'ATION RESEARCHRECORD 158]

53

currenceof Secondary Crasheson rban Arterial Roadways RrcueRrA. Raus Analysisof incidentsand crashesoccurringalong urbanarterialroadwayssuggests that asmany as I 5 percentof the crashesoccurringalong theseroadwaysmay have been,in part, causedby an earlier incident. Toconductthe analysis,datacoveringreportedvehicularcrashes,fires, disablements, traffic enforcement,and other incidentsaffecting roadwayswerecollectedfrom sevencontiguousurbanmunicipalitiesin the NorthemChicago,Illinois, metropolitanregion. Crashesrepresented 35percentofthe incidents;traffic enforcement,30 percent;anddisabled vehicles,27 percent.Othereventsrepresented the remaining8 percent. To associatean initial event with a subsequentcrash,any crash that occurredon roadwaysaffectedby theprimaryevent,within theduration of theprimaryeventplus 15min, andwithin 1600m of thatevent(based on useof geographicinformation systemssoftware)was assumedto havebeenrelated.Analysisof 1,796incidentsin thedatabaseidentified crashesrelatedto 8 I initial incidents.A randomsampleof 97secondary all eventsin the data baseshowedthat only 3.6 percentof the crashes insteadof 15percentmight be expectedto fall within the requiredtemporalandspatiallimits in relationto an earlierevent.This flnding suggestedthat the selectedeventsmay not havebeenrandomoccurrences. Finally,analysisof the secondarycrashesshowedthat I ofevery l0 may havebeenrelatedto an earliercrash;I of every I I to a disabledvehicle, andupto I of every40 to a police stopfor a majortraffic violation.What mustbe studiedin greaterdetail is the degreeof relationshipbetween theinitial eventandthe subsequent crash.This canonly be donethrough bettercrashreporting,in which officers clearly indicateany associated eventsthat may havebeencontributingfactors.

Commonly acceptedamong traffic safety personnelis the assumption that congestion, especially that brought about by incidents, leadsto crashes.However, research in support of this position is lacking.Instead,the assumptionis basedon anecdotalinformation, pailicularly among police who have witnessedcrashesoccurring at or near an incident that they had handled.What is not known is the extentto which crashesoccur around, and are related to, an earlier incident.Recently completed researchon incident managementhas suggestedthat more than l5 percent of all crashesmay have been caused,in pafi, by some earlier incident (1). This paper explores thesesubsequent(called secondary)crashesin greaterdetail. In the literature, only two direct referenceswere found to crashes and congestion. Both were based on observationsmade on British motorways.The earlieststudy, by Owens (2), found that 13 of 75 videotapedcrashes,or 17 percent, were probably related either to congestionalong the roadway or to an earlier incident. Robertset al. (3), examining later data, had opportunity to find similar outcomes. However, both of these studies were performed on limited-access highways in Great Britain and also included contributions to crashes from recurrent congestion.One study of incidents on U.S. freeways alluded to secondarycrashesas part of the problem (4). Measures usedin this last study were not sufficient to determine the extent of the problem. No researchhas examined secondarycrasheson arteNorthwestem University Traffic Institute, Evanston, Ill. 60204

