Click for next page ( 47


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 46
46 CHAPTER 5 ANALYSES AND RESULTS INTRODUCTION using both full-scale testing and computer simulation: Olsen et al. (22) and Holloway et al. (24). In those studies the This chapter summarizes the analyses described in Chap- response of various types of cars traversing a number of dif- ter 4 and their results, as well as relevant analyses from prior ferent curb types was obtained and the information was used studies. In most cases, significantly more detail was provided to assess vehicle stability and to estimate the potential for in one or more appendices or other publications. These barrier override and underride. Roll and pitch displacement- results were used to develop the guidelines that were the pri- time histories and relative bumper trajectorytime history of mary product of this research project, described in Chapter 6. vehicles traversing curbs were collected in their studies. Var- ious impact conditions and curb types were investigated in those studies, and all impact conditions were considered PRIOR STUDIES equally likely since data were not available to discern the most probable impact conditions of crashes. Only the maxi- The analyses of vehicle impact with curbs and curbbarrier mum values of angular displacement and bumper height from combinations conducted in this study were limited to one vehi- the various studies were considered when synthesizing the cle type, a 2000-kg pickup truck. Thus, guidelines based solely data for use in this study. The maximum encroachment angle on the results of those analyses would only be applicable to of both the Olsen et al. study and the Holloway et al. study that one type of vehicle. In order to develop a more general was 20 degrees, whereas the maximum encroachment angle set of guidelines, additional information was needed about used in the current study was 25 degrees. Also, since the the response of a broader range of vehicle types. The liter- vehicle used in the Olsen et al. study was a 1965 Ford four- ature provided an adequate amount of information on the door sedan, those results may not be representative of the response of various types of cars traversing curbs and also a current vehicle fleet. The results and conclusions from the limited amount of information from the results of full-scale study by Olsen et al., however, were similar to those obtained crash tests regarding both cars and pickup trucks impacting in both the Holloway et al. study and the current study. curbbarrier combinations. There are many factors that influence vehicle behavior when traversing curbs, such as abrupt steering caused by the CRASH AND INVENTORY DATA ANALYSES interaction of the front wheels with the curb; loss of contact between the tires and ground; excessive vehicle accelerations; This section describes six analyses of crash and inventory and excessive roll, pitch, and yaw rates of the vehicle during data that were conducted in this study. impact. Although each of these factors may lead to loss of control of the vehicle, all the data that have been collected Extent of the U.S. Curb-Related Safety Problem from full-scale tests and computer simulations suggest that total loss of control was unlikely except in extreme cases. A The goal of this effort was to define the extent of the more important issue, however, may be the effects that these national safety problem related to curbs using FARS and factors precipitate when curbs are placed in combination with NASS-GES data. A more detailed description of this task can roadside hardware (e.g., guardrail, crash cushions, or break- be found in Appendix C. away poles). Even a slight increase in bumper height caused Table 10 presents FARS data for the 199499 period con- by traversing a curb may be sufficient to cause the vehicle to cerning how often "Curb" was noted as the FHE in fatal impact a roadside safety device at a point higher or lower crashes on roads with speed limits of 40 mph (65 km/h) or than normal, which may lead to override or underride of bar- greater for all crashes and SV fatal crashes. Table 11 presents riers or adversely affect the breakaway mechanism of other NASS-GES national estimates for crashes of all severity on roadside devices. these higher-speed roads. These estimates are based on the Two of the studies identified in the literature review GES weighting system, which was applied to the 50,000 cases addressed the issue of override and underride indirectly collected each year. In both tables, the FHE differs from the

OCR for page 46
47 TABLE 10 Fatal crashes with curb as FHE and speed limit 40 mph (65 km/h) or greater Total fatal Crashes with Percent of total Fatal single Fatal SV crashes Percent of fatal Year crashes curb as FHE fatal crashes vehicle crashes with curb as FHE SV crashes (SL>=65 km/h) (SL>=65 km/h) (SL>=65 km/h) (SL>=65 km/h) (SL>=65 km/h) (SL>=65 km/h) 1994 27191 291 1.1% 14520 271 1.9% 1995 28005 387 1.4% 15206 345 2.3% 1996 28464 396 1.4% 15293 359 2.3% 1997 28171 391 1.4% 14906 362 2.4% 1998 28453 325 1.1% 15227 308 2.0% 1999 28527 329 1.2% 15062 307 2.0% MHE in that the FHE was not necessarily the fatality or wide. Clearly, curbs are the initial objects struck in only a injury-producing mechanism. Curb impacts are actually very small portion of fatal or total crashes on the roadways of seldom the MHE in fatal crashes. interest and on all roadways. As shown in Table 10, curbs on higher-speed roads were The remainder of the analyses in this task examined other noted as the FHE in slightly over 1% of all fatal crashes on factors related to these fatal and nonfatal crashes. All were these roads each year; and, while varying somewhat, the fre- restricted to the higher-speed roads (i.e., speed limits of quency and percentage were relatively stable across the 6-year 40 mph [65 km/h] or greater) and to the 1999 FARS and span. Curbs were noted as the FHE in approximately 2% of NASS-GES data. While only the highlights of the findings all SV fatal crashes on these roads. are included here, more detail is presented in Appendix C. Table 11 shows that in terms of total crashes (fatal and nonfatal), curbs were noted as the FHE in fewer than 1% of Curb crashes were more urban than other crashes: 72.3% all GES crashes each year, a rate even lower than that for of the curb-related fatal crashes were on urban roads, fatal crashes. As with fatal crashes, while varying slightly, with 26.7% of the total on urban Interstates or other the frequency and percentage were relatively stable across freeways/expressways. For the GES national estimates, the 5-year span for crashes reported in the GES database. almost half (49.5%) of the total higher-speed curb- Curb impacts were the FHE in approximately 2.5% of the SV related crashes were in urban areas with populations crashes. greater than 100,000, and 71.9% were in areas with pop- These analyses were for roads with speed limits of 40 mph ulations greater than 25,000. The location of these curb- (65 km/h) or greater. If one included all speed limits, and thus related fatal and total crashes differed significantly from urban streets where curbs are standard, the percentages for the location for all SV crashes on these higher-speed both fatal crashes and total crashes would increase, but not to roads: 71.2% of fatal SV crashes occurred on rural a large extent. For example, in the 1999 FARS data, there are roads, and 61% of total SV crashes occurred on road- a total of 599 fatalities in which curb was the FHE on all ways within areas of population less than 25,000. These roadways (regardless of speed limit). This represents approx- findings probably reflect the fact that curbs were more imately 1.6% of the total fatal crashes in 1999. Similarly, the often located in urban areas. 