Analysis of Work Zone Crash Characteristics and Countermeasures (2018)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

NCHRP Project 17-61 35 CHAPTER 3 Effects of Queuing and Crash Countermeasures at Interstate Work Zone Lane Closures Overview The analyses of the work zone crash datasets described in Chapter 2 verify the perceptions held by many practitioners that queues and congestion are a key contributor to rear-end collisions in work zones, especially on high-speed facilities such as freeways and interstates. Meanwhile, as noted in Chapter 2, multiple studies have shown rear-end collisions to be the predominant type of crash occurring in work zones, and often the type of crash that increases most significantly when a work zone is installed on a roadway segment. Unfortunately, past studies have not been able to quantify how significantly a queue caused by a work zone increases crash risk. Conceptually, the greater the amount of time that a queue is present at a location, the greater the chance for a crash caused by a queue to occur. Similarly, a higher number of vehicles encountering a queue over a given duration would be expected to lead to a greater chance of a crash than if fewer vehicles encountered that same queue. The ability to predict the expected increase in crashes that a queue would create would be valuable to work zone designers and project engineers as they make decisions about when to close travel lanes, how long to allow them to be closed, and whether to employ countermeasures to reduce the risk of those crashes occurring. With regards to countermeasures available to reduce crash risks due to traffic queues at work zones, many agencies hire law enforcement personnel and vehicles to deploy within or upstream of the work zone. In one study, researchers found that the presence of enforcement (as could be best recalled by the project engineer) resulted in a 41.5% reduction in crashes overall relative to work zones where enforcement was not present (Chen and Tarko 2012). However, traffic conditions during the times of enforcement were not available and so could not be incorporated into the analysis. More recently, researchers have analyzed the effect of work zone ITS technology combined with portable rumble strips (PRS) upstream of temporary (a single work shift) work zone lane closures where traffic queues were anticipated to occur during some portion of the work shift (Ullman et al. 2016). In their analysis, they concluded that the combination of the real-time end-of-queue warning system (EOQWS) and PRS resulted in a 44% reduction in crashes from what would have occurred had the technologies not been used. In addition, severe crashes and rear-end crashes both appeared to be reduced, consistent with expectations. A major limitation of that analysis, however, was that the actual presence or absence of queues during those temporary lane closures were not considered. Although queues were anticipated during each of the lane closures examined, differences in traffic demands from one day to the next and other factors did result in some lane closures not experiencing traffic queues, and the duration of the lane closures when queues did exist varied from location to location and by day of the week. Fortunately, the corridor from which the data was taken was also instrumented with Bluetooth readers that were configured to measure current point-to-point travel times. The data can provide an estimate of times and locations when and where traffic queues had developed at the temporary work zone lane closures, and were used to examine the effect of queues due to the lane closures affected crashes. In addition, the effect of using portable EOQWS and PRS upon crashes during periods of queuing and periods of non-queuing were also examined.

