The National Academies Press

Currently Skimming:

Pages 38-79

The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.

From page 38... ... 38 4 METHODOLOGY This chapter describes the analysis methodology conducted on the project which comprises five components: 1. the generation of aggregate-level driver and vehicle variables, 2. Read the entire page →
From page 39... ... 39 vehicle information using the QIE method requires the variables to be analyzed to have just a few categories. From crash reports, driver sex can generally be grouped into male, female, or other. Read the entire page →
From page 40... ... 40 If 𝐻𝐻0 holds, the likelihood in equation (4-2) can be simplified to 𝐿𝐿(𝑒𝑒𝑚𝑚 , . Read the entire page →
From page 41... ... 41 Table 4.2 Crash Frequency of Age Group 15-19 and 20-24 Driver Age K A B C O Total 15-19 155 614 4887 7839 43,213 56,708 20-24 507 1990 12,674 20,746 94,333 130,250 Total 662 2604 17,561 28,585 137,546 186,958 Step 3: The proportion of each severity was calculated for each group. In this case, the proportion of K crashes for driver age 15-19 is pi = 155/56708 = 0.0027333. Read the entire page →
From page 42... ... 42 Step 5: The value of the test statistic is compared with the critical value of the Chi-square distribution for specific degrees of freedom. In this case, the degree of freedom is 4 and the critical value of Chi-square is provided in Table 4.5. Read the entire page →
From page 43... ... 43 there are too few Motorcycle/Moped crashes, the team combined this group with the "Others" group in later analysis. The category HV includes the following: • Truck or Truck Tractor • Truck or Truck Tractor With 1 Trailer • Truck or Truck Tractor With 2 Trailers • Truck or Truck Tractor With 3 Trailers • Single Unit Tanker • Truck/Trailer and 1 Tank Trailer • Truck/Trailer and 2 Tank Trailers • School Bus • Other Bus • Emergency Vehicle • Highway Construction Equipment • Spilled Loads • 'Disengaged Tow 4.1.3 Development of Algorithm for Generating Aggregate-Level Data Due to the small number and randomness of crashes by individual road segment or intersection, getting statistically stable estimates of driving population distributions by characteristics requires aggregation at a higher level. Read the entire page →
From page 44... ... 44 Sample Data Analysis Here we use a sample of data to illustrate the above process of calculating aggregate-level driver and vehicle information. The target proportions are male/female driver proportions in the total driving population at the site, the passenger car/truck or bus/other vehicle proportions, and the driver proportions by the age groups defined previously. Read the entire page →
From page 45... ... 45 Table 4.8 Frequencies for Total Not at Fault (NF) Drivers or Vehicles in Each Site ID CNTYRTE BEGMP ENDMP Not at Fault Total Not-at-Fault Drivers Age 15_ 24 Age 25_ 69 Age 70+ PC HV MCO Male Female 1 01001 23 D 59.862 59.926 1 0 1 0 0 0 1 1 0 2 01001 23 D 60.047 60.048 0 0 0 0 0 0 0 0 0 3 01001 23 D 60.049 60.125 2 1 1 0 2 0 0 1 1 4 01001 23 D 60.231 60.397 0 0 0 0 0 0 0 0 0 5 01001 23 D 60.400 60.457 2 0 1 1 1 1 0 2 0 6 01001 23 D 60.680 60.725 2 0 2 0 2 0 0 0 2 7 01001 23 D 61.050 61.126 1 0 1 0 0 0 1 0 1 8 01001 23 D 61.700 61.838 7 1 5 1 4 2 1 5 2 9 01053 17 D 1.047 1.051 0 0 0 0 0 0 0 0 0 10 01053 17 D 1.089 1.195 0 0 0 0 0 0 0 0 0 11 01053 17 D 1.470 2.780 14 1 12 1 10 3 1 8 6 12 01101 08 D 26.102 26.183 0 0 0 0 0 0 0 0 0 13 01101 08 D 27.104 27.280 3 1 1 1 2 0 1 2 1 14 01101 08 D 27.281 27.480 5 1 3 1 4 1 0 2 3 15 01101 08 D 31.188 31.270 5 0 5 0 3 1 1 3 2 16 01101 08 RD 27.564 27.774 2 1 1 0 2 0 0 1 1 Read the entire page →
From page 46... ... 46 Step 2: Aggregate frequencies Within the same route (partitioned by county route number (CNTYRTE) Read the entire page →
