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.
166 A P P E N D I X G This appendix presents the companion tables to the four cross-sectional generalized linear model (GLM) analyses inves- tigating the effect of shoulder rumble strip offset. ⢠Table G-1 presents the GLM results to investigate the effect of shoulder rumble strip placement (edgeline vs. non-edge- line) on SVROR FI crashes based on all site types; it is the companion table to Table 42. ⢠Table G-2 presents the GLM results to investigate the overall effect of shoulder rumble strip placement on SVROR FI crashes across all sites in all states; it is the companion table to Table 43. ⢠Table G-3 presents the GLM results to investigate the effect of shoulder rumble strip offset (at three levels) on SVROR FI crashes based on all site types; it is the companion table to Table 44. ⢠Table G-4 presents the GLM results to investigate the com- bined effect of shoulder rumble strip offset and recovery area on SVROR FI crashes based on all site types; it is the companion table to Table 45. Number of sites, number of site-years, offset, and offset à recovery area statistics for each model are provided in the corresponding Tables 42 through 45. Table G-1: The statistics shown for each roadway type and state (combined or single) include: ⢠Intercept: estimate and standard error ⢠ADT (on natural log scale): estimate, standard error, and p-value (i.e., signiï¬cance level) ⢠Outside RHR: estimate, standard error, and p-value ⢠Overdispersion parameter: estimate and standard error Each regression model is represented by the following equation: Expected total crashes mi yr a b lnADT c R = + à + Ãexp HR d RS PlacementOut + Ã( ) GLM Analysis Results for Effect of Shoulder Rumble Strip Offset and Recovery Area on Safety where a (i.e., intercept), b, and c are the coefï¬cients whose estimates are shown in Table G-1. The companion coefï¬cients for rumble strip placement, d, at two levels (edgeline vs. non- edgeline) as compared to no RS, are shown in Table 42. Table G-2: The statistics shown for all sites and states combined include the estimate for: ⢠Intercept ⢠ADT (on natural log scale) ⢠Outside RHR ⢠Overdispersion parameter The single regression model is represented by the following equation: where a (i.e., intercept), b, and c are the coefï¬cients whose estimates are shown in Table G-2. The variable IRoadway typeÃState is an indicator variable with value 1 for a particular roadway type à state combination in the table, and zero otherwise. The companion coefï¬cients for rumble strip placement, d, at two levels (edgeline vs. non-edgeline) as compared to no RS, are shown in Table 43. Table G-3: The statistics shown for each roadway type and state (combined or single) include: ⢠Intercept: estimate and standard error ⢠ADT (on natural log scale): estimate, standard error, and p-value (i.e., signiï¬cance level) ⢠Outside RHR: estimate, standard error, and p-value ⢠Overdispersion parameter: estimate and standard error Each regression model is represented by the following equation: Expected total crashes mi yr a b lnADT c R = + à + Ãexp HR d OffsetOut + Ã( ) Expected total crashes mi yr a b lnADT c R = + à + Ãexp HR d RS Placement I Out Roadway type Stat + Ã( ){ à à e }
noisrepsidrevORHRedistuOTDAnltpecretnI Roadway type State Estimate SE Estimate SE pâvalue Estimate SE pâvalue Estimate SE Urban freeways PA â6.92 1.15 0.67 0.11 <.0001 0.20 0.07 Combined â8.76 1.46 0.79 0.15 <.0001 0.21 0.11 0.0622 0.19 0.06 MO â7.02 1.79 0.67 0.18 0.0002 0.20 0.07 Rural freeways PAa Combined â7.25 1.47 0.59 0.16 0.0003 0.30 0.10 0.0024 0.44 0.09 MN â9.73 1.47 0.90 0.15 <.0001 0.20 0.07 MO â15.06 3.93 1.55 0.42 0.0002 0.58 0.20 Rural multilane divided highways (nonfreeways) PA â0.05 0.28 0.57 0.32 Combined â5.46 0.77 0.31 0.10 0.0014 0.41 0.06 <.0001 0.84 0.13 MN â4.49 1.20 0.25 0.15 0.09 1.15 0.33 MOa Rural twoâlane roads PAa Combined â5.38 0.79 0.31 0.10 0.003 0.41 0.06 <.0001 0.86 0.14 Rural two-lane roadsb MN â3.83 1.20 0.17 0.15 0.28 1.21 0.46 a LM algorithm did not converge. b Excludes 53 Minnesota nontreatment cross-sectional sites. Table G-1. GLM estimates for SVROR FI crashes based on all sitesârumble strip placement analysis.