1999 weighted GES data for all speed limits indicate curb The MHE in fatal curb crashes was often a rollover, but crashes represent approximately 1.0% of the crashes nation- the MHE in total curb crashes was the curb impact itself. TABLE 11 National estimates of crashes with curb as FHE and speed limit 40 mph (65 km/h) or greater, weighted data Percent of Percent SV Total crashes Crashes with SV crashes SV crashes with crashes with crashes with Year nationwide curb as FHE nationwide curb as FHE curb as FHE curb as FHE (SL>=65 km/h) (SL>=65 km/h) (SL>=65 km/h) (SL>=65 km/h) (SL>=65 km/h) (SL>=65 km/h) 1995 2765377 23680 0.9% 855097 21784 2.5% 1996 2857985 23470 0.8% 899940 21761 2.4% 1997 2839031 20107 0.7% 876545 18981 2.2% 1998 2781930 23908 0.9% 844783 23002 2.7% 1999 2753457 21807 0.8% 835853 20843 2.5%

OCR for page 46
48 When curbs were noted as the FHE on these higher- erty damage to result in a reportable crash but that fatalities speed roads, 38.9% of the 368 vehicles involved in fatal were more likely to result from a rollover. Finally, both fatal crashes were coded as having "Overturn" as the MHE. and total curb-related crashes differed from other SV fatal This was very similar to the 39.6% of all fatal SV crashes crashes on these higher-speed roadways in that they were in which overturn was the MHE. Unfortunately, the data much more likely to occur on urban roads and more likely to did not reveal whether the rollover was related to tripping involve passenger cars rather than SUVs or pickups. on the curb or to an embankment or other object behind the curb. Only 18 (4.9%) of the 368 vehicles in fatal curb- related crashes, of which 15 were motorcycles, were Curb-Related Rollover Risk coded as having the curb as the MHE. In contrast, for and Nature Given a Crash total crashes on these higher-speed roads, only 12.7% of Rollover occurrence and risk is of particular interest when the vehicles were coded as having "Overturn" as the MHE, and 50% were coded as having "Curb" as the curbs are being studied, since the severity of impacts with MHE. The higher percentage of "Overturn" in the fatal this low-profile object would be expected to be related to data was the result of the nature of an overturn: once it whether a vehicle overturned rather than to energy exchange occurs, it is likely to be fatal. in the impact itself (unlike impacts with guardrails, for exam- The curb-related fatal crashes occurred predominately ple). The FARS analysis described in the preceding section at nonjunction locations (80.5%). An additional 10.6% highlighted the fact that the MHE in most fatal curb-related were at interchanges, with the majority of these being crashes is a rollover. This set of analyses was conducted to on ramps. The total crashes were also more likely to be further examine the incidence and nature of rollover in curb- at nonjunction locations, but not to the same extent related crashes. (53%). Here, approximately 26% were at intersections, In order to help ensure that the curb was directly related to and an additional 11.6% of the total crashes were at the rollover under study, databases chosen for use had to have interchanges, mostly on ramps. a sequence of events that would allow examination of only Pavement conditions (e.g., dry or wet) for fatal and total those rollovers preceded by a curb impact. NASS-CDS data curb crashes were very similar to those for all SV fatal and the HSIS data from both Michigan and Illinois included and total crashes: 90.3% of the curb-related fatal crashes such a sequence variable and were thus used in the analyses. occurred on dry pavement, 7.9% on wet pavement, and This subsection presents the results of the NASS cases and the 1.2% in snow/slush/ice; 74.7% of the curb-related total Michigan and Illinois cases separately. A detailed description crashes occurred on dry pavement and 21.8% on wet of this analysis can be found in Appendix D. pavement. There were only subtle differences in the vehicle maneu- vers prior to the crash (e.g., "going straight," "changing NASS-CDS Analysis and Results lanes," or "turning") for the curb and total SV crashes for both the fatal and GES samples. NASS-CDS data for the 199799 period were used in the Vehicle types in curb-related fatal and total crashes did analysis. Using the investigator-supplied sequence of events, differ somewhat from vehicles in the comparable SV cases were chosen that involved at least one impact with a groups on these higher-speed roadways; they were more curb on roads with posted speed limits of 40 mph (65 km/h) likely to be passenger cars (and motorcycles for the fatal or greater. The resulting sample was very small, particularly subset), and somewhat less likely to be SUVs or pickup for cases involving rollover. In the 3 years, there were 101 trucks. While these vehicle-related findings might be SV crashes involving a curb, and 38 of these involved a roll- related to differential exposure (e.g., more passenger over somewhere in the sequence. Of primary interest were cars on urban roads where more curbs were located), those impacts in which a curb was the first event in the they do not seem to indicate greatly increased curb- sequence (92 of the 101 curb-involved cases) and in which a related problems for SUVs or pickups. Again, this con- rollover immediately followed the curb impact and thus was clusion is tenuous given the lack of exposure data in both assumed to be related to it. the FARS and GES files. As noted earlier, the NASS-CDS data are from an unequal probability sample extracted from police reports from across In summary, curb-related fatal crashes on roadways with the nation and overrepresent more severe crashes. The data speed limits of 40 mph (65 km/h) and above represented a can be presented in two forms, unweighted and weighted. very small percentage of total fatal crashes (approximately The unweighted, or raw, data represent the actual number of 1%). Curb-related total crashes represented an even smaller cases in the sample. The weighted data represent the total percent of all crashes (less than 0.5%). Curbs were very sel- number of such cases that would have occurred nationwide, dom the MHE in fatal crashes (approximately 5%), but much given that the sample and the assigned weights are accurate, more likely to be the MHE in total curb-related crashes that is, given that the sample cases as a whole do in fact (53%). This implies that curb impacts caused enough prop- reflect the national incidence of all such crashes. The weight

