Cover Image

Not for Sale

View/Hide Left Panel
Click for next page ( 8

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 7
7 TABLE 2-2 PRPM replacement cycle TABLE 2-4 Suggested replacement cycles for the state of Indiana for PRPMs for the state of Texas Number of lanes ADT (veh/day) Replacement cycle ADT (veh/day) Replacement cycle (years) (years) Two Fewer than 5,000 4 More than 50,000 1 5,000 to 15,000 3 More than or equal to 10,000 23 More than 15,000 2 Fewer than 10,000 34 Four or more Fewer than 10,000 4 10,000 to 30,000 3 30,000 to 75,000 2 The study examined the change in nighttime crashes (from 6:00 p.m. to 5:59 a.m.) and used daytime crashes at the More than 75,000* 2 same sites as a control group. In the crash data from 1975 * These roadways should be inspected at least once each year. to 1980, there were 223 before-installation crashes and 391 after-installation crashes at the selected sites. For approxi- 2.2 REVIEW AND ASSESSMENT mately 68 percent of the sites, no crashes were reported for OF KNOWLEDGE ABOUT the 6 years of data analyzed. A log-linear model was fit to the THE SAFETY EFFECT OF PRPMS data stratified by the year of installation, daytime-versus- nighttime crashes, and before-versus-after installation time This section critically reviews literature before summariz- period. Overall, nighttime crashes were estimated to have been ing any methodological problems arising from past research. reduced by 22 percent compared with daytime crashes at the The literature review focuses on the relatively few studies same sites. that have been conducted from about 1980 to date, since A disaggregate analysis by year of installation revealed that older studies are less likely to be relevant in terms of findings sites modified in 1976 and 1977 had reductions of 33 percent and methodologies employed. and 32 percent, respectively. However, sites modified in 1978 showed a 53-percent increase in nighttime crashes, an effect that could not be explained. Perhaps sites most worthy of 2.2.1 Review of Literature PRPM installation, and therefore most likely to yield safety Seven evaluation studies were reviewed and are summa- benefits, were treated earlier in the program. Single-vehicle rized herein. crashes were estimated to have been reduced by 12 percent The first study to be reviewed was undertaken by Wright more than other nighttime crash types were. These reduc- et al. (13). This study evaluated the safety effects of reflec- tions were found to be independent of ADT and curvature, tive raised pavement markers in Georgia. From 1976 to 1978, although it should be remembered that all curves had at least the Georgia DOT installed reflectorized pavement markers 6 degrees of curvature. (both raised and recessed markers) on the centerlines of 662 Kugle et al. (11) collected 2 years of before data and horizontal curves, all of which were in excess of 6 degrees of 2 years of after data at 469 Texas locations varying in length curvature. At some locations, warning signs, chevron mark- from 0.2 to 24.5 miles (0.32 to 39.4 km). PRPMs were ers, or other delineation devices that were intended to pro- installed between 1977 and 1979. Sixty-five percent of the vide guidance to drivers were also installed. These additional locations were on two-lane roads; 33 percent on four-lane devices may have affected the analysis's results. For each roads; and the remaining on three-, five-, or six-lane roads. curve studied, the location, length, degree of curvature, year Seventeen sites were subsequently omitted from analysis of installation, ADT by year, and annual crash frequency by because they were resurfaced after PRPM installation, which type (single-vehicle or other) and time of day (day or night; would likely have influenced crash risk. Crashes were sub- daytime: 6:00 a.m. to 5:59 p.m.) were collected. Locations classified for analysis by wet weather/dry weather, daytime/ were monitored 200 ft (61 m) in both directions beyond the nighttime, and fatal/injury/property-damage-only (PDO). curve in the belief that curve-related, single-vehicle crashes Comparison of wet-versus-dry crashes excluded conditions often take place beyond the end points of curves. such as muddy and snowy, but these crashes were included in the total nighttime-versus-daytime analysis. Daytime crashes included daytime, dawn, and dusk crashes, while nighttime TABLE 2-3 When to schedule PRPM system crashes included crashes with and without street lighting pres- maintenance for the state of Texas (based on nighttime inspection) ent. Crash types potentially affected by PRPMs--namely head-on, sideswipe, and run-off-road--were identified for a For markers spaced at... Maintenance should be scheduled separate analysis. In addition, ADT and the number of wet as soon as possible if... weather days were recorded for each location during the 80 ft (24 m) Fewer than two markers are visible analysis periods. Three evaluation methods were used and 40 ft (12 m) Three or fewer markers are visible are described below.

