National Academies Press: OpenBook

Analysis of Work Zone Crash Characteristics and Countermeasures (2018)

Chapter: Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes

« Previous: Chapter 3 Effects of Queuing and Crash Countermeasures at Interstate Work ZoneLane Closures
Page 46
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 46
Page 47
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 47
Page 48
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 48
Page 49
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 49
Page 50
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 50
Page 51
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 51
Page 52
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 52
Page 53
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 53
Page 54
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 54
Page 55
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 55
Page 56
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 56
Page 57
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 57
Page 58
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 58
Page 59
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 59
Page 60
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 60
Page 61
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 61
Page 62
Suggested Citation:"Chapter 4 Statistical Modeling of Work Zone Features Upon Crashes." National Academies of Sciences, Engineering, and Medicine. 2018. Analysis of Work Zone Crash Characteristics and Countermeasures. Washington, DC: The National Academies Press. doi: 10.17226/25006.
×
Page 62

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 45 CHAPTER 4 Statistical Modeling of Work Zone Features Upon Crashes Overview Another focus effort under this research project was to gather the appropriate crash, roadway, and work zone feature data at a robust number of work zones nationally, and use state-of-the-art crash analysis techniques to produce CMFs that describe how certain high-priority features affect crashes. Many efforts have been undertaken in recent years to use statistical modeling techniques to develop predictive and explanatory models of work zone crashes (for example, see Khattak et al. 2002, Srinivasan et al. 2008, or Sun et al. 2016). For the most part, those efforts have focused on major work zone exposure indicators such as project length, location, and duration. In some instances, driver and weather factors have also been examined. One study did explore the various work zone design elements (lane and shoulder widths or closures, addition of positive protection to separate work activities from travel lanes, lane shifts and splits, etc.) that affect crashes (e.g., Chen and Tarko 2012), but was limited to a sample of work zones within a single state. A national survey of practitioners was conducted under phase I of this project to determine which work zone features were of greatest interest for purposes of estimating their safety impacts. Factors that were of primary interest in this effort were the following:  Number of lanes open and closed through the work zone  Widths of travel lanes through the work zone  Widths of shoulders (inside and outside) through the work zone  Whether barrier was placed at the edge of the shoulder (or immediately adjacent to the travel lanes if there was no shoulder)  Locations of any lateral lane shift Additional factors (presence of work zone intelligent transportation systems, use of enforcement, use of rumble strips, etc.) were also of interest, but examples of their use were not identified in the final dataset obtained through this effort. Although it is possible that some of these may have been present at one or more of the projects, no documentation or indication to that effect was uncovered during the data collection phase of this effort. Database Development With the assistance of the project panel, other contacts, and project information posted online, the research team queried various state departments of transportation (DOTs) to obtain information about ongoing construction projects. The research team was ultimately able to obtain data describing projects in the States of Ohio, Texas, Utah, Virginia, and Washington. The following data sources are needed to obtain a complete description of a construction project:

NCHRP Project 17-61 46  Work zone drawings, including maintenance of traffic plans (to show the locations of longitudinal barriers throughout the work zone) and typical cross sections (to obtain the key cross-sectional variables such as lane and shoulder width). These drawings are needed for both the “before” period and each phase of the “during” period.  Narratives, notes, and logs describing work zone operations and layouts.  Aerial photographs, which were obtained from Google Earth.  Construction schedule, including Gantt charts that provide the start date, end date, and duration of each construction phase.  Traffic volumes.  Crash data, for both the “before” and “during” periods.  Stationing and mileposts that are used for linear referencing in both the work zone drawings and the state’s traffic volume and crash databases. Among these sources, the work zone drawings required the largest amount of effort. The research team reviewed the work zone drawings to divide each construction project into homogeneous segments based on the cross-sectional widths provided in the typical cross sections in the construction plans, and then subdividing these segments where needed because of other noteworthy changes in characteristics. This process was conducted for both the “before” period and each of the construction phases in the “during” period. The process yielded a database containing one record for each roadway segment in each phase of construction. For example, if a freeway construction project consisted of three phases of work activity, each segment of the freeway would be represented four times in the database (before, phase 1, phase 2, and phase 3). Beginning and ending mileposts were carefully tracked and recorded because homogeneous segment break points were not always identical across the construction phases. Additionally, the work zone drawings were typically referenced in terms of stations while the state databases for crashes and volumes were typically referenced in terms of mileposts. The research team maintained a visual record of the stations and mileposts along each construction project by plotting pins on aerial photographs in Google Earth (see Figure 15). The research team also used aerial photographs to obtain information about ramp entrance and exit presence and length on each segment. The research team obtained construction schedules from several sources. Some of the construction projects in the database were simple, short-term projects like pavement rehabilitation, and did not have multiple phases. For these projects, the starting and ending dates were adequate to define the “during” period, and these dates were often available on public project information pages on the state DOT web sites. The research team was also able to obtain lists of project and their starting and ending dates from some state DOT contacts. For the more complex projects that had multiple phases, the research team obtained scheduling documents from state DOT contacts. These documents included Gantt charts and contractor notes. It must be noted that the accuracy of these sources is affected by the contractors’ and state DOT’s ability to maintain accurate, current descriptions of activities as they actually occurred. The research team was limited to using the construction plans and Gantt charts as the best-available descriptions of project activity. After extracting the geometric data from the work zone drawings and the phase durations from the construction schedules, the research team merged these data with the traffic volume and crash databases that were provided by the state DOTs. The assembled database contained the data elements listed in Table 8.

NCHRP Pr Figure 15 oject 17-61 . Work Zone Linear Refe rencing. 47

NCHRP Project 17-61 48 Table 8. Work Zone Database Elements Category Data Element Source Schedule Period (before, during phase number) Gantt charts, contractor notes, state DOT web sites Phase duration Geometry Number of lanes and lane closures Work zone drawings, aerial photographs, street-level photographs Lane width Shoulder width Median width Longitudinal barrier length and offset Horizontal curve presence Lane shift presence Lane add or drop presence Crossover presence Ramp entrance or exit presence Ramp speed-change lane presence (speed-change lane versus lane add/drop) Ramp speed-change lane length Ramp side (left or right) Volume Traffic volumes State databases, state DOT web sites Traffic control Speed limit Work zone drawings, contractor notes, street-level photographs Crashes Crash location and date State databases Crash type (multiple-vehicle or single-vehicle) Crash severity (K, A, B, C, PDO) A brief description of each project in the database is provided in Table 9. Some of the projects involved adding capacity, others involved repaving or rehabilitation, and others involved changes like adding shoulders or improving drainage. All of the projects were located on freeways. Some of the projects required lane closures, while others did not. Table 10 provides sample sizes for the database in terms of several measures – length, time duration, number of mile-months, and total crash count. The crashes included in the database are crashes that occurred on the freeway mainline segments; crashes on ramps, collector-distributor roads, or frontage roads were not included. In terms of mile-months, roughly half of the dataset is represented by the Texas projects, and roughly 60% of the dataset corresponds to “before” conditions. Overall, AADTs ranged between 5,000 and 70,000 vehicles per day (vpd) on the four-lane segments and between 50,000 and 150,000 vpd on the six-lane segments. The ranges of the key geometric variables are presented in terms of minimum, mean, and maximum values in Table 11. Cross-sectional width is generally smaller, and barrier lengths are generally larger, in the during periods than in the before period. These trends are consistent with typical work zone maintenance-of-traffic practices, which involve narrowing cross-sectional widths and adding temporary barriers to separate traffic from work areas. Meanwhile, Table 12 presents the pre- and during-work zone average crash rates by state, collision type, and severity. Overall, one sees that average crash rates increased in Ohio, Texas, and Virginia, but were down slightly at the work zones examined in Utah and Washington. Interestingly, single vehicle crash rates decreased on average at work zones (relative to pre- work zone conditions) in all of the states, whereas multi-vehicle crash rates were slightly higher. Based