rial roadways,nor has an attemptbeenmade to validate the findings of Owens. Several difficulties are encounteredin attempting to determine whether a crash was the result of an earlier event or simply a chance occurrencein time and space.First, police crash reports, which are the source of most crash data, rarely include information about eventsthat may have beencontributing factors even when thosefactors are observableto the crash investigator (this omission may be due to concem over later court actions). When such information is provided, contributory events are limited to the condition of the driver, immediate roadway, and environment (e.g., weather). That the driver is maneuvering in traffic congestion arising from an earlier car fire, for example, would not be observedor reported. A seconddifficulty lies in establishingwhich parameterslink events. At what distance from an incident, and how long after the incident occurred, can a secondary crash be considered related? Even establishing parameters leaves a need to confirm the relationship between incident and secondary crash. As noted, current police traffic crash investigation rarely lends itself to determining the relationship between the two events. A valid space-time surface for associating an event and a subsequentcrash has not been identified in the literature. The time over which an incident on an arterial roadway can affect traffic has not been studiedextensively,in part becausethe capacity reduction resulting from an incident is not known. Moreover, the lingering effects of an incident are not known. Work by Lindley and Tignor (5) on getaway flows from a freeway suggests that the effects of an incident on a freeway can last from 5 min to more than 60 min after clearance, depending on the nature of the incident and the traffic volume at time of occurrence. For this paper, an incident is assumedto affect traffic for the duration ofthe event plus 15 min (called "time of effect" and expressedas /" - tr"r+ 15, where /1.,is the time taken to clear the first event). For a crashto be included in the study, it had to occur within the time of effect (1")of an earlier incident. The use of l5 min beyond the duration of an incident was based partly on the getaway times provided by Lindley and Tignor (5). It also came from personalobservationsof incidents within the region of study. One difficulty in predicting clearance time for arterial roadways is that, becauseof the availability of sufficient altemate routes,drivers who ordinarily might have to passthe location of the incident may have taken an alternate route and therefore are no longer affected. Drivers also may have altered their trip by adding anotherdestination(e.g., a stop at a conveniencestore) and as a result are no longer part ofthe queueto be dispersed.Knowing the extent to which drivers remain in or leave a queueon the urban roadway not only would be important for the study of secondary accidents, but also would provide a valuable parameter for models of urban travel and congestion.

iq

ilq

I

;

i{

,i '1

) I l\

RECORD1581 RESEARCH TMNSPORTATION

54 No guidance is available for linking two events spatially. Obviously, if the two events occur at the same location within /,., they probably are related (other conditions could be suggestedto argue that even this time-spacerelationship is not a causalone)' It is more difficult, however, to establishhow far from the original location the effect ofthe incident lasts.The distanceshouldrelatestrongly to traffic volume. At low volume, the only drivers likely to be affected are those in the immediate vicinity. On the other hand, at peak periods, an incident might affect traffic over several kilometers. On-site observationsofcrashesoccurring during periodsofthe heaviestflow suggestthat an effect over 1600 m is not unreasonableassumption. For this analysis,distanceof effect (4) from an event was assumed to be less than 1600 m, or d" < 1600 m. Figure 1 displays the surfaceencompassedby r" and d" where the initial incident is represented by the point t, d = 0. Subsequent crashesare representedby dots. Crashesat Points A and B probably were related to the initial event; they occurred at the samelocation and within the duration of the first event. Crashesat Points D and E could have been causedin part by that event; separationin both time and distancelies within the assumptions.The remaining crashes,at Points C, F, and G, are assumedto be independent. One finding from analyzing the crash data for crashesthat fall within the time and distancecriteria was that in most casesthe distance separatingthe initial event and the secondarycrash was less than 800 m. When a greater separationoccurred, one of two situations also had occurred..In the first, more than one secondarycrash existed and the second crash lay within 800 m of the earlier crash but further from the precipitating incident. In the other situation, ," lasted more than 60 min and the crash occurred close to the end of that period. In the caseof multiple crashes,both secondarycrashes were assumed to be related to the first event. This assumption reduces the number of primary incidents and makes the findings perhapsmore conservativethan might actually be the case. A study of incident managementalong arterial roadways in the Chicago suburban area provided data about police response to crashes, traffic violations, disabled vehicles' and other incidents for seven contiguous urban communities in Northern Cook County, Illinois. For each incident, the data indicated incident class, date of occurrence, address of incident, time the call was received, and time the scene was cleared. Because of the availability of telephones and cellular phones in urban communities,