OCR for page 46
49 for each case is assigned by NASS. The weighted estimates estimate might be approximately 10%, based on the weighted are considered reliable when a large number of cases is being and truncated-weight distributions. analyzed, but there are serious questions concerning the reli- These curb-related cases were also examined to see if ability of these estimates when relatively small samples are information could be extracted concerning vehicle impact being studied, as is the circumstance here. speed and whether the vehicle impacting the curb was track- Table 12 shows NASS-CDS data for the frequency of ing or nontracking. Unfortunately, the data did not provide overturn in SV crashes in which the curb impact was the first such information. event in the crash sequence. In this sample, three cases have extremely high weights, and those cases largely determine figures in the "Weighted" column. For this reason, a trun- Michigan and Illinois Analyses and Results cated version of the weighted data is also presented; it gives The HSIS databases for Illinois and Michigan used in this estimates based on the sample excluding the three very high analysis included SV crashes that occurred in 1996 and 1997 weight cases. on sections of roadways with curbs and posted speed limits at As shown in Table 12, the unweighted data indicate that in or above 40 mph (65 km/h). All such crashes were included 17% of these SV crashes in which a curb was the first event, in which a vehicle impacted a curb or another fixed object as an overturn occurred immediately after the curb impact. In an the FHE or the first "substantial" harmful event. The latter additional 21% of these cases, an overturn occurred at some subset included crashes in which the curb impact was not the point in the crash sequence but could not be attributed to the first event, but was only preceded by nonobject events such curb impact. When the full NASS weights are applied to the as "uncoded or errors," "loss of control," or "ran off road same data, an overturn occurred immediately after the curb left/right." For each case, it was also noted whether the vehi- impact in 9% of the cases. An overturn occurred subsequently cle that struck the curb or fixed object was involved in an in the crash sequence but could not be attributed to the curb overturn subsequent to the impact with the curb or fixed object. impact in an additional 6% of the weighted crashes. When the For comparison with the NASS-CDS data, the first analysis truncated weights are used, the overturn occurred in 13% of involved the overall rollover percentage for Michigan and for cases immediately after curb impact and in an additional 8% Illinois. The crashes were then categorized by land use (i.e., of the cases as a later event in the crash sequence. urban or rural) and roadway classification (i.e., Interstate or The rollover cases were further examined to see if the non-Interstate) for each state. Since assigned operating speed investigator noted "Curb" as the "Tripping object" in the for the roadway where the crash occurred is a combination of 16 cases in which the rollover immediately followed the curb land use and roadway class, distributions of rollover per- impact, as one would expect. This was only true in ten cases, centages were generated for four operating speed categories with three other cases having "Ground" as the tripping object within each state. The assigned operating speeds were based and the remaining three being uncoded. on results from the analysis described in Appendix B. Thus, there is some lack of certainty concerning the percent Table 13 presents the rollover percentages for each state. of SV curb impacts resulting in an overturn. It would appear In the Michigan data set, 5% of the SV curb crashes resulted that such overturns occur in at least 7% of the curb impacts in a subsequent overturn, while in Illinois, 8% subsequently (based on the weighted data where the investigator noted overturned. While the percentage of overturns in curb crashes "Curb" as the tripping object), and may be attributed to the is the same as for other objects in Michigan, the percentage curb impact in as many as 17% of the cases (based on the of overturns in curb crashes is higher than for other objects unweighted data, and assuming all overturns immediately fol- in Illinois (8% versus 2%). Both are in the same range as, but lowing the curb impact were caused by the curb). The best slightly lower than, the 10% best estimate from the CDS data. TABLE 12 Frequency of overturn in NASS-CDS SV curb impacts in which the curb impact was the first event in the crash sequence Incidence of Number of vehicles Percent of vehicles overturn No Full "Truncated" No Full "Truncated" weighting weighting weighting weighting weighting weighting Did not overturn 57 30178 19530 61.96% 85.57% 79.33% Overturn immediately 16 3109 3109 17.39% 8.82% 12.63% following curb impact Overturn, not immediately 19 1978 1978 20.65% 5.61% 8.04% following curb impact Total 92 35265 24617 100.00% 100.00% 100.00%

OCR for page 46
50 TABLE 13 Frequency of overturning vehicles in SV crashes in which either a curb or another fixed object was struck (Michigan and Illinois data, 199697) FHE = curb impact FHE = other fixed object impact State Did not Percentage Did not Percentage Overturned Overturned overturn overturns overturn overturns Michigan 1,487 83 5% 6,156 305 5% Illinois 361 30 8% 1,969 36 2% Total 1,848 113 6% 8,125 341 4% Similar tables of rollover percentages for curb and other- curb crashes resulting in rollover appeared to increase as the fixed-object crashes were also produced for urban and rural assigned operating speed increased. Interstate and non-Interstate roadway classes. The samples of Illinois curb-related crashes were too small to be meaningful except for the urban non-Interstate category. The sample for Summary the Michigan curb-related crashes in the rural Interstate cate- The early analyses of curb-related crashes on higher-speed gory was also too small. In the other three categories (i.e., rural roads indicated a relatively high frequency of "rollover" in non-Interstate, urban Interstate, and urban non-Interstate), the FARS fatal curb-related crashes (40%), and a significant, Michigan data indicated that the curb-related rollover per- though lower rollover percentage in the GES (all crashes) centages were very similar to the rollover percentages for data (13%). Since neither of these databases included a other objects. For the urban non-Interstate higher-speed roads, sequence of events allowing a better link between the roll- the Illinois