Design Guidance for Channelized Right-Turn Lanes (2014)

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.

Chapter 5. Safety Analysis of Channelized Right Turns This chapter presents the results of a safety analysis of channelized right-turn lanes. A safety analysis of channelized right-turn lanes, in comparison to other right-turn treatments, was undertaken because of concerns expressed at the outset of the research that channelized right-turn lanes might experience more crashes (and, in particular, more pedestrian crashes) than other right-turn treatments. To investigate this concern, two key questions related to safety at channelized right-turn lanes were addressed in the safety analysis: • What is the safety performance of channelized right-turn lanes? Specifically, how does the safety performance of intersection approaches with channelized right-turn lanes compare to that of intersection approaches with conventional right-turn lanes or shared through/right-turn lanes? • What safety benefits would be lost if channelized right-turn lanes were not used? An overall comparison was performed of the safety performance of intersection approaches with channelized right-turn lanes to intersection approaches with other right-turn treatments. Specifically, a cross-sectional analysis was conducted to compare the crash experience among: • Intersection approaches with channelized right-turn lanes • Intersection approaches with conventional right-turn lanes • Intersection approaches with no right-turn treatments (shared through/right-turn lanes) The cross-sectional analysis involved comparing mean and median crash frequencies and rates for each of the three intersection approach types and developing negative binomial regression relationships for crash frequencies as a function of traffic volume. The crash data analyses looked separately at motor-vehicle crashes and pedestrian crashes. Database Development 5.1 Seven (7) years (1999 to 2005) of motor-vehicle and pedestrian crash and volume data were obtained for a total of 103 four-leg signalized intersections in Toronto, Ontario, Canada. The Toronto data represent a unique resource because they include both vehicles turning movement volumes and pedestrian crossing volumes by intersection approach, as well as crash data that can be classified by intersection approach and turning movement. These are the same data that were used to develop the pedestrian safety prediction model for Chapter 12 of the Highway Safety Manual (HSM) (33, 34). Initially, a comparison of intersections with and without channelized right-turn lanes was planned, but the intersections with channelized right-turn lanes did not have a consistent pattern of approaches with and without channelized right-turn lanes. That is, some intersections had a 61

channelized right-turn lane on one approach; others had a channelized right-turn lane on two or more approaches. Therefore, the analysis was conducted at the intersection approach rather than at the intersection level. The number of intersection approaches of each right-turn treatment type included in the analysis was: • Shared through/right-turn lane (designated as STR): 217 intersection approaches • Conventional right-turn lane (designated as RTL): 95 intersection approaches • Channelized right-turn lane (designated as CRT): 83 intersection approaches Cross-Sectional Crash Analysis Approach 5.2 The safety analysis focused on right-turn crashes involving motor vehicles and/or pedestrians; separate analyses were conducted for motor-vehicle crashes and pedestrian crashes. The effect of right-turn channelization on motor-vehicle and pedestrian crashes of interest was estimated by means of a cross-sectional analysis in which a single statistical model, including an indicator variable for approach type (STR, RTL, or CRT) and right-turn motor-vehicle and pedestrian volumes, was developed using all three approach types. A negative binomial (NB) regression model was used, with the general form as follows, for motor-vehicle and pedestrian crashes, respectively: (1) or (2) where: NMV = predicted number of motor-vehicle crashes per year per approach NPed = predicted number of pedestrian crashes per year per approach IType(i) = indicator (0,1) variable for approach type i, i = 1, 2, or 3; for STR approaches, the value of the coefficient is b1; for RTL approaches, the value of the coefficient is b2; for CRT approaches, the value of the coefficient is 0 Vol1 = right-turning motor-vehicle (MV) volume (24-hour count); same turning movement for all analysis models Vol2 = MV volume (24-hour count); turning movements represented by Vol2 vary with analysis model Vol3 = MV volume (24-hour count); turning movements represented by Vol3 vary with analysis model VolPed = pedestrian volume crossing intersection approaches of interest (24-hour count) a, bi, c, d, e = regression coefficients ln() represents the natural logarithm function Figure 28 illustrates all of the possible vehicle turning movements at a typical intersection, where the right-turn movement, designated in red (Movement 3), represents the right-turn movement at a particular intersection approach. The other turning movements at the intersection are numbered as shown in the figure in relation to the intersection approach being analyzed. A )]Volln(e+)Volln(d+)Volln(c+Ib+aexp[=N 321)i(TypeiMV )]Volln(d+)Volln(c+Ib+aexp[=N ped1)i(TypeiPed 62

