CHAPTER 5. FINAL ANALYSIS
5.4 Final Regression Analysis
5.4.1 Additional roadway data form RCI
5.4.1.1 Additional RCI categorical variable setup
Many variables in RCI were already categorical in nature, other were transformed from numerical to categorical. A description and reasoning behind these categorizations is presented in this section (categorical variables) and the next section (numerical variable transformations).
The lighting conditions on multilane high-speed corridors are further described by a categorical variable in RCI that describes the density of lighting poles in a road section. Both LIGHTCDE (non-high mast lighting pole density) and LIGHTING (high mast lighting pole density) have three categories: yes, partial and none. The categories are defined in Table 5-40 below, which shows the different values for each category in high mast and non-high mast lighting. High mast lighting has various advantages including increased coverage area and better uniformity to avoid driver glare effects. In addition, high masts are generally relocated out of the clear zone and with significantly lower amounts per mile, may also contribute to improved roadside safety performance. In Florida, high mast lighting is primarily limited to interchange locations far from developed areas and some older sites due to light trespassing issues. The implications of these design differences will be discussed later in this section.
Table 5-40 Definitions of the lighting conditions of roads from RCI
Codes Non-high mast (LIGHTCDE) High mast (LIGHTING) N One or none light poles exists
in the section One or none light poles exists in the section
Partial lighting exists Partial lighting exists P (Rates of 4-24 lights poles per
mile)
(Rates of 4-9 light poles per mile)
Full lighting exists Full lighting exists Y (Rates of 25 lights poles per
mile or more)
(Rates of 10 light poles per mile or more)
The Access Management Class variable (nACMANCLS) has a category 1 for limited access and classes 2 through 7 for the multilane non-limited access. Crashes on roads with category one (limited access) was not found in the final sample, as expected. The variable coding used in the final analysis follows the codes 2-7 as defined in Table 5-41 below and one category for not applicable locations without an access classification by FDOT. Class codes 2-4 are defined by less dense land uses with intersection spacing of half a mile (lower density urban areas) and these categories were joined in the models, as the model building process dictated. In the end, the variable had five levels: class 2-4, class 5, 6, 7 and 9 (not applicable). The last level (9) was created for those road sections without access class codes.
Table 5-41 Definitions of the access class codes (from Rule 14-97) of multilane roads in Florida (FDOT, 2007) Median Openings (ft) 2 Restrictive w/ Service
Roads 2,640 1,320 2,640
3 Restrictive 2,640 1,320 2,640
4 Non-Restrictive 2,640
No changes in the urban size (nURBSIZE) variable coding had five levels based on the area population: rural, small urban, small urbanized, large urbanized and metropolitan. The rural/urban binary variable in the model was substituted by urban size to investigate its effects on the model. The urban size did not perform as well as the binary rural/urban classification in the injury severity model. It could not be present simultaneously with the rural/urban variable due to collinearity issues (which were evident during model building), thus it was not entered into the final model.
Two variables related to the road geometric design were expected to contribute to improved intersection models. The auxiliary lane type (nAUXLNTYP) indicates the type of auxiliary lane (if present) in the road section or intersection where the crash occurred. The main types of auxiliary lanes tested were left turn, right turn, bus, merging (outside), merging (inside) and parking lanes. In the final models (segment and non-signalized) only the merging (outside), merging (inside) and parking lanes were significant. This variable is complemented by the auxiliary lane number (nAUXLNUM), which depicts the number of auxiliary lanes present (if applicable) in the road section or intersection where the crash occurred. The combination of these two variables was tested in all the models, with limited success.
Friction courses are applied to roads with heavy traffic volumes and high speed limits.
The type of friction course has been evolving during the past decades and the field nFRICTCSE records 8 types of friction courses, as defined by the FDOT. No crash data was found for the newer friction courses (FC-9.5 and 12.5). There are two general types of friction courses currently used by FDOT: dense graded and open graded. Their thickness is controlled by specification through the minimum and maximum spread rate. Generally friction course type 5 (FC-5) is specified for multilane roadways with speed limits greater than 45 mph. FC-6 mixes
(dense graded) are typically specified for roadways with posted speed limits less than or equal to 45 mph, but were non-significant in the injury severity model essentially due to its low sample size. Older dense graded friction courses (FC-1 and FC-4) were found significant and are discussed in more detail in subsequent sections. The dominant type of friction course for the roads in which crashes were reported was FC-2 as shown in Table 5-42 below. Preliminary analysis and model building led to five categories: FC-2 (base), FC-1, FC-4, FC-5 and none or other types.
Table 5-42 Types of friction courses related to crash involvements in multilane high speed roads Type Friction Course Frequency Percent
0 (none or null) 6767 6.30
1 (FC 1) 6736 6.27
2 (FC 2) 48354 45.00
3 (FC 3) 14022 13.05
4 (FC 4) 24618 22.91
5 (FC 5) 2885 2.68
6 (FC 6) 4067 3.79
Total 107449 100.00
An attempt was made to find correlations between friction course, skid resistance and severe injury to driver involvement ratios. Additional analysis of the interaction plot in Figure 5-6 below show some important correlations that should be taken into account when interpreting the results of the injury severity analysis models. First, older friction courses (FC-2 and FC-4) exhibited the highest severe injury ratio for skid resistance values of 35 and over. For road sections (or intersections) with FC-4, severe injuries account for 2.50% of total involvements for skid numbers over 44, compared to only 0.14% for all skid resistance values. For road sections (or intersections) with FC-2, severe injuries account for 3.18% of total involvements for skid numbers ranging from 35 to 44, but is also high (3.10%) for the rest of the skid resistance values.
All involvements
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035
None 1 2 3 4 5 6
Friction Course
Proportion of driver involvem
Skid no. 1-25 Skid no. 26-34 Skid no. 35-44 Skid no. >44 All skid no.
Figure 5-6 –Severe injury ratio to all involvements by skid resistance and friction course
Roads with friction courses FC-2 and FC-4 accounted for 46% and 23% of the total severe involvements. Similarly for the wet pavement crashes roads with FC-2 and FC-4 accounted for 50% 20% of the severe injury involvements, as shown in Figure 5-7. There is a clear tendency of increased total and severe injury involvements at locations with older friction courses. Decreasing skid resistances of older friction courses (polishing effects) is an important concern for skid hazard prevention programs. However, roads under wet pavement hazards are considered when at least 25% of the crashes are related to wet pavement. If there is a systematic decrease in friction resistance on multilane high-speed corridors with older friction courses, it is not necessarily captured at the district level. The injury severity models only showed a trend by land use.
Wet pavement involvements
Figure 5-7 –Severe injury ratio to wet pavement involvements by skid resistance and friction course
Another important concern is the frequency of severe injuries on roads with skid numbers 1-25 (deemed unsafe) for both total and wet pavement crashes, as shown in Table 5-43 below.
Driver crash involvements on locations with low skid resistance accounted for 28.56% of the severe injuries (total crashes) and 30.87% of the severe injuries (wet pavement crashes). The nature of these crashes, especially for wet pavement, should be further investigated.
Table 5-43 Percent of driver involvements for total and wet pavement crashes by skid resistance Percent of driver involvements
Total Wet Pavement
Skid Resistance
Non-severe Severe Non-severe Severe
1-25 34.35 28.65 37.72 30.87
26-34 0.61 0.49 0.93 0.61
35-44 57.07 61.05 55.16 60.53
>44 7.97 9.81 6.19 7.99
All numbers 100 100 100 100