data indicated a higher curb-related rollover per- over and the curb or other-fixed-object impact, NASS-CDS, centage than was found for other objects that are struck first Michigan, and Illinois data were analyzed. The relatively (7% versus 1%). When the data from the two states were small sample size of curb-related crashes on higher-speed combined, the curb-related rollover percentage was slightly roads and the issue of weighting led to difficulties in draw- higher on urban Interstates than on urban non-Interstates (8% ing firm conclusions from the 199799 NASS-CDS data. versus 5%). The Michigan and Illinois data provided somewhat larger The final analysis involved curb-related crashes classified samples. Even though conclusions were difficult with the by roadway operating speed for their crash location (see CDS data, the "combined estimate" of 10% rollover was sim- Table 14). As can be seen from Tables 13 and 14, the Illinois ilar to, but slightly higher than, the rollover estimates from curb-related rollover percentages were consistently higher Michigan and Illinois. The Michigan data indicated that the than the corresponding Michigan percentages. This could curb-related rollover percentages were very similar to the have been related to curb design or placement standards or to rollover percentages for other objects for the three roadway differences in crash reporting between the two states. If non- categories where adequate samples were found (i.e., rural injury crashes in Illinois were systematically reported less non-Interstate, urban Interstate, and urban non-Interstate often than in Michigan, the rollover percentage for Illinois roads). The Illinois data for urban non-Interstate roads, the would be higher since rollover crashes, which are more likely only category with adequate sample size, indicated a higher to result in injuries, were most likely to be reported fully in curb-related rollover percentage than was found for other both states. In both Illinois and Michigan, the proportion of objects (7% versus 1%). Finally, because of the small sample TABLE 14 Frequency of overturning vehicles in SV curb crashes categorized by roadway operating speed (Michigan and Illinois data, 199697) Assigned Michigan Illinois operating Did not Percentage Did not Percentage speed Overturned Overturned overturn overturns overturn overturns NA 50 4 7% 46 1 2% 40-49 mph 193 6 3% 59 6 9% 50-59 mph 633 28 4% 172 13 7% 60-69 mph 597 40 6% 75 8 10% 70-79 mph 30 5 14% 9 2 18% Total 1503 83 5% 361 30 8%

OCR for page 46
51 sizes and poor quality of the raw data available for these guardrail section, there could be multiple records referring to cases, it was not possible to extract further information from the same milepost on the highway, as a result of guardrails the NASS-CDS data on vehicle tracking/nontracking prior to being on each side of the road or even in the median as well curb impact or vehicle impact speed, both of which would be as on each side of the road. In contrast, the Michigan Road- presumably related to rollover risk. way Inventory File was organized by homogeneous segments In summary, rollover after impacts with curbs appears to of roadway, a new segment beginning when any change be a relatively low-frequency occurrence in all crashes. How- occurred in a major variable (e.g., divided/undivided, shoul- ever, it remains a problem worthy of design attention due der width/type, or lane width). When divided highways were to the severity of rollovers, as demonstrated by the higher present, there were separate inventory items for each direc- rollover percentages in fatal curb-related crashes. tion of travel, but the record included both directions. The presence of curbs on the roadway was found under the "shoul- der type" variable, and there were either two or four shoul- Crash, Injury, and Rollover Rates for ders on each homogeneous segment, depending on whether Guardrail Sections with and without Curbs the roadway was undivided or divided. Finally, the Michigan crash file had information on the crash milepost and the Since rollover after striking a curb on higher-speed roads direction of travel of each of the vehicles, but did not specify could be a significant cause of injury, different data sources the side of the divided roadway on which the crash occurred. were examined in an attempt to gather more information on The complicated nature of the guardrail file resulted in a rollover in the presence of curb and curbguardrail combi- complex data screening and merging effort involving a series nations. The analyses described previously concerned the of decisions (e.g., how to properly link crashes that are not risk of rollover once a crash has occurred and therefore used mileposted to different sides of the roadway). The product of crash data and a rollover subset within that data. The basic the significant data-preparation effort was an analysis file goal of the analysis described in this section was to examine containing 199394 crash counts and 1992 AADT and other curbguardrail-related crash risk and rollover risk, which is roadway characteristic data (e.g., the presence of a curb) similar to "crash rate" and "rollover rate," per passing vehi- linked to directional segments of 1992 guardrail for three cle for segments of highway with guardrails and segments highway classes: urban freeways, urban multilane divided with curbguardrail combinations. These guardrail and curb roads, and urban multilane undivided roads. Only these classes guardrail sections were not compared to roadway sections contained sufficient mileage of both guardrail-only sections without a curb or guardrail since reporting of crashes on the and curbguardrail combination sections, and even in these latter section is a function of the nature of the roadside classes the total directional mileage was limited (e.g., only beyond the shoulder, which is not in any roadway inventory 15 total miles of curbguardrail combination sections on urban file. A more detailed description of this effort can be found freeways). in Appendix E. Two types of analysis were conducted: simple comparison To examine guardrail-related crash and rollover risk or of guardrail versus curbguardrail crash rates per million rate per passing vehicle, a database was needed that allowed vehicle-miles of passing vehicles within each of the three identification of specific segments of roadway with guard- roadway classes, and regression modeling (i.e., Poisson and rails and with curbguardrail combinations that could be negative-binomial) in which AADT and other factors were linked with run-off-road crash and rollover counts, AADT, better controlled for. Details of both analyses are presented and other characteristics of the roadway, such as road classi- in Appendix E. fication, curvature, and speed limit. By definition, when one The crash rates developed for total SV crashes, injury- is attempting to compute "risk" or "rate," the analysis record producing SV crashes, and