especially along arterial roadways, an assumption was made that the time police received a call was close enough to the time the incident occurred to be used as the start of the incident. In a recent study, Willmink and Immers (6) determined that the use of cellular telephonesand other radio communications resulted in a delay of approximately 2 minutes between the occurrence of an incident and its reporting. Time for responding to and handling a call was available from the dispatch records. Duration ofincident was measuredas the difference between when the call was received and when the officer finished. However, a scene can be cleared before an officer finishes the call. Data from Northwest Central Dispatch (NWCD), which representedmost of the incidents in the data base,provided sufficient documentation to estimate the time at which the roadway was cleared, even when this time occurred before the officer was finished with the incident. Becausean addresswas available for the incidents, they could be coded to a geographicinformation system (GIS). The geocoding of each incident was done by Maplnfo software with geographically coded streetfiles. Once the incidents were coded,distancesbetween points could be measured.The processselecteda crash and all other incidents within 1600m of that crash.Examination of theseselected incidents showed casesin which /" of an earlier incident extended beyond the starting time for the crash. If the crash also fell along a roadway affected directly by the primary incident, it was added to the baseofsecondary crashes.When the possibleprecipitating event was a crash, further checking was done to determine whether this crash, too, might have arisen from an even earlier incident. This processcontinued until all crasheswere reviewed. By using the assumptionfor duration and spatial separation'the analysis of secondarycrashesin the selectedurban areauncovered 97 crashesthat could be linked to earlier events.All of the crashes except one occurred during hours of peak travel. Given that 627 crashes were recorded in the data base, the finding suggeststhat during peak travel periods as many as I 5 percentof all crashesmay be related to an earlier event. This value closely approachesthat found by Owens in his study of a British motorway (17 percent); the study by Owens also included crashes related to recurrent congestion.

RELATIONSHIP AND CRASIIES

AMONG INCIDENTS

Base Data Collected

0,0

+I lsi

FIGURE 1 Relation of secondarycrashesto an initial event.

Data usedfor the study covered the period January9' 1995,through February 5, 1995. Although the period was not a preferred one, it was one for which data were available and, fortunately, in which there was no severewinter weatherthat could have confoundedrelationships with crash data. Roadways in the area were free of snow and ice during the period. This report covers only those incidents occurring along arterial roadways during the 28-day period. Table I presentsa crosstabulation of the data by community and by generalclassification of the event. All incidents were classified either as a crash (a further distinction is made among crasheswith property damageonly, injuries, and other or injuries unknown), traffic enforcement,disabled vehicle, fire, or other. The last classification included malfunctioning traffic signals and railroad crossing gates,materials spillage, and other road blockage. Not all stops for traffic violatioris were included, only those in which the police offi-

TABLE I

Arterial IncidentsShowingCity and Primary Classification All Incidents

lity \lame

|uffalo Grove llk Grove Village v{ount Prospect talatine )rospectHeights

Disabled

Stop

Vehicle

11

169 61 lll 101 100 37

\rlington Heights

Traffic Fire

Crash

A

6 t2 9 a

4R

lolline Meadows

627 34.9E,

\ll Cities )ercent ofTotal

44 1lA

cer communicateddirectly with the public safety telecommunicator (thesestops were generally the more serious ones, such as driving under the influence). A large number of the communications relating to traffic stopsare done computer to computer and would not be part of the telecommunicatordata base. The base data contain 1,796 incidents. Of these, crashesrepresented 35 percent; traffic stops, another 30 percent; and disabled vehicles, 27 percent; fire and other categories account for the remaining 8 percent. Of the 62'7 crashes in the base, 75 percent resulted in property damage only. An additional 19 percent involved an injury, and, for 6 percent, the existence of injury was unknown. This last category was included becausethe dispatcher entry showed "injury unknown" in the dispatch record and the record subsequentlywas not changed.

General Relationship Between Incidents and Subsequent Crashes The first step in the analysis was to determine the general relationship betweenincidentsand subsequentcrashes.Given that an incident occurred, what was the temporal and spatial separation from the first occurring crash?Moreover, what percentageof subsequent crashesfell within both r, and d"? To perform this analysis, a l0 percent sampleof incidents was selectedfrom the base. A1l incidents were counted until the 10th was reached,then the next crash occurring was considereda secondarycrash.Because of the distribution of the 627 crashesin the data base, more than one set of 10 eventscould be counted before a crash was found. Therefore, the final sample contained 116 events. occurring between the hours of 0600 and 2200, with which a subsequent crash was associated.As shown in Table 2, 50 of the first events were crashes(none ofthe crashesselectedservedasboth a primary