NCHRP Project 17-61 49 on the differences in average before and during work zone PDO crash rates, it appears that work zone crashes tended to be more severe in Ohio, Utah, and Virginia, but less severe in Texas and Washington. Table 9. Construction Project Descriptions State Project Number of Lanes Description Before During Ohio I-75 4 4 Widening from 4 lanes to 6 Texas I-35 West 4 4 Widening from 4 lanes to 6 I-35 Waco 4 4 Widening from 4 lanes to 6 I-35 Lorena 4 4 Widening from 4 lanes to 6 I-35 Bruceville- Eddy 4 4 Widening from 4 lanes to 6 I-35 Troy 4 4 Widening from 4 lanes to 6 I-35 Salado 4 4 Widening from 4 lanes to 6 Virginia I-64 Low Moor 4 3 Drainage installation, grading, and paving I-95 Hanover County 6 6 Bridge replacement and ramp construction I-81 Christiansburg 4 4 Truck climbing lane construction and median and shoulder upgrade I-95 Richmond 6 6 Bridge rehabilitation I-95 Dumfries 6 5 Shoulder improvements and auxiliary lanes Washington I-90 4 4 Asphalt lane and shoulder paving I-82 4 4 Repaving asphalt pavement Utah I-15 Beaver 4 2 Paving I-70 Sevier 4 2 Concrete cutting and pouring works I-80 Summit 4 2 Concrete pavement reconstruction I-80 Summit 4 2 Barrier replacement I-80 Tooele 4 2 Concrete maintenance I-80 Summit 6 4 Barrier and lighting installation

NCHRP Project 17-61 50 Table 10. Sample Sizes State Period Average AADT Length, mi Duration, months Number of mile-months Total crash count All All 47329 660 10807 8176 7652 Before 43977 134 6628 6019 4512 During 56687 526 4179 2157 3140 Ohio All 54382 35 214 429 617 Before 55112 7 130 320 332 During 52235 27 84 109 285 Texas All 59925 409 9284 4228 5191 Before 59354 44 5593 2547 2599 During 60788 365 3691 1681 2592 Utah All 17714 137 309 1886 410 Before 17790 49 268 1683 387 During 17086 88 42 204 23 Virginia All 79354 48 780 775 1156 Before 79087 18 432 648 932 During 80714 30 348 127 224 Washington All 17888 32 220 857 278 Before 17874 16 206 821 262 During 18206 16 14 36 16 Table 11. Geometric Variable Ranges Period Statistic Cross-sectional width, ft Barrier length, mi Non- traversable median Total median Lane Right shoulder Left shoulder Right Left Before Min. 2.0 8.0 12.0 5.0 4.0 0.0 0.0 Mean 12.6 41.4 12.0 10.4 5.6 0.1 0.9 Max. 400.0 410.0 12.0 18.0 12.0 2.5 13.0 During Min. 2.0 2.0 11.0 0.0 0.0 0.0 0.0 Mean 14.8 40.2 11.9 8.6 6.2 0.1 0.5 Max. 400.0 405.0 12.5 17.0 14.0 5.6 13.0