SV rollover crashes are shown in needed is a segment of highway, not a crash, since one must Table 2 in Appendix E for all three roadway classes. How- also include segments of highway which have had no roll- ever, due to the small number of such crashes, only the rates overs or crashes. The only database available for such an related to urban freeways appeared to be somewhat mean- analysis, and probably one of very few such databases in the ingful; these are shown in Table 15. nation, was the Michigan HSIS database. While most states For the urban freeways, it appeared that the total run-off- have roadway inventory files that include AADT and details road rate was slightly lower on guardrail-only segments than of the cross-section of the roadway to the edge of the shoul- on curbguardrail segments (0.175 versus 0.195 crashes per der, few include any information on guardrails. Michigan million vehicle-miles passing). This may have been due to had developed and maintained a separate guardrail inventory the presence of the curb as another object to strike on the file up through 1992. Each record identifies a section of roadside, or to other factors that were not accounted for in guardrail. The inventory provided details of location (i.e., these analyses (e.g., speed limit). Perhaps of more impor- side of the highway), beginning and ending milepoints, and tance, but with the same caveats, the injury crash rate in such details of the guardrail such as type, end treatment, and off- crashes was also slightly higher where there was a curb pres- set from the pavement edge. Since there was one record per ent in addition to the guardrail. Finally, the rollover rate was

OCR for page 46
52 TABLE 15 Descriptors, crash frequencies, and crash rates (e.g., curves and speed limits) strengthens these findings. The per million vehicle-miles passing for guardrail-only and injury-crash model was considered to be a limited surrogate curbguardrail segments on urban freeways in Michigan for a model of rollover crashes. The presence of a curb (1992 inventory data and 199394 crash data) appeared to increase the crash frequencies on these urban Guardrail only Curbguardrail freeway segments. Total Mileage 186.64 15.01 Average AADT 45,247 78,717 Total MVMT per Side 3064.1 442.1 Total SV Crash Freq. 537 87 Curb-Crash Severity Modeling Total SV Crash Rate 0.175 0.197 SV Injury Crash Freq. 139 29 The analyses described in the preceding two sections were SV Injury Crash Rate 0.045 0.066 related to rollover, one of the most important factors predict- SV Rollover Crash Frequency 31 5 SV Rollover Crash Rate 0.010 0.011 ing crash severity. The analysis in this task was designed to provide additional information on curb-crash severity in both rollover and nonrollover crashes. As noted previously, it is difficult to study either crash occurrence or severity of curb essentially the same for guardrail-only and curbguardrail impacts since the vehicle almost always overrides the curb, sections based on the simple rates. and both the occurrence of a reported crash and the resulting However, the comparison of rates such as these can be severity are often defined by what is behind the curb. Unfor- misleading unless the rates are from highway segments tunately, there was no good inventory of the area behind the with essentially the same AADT, because the relationship curb in even the best databases (e.g., the HSIS roadway between crashes and AADT is not linear in nature. The sec- inventory files). ond analysis, statistical modeling, was intended to account The goal of this task was to compare the severity of all SV for this issue. Poisson and negative-binomial models were curb crashes with the severity of SV noncurb crashes (i.e., developed to predict both SV crash and SV injury crash fre- crashes with other roadside objects) to determine whether quency on urban freeways as a function of a number of pre- they differed under similar conditions. Since curb and non- dictor variables. Unfortunately, rollover crashes could not curb crashes do not always occur under similar conditions, be analyzed separately due to the small sample size. Pre- conditions were controlled through regression-type model- dictor variables analyzed included curb presence, segment ing. To ensure that the curb was related to the subsequent length, AADT, horizontal curve presence, speed limit, and injury or rollover, a database was needed that included a guardrail offset. Tables 3 and 4 in Appendix E provide the sequence of events. To equalize the roadside behind the curb detailed results. Because the Poisson results were similar to with the roadside for noncurb crashes to the extent possible, the negative-binomial results, and since the latter is consid- a subsample was needed of noncurb crashes that occurred in ered more appropriate, only the negative-binomial results areas similar to the curb crashes (i.e., crashes occurring on a are summarized here. "curb-type" roadway, but without a curb present in the crash). In almost all cases, the predictor variables in the model of These requirements led to the use of the 1996-97 Michigan both total SV and injury SV crashes exhibited logical behav- HSIS database, which contained both crash and roadway ior (e.g., crashes increased with AADT and segment length inventory information. A more detailed description of this and decreased with increasing guardrail offset). Of most effort can be found in Appendix F. interest, the presence of a curb with the guardrail signifi- The curb crashes included in the dataset were SV crashes cantly increased both total and injury SV crashes when all involving a vehicle striking a curb as the first or first mean- other factors were held constant. The total SV crash model ingful event in the sequence of events. Note that first mean- predicted 0.1525 crashes per mile on average, when the inde- ingful included curb impacts as a second, third, or fourth event pendent variables were held at their means. Crashes increased if all of the preceding events were nonobject/nonrollover by 0.0640 per mile (42%) when a curb was present or when events such as "loss of control" or "run off road." The noncurb a curb was added to a guardrail. The injury crash model, crashes occurred on segments of roadway with a curb present which was considered to be a surrogate of rollover crashes, on at least one side of the roadway (i.e., opposite shoulder or predicted 0.0416 injury crashes per mile, increasing by 0.0238 median) according to the roadway inventory data, but not (57%) when a curb was added. where the crash occurred, based on the absence of curb in the In summary, both the simple rate comparisons and the sequence of events. Crashes in which curb impacts were Poisson and negative-binomial models indicated that on urban noted as an event following an impact with another fixed freeways, segments with both guardrails and curbs were object or a rollover were deleted from the data set. In all more likely to have both SV crashes and injury crashes than cases, the analyses were restricted to roadways with posted segments with only guardrails. While it was not possible to speed limits of 40 mph (65 km/h) or greater. "Overturn" was control for all potentially confounding variables, the fact that captured as an event, and by definition, followed the curb the models statistically control for exposure variables (e.g., impact in the curb-crash set. Other variables captured for segment length and AADT) and geometric/design variables analysis included the speed limit, assigned operating speed,