OCR for page 7
8 The first evaluation method involves calculating the cross- showed a nonsignificant increase. The authors noted that product ratio as an overall measure of effectiveness. This roughly half of the sites showed a reduction in both nighttime method aggregates data from all sites and does not consider and wet weather crashes, but roughly 10 percent of the sites noncrash factors, such as ADT. The cross-product ratio mea- showed very large increases in total crashes, which may have sures the relative change in the crash type of interest com- unfairly skewed the overall results. pared with a control group of crashes believed not to be Mak et al. (14) conducted a study using the same Texas affected by the measure of interest. The control group can locations as Kugle et al. (11) to reevaluate the safety effect therefore be used to control for factors such as changes in of PRPMs on nighttime crashes. This study screened the ADT and other changes over time that affect crash risk. Night- original database of 469 locations and eliminated those that time crashes were compared with daytime crashes and wet underwent major modifications other than the PRPM instal- weather crashes compared with dry weather crashes. Using lation during the evaluation period, modifications that may nighttime-versus-daytime crashes for illustration, the cross- have influenced the previous study results. Several other product ratio, T, is calculated as locations were not included in the new analysis because they experienced no crashes in either the 2-year before period or x11 x 22 the 2-year after period. After screening for these criteria, only T= (2-1) x12 x 21 87 of the original 469 locations remained for further analysis. The new analysis focused on individual locations. The day- Where time crashes were again used as a comparison group to account x11 = Crashes in the after period during nighttime, for any factors that may have influenced crash frequency x12 = Crashes in the before period during nighttime, between the before and after periods but that were not related x21 = Crashes in the after period during daylight, and to the PRPM installation. However, daytime crashes did not x22 = Crashes in the before period during daylight. include crashes occurring during dusk or dawn; dusk and dawn crashes that were eliminated from the analysis were reported The change in crash frequency due to treatment is esti- to be about 13 percent of the total crashes. A statistical pro- mated as cedure, based on the cross-product ratio, was used to measure the effect of PRPMs at individual locations, Z. This procedure Percent Change = 100(T - 1) (2-2) is based on a test statistic: The second method, called Gart's procedure, calculates the ln(T ) Z= (2-3) cross-product ratio at each individual location and weights 1 x11 + 1 x12 + 1 x 21 + 1 x 22 each estimate by the total number of crashes at each site for a weighted average estimate of treatment effectiveness. This Where allows the higher-crash locations to exert more influence on x11 x 22 the estimated effectiveness. T= , The third method uses logistic regression, which can x12 x 21 include the influence of factors other than the PRPMs in the estimation of effectiveness. The probability of an individual x11 = Crashes in the after period during nighttime, location experiencing a nighttime crash is modeled as a func- x12 = Crashes in the before period during nighttime, tion of time (before or after installation), ADT, and number x21 = Crashes in the after period during daylight, and of lanes. This procedure provides an estimate of effective- x22 = Crashes in the before period during daylight. ness adjusted for site differences in ADT and number of lanes. Kugle et al.'s analysis of wet weather crashes also Z was calculated for each location. If there are no safety included the number of wet weather days as a variable in the effects, Z will be normally distributed with a mean of 0 and model (11). a variance of 1. A positive value of Z would indicate an The results of the three methods provided different numeric increase in nighttime crashes relative to daytime crashes, a results, but all methods indicated the same trend for all negative value indicates a relative decrease, and a value of zero crash types. The cross-product analysis indicated a 15-percent indicates no change. Of the 87 locations, 56 (64.4 percent) increase in nighttime crashes and a nonsignificant 1.4-percent showed a relative increase in nighttime crashes, 30 (34.5 per- decrease in wet weather crashes. Gart's procedure indicated cent) showed a relative decrease, and 1 (1.1 percent) showed a 31-percent increase in nighttime crashes and a nonsignifi- no change. Using a confidence level of 10 percent to check cant 1-percent decrease in wet weather crashes. Logistic mod- for significance, 4 locations (4.6 percent) showed signifi- eling also indicated a significant increase in nighttime and a cant reductions in nighttime crashes relative to daytime nonsignificant decrease in wet weather crashes. These effects crashes, 9 (10.3 percent) showed significant increases, and were found to be consistent for all crash and severity types 74 (85.1 percent) showed nonsignificant changes in night- with the exception of wet weather sideswipe crashes, which time crashes relative to daytime crashes.