168 where a (i.e., intercept), b, and c are the coefï¬cients whose estimates are shown in Table G-3. The companion coefï¬cients of offset distance, d, at three levels as compared to no RS, are shown in Table 44. Table G-4: The statistics shown for each roadway type and state (combined or single) include: ⢠Intercept: estimate and standard error ⢠ADT (on natural log scale): estimate, standard error, and p-value (i.e., signiï¬cance level) ⢠Outside RHR: estimate, standard error, and p-value ⢠Overdispersion parameter: estimate and standard error Each regression model is represented by the following equation: Expected total crashes mi yr a b lnADT c R = + à + Ãexp HR d Offset RAOut + à Ã( ) where a (i.e., intercept), b, and c are the coefï¬cients whose estimates are shown in Table G-4. The companion coefï¬- cients, d, for the combination offset à recovery area, at five levels as compared to no RS with narrow shoulders, are shown in Table 45. Tables G-1 through G-4: For states that treat both sides of a divided highway as separate sites (i.e., Missouri and Pennsylvania), the RHR variables in the models represent the values for a single side of the divided highway. When both sides of a divided highway are treated as a single site (i.e., Minnesota sites), the RHR variables in the model represent average values for both directions of travel. Similarly, the RHR variable in the models for rural two-lane roads represents the average RHR for both sides of the roadway. No GLM results are shown in those cases where the algo- rithm did not converge. Empty cells in those cases where the GLM algorithm did converge indicate that the corresponding coefï¬cient estimate is not statistically signiï¬cant at the 0.15 level or that the coefï¬cientâs sign is not in the expected direction. Roadway type State Intercept (or state effect) lnADT Outside RHR Overdispersion Urban freeways PA â3.70 0.62 0.07 MO â4.29 0.66 0.12 Rural freeways PA â6.82 0.87 0.27 MN â6.00 0.82 0.07 MO â11.11 1.42 0.13 Rural multilane divided highways (nonfreeways) PA â2.14 0.38 0.39 MN â0.93 0.06 0.43 MO â6.01 0.45 0.98 Rural twoâlane roads PA â3.01 0.25 â0.0008 0.29 Table G-2. GLM estimates for SVROR FI crashes based on all sitesâ overall rumble strip placement analysis.
noisrepsidrevORHRedistuOTDAnltpecretnI Roadway type State Estimate SE Estimate SE pâvalue Estimate SE pâvalue Estimate SE Urban freeways PA â5.85 1.21 0.60 0.11 <.0001 0.18 0.07 Combined â0.54 0.26 0.07 MO â7.02 1.79 0.67 0.18 0.0002 0.20 0.07 Rural freeways PA â0.54 0.24 0.11 Combined â6.59 1.49 0.60 0.16 0.0003 0.29 0.10 0.004 0.43 0.09 MN â9.73 1.47 0.90 0.15 <.0001 0.20 0.07 MO â15.11 3.91 1.55 0.42 0.0002 0.58 0.20 Rural multilane divided highways (nonfreeways) PA 0.81 0.00 0.48 0.30 Combined â5.45 0.78 0.34 0.11 0.0021 0.39 0.07 <.0001 0.81 0.12 MN â4.50 1.33 0.25 0.19 0.1774 1.14 0.33 MO â0.65 0.75 0.23 Rural twoâlane roads PA â0.89 0.59 0.15 noisrepsidrevORHRedistuOTDAnltpecretnI Roadway type State Estimate SE Estimate SE p-value Estimate SE p-value Estimate SE Urban freeways PA â6.92 1.15 0.67 0.11 < .0001 0.20 0.07 Combined â8.73 1.51 0.78 0.16 < .0001 0.21 0.12 0.068 0.19 0.06 MO â6.74 1.77 0.65 0.18 0.0004 0.20 0.07 Rural freeways PAa Combined â7.36 1.56 0.60 0.17 0.001 0.29 0.10 0.003 0.43 0.09 MN â8.90 1.47 0.82 0.15 < .0001 0.19 0.07 MO â15.06 3.93 1.55 0.42 0.0002 0.58 0.20 Rural multilane divided highways (nonfreeways) PAa Combined â5.48 0.77 0.31 0.10 0.001 0.42 0.06 <.0001 0.84 0.13 MN â4.32 1.20 0.23 0.15 0.123 1.13 0.33 MOa Rural two-lane roads PAa a GLM algorithm did not converge. Table G-3. GLM estimates for SVROR FI crashes based on all sitesâoffset analysis. Table G-4. GLM estimates for SVROR FI crashes based on all sitesâcombined rumble strip offset and recovery area.