance. AASHO, Washington, DC, Phlimtim1 nf this [IO[ler.i[lnn.inred hy Cnmmittee nn U.ier lnfnrmatim1 Systems.