NCHRP Project 17-61 51 Table 12. Crash Rate Comparison Between Pre- and During-Construction Periods State Period Crash rate (crashes per million vehicle-miles) Total By Collision Type By Crash Severity MV SV K A B C PDO OH Before 0.35 0.530 0.470 0.000 0.018 0.060 0.096 0.825 During 1.26 0.611 0.389 0.004 0.025 0.084 0.102 0.786 TX Before 0.56 0.633 0.367 0.007 0.018 0.098 0.142 0.735 During 0.81 0.666 0.334 0.005 0.014 0.084 0.143 0.754 UT Before 0.52 0.217 0.783 0.021 0.006 0.088 0.088 0.798 During 0.39 0.304 0.696 0.000 0.000 0.087 0.174 0.739 VA Before 0.60 0.761 0.239 0.002 0.067 0.071 0.152 0.708 During 0.64 0.764 0.236 0.005 0.049 0.159 0.082 0.703 WA Before 0.38 0.519 0.481 0.008 0.019 0.118 0.137 0.718 During 0.30 0.545 0.455 0.000 0.091 0.000 0.182 0.727 All Before 0.55 0.613 0.387 0.006 0.027 0.090 0.137 0.740 During 0.80 0.663 0.337 0.005 0.018 0.088 0.136 0.753 Statistical Analysis Methodology The research team then focused on developing a cross-sectional statistical model describing crash frequency in a given segment before and during work zone as a function of various geometric and work zone variables. The effort started with a model functional form described as follows: ௜ܰ ൌ ௕ܰ௔௦௘,௜ ൈ ܥܯܨଵ ൈ ܥܯܨଶ ൈ …ൈ ܥܯܨ௡; i =4 or 6 lanes Where, ௜ܰ = predicted annual average crash frequency for model i (i =4 or 6 lanes ௕ܰ௔௦௘,௜ = predicted annual average crash frequency at base conditions as described below ܥܯܨଵ, ܥܯܨଶ, …ܥܯܨ௡ = crash modification factors for various road segment features (1, 2, …, n) The predictive model calibration process consisted of the simultaneous calibration of four-lane and six- lane freeway models and CMFs using the aggregate model. Note that the number of lanes is the lane count during the pre-work zone (before) conditions. The simultaneous calibration approach was needed because all CMFs were common to four-lane and six-lane freeways. Fixed state-effects were used to capture the differences between states that are not possible with just the variables in the model. Work zones were allowed to have different state fixed effects than the normal roadway segments. Different functional forms were examined with various combinations of variables and the form presented below reflects the findings from several preliminary regression analyses. ܰ ൌ ௕ܰ௔௦௘ ൈ ܥܯܨ௟௪ ൈ ܥܯܨ௟௦௪ ൈ ܥܯܨ௥௦௪ ൈ ܥܯܨ௟௦ ൈ ܥܯܨ௠௪ ൈ ܥܯܨ௜௕ ൈ ܥܯܨ௢௕ ൈ ܥܯܨ௟௖

NCHRP Project 17-61 52 with, ܥܯܨ௟௪ = ݁௕೗ೢሺௐ೗ିଵଶሻ CMFlsw = eblswሺWls-6ሻ ܥܯܨ௥௦௪ = ݁௕ೝೞೢሺௐೝೞିଵ଴ሻ ܥܯܨ௟௦ = ݁௕೗ೞூ೗ೞ ܥܯܨ௠௪ = ሺ1.0 െ ௜ܲ௕ሻ݁௕೘ೢሺௐ೘ିଶௐ೗ೞିସ଼ሻ ൅ ௜ܲ௕݁௕೘ೢሺଶௐ೔೎್ିସ଼ሻ ܥܯܨ௜௕ = ݁௕೔್ூ೔್ ܥܯܨ௢௕ = ݁௕೚್ூ೚್ ܥܯܨ௟௖ = ݁௕೗೎ூ೗೎ where, ܥܯܨ௟௪ = lane width CMF ܥܯܨ௟௦௪ = left shoulder width CMF ܥܯܨ௥௦௪ = right shoulder width CMF ܥܯܨ௟௦ = lane shift CMF ܥܯܨ௠௪ = median width CMF ܥܯܨ௜௕ = inside (median) barrier CMF ܥܯܨ௢௕ = outside (roadside) barrier CMF ܥܯܨ௟௖ = lane closure CMF ௟ܹ = average lane width, ft ௟ܹ௦ = average left shoulder width, ft ௥ܹ௦ = average right shoulder width, ft ܫ௟௦ = lane shift presence indicator variable (= 1.0 if present, 0.0 if absent) ௠ܹ = median width, ft ௜ܹ௖௕ = distance from edge of inside shoulder to barrier face, ft ܫ௜௕ = inside barrier presence indicator variable (= 1.0 if present, 0.0 if absent) ܫ௢௕ = outside barrier presence indicator variable (= 1.0 if present, 0.0 if absent) ܫ௟௖ = lane closure presence indicator variable (= 1.0 if present, 0.0 if absent) Three types of models were developed. In the first model, for both “before” and “during” conditions, the intercept and ADT coefficients are forced to be the same but the CMFs are allowed to be different. This means, for base conditions (i.e., 6-ft inside shoulder, 10-ft outside shoulder, 60-ft median width including inside shoulders, no barriers, no lane closure or lane shift), the work zone will behave similar to a normal road: ௕ܰ௔௦௘ ൌ ܮ ൈ ݊ ൈ ݁௕బା௕ೌ೏೟ ୪୬ሺ஺஺஽்ሻା௕ೕ௦௧௔௧௘ where, L = length of work zone segment, n = time (in months)/12, AADT = annual average daily traffic on the segment, state = state in which the project was located if not in Texas, and bj = calibration coefficient for variable j.