OCR for page 7
9 The effect of PRPMs on crash severity was also studied at Griffin (15) analyzed the same data as Mak et al. (14), which 37 locations that had a minimum of 30 crashes in the before is a subset of the Texas data originally used by Kugle et al. (11) or after period or in the two periods combined. Two severity using a different statistical approach. Griffin quantified the indexes were calculated for each site separately for nighttime safety effect of PRPMs on nighttime crashes at 86 locations and daytime crashes: using daytime crashes as a control group. One of the loca- tions used in the previous analysis was not included in this Severe = (percent fatal or incapacitating-injury crashes study because it could not be located. The overall, or aver- in after period) - (percent fatal or incapacitating-injury age, effect of PRPM installation on nighttime crashes was crashes in before period). estimated by calculating a weighted log odds ratio. The log Injury = (percent fatal, incapacitating-injury, or non- odds ratio, L, is calculated by taking the natural logarithm of incapacitating-injury crashes in after period) - (percent T, defined previously in Equation 2-1. fatal, incapacitating-injury, or non-incapacitating-injury The weight for the log odds ratio at each site, w, is calcu- crashes in before period). lated as 1 A logit model was then used to test for statistically signif- w= (2-4) icant differences using the daytime crashes as a comparison (1 x11 + 1 x12 + 1 x21 + 1 x22 ) group. None of the 37 sites showed a significant change in the percentage of severe crashes, perhaps due to low numbers Where of these crashes. For injury crashes, 4 locations showed a sig- x11 = Crashes in the after period during nighttime, nificant decrease in nighttime crash severity, 1 showed a sig- x12 = Crashes in the before period during nighttime, nificant increase, and 32 showed no significant change. x21 = Crashes in the after period during daylight, and Locations that showed a significant increase (nine total) or x22 = Crashes in the before period during daylight. decrease (four total) in crash frequency were further exam- ined in an attempt to identify crash characteristics that might The weighted log odds ratio, Lavg, is thereafter calculated as be associated with PRPMs. Not enough data were available for statistical tests, but an examination of the relative pro- Lavg = wL (2-5) portions between the before and after periods indicated that w for the sites showing a significant increase in crashes, the pro- portion of nighttime multivehicle crashes increased and the Where proportion of nighttime fixed-object crashes decreased. For the locations showing a significant decrease in crashes, this L = ln(T) same increase in multivehicle crashes and decrease in fixed- object crashes was found for the daylight hours but not for The average effect is equal to the antilogarithm of Lavg, and nighttime hours. For both groups, the proportion of nighttime the standard error of L, Lse, is equal to crashes occurring on horizontal curves greater than 2 degrees increased. 1 Lse = (2-6) A number of roadway characteristics were also examined w for their effect on the influence of PRPMs using the same groups of nine (significant increase) and four sites (significant Using this methodology, the expected change in nighttime decrease), but no strong evidence that any of the variables crashes following the installation of PRPMs was estimated interacted with PRPM installation was found. The examined to be a 16.8-percent increase, with the 95-percent confidence characteristics included the following: limits between a 6.4- and 28.3-percent increase. Pendleton (16) used both "classical" and empirical Bayes Intersection type (none, interchange, T-intersection, four- before-and-after methods for evaluating the effect of PRPM leg intersection, or multiple intersection), nighttime crashes on undivided and divided arterials in Michi- Whether the roadway was within the city or outside the gan. Seventeen locations totaling 56 miles (90 km) served as city, installation sites, and 42 sites totaling 146 miles (235 km) Horizontal curvature (less than 1 degree, 1 to 3 degrees, were used as control sites where PRPMs were not installed. or more than 3 degrees), Crash data for 2 years prior to installation and 2 years after Grade (less than 3 percent or more than 3 percent), installation were used for two categories of analysis. The first Structures (none, culvert, or bridge), category used as a control group daytime crashes at the instal- Number of lanes (less than or equal to four, or more than lation sites, which were assumed to be unaffected by the instal- four), and lation of PRPMs. Daytime crashes did not include crashes Whether the roadway was divided or undivided. that occurred 1 hour before and 1 hour after both sunrise and

OCR for page 7