multiple-vehicle right-turn crash involving the intersection approach being analyzed by definition includes a right-turning vehicle (Movement 3). The other involved vehicle could potentially be making any other turning movement at the intersection, but the right-turning vehicle is more likely to conflict with certain turning movements (Movements 1, 2, 3, 7, and 11) than with others. The research team did not want to presuppose which movements would most likely conflict with the right-turning vehicle without evaluating all possible movements first. Therefore, a multi-tiered analysis approach was conducted in which all movements were initially included in the analysis, and then each subsequent analysis became more focused on the movements most likely to conflict with the right-turn movement in question (Movement 3). Each individual analysis is referred to in the following discussion as an “analysis model.” In all, nine different analysis models were investigated—labeled as Analysis Models 1, 2, 3, 4, 5, 6a, 6b, 6c, and 7— three of which are only slight variations of one another. A detailed description of each analysis model is provided later in this section. Figure 28. Intersection Turning Movements Relative to an Intersection Approach (Approach 1) with a Specific Right-Turn Treatment (CRT, RTL, or STR) 63

The following sections present the statistical analysis results, separately for each analysis model. Each section presents the following information: 1. Summary of analysis inputs and criteria: this includes a discussion of the choice of vehicle maneuvers, initial impact types, directional vehicle movements, and aggregated motor-vehicle volumes considered; exceptions are also discussed 2. Mean and median motor-vehicle and pedestrian volumes (based on 24-hour counts), separately for each intersection approach type 3. Motor-vehicle and pedestrian crash statistics—minimum, maximum, and sum of crashes in the seven-year period, separately for total and fatal-and-injury (FI) crashes and for each intersection approach type 4. Regression results, separately for total and FI crashes and for each intersection approach type 5. Statistical comparison of crash rates between CRT and STR or RTL approaches (these results are only provided when the right-turn treatment had an overall statistically significant effect on safety), separately for total and FI crashes 6. Predicted yearly crash counts, separately for total and FI crashes and for each intersection approach type Cross-Sectional Crash Analysis Results 5.3 5.3.1 Analysis Model 1 Results Analysis Model 1 includes consideration of all possible vehicle maneuvers at each intersection and all possible initial impact types involving Movement 3 and any other vehicle, including single-vehicle crashes involving only Movement 3. While certain conflicts appear to be highly unlikely (e.g., a conflict between Movements 3 and 9), this first analysis was conducted to represent a comprehensive comparison of the safety performance of the three intersection approach types (CRT, RTL, STR) considering all possible maneuvers and impact types. Pedestrian crashes are excluded. The layout for this analysis model is presented in Figure 29. The graphic in Figure 29 shows all vehicle turning movements, labeled 1 through 12, at a typical four-way intersection. Each such intersection provided four intersection approaches as separate observations for the statistical analysis, with each approach categorized depending on its right-turn treatment. Approach 1 always refers to the analysis approach considered as the primary approach for a particular observation—it can therefore be the NB, WB, SB, or EB intersection approach. The information in Figure 29 (and similar subsequent figures) is explained from the perspective of Approach 1. The three motor-vehicle volumes used for Analysis Model 1 in Equation 1 are explained in Figure 29, based on the 12 vehicle movements as follows: 64