OCR for page 46
53 functional class, weather and light conditions, road surface, When the rollover variable and all other variables except right shoulder type, highway area type, vehicle type, curve curb presence were held constant, curb impacts were slightly code, and terrain. less severe, or at least no more severe, than crashes with other As detailed in Appendix F, when one is attempting to objects. model differences in the full distribution of crash severity As noted earlier, there is no guarantee that the roadsides (for example, by use of the KABCO injury scale: K = killed; for these curb and noncurb crashes were similar, and if not, A = severe injury; B = moderate injury; C = minor injury; and the severity difference found, or the lack thereof, may have O = no injury) as a function of other variables (such as been confounded by these unknown differences. However, curb/noncurb or speed limit), the most appropriate model is given that this analysis was restricted to the most similar an ordinal regression model. Two common forms are the locations possible, those with a curb on at least one side of ordered logit and probit. In this case, the logit form proved to one roadway, the conclusion that curb crashes are at least no be most appropriate. more severe, and probably less severe, than noncurb SV Models predicting severity were developed using the above crashes appeared to be supported by the data. set of variables. The primary model included speed limit and an urban/rural variable based on functional class as two of the independent predictors. A subsequent model used assigned Nature of Curb Impacts roadway operating speed instead of these two variables, since One of the goals of the crash-data analysis effort was to assigned operating speed was a direct function of speed limit develop or extract information from real-world crash data within functional class. Both of these models contained roll- that might be useful in defining inputs to the simulation and over as a predictor and thus allowed controlling for rollover crash-testing analyses. This was a two-part task. in examining curb versus noncurb severity. The speed limiturban/rural model indicated that the effect of hitting a curb on injury severity was negative (i.e., Crash Reconstruction Data it lowered injury severity), although this effect was only mar- ginally significant (at the 8% level). The model, which con- The first part of this effort involved the analysis of NASS- trolled for many other variables, showed that injury severity CDS data to extract information on the specific nature of curb- was higher in the following cases: related impacts in the real world (e.g., angle of impact, speed, tracking/nontracking). NASS-CDS was the only national data- The vehicle rolled over; base of crash reconstruction data where such detail was cap- The crash occurred on an urban rather than rural roadway; tured. Police data did not include such information. Two The weather was clear or cloudy, not foggy, raining, sources of CDS data were used: basic data downloaded snowing, sleeting/hailing, or severely windy; from the NHTSA website and enhanced data obtained from The road surface was dry, rather than wet, muddy, snowy, TTI. Details of the data, analyses, and results are found in slushy, or with debris; Appendix G. The vehicles involved were trucks, buses, motorcycles, Initially, NASS-CDS data for the 199799 period were motor scooters and mopeds, rather than passenger cars, downloaded from the NHTSA web site for analysis. The vans, or pickups; 11 separate files for each year (e.g., vehicle exterior file and The crash occurred on level terrain; event file) were combined into usable vehicle-based analysis The crash occurred on a curve; or files that allowed examination of the sequence of events for The posted speed limit was at the higher end of the each vehicle in a crash and determination of when the curb 4065 mph range. was struck and what occurred after that impact. The NASS- CDS data contained up to 22 events (e.g., "hit curb") for each The results of the second model in which assigned operat- vehicle involved in a crash. However, detailed information ing speed replaced speed limit and rural/urban variables indi- such as impact speed and impact angle was only recorded for cated that many of the same predictors were significant. one event in each crash, the event that caused the largest However, in this case, while assigned operating speed was a change in velocity. Data on the direction of force and defor- significant predictor, curb presence was no longer significant. mation extent were recorded for the events that caused the This model implied that there was no difference in crash highest and second-highest changes in velocity. severity between the curb and noncurb crashes. Examination of the data after preparation indicated a major In conclusion, although the two models differed somewhat problem in the sample of curb-related cases: no impact angle in their estimates of the effect of curbs on severity, both or speed data were present, probably because CDS places implied that in locations where curbs might be located, SV higher priority on reconstruction of vehicle-to-vehicle crashes involving curbs were clearly no more severe than impacts than fixed-object impacts. Of the 473 cases in which crashes involving other roadside objects. Rollover was an a curb impact was one of the events, including 32 cases in important predictor of injury, perhaps the most important. which the curb was the highest-change-in-velocity event,

OCR for page 46