All Other

155 55

132

r27

133 84 47 40

118 M 18 t7 534 29.7q,

)J

491

n3E

lncidentq

39 8 t7 t2 18 6

506 183 394

100

r796

3n 218 103 65

lrash

fU

)ther 'otal

66 116

t0.l 7.4

Tim 111 81.8

8.6

94.5

Distancr

Note: Distance in kilometers; Time in minutes.

lOO.Oq'

Potentially Associated Events and Crashes All 1,796 incidents occurring on arterial roadways were reviewed by using the time and spatial standardsnoted. When a crash was found to have occurred within r" of an earlier incident, the distance separation was measured on the basis of its latitude-longitude

toO.o%

80.0% I

fffit"d cumulative I

7 +o.oco O 2O.jVa

Seoarationin Frequenc'

2l.gEl 18.zEl n.9l 5.7E1 ?.6q,

and a secondary event), and the remaining 66 fell into other classifications such as traffic stops. The average separation between the incident and the first subsequentcrash was 8.6 km and 95 minutes. The average distance between the initial event and subsequent crash did not meet the separation criterion of 1600 m or less required by the original assumption. A secondstep was to reduce the sample of principal eventsto eliminate any crash that occurredbeyond 15 minutes after the end of the first event (r,). This stepwas taken to yield only eventsthat fit within the required temporal limits (i.e., duration of the fust event plus 15 minutes),to determinewhat percentageof crashesfell by chance within the 1600 m separation. The average distance of the first crash that occurred within /" remained 8.6 km. Fewer than 4 percent of the crashesoccurred within 1600 m, as shown in Figure 2. A random pattern of incidents and secondary crashes should have yielded l0 percent of secondarycrashesoccurring between 0 and 1600 m. Yet, only at a separationof 9.6 km did the cumulative percent of crashesexceedthe expectedcumulative percent.

I percenr I

nitial llassification

28.2E t0.2E,

5.6q, 100.09

B ffi.0v" TABLE 2 Sampleof Incidentsand SubsequentCrashes (All Primary IncidentsOccurring Between6:00 a.m. and 10:00p.m.)

Percent ofTotal

o.o%

0.0

FIGURE 2

1.6

4.8 6.4 8.0 9.6 3.2 Distancefrom Earlier Eventin Km

First crash after an earlier event.

RESEARCHRECORD1581 TRANSPORTATION

56

TABLE 3

Secondary Crashes FromPrimaw Event

Average -l

acqification

Jrash,Injury lrash,PDO lrash,Other lotal )ct. of Database

Freouencv

16 ?R J

97

Duration (minutes)

7t . 7 56.0 38.7 58.1

Pct. of Total

Database

16.5q, 80.4q, 3.1q, 100.09

Pcr.ent

Time l'minrrfes

Distance t'milcc l

D.Oq' 75.0q, 6.O5,

100.09

36.4

o.4

s.5q,

points, but along the roadways.If the subsequentcrashlay within 1600 m and along a roadway that could be affectedby the initial event, an assumptionwas made that the event and the crash may have been associated.In severalcases,more than one crash was associatedwith an initial event,either becausethe next crashmet the criteria with respecteither to the primary incident or to an earlier secondarycrashfor that incident.