NCHRP Project 17-61 53 The research team also tried describing the segment length as a variable instead of an offset (i.e., instead of L, it is used as Lβ). The β parameter value was equal to 0.99 for the pre-work zone (before) period and 1.01 for the during work zone period. Thus, it was concluded that the segment length should be used as an offset rather than as a variable. In the second model, for both “before” and “during” conditions, the intercept and ADT coefficients are forced to be the same but an added effect of work zone is introduced and the CMFs are different. This means, for base conditions (i.e., 6-ft inside shoulder, 10-ft outside shoulder, 10-ft median width, no barriers, no lane closure or lane shift), the work zone will still have more crashes than a normal road. ௕ܰ௔௦௘ ൌ ܮ ൈ ݊ ൈ ݁௕బା௕ೌ೏೟ ୪୬ሺ஺஺஽்ሻା௕ೢ೥ା௕ೕ௦௧௔௧௘ In the third model, for “before” and “during” conditions, the intercept and ADT coefficients are allowed to be different in addition to different CMFs. This means, for base conditions (i.e., 6-ft inside shoulder, 10-ft outside shoulder, 10-ft median width, no barriers, no lane closure or lane shift), the work zone will still have more crashes than a normal road plus the crash trend is different with the change in ADT (i.e., crash risk also changes). ௕ܰ௔௦௘_௪௭ ൌ ܮ ൈ ݊ ൈ ݁௕బ_ೢ೥ା௕ೌ೏೟_ೢ೥ ୪୬ሺ஺஺஽்ሻା௕ೕ௦௧௔௧௘ The inverse dispersion parameter, K (which is the inverse of the over dispersion parameter k), is allowed to vary with the segment length. The inverse dispersion parameter is calculated using: ܭ ൌ ܮ ൈ ݁௞ where, ܭ = inverse dispersion parameter ݇ = calibration coefficient for inverse dispersion parameter The mixed nonlinear regression procedure (NLMIXED) in the Statistical Analysis System (SAS) software was used to estimate the proposed model coefficients. This procedure was used because the proposed predictive model is both nonlinear and discontinuous. The log-likelihood function for the NB distribution was used to determine the best-fit model coefficients. Modeling Results Table 13 through Table 15 present the results of the analysis using each of the three models described above. Unfortunately, researchers were unable to develop statistically valid estimates of any of the geometric variables that were of primary interest in this analysis under any of the three modeling approaches. A few of the variables yielded parameter estimates that appeared reasonable and were within expected range, but were not statistically significant. For example, depending on the model results used, the effect of a lane shift on crashes equated to a CMF of between 1.165 and 1.185, comparable to a CMF value of 1.213 for lane shifts reported elsewhere (Chen and Tarko, 2012). Conversely, parameter values for other variables were highly counter-intuitive. For example, the marginal effects of increasing lane widths in work zones in model 1 was positive, suggesting that crashes increased as lane widths increased. Similarly, model 1 results regarding the effect of median and shoulder widths indicate that these variables have little effect upon work zone crashes, despite the fact that both have an effect upon crashes in normal (non-work zone) environments as illustrated by the roadway segments used in this analysis as well as the CMFs that have been developed for use in the HSM (see Figure 16 and Figure 17).