Figure 29. Summary of Analysis Model 1 Inputs and Criteria 65

• Vol1 always represents the right-turning MV volume (i.e., Vol1 corresponds to vehicle Movement 3, indicated in red) • Vol2 always represents the sum of the remaining MV volumes of the same direction included in the model (i.e., Vol2 corresponds to the sum of vehicle Movements 1 and 2) • Vol3 always represents the sum of the MV volumes of the other directions included in the model (i.e., Vol3 corresponds to the sum of vehicle Movements 4 through 12) Initial impact types and possible vehicle maneuvers are indicated in the subtables in Figure 29. The possible vehicle maneuvers are those used in standard Ontario crash data format. The subtable in Figure 29 (and in subsequent figures with the same format), entitled “vehicle maneuvers for crashes included in the analysis,” defines which crashes were considered in the analysis. Vehicle 1 always represents a right-turning vehicle. Vehicle 2 identifies the vehicle maneuvers considered for the second crash-involved vehicle in multiple-vehicle crashes. The accompanying text in the figure identifies whether single-vehicle crashes involving a right- turning vehicle were included or excluded. Analysis Model 1 uses the largest dataset (with respect to total MV crash count) of all the models considered in the cross-sectional crash analysis. Volume statistics (mean and median) are shown in Table 13, where Vol1, Vol2, and Vol3 are as defined in Figure 29. Total and FI crash statistics are shown in Table 14. Table 13. Mean and Median Motor-Vehicle Volumes by Intersection Type—Analysis Model 1 Intersection approach type Number of approaches Mean MV volumes (24-hr counts)a Median MV volumes (24-hr counts)a Vol1 Vol2 Vol3 Vol1 Vol2 Vol3 STR 217 1,435 10,208 36,209 1,091 9,118 30,891 RTL 95 2,274 14,507 47,841 2,169 13,845 50,256 CRT 83 1,799 12,070 42,770 1,581 12,189 42,211 a Vol1, Vol2, and Vol3 are as defined in Figure 29. Table 14. Seven (7)-Year Total and FI Motor-Vehicle Crash Counts by Intersection Approach—Analysis Model 1 Intersection approach type Number of approaches 7-year total crash counts 7-year FI crash counts Minimum Maximum Sum Minimum Maximum Sum STR 217 0 10 365 0 3 52 RTL 95 0 14 276 0 3 36 CRT 83 0 13 185 0 5 38 The regression results for the total and FI models are summarized in Table 15. These include the regression coefficients [see Equation (1)] and their 90 percent confidence limits. The Type 3 p-value provides the significance level of each parameter in the model; the last column indicates whether the parameter is statistically significant at the 90 percent confidence level. 66

Table 15. Regression Results for Motor-Vehicle Crash Models—Analysis Model 1 Parametera Regression coefficientsb Type 3 p-value Parameter significant at 90% confidence level? Estimate 90% lower confidence limit 90% upper confidence limit Total motor-vehicle crashes—Analysis Model 1 Intercept –13.49 –15.99 –11.01 Intersection approach type STR –0.04 –0.25 0.17 0.67 No RTL 0.07 –0.16 0.29 CRT 0 Vol1 0.38 0.25 0.51 < .0001 YES Vol2 0.15 0.01 0.29 0.09 YES Vol3 0.76 0.50 1.03 < .0001 YES Dispersion 0.35 0.25 0.47 Fatal-and-injury motor-vehicle crashes—Analysis Model 1 Intercept –14.72 –20.21 –9.45 Intersection approach type STR –0.38 –0.79 0.04 0.26 No RTL –0.39 –0.84 0.06 CRT 0 Vol1 0.58 0.29 0.87 0.0007 YES Vol2 0.12 –0.18 0.44 0.54 No Vol3 0.62 0.05 1.20 0.07 YES Dispersion 0.71 0.25 1.36 a Vol1, Vol2, and Vol3 are as defined in Figure 29. b Using the model form in Equation (1). Based on this analysis model, the right-turn treatment has no statistically significant effect on total crashes (p = 0.67) or FI crashes (p = 0.26). The coefficients (third column) for the three types of intersection approaches are to be interpreted as follows: • The CRT treatment is the base (comparison) treatment in all models and its coefficient is therefore always zero on the log-scale or one on the original scale. • If the coefficient for either STR or RTL approaches is positive, then approaches with that type of right-turn treatment experience, on average, higher crash counts than CRT approaches, all volumes held constant between intersection approach types. • If the coefficient for either STR or RTL approaches is negative, then approaches with that type of right-turn treatment experience, on average, lower crash counts than CRT approaches, all volumes held constant between intersection approach types. The coefficient estimates corresponding to the intersection approach types shown in Table 15 therefore provide a means for ranking the right-turn treatments with respect to their predicted crash counts (all volumes held constant between intersection approach types): the smallest coefficient of the three corresponds to the lowest predicted crash count. This, however, does not imply statistical significance; the last column in Table 15 provides that information. Because the right-turn treatment for Analysis Model 1 was not statistically significant at the 90 percent confidence level, the NB regression analysis was not followed-up with a direct comparison of CRT approaches to either STR or RTL approaches. 67