Secondary Crashes Given the assumptionsregarding time and distanceseparation,with added distance separation for multiple crashes,the results of the analysis showed 97 secondarycrashesfor 81 primary incidents. ThesecrashesrepresentI 5.5 percentof the 627 crashesrecordedfor the sevencommunities.This percentageis substantiallyhigherthan the 3.6 percent found when incidents were selectedfrom a sample. It suggeststhat the secondarycrashesmay not have been randomly distributed. As shownin Table 3, 16 ofthe 97 secondarycrashes,or 16.5percent, were injury producing. This percentageis not significantly different from the percentagefor crashesas a whole, 19.0 percent. Average time from the primary incident until the secondarycrash was 36.4 min with a separationof 600 m. Forty percent of the secondary crashesoccurred within 20 min of the primary event; more than 55 percentfell within the first 30 min. In addition,police spent an averageof 58.1 min handlingthe secondarycrashes.This latter value is important, given that in 45 percentof the cases,the secondary crash occurred more than 30 min after the initial incident. This meansthat even though the initial incidentmay have delayed traffic for 30 min, when a secondarycrashwas associated,the delay increasedfrom 30 to approximately90 mins. Even though separationdistancebetweenincidentand secondary crash was set at 1600 m, approximately35 percentof all crashes occured within the first 200 m. More than 50 percenthad occurred within the first 600 m, as shown in Figure 3. One hypothesistested in the study was that, as the time taken to clear an event lengthened,the distanceseparatingthe primary event from the secondarycrash also would become greater.The underlying assumptionwas that, as time to remove an incident increases, congestionandmore importantlyqueuesincrease.A plot of the time and distance separationsto determine whether any spatial relationshipsexistedshowedno pattem.A largerbaseof secondarycrashes might provide a betterfoundationfor testingthis assumption. A final analysis of secondarycrashesoccurring after the primary incidentshowedthat in 11 of 81 cases(13.6 percent)more than one crash could be linked to a primary event. Of these 11 cases,4 representedthree secondarycrashesand the remaining

7 had two collisions. This finding suggeststhat as frequently as I in 20 crashesthat police handleasprimary events,they may have to handleup to 3 more crashesresultingin part from the first crash. Such activity representsa significant use of tesourcesthat might be otherwiseemployed.

Primary Event The 97 crasheswere related to 8l primary events.Of theseprimary events,60.5 percentwere crashes(but representingonly 27.7 percent ofall events) and another24.7 percentwere disabled vehicles (27.3 percentof all events).Crashesrepresenteda significantly higher percentageof primary events(chi-squareof 27.7, significant at the .01 level) than of all events.Table 4 shows the distribution by classification of primary eventsas a percentageof initial and all incidents. Distribution of primary incidentsby day of the week and time of day also showeddifferencesfrom the distribution of incidents as a whole. Differencesshouldhave beenexpectedgiven that previous researchhas shown a close relationshipbetweencongestion and crashes;a crash is more likely to occur during peak travel hours and on weekdays. The distribution by day of the week showed that more than 86 percent of the primary incidents fell during the period from Monday through Friday (Table 5). On the other hand,fewer than 76 percentof all incidentsoccurredduring the same period. The differencesare significant at the .05 level { c h i - s q u a r=e 1 4 . 9 .d f = 6 ) . Although the chi-squaretest of difference by hour group was not significant,none of the primary eventsoccurredbefore 6:00 a.m. Further, only 5 of the 81 primary incidents occurred after 9:00 p.m. (Table 6) and only one after l0:00 p.m.

E 1iq^

l07a

Distance in Kilometers

FIGURE 3

Secondarycrash distancefrom initial incident.

Primary Event Analysis: Classification of Incident

TABLE 4

Frequency

Average Duration

llassification

lraffic /ehicle )ther

54.1 12.7 29.7 34.8 )6n

49 3 6 20 3

\11Classes

M.9

81

lrash rire

TABLE 6

Pct.of Pct. of Primary All Incidents Incidents 60.5q, 3.7E, 7.4q, 24.7q, 3;7q, 100.09

34.9q, 2.4q, 29;7q, 273q, 56q

100.09

Primarv Event Analvsis: Time of Dav

Start Hour

0000-02s9 0300-0559 0600-08s9 0900-1 159 t200-1459 1500-1759 1800-2059 2100-2359 All Hours