10 sunset, a total of 4 hours per day. The second category used effectively by improving delineation during nighttime wet nighttime crashes at control sites as a control group. Pendleton weather conditions." It further stated that PRPMs should be made the following conclusions: installed only at locations having high frequencies of wet weather, nighttime, guidance-related crashes. Undivided roadways showed an increase in nighttime Orth-Rodgers and Associates, Inc. (18), used the same crashes and divided roadways showed a decrease in night- "odds ratio" methodology as Griffin (15) to evaluate the time crashes when analyzed separately. Whether a high- effects of both raised and recessed pavement markers on way was divided was concluded to be the most significant nighttime crashes at 91 Interstate highway locations in Penn- road characteristic affecting the effectiveness of PRPMs. sylvania. PRPMs were installed at these sites between 1992 Using daytime crashes at treated sites as a comparison and 1995, and crash data from 1991 to 1996 were used in the group yielded larger reductions (or smaller increases) in analysis. Daytime crashes at the same sites were used as a crashes than when nighttime crashes at untreated sites comparison group. Sites that had no crashes in any of the day- were used as a comparison group. The issue of which time or nighttime periods before or after PRPM installation comparison group to use stayed unresolved. were eliminated since a zero value would render the odds The empirical Bayes methodology generally produced ratio meaningless. This omission creates a subtle bias toward smaller reductions (or larger increases) than the simple underestimation of effects if the realization of zero crashes at or "classical" before-and-after methodology. This con- a site in the after period is due to PRPM installation. This clusion usually is an indication that regression-to-the- underestimation is exaggerated by the fact that the after peri- mean was at play and accounted for by the empirical ods were, on average, much shorter than the before periods Bayes methodology. and were therefore more likely to contain zero crashes. Sites Exposure should be properly accounted for, and the in urban and lit areas were also eliminated, assumed by the researchers lamented the fact that the estimates of night- authors as "not good candidates for an analysis of this type." time traffic volume were only approximations. The study Several crash types were excluded because they were con- also revealed the difficulties of using crash rates (crashes sidered to be unrelated to PRPMs (e.g., crashes that hap- per unit of exposure) to control for exposure differences. pened during dusk, dawn, or unknown lighting condition; These difficulties arise from the nonlinear relationship crashes that occurred in weather conditions other than rain or between crashes and exposure that indicates that these "no adverse conditions"; crashes that occurred when the road rates can change because of volume changes and not surface condition was other than dry or wet; and crashes for necessarily because of a treatment. which the impact type was "unknown"). Results indicated a 12.3-percent increase in nighttime New York State DOT (17) undertook a safety assessment of crashes (95-percent confidence limits of 1.1 and 24.8 per- PRPMs in New York to review the DOT's policy on PRPM cent) for all sites, a nonsignificant 1.2-percent decrease for installation. The DOT used a simple before-and-after study locations with raised pavement markers, and a significant design in which numbers of crashes before and after treat- 20.1-percent increase (95-percent confidence limits of 5.5 ment were compared without controlling for other factors. and 36.9 percent) for locations with recessed pavement mark- Two analyses were undertaken using this simple before-and- ers. The authors suspected that the small decrease in night- after design. The first analysis, at 20 sites, targeted PRPMs time crashes due to raised PRPMs might have been because at sections of unlit suburban and rural roadways with pro- there was a positive effect (i.e., a reduction in crash fre- portionately high numbers of nighttime and nighttime wet quency) on the daytime crashes that was used for the com- weather crashes. Overall, there was a nonsignificant decrease parison group. of 7 percent for total crashes, a highly significant decrease of Additional results were obtained for two crash subsets. 26 percent for nighttime crashes, and a significant decrease Nighttime wet condition crashes also showed large increases of 33 percent for nighttime wet weather crashes. Furthermore, from 30 to 47 percent (confidence limits not reported), depend- there was a significant reduction of 23 percent in all guidance- ing on the comparison group of crashes used (daytime wet related crashes, which are crashes resulting from a vehicle condition, nighttime other, or all daytime crashes). Nighttime leaving its assigned travelway (e.g., run-off-road, head-on, wet road sideswipe and fixed-object crashes were estimated to encroachment, and sideswipe). There was also a 39-percent have increased 56.2 percent (confidence limits not reported) reduction in nighttime guidance-related crashes. using nighttime