Using Equation (1) and the regression coefficients shown in Table 15, yearly MV crash counts were predicted for all three approach types. To account for the differences in MV volumes among the three intersection approach types (as shown in Table 13), the three models were applied to all three intersection approaches with MV volumes (Vol1, Vol2, and Vol3) set at the observed mean and median volumes for CRT approaches, as shown in Table 13. The predicted average crash frequencies are shown in Table 16 for each approach type using mean and median volumes, respectively. Table 16. Yearly Motor-Vehicle Crash Predictions—Analysis Model 1 Intersection approach type Yearly crash predictions per approach At mean CRT volumes At median CRT volumes Total motor-vehicle crashes—Analysis Model 1 STR 0.316 0.298 RTL 0.351 0.331 CRT 0.329 0.310 Fatal-and-injury motor-vehicle crashes— Analysis Model 1 STR 0.046 0.042 RTL 0.045 0.041 CRT 0.066 0.061 For Analysis Model 1, the right-turn treatment has no statistically significant effect on total crashes or FI crashes. This may be largely due to the fact that all possible maneuvers and impact types, several of which are unlikely to affect right-turn crashes involving Movement 3, have been included in Analysis Model 1. 5.3.2 Analysis Model 2 Results Analysis Model 2 is similar to Analysis Model 1 with the exception that two turning movements (Movement 9 and 10), which appear to be highly unlikely to conflict with Movement 3, have been omitted from the analysis. Analysis Model 2 includes multiple-vehicle crashes that involve one vehicle making Movement 3 and another vehicle making one of the following movements: 1, 2, 3, 4, 5, 6, 7, 8, 11, or 12. Also, single-vehicle crashes and crashes where the maneuver of the second vehicle was coded as “unknown” or “other” or was left blank have been excluded from the analysis. The layout for this analysis model is presented in Figure 30. Volume statistics (mean and median) corresponding to Analysis Model 2 are shown in Table 17, where Vol1, Vol2, and Vol3 are defined in Figure 30. Total and FI crash statistics are shown in Table 18. The selection of possible vehicle maneuvers for Analysis Model 2 resulted in a considerably smaller number of crashes than that for Analysis Model 1 (compare to Table 14). 68

Figure 30. Summary of Analysis Model 2 Inputs and Criteria 69

Table 17. Mean and Median Motor-Vehicle Volumes by Intersection Type—Analysis Model 2 Intersection approach type Number of approaches Mean MV volumes (24-hr counts)a Median MV volumes (24-hr counts)a Vol1 Vol2 Vol3 Vol1 Vol2 Vol3 STR 217 1,435 10,208 32,647 1,091 9,118 29,215 RTL 95 2,274 14,507 43,124 2,169 13,845 45,493 CRT 83 1,799 12,070 39,288 1,581 12,189 39,023 a Vol1, Vol2, and Vol3 are as defined in Figure 30. Table 18. Seven (7)-Year Total and FI Motor-Vehicle Crash Counts by Intersection Approach—Analysis Model 2 Intersection approach type Number of approaches 7-yr total crash counts 7-yr FI crash counts Minimum Maximum Sum Minimum Maximum Sum STR 217 0 8 221 0 3 32 RTL 95 0 10 182 0 2 18 CRT 83 0 9 115 0 2 20 The regression results for the total and FI models are summarized in Table 19. Based on this analysis model, the right-turn treatment has no statistically significant effect on total crashes (p = 0.48) or FI crashes (p = 0.39). Because the right-turn treatment for Analysis Model 2 was not statistically significant at the 90 percent confidence level, the NB regression analysis was not followed-up with a direct comparison of CRT approaches to either STR or RTL approaches. Using Equation 1 and the regression coefficients shown in Table 19, yearly MV crash counts were predicted for all three approach types. To account for the differences in MV volumes among the three intersection approach types (as shown in Table 17), the three models were applied to all three intersection approaches with MV volumes (Vol1, Vol2, and Vol3) set at the observed mean and median volumes for CRT approaches. The predicted average crash frequencies are shown in Table 20 for each approach type using mean and median volumes, respectively. For Analysis Model 2, the right-turn treatment has no statistically significant effect on total crashes or FI crashes. Again, this may be largely because there are still several maneuvers and impact types included in the analysis that are not necessarily related to the right-turn movement in question. 70