NCHRP Project 17-61 36 Methodology Description of the Study Corridor The Interstate 35 (I-35) Central Texas Expansion Project is a seven year, $2.1 billion highway expansion project currently underway along 96 miles within the Waco District, located between Austin and Dallas. The corridor carries between 55,000 and 111,000 vehicles per day, of which 25% to 35% are truck traffic, with higher percentages of trucks during the nighttime hours. The main purpose of the project is to widen the highway to six lanes in rural areas. The project additionally involves improving the freeway infrastructure and changing the current two-way frontage roads into one-way frontage roads. Most of the work is performed alongside the current freeway or protected by barriers in the median, but occasionally, lane closures are necessitated for some construction activities. To reduce the traffic disruption from these activities, the Texas Department of Transportation (TxDOT) mandated that the temporary lane closures only be performed at night, between the hours of 7 PM and 7 AM. However, traffic volumes are still substantial enough for queuing to occur in some locations on some nights. Given that most of the corridor is rural, queues are unexpected at night by motorists, which raised safety concerns about rear-end collisions within TxDOT. Description of the Queue Warning Crash Countermeasures Because of practical concerns of using traditional work zone ITS technologies that are placed on the roadside and left for the duration of the project, a highly-portable EOQWS concept based on easily- deployable radar speed sensors (to minimize calibration requirements) was selected. The sensors are linked wirelessly to a central data processing unit, as are one or more portable changeable message signs (PCMSs). System logic evaluates the status of the sensors, and automatically displays an appropriate queue warning message based on the distance from the sign to the location of closest sensor detecting slowed or stopped traffic (see Figure 10). In one configuration, sensors are placed at the lane closure merging taper and then at 0.5, 1.5, and 2.5 miles upstream of the taper. A portable changeable message sign is placed 3.5 miles upstream of the taper. For lane closures where longer queues are expected, additional sensors are placed 3.5, 4.5, 5.5, and 6.5 miles upstream of the taper, and a second PCMS is positioned 7.5 miles upstream of the taper. The concept requires minimal calibration (sensors simply need to be aimed towards oncoming traffic properly) and so could be easily incorporated into the setup of the temporary traffic control for the lane closure each night it was needed, and quickly removed the next morning. Another technology of interest to TxDOT was the use of PRS. Although rumble strips in general do not typically have a dramatic effect on vehicle speeds in work zones, it was hypothesized that the tactile and auditory stimuli they provide could be beneficial in gaining the attention of distracted drivers approaching a work zone lane closure. Until recently, transverse rumble trip applicability to work zones was limited to long-term deployments because a portable and reusable rumble strip did not exist. However, one vendor developed a rumble strip product which could be quickly put down and picked up and which was reported to be able to remain in place relatively well under traffic (Heaslip et al. 2010). The PRS do require periodic attention to ensure that they remain in place. For this application, the PRS were put down as part of the temporary traffic control for the lane closure in the evening, and then picked back up when the temporary traffic control was picked up. The PRS were installed before and after the PCMS, shown as the transverse dotted lines in Figure 10. Figure 11 illustrates the selected product deployed approaching a work zone on I-35.

NCHRP Pr F oject 17-61 igure 10. Conceptual operation of the 37 portable queue warning system.