From page 47... ... 47 Table 4.9 Vehicle Type Proportions in Each Site ID CNTYRTE BEGMP ENDMP NF Driver NF Driver sum NF PC NF PC sum NF PC/Total NF HV NF HV sum (HV) /Total 1 01001 23 D 59.862 59.926 1 15 0 9 0.600 0 3 0.200 2 01001 23 D 60.047 60.048 0 15 0 9 0.600 0 3 0.200 3 01001 23 D 60.049 60.125 2 15 2 9 0.600 0 3 0.200 4 01001 23 D 60.231 60.397 0 15 0 9 0.600 0 3 0.200 5 01001 23 D 60.400 60.457 2 15 1 9 0.600 1 3 0.200 6 01001 23 D 60.680 60.725 2 15 2 9 0.600 0 3 0.200 7 01001 23 D 61.050 61.126 1 15 0 9 0.600 0 3 0.200 8 01001 23 D 61.700 61.838 7 15 4 9 0.600 2 3 0.200 9 01053 17 D 1.047 1.051 0 14 0 10 0.714 0 3 0.214 10 01053 17 D 1.089 1.195 0 14 0 10 0.714 0 3 0.214 11 01053 17 D 1.470 2.780 14 14 10 10 0.714 3 3 0.214 12 01101 08 D 26.102 26.183 0 8 0 6 0.750 0 1 0.125 13 01101 08 D 27.104 27.280 3 8 2 6 0.750 0 1 0.125 14 01101 08 D 27.281 27.480 5 8 4 6 0.750 1 1 0.125 15 01101 08 D 31.188 31.270 5 7 3 5 0.714 1 1 0.143 16 01101 08RD 27.564 27.774 2 7 2 5 0.714 0 1 0.143 Read the entire page →
From page 48... ... 48 4.2 CALIBRATION OF THE HSM AND NCHRP 17-62 MODELS AS BENCHMARK 4.2.1 Overall Method The objective of this subtask was to calibrate the HSM (1st edition) Safety Performance Function (SPF) Read the entire page →
From page 49... ... 49 frequency for the corresponding severity levels to generate the adjusted crash frequency for the specific severity levels for each site. We illustrate each step in more detail below using two example facilities. Read the entire page →
From page 50... ... 50 SPF Adjustment Factors AFs are used to adjust the SPF estimate of predicted average crash frequency for the effect of individual geometric design and traffic control features. The AF for the SPF base condition of each geometric design or traffic control feature has a value of 1.00. Read the entire page →
From page 51... ... 51 Table 4.11 AF For Installation of Right Turn Lanes on Intersection Approaches Intersection Type Intersection Traffic Control Number of Non-Stop-Controlled Approaches with Right-Turn Lanes 1 2 3 4 Three-leg Minor road stop control 0.86 0.74 -- -- Four-leg Minor road stop control 0.86 0.74 -- -- Traffic signal 0.96 0.92 0.88 0.85 Lighting The base condition for intersection lighting is the absence of lighting. Read the entire page →
From page 52... ... 52 4.2.4 Roadway Segments Rural Two-Lane Two-Way Roads Safety Performance Functions The effect of traffic volume (AADT) on crash frequency is incorporated through an SPF, while the effects of geometric design and traffic control features are incorporated through AFs. Read the entire page →
From page 53... ... 53 Table 4.13 AF for Lane Width Lane Width AADT < 400 400 to 2000 > 2000 9-ft or less 1.05 1.05+2.81x10-4(AADT-400) Read the entire page →
From page 54... ... 54 Table 4.15 AF for Shoulder Type Shoulder Type Shoulder Width (ft) 0 1 2 3 4 6 8 Paved 1.00 1.00 1.00 1.00 1.00 1.00 1.00 Gravel 1.00 1.00 1.01 1.01 1.01 1.02 1.02 Composite 1.00 1.01 1.02 1.02 1.03 1.04 1.06 Turf 1.00 1.01 1.03 1.04 1.05 1.08 1.11 The AFs for shoulder width and type shown in the tables apply only to the collision types that are most likely to be affected by shoulder width and type: single-vehicle run-off the-road and multiple-vehicle head-on, opposite direction sideswipe, and same-direction sideswipe crashes. Read the entire page →
From page 55... ... 55 Note: Some roadway segments being analyzed may include only a portion of a horizontal curve. In this case, 𝐿𝐿𝑐𝑐 represents the length of the entire horizontal curve, including portions of the horizontal curve that may lie outside the roadway segment of interest. Read the entire page →
From page 56... ... 56 Driveway Density The base condition for driveway density is five driveways per mile. As with the other AFs, the model for the base condition was established for roadways with this driveway density. Read the entire page →
From page 57... ... 57 The value of 𝑃𝑃𝑑𝑑𝑑𝑑𝑑𝑑 can be estimated using the following equation: 𝑃𝑃𝑑𝑑𝑑𝑑𝑑𝑑 = (0.0047∗𝐴𝐴𝐴𝐴) Read the entire page →
From page 58... ... 58 4.3 NEGATIVE BINOMIAL - ORDERED PROBIT FRACTIONAL SPLIT MODELING FRAMEWORK 4.3.1 Mathematical Representation The model structure described presents an approach to jointly model total number of crashes and proportion of crashes by severity. We name it the Negative Binomial - Ordered Probit Fractional Split (NBOPFS) Read the entire page →
From page 59... ... 59 Table 4.16 Joint NB-OPFS Model Results for California (Urban Four-Lane freeways) Variables Count Model Coefficient (t-stat) Read the entire page →
From page 60... ... 60 propensity of crashes at higher traffic volumes, as well as the effect of exposure. Further, heavy vehicle percentage shows a negative impact on the total crash frequency, indicating a lower probability of crashes in urban four-lane freeway segments with higher heavy vehicle activity. Read the entire page →