Table 19. Regression Results for Motor-Vehicle Crash Models—Analysis Model 2 Parametera Regression coefficientsb Type 3 p-value Parameter significant at 90% confidence level? Estimate 90% Lower confidence limit 90% Upper confidence limit Total motor-vehicle crashes—Analysis Model 2 Intercept –16.05 –18.97 –13.19 Intersection approach type STR –0.03 –0.27 0.20 0.48 No RTL 0.12 –0.12 0.37 CRT 0 Vol1 0.42 0.26 0.57 < .0001 YES Vol2 0.13 –0.03 0.30 0.18 No Vol3 0.96 0.65 1.27 < .0001 YES Dispersion 0.30 0.18 0.45 Fatal-and-injury motor-vehicle crashes—Analysis Model 2 Intercept –19.42 –26.64 –12.61 Intersection approach type STR –0.18 –0.69 0.34 0.39 No RTL –0.48 –1.06 0.10 CRT 0 Vol1 0.56 0.18 0.94 0.01 YES Vol2 0.12 –0.25 0.54 0.61 No Vol3 1.02 0.27 1.79 0.02 YES Dispersion 0.49 –0.11 1.48 a Vol1, Vol2, and Vol3 are as defined in Figure 30. b Using the model form in Equation 1. Table 20. Yearly Motor-Vehicle Crash Predictions—Analysis Model 2 Intersection approach type Yearly crash predictions per approach At mean CRT volumes At median CRT volumes Total motor-vehicle crashes—Analysis Model 2 STR 0.195 0.184 RTL 0.227 0.214 CRT 0.201 0.190 Fatal-and-injury motor-vehicle crashes— Analysis Model 2 STR 0.029 0.027 RTL 0.021 0.020 CRT 0.034 0.032 71

5.3.3 Analysis Model 3 Results Analysis Model 3 focuses on those vehicle maneuvers more likely to conflict with the right- turning vehicle. This analysis includes the right-turn movement (Movement 3) and any other movement that could potentially conflict with the right-turning vehicle, either at the departure end of the right turn (Movements 1 and 2) or as the right-turning vehicle merges into traffic on the cross street (Movements 7 and 11). The layout for this analysis model is presented in Figure 31. Volume statistics (mean and median) corresponding to Analysis Model 3 are shown in Table 21, where Vol1, Vol2, and Vol3 are defined in Figure 31. Total and FI crash statistics are shown in Table 22. The selection of possible vehicle maneuvers for Analysis Model 3 resulted in a yet smaller number of crashes than that for Analysis Model 2 (compare to Table 18). Table 21. Mean and Median Motor-Vehicle Volumes by Intersection Type—Analysis Model 3 Intersection approach type Number of approaches Mean MV volumes (24-hr counts)a Median MV volumes (24-hr counts)a Vol1 Vol2 Vol3 Vol1 Vol2 Vol3 STR 217 1,435 10,208 10,323 1,091 9,118 8,937 RTL 95 2,274 14,507 13,220 2,169 13,845 13,697 CRT 83 1,799 12,070 12,786 1,581 12,189 13,743 a Vol1, Vol2, and Vol3 are as defined in Figure 31. Table 22. Seven (7)-Year Total and FI Motor-Vehicle Crash Counts by Intersection Approach—Analysis Model 3 Intersection approach type Number of approaches 7-yr total crash counts 7-yr FI crash counts Minimum Maximum Sum Minimum Maximum Sum STR 217 0 8 172 0 2 20 RTL 95 0 9 154 0 2 17 CRT 83 0 9 86 0 2 14 The regression results for the total and FI models are summarized in Table 23. Based on this analysis model, the right-turn treatment has no statistically significant effect on total crashes (p = 0.28) or FI crashes (p = 1). Because the right-turn treatment for Analysis Model 3 was not statistically significant at the 90 percent confidence level, the NB regression analysis was not followed-up with a direct comparison of CRT approaches to either STR or RTL approaches. Using Equation (1) and the regression coefficients shown in Table 23, yearly MV crash counts were predicted for all three approach types. To account for the differences in MV volumes among the three intersection approach types (as shown in Table 21), the three models were applied to all three intersection approaches with MV volumes (Vol1, Vol2, and Vol3) set at the observed mean and median volumes for CRT approaches. The predicted average crash frequencies are shown in Table 24 for each approach type using mean and median volumes, respectively. 72