54 none included reconstructed impact speed and angle data. speeds, ranged from 15 to 61 mph. Except for the Total Examination of the "direction of force" variable also indi- Weighted group, which was almost totally influenced by one cated that it could not be a measure of "angle of impact." This case with a weight of 10,939, the distributions were some- was verified in subsequent conversations with a NASS inves- what similar. Approximately 50% of the cases had encroach- tigation supervisor, who indicated that the data might be used ment speeds of between 35 and 45 mph. Similar tables related as an indicator of tracking/nontracking vehicles, although to reconstructed encroachment angle, combined speed and that also proved later to be somewhat questionable. angle data, and tracking/nontracking status of the vehicle can Because of these initial data problems, the researchers be found in Appendix G. requested and received enhanced CDS data developed by Both samples of NASS-CDS data available for use in Dr. Roger Bligh of TTI for NCHRP Project 17-11, "Deter- these analyses of speed, angle and tracking/nontracking were mination of Safe/Cost Effective Roadside Slopes and Asso- very small. Thus, the results must be viewed with caution. ciated Clear Distances." In that project, TTI had NASS crash This is particularly true of some of the weighted results in the investigators capture additional data at selected CDS crash TTI data, which were significantly affected by two high- sites, and then reconstructed encroachment speed, angle, and weight cases. Given these important caveats, based on the tracking/nontracking information where possible. They also data available, the following observations can be made: developed a confidence rating for the speed and angle recon- structions (i.e., 1 as low confidence and 10 as high confi- Curb impact speeds ranged from 15 to 61 mph. Approx- dence). TTI staff provided a set of 21 cases in which a curb imately 50% of the cases had encroachment speeds had been struck for use in this study. All these cases were SV between 35 and 45 mph. run-off-road (ROR) collisions; and all occurred on roadways Curb impact angles ranged from 6 to 31 degrees. In the with speed limits of 45, 50, or 55 mph (72, 80 or 89 km/h). majority of the unweighted cases (70%), the angles were Since the widest shoulder width was less than 6 ft, most of 15 degrees or less, with 50% between 11 and 15 degrees. the curbs were apparently near the travel lane. For that rea- The distribution of impact angles in the weighted data son, the encroachment data were expected to provide some was highly dependent on the inclusion of the high-weight indication of the speed and angle distributions for the curb cases, with 26 to 83% of the cases having angles less than impacts. In addition, whether the vehicle was tracking or not 15 degrees and 20 to 80% between 11 and 15 degrees. when it left the roadway was considered to be a good indica- According to the TTI (reconstructed) data, 77% of the tor of tracking during curb impact. vehicles in the unweighted sample and 97% of the vehi- Table 16 presents the reconstructed encroachment speed cles in the weighted were tracking; while in the NASS- data, based on a very small sample of 14 curb-related crashes; CDS data, 51% (unweighted) and 56% (weighted) were since the encroachment speed was "unknown" in 7 of the 21 tracking based on direction of force. crashes, only these 14 crashes were relevant. In addition to the raw frequencies for all cases and a subset of cases with higher confidence ratings (i.e., 5 or higher), percentages within Analysis of Extreme versus Nonextreme Crashes speed categories are presented for the total unweighted data, the total unweighted subset of cases with higher confidence, The second part of this task involved analysis of the the weighted full sample and the weighted high-confidence NASS-GES, Michigan, and Illinois data to determine if cer- subset. The weights in the latter two cases are those provided tain curb-related crash conditions might distinguish extreme in the CDS data for each NASS-CDS case. crashes (those involving fatal or incapacitating injury) from As shown in the table, the encroachment speeds for this nonextreme crashes (those involving property damage only small sample of cases, which are estimates of the curb-impact [PDO]). Three analyses were conducted in this effort: TABLE 16 Encroachment speed distributions from the TTI NASS-CDS sample Speed Unweighted data Weighted data (mph) All cases High-confidence All cases High-confidence 15-20 1 7.1% 0 0.0% 96 0.8% 0 0.0% 20.1-25 0 0.0% 0 0.0% 0 0.0% 0 0.0% 25.1-30 2 14.3% 1 9.1% 11205 92.3% 266 25.0% 30.1-35 1 7.1% 1 9.1% 24 0.2% 24 2.3% 35.1-40 4 28.6% 4 36.4% 361 3.0% 361 34.0% 40.1-45 3 21.4% 2 18.2% 272 2.2% 225 21.2% 45.1-50 0 0.0% 0 0.0% 0 0.0% 0 0.0% 50.1-55 1 7.1% 1 9.1% 82 0.7% 82 7.7% 55.1-61 2 14.3% 2 18.2% 106 0.9% 106 9.9% Unknown 7 0 7 0 Total 21 100.0% 11 100.0% 12153 100.0% 1064 100.0%

OCR for page 46
55 A comparison of NASS-GES curb-related crashes result- ways that might provide guidance to the simulation, crash ing in severe injury with curb-related crashes resulting in testing, and policy development efforts. The data set used was no injury, similar to the one described above. The curb-related group A comparison of NASS-GES severe curb-related crashes comprised SV crashes in which (1) a curb impact was the with severe SV ROR crashes not involving a curb, and FHE, (2) the posted speed limit for the roadway was 40 mph An analysis of Michigan and Illinois severe and non- (65 km/h) or greater, and (3) the most severe injury was either severe curb-related crashes. fatal or incapacitating. The noncurb group included SV crashes meeting the same criteria except the FHE was not a curb Details of the analyses are given in Appendix H. impact. These sets of crashes were first characterized as Inter- states and non-Interstates, and the non-Interstate category was NASS-GES Severe and Nonsevere Curb-Related Crashes. subdivided into multilane divided highways and multilane The GES sample was drawn from the same police agencies undivided highways to try to isolate groups more likely to as the NASS-CDS data described earlier, but the sample was have the same exposure to curb presence. much larger, approximately 50,000 cases per year. All 199599 While a number of different categorizations of the data GES crashes in which the curb was the FHE were divided were used in these comparisons, the findings did not add a into two groups: (1) all crashes involving fatal or incapacitat- significant amount of information to that learned from the ing injury, approximately 10 to 15% of the sample, and (2) all earlier analysis of severe versus nonsevere curb crashes. It PDO crashes, approximately 50% of the sample. The crashes was difficult to identify clear findings because the curb and in each group were categorized by roadway class and speed noncurb crashes might well be occurring at different types limit (i.e., Interstate highways, non-Interstate highways with of locations (i.e., the curb locations could be somewhat dif- speed limits of 40 to 50 mph, and non-Interstate highways ferent from locations of crashes where no curb is involved). with speed limits greater than 50 mph). The sample sizes for In addition, the freeway-related findings were based on these categories are shown in Table 17. very small samples of curb-related crashes. The more con- The severe crashes were compared to the PDO crashes sistent patterns are included in a summary at the end of this within the three roadway types for variables related to crashes subsection. (e.g., relationship to junction), vehicles (e.g., vehicle body type), and roadways (e.g., roadway profile). Michigan and Illinois Severe and Nonsevere Curb-Related These analyses were conducted