Frequency

Pct.of Primary Incidents

0 0 l8 l3 15 19 11 5 8l

0.0% 0.0E, 22.2E, 16.0E( 18.5E( 23.5q. t3.6E, 6.25, 100.0q

Pct. of All Incidents

8.7El 4.5q, t3.3E, 13.4E, 15.8E, 21.2q, n.8q

rc3q, 100.09

Comparison with Incidents Occurring During Peak Periods Becauseall but five primary incidentsoccurredbetween6:00 a.m. and 9:00 p.m., further analysisof incidentswas done with the same rangeof hours.Eliminating incidentsthat occurredfrom 9:00 p.m. through6:00 a.m. reducedthe data baseof all incidentsby 20 percent,from 1,796to 1,448.The proportionofeventsoccurringin each municipality remainedapproximatelythe samewith the reduceddata base.For incidents by classification,reducing the data baseresulted in a greaterpercentageof incidents that were traffic crashes,increasing thesefrom 34.9 to 40.5 percent,and a smaller percentageof traffic stops,decreasingthesefrom29.7to22.l percent.The differences between classificationsfor all incidents and primary incidents remained significant. An analysisof the primary incidentsas a percentageof all incidentsand ofthose occurringduring peaktravel is shownasTable7. When basedon all incidents,primary incidentsrepresentedbetween 3.1 (Mount Prospect)and'7.7 (Rolling Meadows) percentof the incidentshandled.When only those incidentsoccurring between 6:00 a.m. and 9:00 p.m. were included, the percentageincreased from a low of 4. 1 percentin Elk Grove Village to a high of 9.3 percent in Rolling Meadows. This latter value suggeststhat for approximately every 10 incidents occurring in Rolling Meadows, one secondarycrashoccurred. A more interestingcomparisonlies in classificationof primary incident. For the reduced data base, 8.6 percent of fires (mostly vehicle fires) and 8.4 percent of vehicle crashes,or I of every I I incidents,had one or more secondarycrashesassociated.The percentagefor crashesshould be adjustedfor thoseconsideredto be secondary to the initial crash. Removing the 92 secondary crashesthat occurred from 6:00 a.m. through 8:59 p.m. from the baseof all crashesresultsin 494 (586 -9D crashesthat could have servedas a primary incident. Therefore, given that 49 crasheswere

TABLE 5

Primary Event Analysis: Day of Week

Day of Week iunday vlonday luesday

Freouencv

5 19 15

Mednesday lhursday rriday iafilrdav

\ll Davs

19 12 r) 81

Pct.of Primary lncidents

Pct. of All Incidents

6.2q1 23.5q1 18.5El 6.2E, 23.5E 14.8E '7.4q

0.4q, 3.69, 43q,

100.09

100.09

4 )q,

8.2q, 5.59, 3.89,

consideredprimary incidents, a secondarycrashmay have occurred for every l0 crasheshandled. Disabled vehicleswere responsiblefor 4.8 percentof the primary incidents,or I of 20. Of greaterimportance,however,is that approximately I of every 50 traffic stops may have contributed to a later crash.In this latter case,the relationshipwas very time-of-day dependent.During peak travel pedods,one municipality had an associated crashfor every 20 traffrc stops.On the other hand,one crashoccurred for every 100 traffic stopsduring off-peak periods.

WHAT THE ANALYSIS INDICATES This analysisof crashesrelatedin time anddistanceto previousincidents is an attempt to addressthe extent to which secondarycrashes may occur on urban arterial roadways. It was basedon the premise that if a crashocculred within a given period of time after an earlier event and within 1600 of that event, the first event may have been a contributory factor. By using Gls-based mapping software,each crash was locatedand a possibletriggering incident sought.This processwas used becauseinformation about crashesgatheredby police agenciesdoes not provide a detailed narrative that would provide an opportunity to link an incident and a later crash. The resultsofthe analysissuggestthat more than 15 percent(97 of 627) of all crashesreportedby police may havebeena result of something that occuned earlieron the roadway.Ifthe findingshold, evenat a somewhat lesser level than 15 percent, they suggestthat prompt responseto, and clearanceof, incidentson roadwayscan be important in reducingthe effect ofincidents on other traffic. In somecases,the incident can be removedfrom the roadway so that it doesnot havean opportunity to affect traffic. For example,often a ffaffic law violator can be stoppedin sucha way that the vehicle andthe police car areoff the roadway (e.g.,in a parking lot or on a side street).Even ifthe incidentcannotbe eliminated,an assumptionthat astime increasessodoes the likelihood of a secondarycrash cannot be discarded.Therefore, respondersneedto considermanagingthe incidentsto reducethe time alongwith the effect that incidenthason othermotoristswho may subsequentlybe involved in a collision. The arterialincident management study produced a number of recommendationsthat police and other responderscould employ to reducethe effectsofincidents (7). Reducing secondarycrashesis impoftant, if for no other reasonthan it will reducethe police personnelrequiredto handlecrashinvestigation. What is needednow is a more thorough investigationof cause and effect. Part of the effort is to establisha better data base,which provides information about all actions taken by police and fire