NCHRP Pr Figure 11 Descript Since 2 the corrid expected since the queue wa portable E of potenti nighttime data, for w over 400 closures w Determin Queuin Bluetooth Bluetooth likewise a that vehic used to de Each o Bluetooth segments upstream. assess wh oject 17-61 . Portable r ion of the A 012, a queue or. This ana traffic capaci beginning of s expected. F OQWS be d al queued c lane closure hich every l nights of la ith safety tre ing When Tr g was determ -enabled ele device iden ssociates a t le. This trav tect queuing f the lane cl sensors are s were identif Data for th ether the tra umble strips nalysis Me prediction an lysis compar ty of the clo 2013, TxDO or lane closur eployed in co onditions. I s during that ane closure h ne closures w atments, wer affic Queues ined using B ctronic devic tifiers. Wh imestamp an el time data i that may hav osures had m imilarly geo- ied that enc ese segments vel time thro used upstre thodology alysis was u ed the expec sure and det T mandated es predicted njunction wi n 2012, no t year provide ad traffic saf ithout safet e analyzed. Were Prese luetooth sens es in passin en a sensor d the travel t s then aggreg e occurred at ile markers located withi ompassed th on that nigh ugh that seg 38 am of temp ndertaken for ted traffic de ermined if qu that PRS be to cause queu th PRS to pr raffic safety d a controlle ety treatment y treatments nt ors deployed g vehicles a downstream ime through ated and ave specific loca associated w n the corrido e lane closu t were extrac ment increas orary lane c all lane clos mand during euing was p deployed ir ing, TxDOT ovide maxim treatments w d baseline fo s of some ki deployed, a along the I-3 nd associate encounters the roadway raged every tions along th ith its start r. For each c re and appr ted from the ed by enoug losures on I- ures propose the hours of robable. Fo respective of additionally um possible ere deploye r comparison nd deployed. nd another 7 5 corridor. T timestamps the same un segment can fifteen minu e corridor. and end po losure on eac oximately th archived da h to suggest 35. d by contract the closure r all lane clo whether or mandated th warning to d d, and temp to the 2013 For this ana 00 nights of hese sensors with the u ique identif be calculate tes, and the r ints. Each o h night, Blue ree to five tabase and us that queuin ors in to the sures not a at the rivers orary -2016 lysis, lane ping nique ier, it d for esults f the tooth miles ed to g had

NCHRP Pr developed travel tim minute tim present at average tr consecutiv lane closu 5-10 mph times wou magnitude closure m speeds thr The ho were then lane closu analysis f  hours w  hours w  hours w  hours w  hours w  hours w Figure 12 closure s oject 17-61 . If it had, th es over the a e intervals) the lane clos avel times in e 15-minute res were app while traver ld not incre across the erge taper wo ough the lane urs and porti recorded, an re. Consequ or the set of n hen queues w hen no queu hen queues w hen queues w hen queues w hen queues w . Illustratio egment. e duration w nalysis segm . Each grap ure during ea creased by a time periods roximately tw sing the lane ase more tha two miles. C uld be enoug closure) to e ons thereof w d the other h ently, there ighttime lane ere present es were prese ere present ere not pres ere present ere not pres n of the man hen queuing ent of each l h was then ch 15-minute t least 120-se . The 120-se o miles in le closure even n this 120-s onversely, e h to increase xceed the 12 hen queuing ours of the la existed the fo closures exa and no safety nt and no saf and only PRS ent and only and EOQWS ent and EOQ ual queue 39 was present ane closure w reviewed ma interval. Th conds (label cond thresho ngth. Recog if no queui econd thresh ven a small travel times 0-second thre was determ ne closure w llowing six mined: treatments w ety treatmen were deploy PRS were de and PRS we WS and PRS detection pr was also esti as graphed nually to as is manual pr ed as "moun ld was selec nizing that v ng was prese old due sole (i.e., 0.25 m by enough (i shold. ined to exist ere designate major catego ere deployed ts were deplo ed ployed re deployed were deploy ocess for ea mated. To ac over time (a sess if and w ocess looked tains" in Figu ted because t ehicle speeds nt, it was es ly to a speed ile) queue fo n conjunctio at each nigh d as non-que ries of condi yed ed ch tempora complish thi ggregated int hen queues for regions w re 12) acros he majority normally dro timated that reduction o rming at the n with the red ttime lane cl uing times fo tions availab ry nighttime s, the o 15- were here s two of the pped travel f this lane uced osure r that le for lane