NCHRP Project 17-61 54 Table 13. Results of Analysis for Model 1 Variable Parameter Estimate Standard Error Pr > |t| Intercept (for 4-or 6-lane facilities) b0,4 -12.795 1.113 <.0001 b0,6 -12.603 1.153 <.0001 AADT badt 1.410 0.102 <.0001 Dispersion (for 4-or 6-lane facilities) k4 2.124 0.113 <.0001 k6 0.701 0.177 <.0001 Lane width blw,during 0.120 0.198 0.543 Left shoulder width blsw, pre -0.029 0.038 0.443 blsw, during -0.004 0.012 0.763 Right shoulder Width brsw,pre -0.109 0.050 0.030 brsw,during -0.019 0.017 0.259 Lane shift bls 0.163 0.152 0.284 Median width bmw, pre -0.001 0.001 0.315 bmw, during -0.002 0.002 0.235 Inside barrier bib, pre -0.197 0.143 0.168 bib, during 0.105 0.125 0.402 Outside barrier bob, pre -0.606 0.655 0.355 bob, during 0.256 0.171 0.134 Lane closure blc 0.473 0.576 0.412 State fixed effects for pre- construction conditions bWA, pre 1.101 0.415 0.008 bVA, pre -0.716 0.204 0.001 bOH, pre 0.306 0.251 0.223 bUT, pre 0.783 0.231 0.001 State fixed effects for during-construction conditions bWA, during 0.644 0.364 0.077 bVA, during -1.022 0.210 <.0001 bOH, during 0.649 0.190 0.001 bUT, during -1.935 0.749 0.010 Researchers did find that the state in which the work zone was located had a significant effect upon crashes in both the pre-work zone and during-work zone conditions. Relative to the work zones from Texas (which was defined as the base case in the model), projects in Ohio and Washington experienced a greater propensity for crashes than those in Texas both pre- and during the work zones, whereas project locations in Virginia experienced fewer crashes pre- and during-work zones. Somewhat counterintuitively, project locations in Utah experienced a greater propensity for crashes than the Texas locations in the pre-work zone condition, but fewer crashes in the during-work zone condition.

NCHRP Project 17-61 55 Table 14. Results of Analysis for Model 2 Variable Parameter Estimate Standard Error Pr > |t| Intercept (for 4-or 6-lane facilities) b0,4 -12.509 1.130 <.0001 b0,6 -12.318 1.167 <.0001 ADT badt 1.344 0.105 <.0001 WZ added effect bwz 0.683 0.207 0.001 Dispersion (for 4-or 6-lane facilities) k4 2.143 0.113 <.0001 k6 0.712 0.177 <.0001 Paved width in each direction bpw, pre -0.017 0.025 0.490 bpw, during -0.009 0.008 0.278 Lane shift bls 0.144 0.147 0.326 Median width bmw, pre -0.001 0.001 0.388 bmw, during -0.004 0.002 0.049 Inside barrier bib, pre 0.292 0.200 0.146 bib, during -0.144 0.154 0.350 Outside barrier bob, pre -0.129 0.629 0.837 bob, during 0.185 0.171 0.280 Lane closure blc 0.554 0.577 0.337 State fixed effects for pre- construction conditions bWA, pre 0.634 0.374 0.090 bVA, pre -0.783 0.228 0.001 bOH, pre 0.548 0.250 0.029 bUT, pre 0.449 0.243 0.065 State fixed effects for during-construction conditions bWA, during 0.707 0.362 0.051 bVA, during -0.920 0.195 <.0001 bOH, during 0.646 0.164 <.0001 bUT, during -1.954 0.731 0.008 The state-by-state effects likely reflect a wide range of other, unmeasured variables affecting crashes in the pre- and during-construction conditions and which interacted and confounded with the work zone variables of interest, yielding the lack of statistical significance of those variables in the model. In addition, the lack of statistical significance of the work zone variables of interest is also due in large part to the limitations of the available projects from which the dataset was developed. Although the research team desired to locate projects that spanned a good range of values for each of the variables of interest, the reality was that the projects available for analysis did not provide this range for most of the variables. In addition, changes in several of the variables were correlated, confounding their effects. For example, projects where a full inside shoulder was retained while the outside shoulder width was reduced were not available (the reverse of this condition was similarly not available). Likewise, projects where travel lanes were narrowed but where median and outside shoulders were maintained were not available. From a practical perspective, it makes sense why this confounding would occur. Typically, projects are completed in a manner that maximizes contractor productivity while maintaining traffic to the minimum requirements put forth by the owner agency. It would not make sense for the agency or the contractor to create work zones with some of the combinations of values of the variables listed above because they