dry road sideswipe and fixed-object crashes as The second analysis looked at PRPMs installed non- a comparison group. Not much emphasis was placed on these selectively over 50 long sections of highway. The analysis additional results since these increases could be exaggerated revealed that nighttime crashes were reduced by a nonsignifi- by a positive effect of PRPMs on the comparison sites. cant 8.6 percent, that total crashes were reduced by a statis- Table 2-5 summarizes the review of seven relevant evalu- tically significant 7.4 percent, and that nighttime wet weather ations of the safety effects of PRPMs, measured in terms of crashes increased by a nonsignificant 7.4 percent. Thus, New reductions or increases in crashes (two of the seven studies York State DOT recommended that PRPMs be installed selec- are re-analyses of subsets of data previously analyzed). All tively "when their use is likely to reduce crash frequency cost but one of the studies listed in Table 2-5 used daytime crashes

OCR for page 7
11 TABLE 2-5 Summary of literature on the safety effectiveness of PRPMs Study Ref. Site Type Installation I Installation Sample Sizes Dependent Indepen dent Comparison Other Notes Estimated Effects /Location Location Period for Treatment Variable Variables Group B Before- and Analyzed Period Length Comparison A After- Groups Period Length Wright et al. Horizontal Centerline I 1976 -1978 Treatment Total ADT, degree of Total Both raised 22% reduction in 1982 (13) curves on B 1 to 3 years 662 locations nighttime curvature daytime and recessed nighttime crashes; Georgia two-lane Comparison crashes crashes reflective single-vehicle A 2 to 4 years highways same as markers were crashes reduced in excess of treatment group used; at some 12% more than 6 degrees locations other nighttime of warning crashes; reductions curvature signs, independent of ADT chevron or horizontal markers or curvature for curves other with degree of curve guidance greater than 6 devices were installed Kugle et al. Two- , Does not I 1977 -1979 Treatment Total ADT, number of Total None 15 to 31% increase 1984 (11 ) three-, specify B 2 years 452 locations nighttime lanes, number of daytime in nighttime crashes; Texas four-, five -, Comparison crashes, wet weather days crashes no significant effect A 2 years and six- same as some on wet weather lane treatment group analysis by crashes roadways crash and severity Mak et al. Two- , Does not I 1977 -1979 Treatment 87 Total Intersection type, Total Used a subset 4.6% of locations 1987 (14 ) three-, specify B 2 years locations nighttime within/outside daytime of the data wed significant Texas four-, five -, A 2 years Comparison crashes, city, horizontal crashes from Kugle reductions, 10.3% and six- same as some curvature, grade, et al., 1984 showed significant lane treatment group analysis by structures, (11 ) increases, 85.1% roadways crash and number of lanes, showed severity divided/ nonsignificant types undivided effects Griffin, 1990 Two- , Does not I 1977 -1979 Treatment 86 Total None Total Used a subset 16.8% increase in (15 ) three-, specify B 2 years locations nighttime daytime of the data nighttime crashes, Texas four-, five -, Comparison crashes crashes from Kugle with the 95% A 2 years and six- same as et al., 1984 confidence interval lane treatment group (11 ) between a 6.4 and roadways 28.3% increase. Pendleton, Divided Centerline I 1989 Treatment 17 Total Divided/ Total None No significant 1996 (16 ) and on B 2 years locations nighttime undivided and daytime effect, direction of Michigan undivided undivided totaling crashes VMT (vehicle crashes, effect positive or A 2 years arterials arterials, 56.11 mi miles traveled) total negative dependent lane lines on (90.3 km) used in empirical nighttime on method used and divided Comparison Bayes analysis crashes at access control arterials 42 locations comparison totaling sites 146.28 mi (235 km) New York Suburban Does not I unknown Selective Total None None Regression to 26% decrease in State DOT, and rural specify B unknown Installation: crashes, the mean is nighttime crashes 1989, 1997 roadways Treatment 20 total cited as being when placed A unknown (17, 19) locations nighttime a factor selectively, no New York totaling 26 mi crashes significant effect (41.84 km) when installed Comparison nonselectively none used Nonselective Installation: Treatment 50 locations Comparison none used Orth-Rodgers Interstate Does not I 1992 -1995 Treatment 33- Total None Total Both raised 18.1% overall and Associates, highways specify B 1-3 years 76 locations nighttime daytime and recessed increase in Inc., 1998 in rural depending on crashes, crashes, reflective nighttime crashes, A 1-3 years (18 ) non- crash type nighttime daytime wet markers were nighttime wet Pennsylvania illuminated studied wet road, road, used condition crashes areas Comparison nighttime daytime wet increased from 30 to same as wet road road 47%, nighttime wet treatment group sideswipe sideswipe or road sideswipe or fixed-object fixed-object fixed-object increased by 56.2%