NCHRP Project 17-61 40 Generally speaking, queues only developed for a portion of the nighttime lane closures when queuing did occur, and for many of the nights, no queues at developed at the lane closures. As a result, the dataset was heavily weighted towards non-queuing hours. Table 5 provides a summary of the hours associated with each of the above six categories. On nights when a traffic queue did form at a lane closure, the queue typically lasted three to five hours before dissipating. The data for the no safety treatments-queued conditions and the only PRS-queued conditions actually represents approximately 120 lane closures each, and the both EOQWS & PRS deployed – queued conditions comes from approximately 206 lane closures. Table 5. Sample Sizes Analysis Category Total Hours No safety treatments - queued conditions 491.5 No safety treatments - non-queued conditions 3941.5 Only PRS deployed - queued conditions 491.5 Only PRS deployed - non-queued conditions 2674.5 Both EOQWS & PRS deployed - queued conditions 1763.5 Both EOQWS & PRS deployed - non-queued conditions 827.5 Safety Analysis Methodology As stated previously, the location of lane closures changed continuously throughout the corridor as contractors worked on and completed various tasks within each project. The fact that the lane closures when no safety treatments were deployed were not necessarily at the same locations as those where PRS only or the combined EOQWS and PRS treatments were deployed meant that it was not appropriate to simply compare crashes occurring under each of the above six categories. Therefore, the analysis approach ultimately adopted was to compare expected baseline crashes (had the work zone and nighttime lane closures not been present in the corridor) to crashes that actually occurred during each of the six categories of interest. To accomplish this, the corridor was divided into a set of discrete homogenous segments and Empirical-Bayes analysis methodologies applied for years 2003-2009, using the Enhanced Interchange Safety Analysis Tool (ISATe) (Bonneson et al. 2013) which is based on models published in the 2014 supplemental document to the HSM. The result of that analysis was an expected number of crashes/year in each segment. The average proportion of crashes occurring during the same nighttime hours that lane closures were allowed during construction was then calculated from the calibration data, and applied to each of the segments to obtain the expected nighttime crashes/year for that segment. Dividing these yearly values by 365.25 yielded an expected number of crashes/night within each segment along the entire corridor. To estimate the expected crashes per night (ECN) for each lane closure included in the dataset, a lane closure section was selected that included the closure length itself plus a section that extended five miles upstream of the beginning of the lane closure. The ISATe segments corresponding to the lane closure and upstream section was determined, and the expected number of crashes from the ISATe analysis corresponding to that lane closure section determined. In cases where the lane closure section began or ended within one of the discrete homogenous segments, the expected crash/night value was adjusted based on the proportion of the segment overlaid by the lane closure section. Figure 13 illustrates the estimation process. Finally, the expected crashes per night were then further subdivided according to the proportion of crashes occurring each hour per night to yield an expected number of crashes during each hour per night within that lane closure section.

NCHRP Pr Figure 13 Result As show deployed Likewise, than the n TxDOT a when the given nigh dataset fo conditions Table 6 in each c occurring during ho whereas th present. T crashes du zone or la what wou closures h work zon deployed the no sa EC w EC Se mod X oject 17-61 . Method of s n in Table experienced the number umber of hou ttempted to d lane closures t, there were r the combin than hours w present the ategory if th during those urs when qu e actual cras he increase ring hours a ne closure w ld have norm ad not occurr e and lane cl and queues w fety treatmen Nlane closure sec here, N = Expect gment i = ith el = Proportion estimating e 5, the nightti many more of hours whe rs when only eploy both t were deploy occasions w ed EOQWS hen no queu expected base e work zone hours at the eues were pr hes were onl is particularly nd locations ere present. ally been exp ed. One see osures during ere present. t condition. tion = x*ECN ed Crashes/N Homogeneo of ISATe se xpected cra me lane clos hours of n n PRS only w PRS were d he EOQWS ed. Howeve hen queues d and PRS dep es were prese line crashes and lane cl lane closure esent were