NCHRP Project 17-61 56 would not have a positive effect on productivity or efficiency, and in some cases could actually detract from that. Given that the lane width and shoulder widths were not statistically significant, the total pavement width in each direction was used as a variable instead, in models 2 and 3. The base condition for this variable is 40 feet for 4-lane segments and 52 feet for 6-lane segments. Table 15. Results of Analysis for Model 3 Variable Parameter Estimate Standard Error Pr > |t| Intercept (normal) (for 4- or 6-lane facilities) b0,4,pre -11.231 1.270 <.0001 b0,6,pre -9.438 1.551 <.0001 ADT (normal) badt,pre 1.248 0.116 <.0001 Dispersion (normal) (for 4- or 6-lane facilities) k4,pre 2.537 0.188 <.0001 k6,pre 0.350 0.144 0.016 Intercept (work zone) (for 4-or 6-lane facilities) b0,4,during -10.036 2.829 0.000 b0,6,during -9.987 2.851 0.001 ADT (work zone) badt,during 1.164 0.256 <.0001 Dispersion (work zone) (for 4-or 6-lane facilities) k4,during 2.005 0.128 <.0001 k6,during 1.595 0.269 <.0001 Paved width in each direction bpw, pre -0.027 0.024 0.253 bpw, during -0.006 0.008 0.429 Lane shift bls 0.148 0.152 0.329 Median width bmw, pre -0.002 0.001 0.077 bmw, during -0.004 0.002 0.090 Inside barrier bib, pre -0.183 0.187 0.330 bib, during -0.071 0.156 0.648 Outside barrier bob, pre 1.423 0.633 0.025 bob, during 0.168 0.171 0.327 Lane closure blc 1.291 0.615 0.036 State fixed effects for pre- construction conditions bWA, pre 0.166 0.382 0.665 bVA, pre -1.836 0.284 <.0001 bOH, pre 0.319 0.221 0.150 bUT, pre 0.363 0.248 0.143 State fixed effects for during-construction conditions bWA, during 0.501 0.439 0.254 bVA, during -0.917 0.191 <.0001 bOH, during 0.606 0.175 0.001 bUT, during -1.927 0.719 0.008

NCHRP Pr Figure 16 test secti Figure 17 test secti oject 17-61 . Estimated ons (based . Estimated ons (based effects of on model 1) effects of on model 1) median shou as compared outside sho as compared 57 lder width to the HSM ulder width to the HSM before and d CMF. before and d CMF. uring the w uring the w ork zones a ork zones a t the t the

NCHRP Project 17-61 58 Because of the lack of statistical significance of the work zone variables computed in this analysis and the likely reasons for this as outlined above, the research team concluded that the best sources of estimates of the potential effect of these variables upon work zone crashes will continue to be the CMFs included in the HSM for permanent roadway conditions, as well as some of the CMFs included in the CMF Clearinghouse. Although this analysis did not yield specific CMFs for the individual variables, the results do provide general models that can be used by practitioners who do not have site-specific data or other means of estimating the number of crashes expected to occur in their interstate/freeway work zones. These models can serve as a generic safety performance function (SPF) describing the basic effect of a work zone upon crashes. Using the parameters for the intercept and AADT from Table 15 (model 3), the following functions are recommended: 4lanes ሺwork zoneሻ: ௕ܰ௔௦௘ ൌ ܮ ൈ ݊ ൈ ݁ିଵ଴.଴ଷ଺ାଵ.ଵ଺ସ ୪୬ሺ஺஺஽்ሻ 6lanes ሺwork zoneሻ: ௕ܰ௔௦௘ ൌ ܮ ൈ ݊ ൈ ݁ିଽ.ଽ଼଻ାଵ.ଵ଺ସ ୪୬ሺ஺஺஽்ሻ The base work zone conditions for the models are:  Pavement width = 40 feet in each direction for 4-lane segments and 52 feet in each direction for six-lane segments (equal to 12-ft lanes, 6-ft inside shoulder, and 10-ft outer shoulder)  No lane shifts present  No lane closures present  Median width of 60 feet, which includes the inside shoulder width of 6-ft in both directions  No longitudinal barriers present These equations are plotted in Figure 18 and Figure 19 (labeled as NCHRP 17-61, Work Zone). Also plotted in those figures for comparison purposes are the combined safety performance functions for normal conditions included in the HSM (both Rural and Urban), and the results of the statistical models from Table 14 and Table 15 for the pre-work zone (NCHRP 17-61, Normal Non-Work Zone) conditions from the project locations in this study. The model structure and parameters for model 3 provided the best fit of the data for the work zone conditions on both the four-lane and six-lane facilities, and for the pre-work zone conditions on four-lane facilities. However, for pre-work zone conditions on six-lane facilities, the model 2 structure worked better. Overall, one sees that the models for normal conditions match the HSM functions for normal roadway very well, providing reasonable confidence about the validity of the project locations and data. For the four-lane roadways, the function from this study for the pre-work zone conditions falls between the rural and urban functions. This result was as would be expected, as the project locations came from both rural and urban facilities nationally. For the six-lane roadways, the pre-work zone function for this study mimicked fairly closely the urban six-lane roadway function from the HSM. This again was consistent with expectations, as the six-lane roadway project locations used in this study came from primarily urban areas.