TRANSPORTATION RESEARCH RECORD 1581

TABLE 7

Comparison of Primary Incidents with All Incidents Peak

Citv

Ail

Travel

Incidents

Incidents*

vlount Prospect )alatine

506 183 394 327 218

)rospect Heights

TUJ

\rlington Heights ]uffalo Grove ilk Grove Village

tollins Meadows total

65 1796

Percent

Primary Incident

410 138 340 240 181 85 54 1448

25 10 14 10 13

586 35 320 4t7 90

49

Percent of All

I

5

8t

of Peak Travel

4.9E, 5.5E1 3.65, 3.91 6.0q' 3.9E 7.7E 4.5E,

6.1E, 7.2q, 4.lq 4.29, 7.2E 4.7E, 9.35, 5.65,

7.8q, 6.8q, 1.9, 4.1q, 3.0q, 4.5q,

8.4ot, 8.6q,

llassification lrash lire itop r'ehicle )ther

ozI

44 534 491 100 179(

ncidents t448 x Incidents occurring between 0600 and 2059 daily.

respondersand which has sufficient information to allow computing of the time when the incident started and the time when it was cleared. Further study also could benefit from better police reporting of crashes,which calls attention to earlier events when applicable and is more specific about traffic conditions and possible contributing factors.

ACKNOWLEDGMENTS This study was funded by the Illinois Departmentof Transportation (IDOT) and the CongestionMitigation for Air Quality funds of the Environmental Protection Agency. Special support came from District I of IDOT under leadership of Duane P. Carlson, District Engineer. Terry Rammacher,Arterial Traffic OperationsManager, served as project manager and provided important guidance and support throughout the work. Joining him were Arland (Ted) Smith and JosephMcDermott, who added their respectiveexpertise. Additional credit also is given to police and fire agenciesin the study area,especially the Northwest Central Dispatch. Background literature was assembledby Hema Ramachandran,Dorothy Ramm, and Mary McCreadie at Northwestern's TransportationLibrary.

3 6

20 J

81

r.9s, 4.8q, 3 \q, 5.6q,

REFERENCES l. Raub, R. A., R. E. Lucke, J. L. Schofer, C. H. Mcl-ean, and E. Bard. Managing Incidents on Arterial Roadways. Final Report. Northwestern University Traffic Institute, Evanston, I11.,Feb. 1996. 2. Owens, D. Trffic Incidents on the Ml Motorway in Hertfordshire. Transport and Road Research Laboratory, Crowthorne, Berkshire, England, 1978. 3. Roberts,N. P., S. A. Webb, and G. Coe. Incidents onMotorways. Trffic Engineeing and Control,Yol.35, No. 10,Oct. 1994,pp.550-554. 4. Lai, 4., D. Christianson,and S. Porter.1--l5WIncident Managementand Impact of Incidents on Freeway Operations. Report TE-82l04. FHWA, U.S. Department of Transportation, Jan. 1982. Available from NTIS, Springfield, Va. 5. Lindley, J. A., and S. C. Tignor. Getaway Flow Rates for Freeway Incident and Geometric Bottlenecks.Public Roads, Vol.43, No. I, June 1919,pp. 1-7. 6. Willmink, I. R., and L. H. Immers. Deriving Incident ManagementMeasuresUsing Incident Probability Models and Simulation. ln Transportation Research Record 1554, TRB, National Research Council, Washington, D.C., 1996, pp. 196-203. 7. Raub, R. A., J. L. Schofer,and R. E.Lucke. Handling Incidents on Urban Roadways: A Summary of Issues and Recommendalior?s. Northwestern Universitv Traffic Institute. Evanston.IIl.. Feb. 1996. Publication of this paper sponsored by Committee on Traffic Records and Accident Analysis.