s y somewhat evident for where only This repres ected in tho s that crashes times when However, th During que Segment i-1 + EC ight us segment gment within 41 shes per nig ures during 2 on-queued t ere deploye eployed and and PRS on r, due to the id form whe loyment con nt. that would h osures had n locations. On ubstantially greater than b the no safety 3.152 crashes ents the equi se locations d are much hig only PRS w e proportiona ued conditio NSegment i + E in correspond the analysis ht within la 012 when n raffic condi d when no q queues forme nights when stochastic na n only PRS w ditions cons ave occurred ot occurred, e sees that t greater than aseline durin treatment co would have valent of a 4 uring those t her than wo ere deployed l increases ar ns when on CNSegment i+1 ing ISATe section ne closure a o safety trea tions than q ueues were p d. This was queues were ture of traffi ere deployed isted of mor during the h as well as he number of the baseline g hours whe ndition, whi normally oc 71% increas imes if the w uld have been or both EOQ e substantial ly PRS were nalysis sect tments were ueued condi resent was g to be expect expected to c demands o . Conversel e hours of qu ours and loca the actual cr crashes occu expected cra n queues we ch experienc curred if no e in crashes ork zone and expected w WS & PRS ly smaller th deployed, a ion. being tions. reater ed, as form n any y, the eued tions ashes rring shes, re not ed 18 work from lane ith no were an for ctual

NCHRP Project 17-61 42 crashes were 277% higher than would have been expected if no work zone or lane closure had occurred. During queued conditions when EOQWS & PRS were deployed, actual crashes were 331% higher. Table 6. Expected and Actual Crash Results Analysis Category Crashes Expected if No Work Zone & Lane Closure Actual Crashes During Lane Closures No safety treatments - queued conditions 3.15 18 No safety treatments - non-queued conditions 22.61 29 Only PRS deployed - queued conditions 3.45 13 Only PRS deployed - non-queued conditions 13.69 18 Both EOQWS & PRS deployed - queued conditions 4.17 18 Both EOQWS & PRS deployed - non-queued conditions 9.14 10 The relative effect of the two safety treatments under both queued and non-queued conditions can be computed using an odds ratio analysis as outlined in available guidance on computing crash modification factors (Gross et al. 2010): ܥܯܨ்௥௘௔௧௠௘௡௧ ൌ ܶܣ்௥௘௔௧௠௘௡௧ ܶܧே௢ ்௥௘௔௧௠௘௡௧ܶܧ்௥௘௔௧௠௘௡௧ܶܣே௢ ்௥௘௔௧௠௘௡௧ ቀ1 ൅ 1ܶܧே௢ ்௥௘௔௧௠௘௡௧ ൅ 1 ܶܧ்௥௘௔௧௠௘௡௧ ൅ 1 ܶܣ்௥௘௔௧௠௘௡௧ቁ ܵܧሺܥܯܨொ௨௘௨௘ ሻ ൌ ඪ ܥܯܨ்௥௘௔௧௠௘௡௧ଶ ቀ 1ܶܣ்௥௘௔௧௠௘௡௧ ൅ 1 ܶܧே௢ ்௥௘௔௧௠௘௡௧ ൅ 1 ܶܧ்௥௘௔௧௠௘௡௧ ൅ 1 ܶܣே௢ ்௥௘௔௧௠௘௡௧ቁ ቀ1 ൅ 1ܶܧே௢ ்௥௘௔௧௠௘௡௧ ൅ 1 ܶܧ்௥௘௔௧௠௘௡௧ ൅ 1 ܶܣ்௥௘௔௧௠௘௡௧ቁ Where, CMFTreatment = crash modification factor representing the proportional effect of the treatment upon crashes during nighttime lane closures for the particular traffic condition of interest (queued or non-queued) SE(CMFTreatment)= standard error of CMFTreatment TATreatment = total crashes actually occurring when the treatment was deployed for the particular traffic condition of interest (queued or non-queued) TANo Treatment = total crashes actually occurring when the treatment was not deployed for the particular traffic condition of interest (queued or non-queued) TETreatment = total crashes expected when the treatment was deployed for the particular traffic condition of interest (queued or non-queued) TENo Treatment = total crashes expected then the treatment was not deployed for the particular traffic condition of interest (queued or non-queued) Table 7 presents the CMFs associated with the PRS only treatment and the EOQWS and PRS combined treatment for both traffic conditions. During hours when no traffic queues were present, neither the PRS only nor the combined EOQWZ and PRS treatment resulted in a statistically significant CMF (although the values were both slightly less than 1). Conversely, during hours when a traffic queue