From page 62... ... 62 count models for each severity level, we assume that the observed crash counts at different severity levels each follow a negative binomial distribution with μi as the mean crash count. Read the entire page →
From page 63... ... 63 Table 4.20 Variables Used in the Univariate Count Models Variables Explanations AADT Annual Average Daily Traffic LW < 12 Dummy that equals 1 if Lane width < 12 ft; 0 if Lane width is 12 ft or greater SW Shoulder width (ft) ; continuous variable MW Median width (ft) Read the entire page →
From page 64... ... 64 Table 4.21 Parameter Estimates for California Urban Four-Lane Freeway Segments (Univariate) Variables KABCO (1357) Read the entire page →
From page 65... ... 65 Table 4.22 Example Data for Predicting Crashes on Urban Four-Lane Freeway Segments (Univariate) Variables Segment 1 Segment 2 Segment 3 Segment Length (mile) Read the entire page →
From page 67... ... 67 population. In the multilevel modeling framework, the driver and vehicle characteristics are further considered in the discrete outcome model at the crash-level, instead of just at the site-level. Read the entire page →
From page 68... ... 68 4.5.3 Discrete Outcome Model In the discrete outcome model, the dependent variable is the injury severity level of a crash, and the predictors are the crash-level driver and vehicle information along with site-level roadway characteristics, including AADT. Based on our investigation of four distinct types of models (Model 1 - Partial Proportional Odds model with individual effects of driver and vehicle characteristics as predictors, Model 2- Partial Proportional Odds model with the top groups as predictors, Model 3 - Ordered Logit model with the top groups as predictors, and Model 4 - Ordered Probit model with the top groups as predictors) Read the entire page →
From page 69... ... 69 𝑀𝑀𝑛𝑛𝑙𝑙𝐿𝐿 = ∑ ∑ 𝑎𝑎𝑚𝑚𝑚𝑚log [𝐹𝐹�𝛿𝛿𝑚𝑚 − 𝒙𝒙′ 𝒋𝒋𝜷𝜷� − 5𝑚𝑚=1𝑛𝑛𝑚𝑚=1 𝐹𝐹�µ𝑚𝑚−1 − 𝒙𝒙′ 𝒋𝒋𝜷𝜷�] Read the entire page →
From page 70... ... 70 effects of independent variables on the proportions of each driver group, and hence are easier to interpret. Therefore, the alternate parameterization method is used in estimating the Dirichlet regression models in this study. Read the entire page →
From page 71... ... 71 Table 4.25 Description of Variables Used in Developing the Multilevel Discrete Outcome Model Dataset Variable Name Notation Variable Type Description (Unit) Crash Injury Severity Severity Severity Ordered K, A, B, C, O Roadway Geometry Lane Width LW<12' Dummy 1 if lane width is less than 12', 0 otherwise Median Width MW Continuous Median width (ft) Read the entire page →
From page 72... ... 72 Modeling and Results As the same univariate count model is used for total crash counts, the parameter estimates for total crash count model can be found in Table 4.21 for California Urban four-lane freeway segments. The parameter estimates for the connection model and the discrete outcome model are shown in Table 4.26 and Table 4.27, respectively. Read the entire page →
From page 73... ... 73 Table 4.26 Parameter Estimates of Connection model for Urban Four-Lane Freeway Segments Variables G1 (Pc_M_ Mid) Read the entire page →
From page 74... ... 74 Table 4.27 Parameter Estimates of the Ordered Logit Discrete Outcome Model for Four-Lane Urban Freeways Variables OL (Full) OL (Reduced) Read the entire page →
From page 75... ... 75 Table 4.28 Example Data for Predicting Crashes on Urban Four-Lane Freeway Segments (Multivariate) Variables Segment 1 Segment 2 Segment 3 Segment Length (mile) Read the entire page →
From page 76... ... 76 For segment 1, proportion of G1 is exp(𝑒𝑒1.𝛽𝛽.1) Read the entire page →
From page 78... ... 78 Total crash count is 10.78 (Table 4.29) Proportion of total crashes for G1 is 0.055 (Table 4.30) Read the entire page →
From page 79... ... 79 The MAD gives a measure of the average magnitude of variability of prediction. Smaller values are preferred to larger values. Read the entire page →

From page 38...

... 38 4 METHODOLOGY This chapter describes the analysis methodology conducted on the project which comprises five components: 1. the generation of aggregate-level driver and vehicle variables, 2.