NCHRP Pr Figure 18 Figure 19 oject 17-61 . Predicted . Predicted crashes on crashes on 4-lane freew 6-lane freew 59 ay segments ay segments . .

NCHRP Pr R presence o roadway s models w nonlinear from pre- illustrated In abso work zone equations 4-lane f ܹܼܥܯܨସ 6-lane f ܹܼܥܯܨ Figure 2 segments oject 17-61 elative to pr f a work zon egment as co ere construct . Specifically work zone c graphically i lute terms, th presence sti are as shown acilities: ି௟௔௡௘௦ ൌ ݁݁ acilities: ଺ି௟௔௡௘௦ ൌ ݁݁ 0. Average . e-work zone e on a roadw mpared to th ed, the analy , work zone onditions on n Figures 20 ough, the nu ll does increa below. ିଵ଴.଴ଷ଺ାଵ.ଵ଺ସ ିଵଵ.ଶଷଵାଵ.ଶସ଼ ିଽ.ଽ଼଻ାଵ.ଵ଺ସ ିଵଶ.ଷଵ଼ାଵ.ଷସସ effect of w (normal) co ay segment i e crashes exp sis results als s are predict higher AA and 21. mber of add se as AADT ୪୬ሺ஺஺஽்ሻ ୪୬ሺ஺஺஽்ሻ ୪୬ሺ஺஺஽்ሻ ୪୬ሺ஺஺஽்ሻ ork zones 60 nditions, Fig ncreases the ected if a wo o suggest th ed to result i DT facilities itional crashe increases. E upon crash ure 18 and number of cr rk zone is n at the effect n a smaller p than on low s per mile p xpressed as a es predicte Figure 19 a ashes expect ot present. B of a work zo roportional i er AADT f er year that a work zone C d on 4-lane lso show tha ed to occur o ecause of ho ne upon cras ncrease in cr acilities. T re associated MF, the res freeway t the n that w the hes is ashes his is with ulting

NCHRP Pr A pract where a w and apply Figure segme oject 17-61 itioner who a ork zone is it to the base 21. Averag nts. lready has a planned coul line to estima e effect of baseline estim d calculate th te the numbe work zone 61 ate of crash e appropriat r of crashes s upon cra es normally o e expected w expected duri shes predic ccurring on ork zone CM ng the work ted on 6-la the section o F for that se zone. ne freeway f road ction

Next: Summary and Research Recommendations »
Analysis of Work Zone Crash Characteristics and Countermeasures Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB's National Cooperative Highway Research Program (NCHRP) Web-Only Document 240: Analysis of Work Zone Crash Characteristics and Countermeasures documents the research results of multiple analyses focused on developing an improved understanding of work zone crash characteristics and countermeasure effectiveness used to produce NCHRP Research Report 869: Estimating the Safety Effects of Work Zone Characteristics and Countermeasures: A Guidebook.

The guidebook provides practitioners who develop phasing and staging plans for temporary traffic control through work zones with guidance to evaluate the safety impacts of their plan decisions. There is limited data on work zone crashes and fatalities that address trends, causality, and the best use of resources to improve work zone safety. This guidebook provides clearer guidance to encourage the use of data-driven, comprehensive, collaborative planning approaches for the selection and implementation of effective countermeasures to improve work zone safety.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!