NCHRP Project 17-61 43 was present, both treatments yielded significant reductions in crashes relative to what would have been expected had the treatments not been deployed. Interestingly, the PRS only treatment achieved essentially the same crash reduction as the EOQWS and PRS combined, yielding 60.3% and 53.2% reductions in crashes, respectively. In other words, deploying both the EOQWS and PRS together had almost the same effect as did deploying PRS only when traffic queues were present. It should be noted that there may be some bias in the data and these results since the PRS only condition was deployed when queues were not expected, whereas the EOQWS and PRS together were deployed where queuing was expected. It is possible that some of the queuing for the PRS only deployments may be due to unusual traffic conditions such as incidents or unusually high traffic volumes on a particular night, and so comparison to the baseline queued condition may overstate the effect of the treatment. Table 7. Effects of Safety Treatments Safety Treatment Traffic Condition Crash Modification Factor Standard Error P-Value Only PRS deployed Non-queued 0.890 0.377 0.771 Queued 0.397 0.265 0.023 EOQWS and PRS deployed Non-queued 0.717 0.353 0.424 Queued 0.468 0.301 0.078 Aggregating the results across both treatment types, it also appears that utilizing the safety treatments reduced the severity of crashes that did occur at these nighttime lane closures. As shown in Figure 14, 50 percent of the crashes occurring during hours when a traffic queue was present during a nighttime lane closure but no safety treatment was deployed involved injuries or fatalities (i.e., were considered severe crashes), whereas only 16.1% of the crashes during hours of traffic queuing when the safety treatments were deployed were of the severe category. For comparison purposes, the percentage of crashes occurring during hours when no traffic queues were present that were of the severe variety were essentially the same, 28.6% and 31.0% with and without safety treatments deployed, respectively. Discussion It is important to be cautious when interpreting the results of this analysis. Although the dataset itself is comprised of several hundred nights of temporary lane closures, it does only represent a single corridor with multiple projects that were ongoing simultaneously over several years. In addition, although the results of the analysis did yield statistically significant conclusions, the number of crashes included in the analysis is very limited (as shown in Table 5). Finally, the location of the crash relative to the traffic queue was not examined as part of this analysis. It is likely that not all crashes occurring during the queued traffic conditions occurred at the upstream end of the queue. Rather, some of the crashes may have been lower-speed rear-end and sideswipe collisions within the queued section. Despite this limitation, though, the results of the analysis do provide useful insights into the magnitude of increases in total crashes due to work zone lane closures when queues are and are not present. This analysis focused strictly on the total crash reduction effectiveness of the two safety treatments being deployed during nighttime lane closures on the I-35 corridor. However, both types of treatments have unique cost and deployment characteristics that also need to be taken into consideration when deciding what type of safety treatment to deploy. PRS tend to have lower capital cost requirements than EOQWS, but have higher deployment and retrieval costs. PRS require workers to be out in the travel lanes for deployment and retrieval, which increases safety risks (EOQWS are deployed on the roadside and typically do not require workers to be out in a travel lane). Consequently, practitioners must weigh

NCHRP Pr these cons (if any) to Figure 14 queuing. The res transporta imply, de not likely PRS depl deploying offset by be such a reduction oject 17-61 iderations in utilize on a g . Effect of ults of this tion manage ploying these yield much o oyments that , and retrievi a reduction in major contri effects of the addition to t iven project. safety trea analysis also ment plannin safety treatm f a safety ben require work ng these trea crashes. Ho butor to cras se counterme he costs of th tments upon illustrates g as well as ents at loca efit, and may ers to be out tments at loc wever, given hes, the impo asures is sub 44 e treatments crash sev the importan throughout tions where q actually inc in the trave ations where that queues rtance of the stantial. when makin erity during ce of good the duration ueues are no rease safety r l lanes). In e queues are n and congest se results in g decisions o hours of q traffic impac of the proje t anticipated isks for work ssence, the c ot anticipated ion at work z quantifying n which trea ueuing and t analyses d ct. As the r to develop w ers (especial osts of prov will not like ones was fou the potential tment non- uring esults ould ly for iding, ly be nd to crash