Figure 31. Summary of Analysis Model 3 Inputs and Criteria 73

Table 23. Regression Results for Motor-Vehicle Crash Models—Analysis Model 3 Parametera Regression coefficientsb Type 3 p-value Parameter significant at 90% confidence level? Estimate 90% Lower confidence limit 90% Upper confidence limit Total motor-vehicle crashes—Analysis Model 3 Intercept –15.01 –17.51 –12.59 Intersection approach type STR 0.00 –0.26 0.27 0.28 No RTL 0.21 –0.06 0.48 CRT 0 Vol1 0.58 0.41 0.76 < .0001 YES Vol2 0.38 0.20 0.57 0.0005 YES Vol3 0.56 0.37 0.75 < .0001 YES Dispersion 0.31 0.17 0.48 Fatal-and-injury motor-vehicle crashes—Analysis Model 3 Intercept –5.16 –5.16 –5.16 Intersection approach type STR –0.09 –0.09 –0.09 1 No RTL –0.08 –0.08 –0.08 CRT 0 Vol1 0.19 0.19 0.19 1 No Vol2 0.02 0.02 0.02 0.02 YES Vol3 0.08 0.08 0.08 1 No Dispersion –0.50 –0.50 –0.48 a Vol1, Vol2, and Vol3 are as defined in Figure 31. b Using the model form in Equation (1). Table 24. Yearly Motor-Vehicle Crash Predictions—Analysis Model 3 Intersection approach type Yearly crash predictions per approach At mean CRT volumes At median CRT volumes Total motor-vehicle crashes—Analysis Model 3 STR 0.162 0.157 RTL 0.200 0.194 CRT 0.162 0.157 Fatal-and-injury motor-vehicle crashes—Analysis Model 3 STR 0.054 0.053 RTL 0.055 0.054 CRT 0.059 0.058 For Analysis Model 3, the right-turn treatment has no statistically significant effect on yearly predictions of total or FI crashes. However, while not statistically significant, the yearly 74