using only unweighted GES Crashes. Analyses similar to those described above were con- data because severity was the predominant weighting variable ducted with the 199697 Michigan and Illinois HSIS data. used in weighting. However, to verify the unweighted results, Criteria similar to those described for the GES analyses were a set of analyses was conducted of the severe curb and non- employed: curb crashes using weighted data. These analyses indicated that the overwhelming majority of the variables analyzed had very similar distributions for the weighted and nonweighted A crash either involved at least one vehicle that struck a data, for both the severe and PDO crashes. Generally, the dis- curb somewhere in the sequence of events or occurred tributions were within 2% of each other, and those outside on a segment of roadway equipped with curbs accord- this range exhibited differences of less than 5%. Therefore, ing to the roadway inventory data; using the nonweighted data appeared to be an appropriate Either the FHE was an impact with a curb or the second, method of comparison. third, or fourth harmful event was an impact with a curb Some consistent findings were noted and are included in and the preceding events were nonimpact events such as the summary at the end of this subsection. "uncoded or errors," "loss of control," "ran off road left," or "ran off road right;" NASS-GES Severe Curb-Related and Noncurb-Related The posted speed limit was at least 40 mph; and SV Crashes. This second GES analysis compared extreme The maximum injury in the crash was either a fatality or curb-related crashes to the larger group of extreme ROR an incapacitating injury (K or A on the KABCO injury crashes not involving a curb as the FHE. Again, the goal was scale used by most police departments) for the severe to see if these high-injury curb and ROR crashes differed in impacts or PDO for the nonsevere impacts. TABLE 17 Sample sizes for NASS-GES analysis of extreme crashes in which the curb was the FHE Fatal & severe injury PDO Roadway class and speed limit crashes crashes Interstate Highway (All Posted Speeds) 10 64 Non-Interstate Highway (Posted 40-50 mph) 105 428 Non-Interstate Highway (Posted over 50 mph) 17 113

OCR for page 46
56 These severe and nonsevere curb crashes were categorized Severe curb crashes on Interstates and higher-speed further as Interstate and non-Interstate crashes. The sample non-Interstates were more likely to be in urban areas. sizes for severe and nonsevere curb crashes on Interstates and This could reflect such factors as high-speed roadside non-Interstates are presented in Table 18. Because of the encroachments at which more barrier curbs are present extremely small sample size of Interstate crashes in Illinois, and the higher severity of curb crashes on interchange those were not analyzed. ramps, more ramps being located in urban areas. Design Severe curb crashes differed from the nonsevere crashes and placement may therefore be even more critical on on Michigan Interstates by occurring more often on ramps, higher-speed roads in urban areas. in good weather, and involving alcohol use and motorcycles. Severe curb crashes were somewhat more likely to Curbs were more often the MHE in the nonsevere Interstate occur in clear weather on dry roads than less severe crashes, with rollover and other impacts being the MHE in crashes were. the severe crashes. The non-Interstate findings were some- There was little difference between the curb and non- what similar. Curb impacts on these roads were more likely curb groups with respect to violations cited or whether to be in urban areas, regardless of severity. Severe crashes the crash was considered speed related. The pattern of differed from nonsevere crashes in both states by occurring which of the groups had higher proportions varied by more often in clear weather on dry roads. The Michigan data roadway type. However, the Michigan data appeared to again indicated that the curb was less likely to be the MHE indicate more alcohol use in the severe crashes. There were no major differences between the frequency in the severe crashes and that more alcohol use and more motorcycles were involved in the severe crashes. Illinois of rollovers in the severe curb-related and SV ROR data for the non-Interstates indicated that the severe crashes crashes. The percentage of rollover was relatively high in occurred slightly more often at night. both groups (18% and 70%, respectively). As expected, the mechanism for the rollover differed between the two groups, being the curb in the curb-related crashes. Comparison of Findings from the NASS-GES and State Curbs were problematic for motorcycles. Analyses. The analyses described above examined a wide variety of crash-related factors that might differentiate among severe curb crashes, nonsevere curb crashes, and severe non- curb crashes. Both NASS-GES, a national database, and state- Summary of Crash and Inventory Data Analysis level data from Michigan and Illinois were examined. While there were some subtle differences among the results, there This section has described the analysis efforts involving were some rather consistent findings related to curb design real-world crash data that were included in this project. The and placement: goals of these analyses were (1) to better characterize safety problems associated with curb and curbbarrier combina- tions on high-speed roadways, and (2) to provide leads to the Overturn was an important variable in terms of severe crash testing and simulation efforts that were conducted in injury causation. Curb designs or placement that decreases other parts of this project. All efforts were ultimately aimed the probability of overturn are clearly important. Impact speed and angle were important; more severe at the development of the design guidelines. The major findings concerning extent of the problem, curb-related crashes occurred at higher speeds, on grades, and on curved alignments. curb-crash characteristics, and leads to simulation and crash Both in comparison with other curb crashes (in GES and testing efforts include the following: state data) and in comparison with other SV ROR crashes on freeways, severe curb crashes were more often related Curb-related crashes on roadways with speed limits of to ramps. This could simply be because curbs were more 40 mph (65 km/h) and above represented a very small likely to be located on ramps than on other road segments percentage of either total fatal crashes (1%) or all crashes of freeways. However, it does underline the need for for- (0.5%). The importance of the curb problem stems from giving curb designs on interchange ramps. the potential for rollover following impact. TABLE 18 Sample sizes of severe and nonsevere curb crashes in 1996 and 1997 Michigan and Illinois HSIS Data Michigan Illinois Severe (K+A) Nonsevere (PDO) Severe (K+A) Nonsevere (PDO) Interstate 17 185 2 26 Non-Interstate 63 1171 37 332 Total 80 1356 39 358