total crash predictions tend to show that CRT and STR approaches have similar safety performance (0.162 crashes per year per approach based on mean CRT volumes or 0.157 crashes per year per approach based on median CRT volumes). 5.3.4 Analysis Model 4 Results Analysis Model 4 focuses on those vehicle maneuvers more likely to conflict with the right- turning vehicle (Movement 3) as it merges into traffic on the cross street (Movements 7 and 11). Thus, Analysis Model 4 includes multiple-vehicle crashes that involve one vehicle making Movement 3 and another vehicle making Movement 7 or 11. The objective of this analysis was to assess whether CRT approaches experience more crashes than RTL or STR approaches as the right-turning vehicle merges with the cross street, particularly since a CRT approach positions the driver at more of a skew angle at the point of the merge. The layout for this analysis model is presented in Figure 32. Volume statistics (mean and median) corresponding to Analysis Model 4 are shown in Tables 25, where Vol1 and Vol3 are defined in Figure 32. Total and FI crash statistics are shown in Table 26. The selection of possible vehicle maneuvers for Analysis Model 4 resulted in a yet smaller number of crashes than that for Analysis Model 3 (compare to Table 22). Note that the 7-year FI crash counts are very low for this scenario; indeed, at most one FI crash occurred at any single approach over the 7-year study period. Table 25. Mean and Median Motor-Vehicle Volumes by Intersection Type—Analysis Model 4 Intersection approach type Number of approaches Mean MV volumes (24-hr counts)a Median MV volumes (24-hr counts)a Vol1 Vol2 Vol3 Vol1 Vol2 Vol3 STR 217 1,435 10,323 1,091 8,937 RTL 95 2,274 13,220 2,169 13,697 CRT 83 1,799 12,786 1,581 13,743 a Vol1 and Vol3 are as defined in Figure 32. Table 26. Seven (7)-Year Total and FI Motor-Vehicle Crash Counts by Intersection Approach—Analysis Model 4 Intersection approach type Number of approaches 7-yr total crash counts 7-yr FI crash counts Minimum Maximum Sum Minimum Maximum Sum STR 217 0 3 56 0 1 7 RTL 95 0 5 75 0 1 7 CRT 83 0 5 41 0 1 6 75

Figure 32. Summary of Analysis Model 4 Inputs and Criteria 76

The regression results for the total and FI models are summarized in Table 27. The algorithm to estimate the regression coefficients for the FI crash model did not converge, therefore no model for FI crashes for Analysis Model 4 is available. Based on this analysis model, the right-turn treatment has a statistically significant effect on total crashes (p = 0.01). Table 27. Regression Results for Motor-Vehicle Crash Models—Analysis Model 4 Parametera Regression coefficientsb Type 3 p-value Parameter significant at 90% confidence level? Estimate 90% Lower confidence limit 90% Upper confidence limit Total motor-vehicle crashes—Analysis Model 4 Intercept -14.47 -17.69 -11.49 Intersection approach type STR -0.34 -0.71 0.03 0.01 YES RTL 0.25 -0.10 0.62 CRT 0 0.00 0.00 Vol1 0.83 0.58 1.08 < .0001 YES Vol3 0.60 0.31 0.91 0.0004 YES Dispersion 0.30 0.06 0.63 Fatal-and-injury motor-vehicle crashes—Analysis Model 4 No regression model available a Vol1 and Vol3 are as defined in Figure 32. b Using the model form in Equation 1. For Analysis Model 4, the right-turn treatment was statistically significant at the 90 percent confidence level. In this analysis, CRT approaches have a lower estimate of total crashes than RTL approaches, but a higher estimate than STR approaches. Since the right-turn treatment for Analysis Model 4 was statistically significant at the 90 percent confidence level, the NB regression analysis was followed-up with a direct comparison of CRT approaches to either STR or RTL approaches; these comparisons are summarized in Table 28 for total crashes. Although the overall effect of the right-turn treatment was statistically significant, neither one-to-one comparison was statistically significant at the 90 percent confidence level (p = 0.13 for CRT vs. STR; p = 0.24 for CRT vs. RTL). The second column in Table 28 indicates whether average total crash rate for CRT approaches is lower than that for either of the other two approach types. Each answer is simply based on the comparison of the corresponding parameter estimates shown in Table 27—a lower estimate corresponds to a lower crash rate (this is just a mathematical assessment, not a statistical one). 77

Table 28. Contrast Results for Motor-Vehicle Crash Models—Analysis Model 4 Comparison Crash rate lower at CRT? Chi2 p-value Contrast significant at 90% confidence level? Total motor-vehicle crashes—Analysis Model 4 CRT vs. STR No 0.13 No CRT vs. RTL Yes 0.24 No Fatal-and-injury motor-vehicle crashes—Analysis Model 4 No regression model available Using Equation (1) and the regression coefficients shown in Table 27, yearly MV crash counts were predicted for all three approach types. To account for the differences in MV volumes among the three intersection approach types (as shown in Table 27), the three models were applied to all three intersection approaches with MV volumes (Vol1 and Vol3) set at the observed mean and median volumes for CRT approaches. The predicted average crash frequencies are shown in Table 29 for each approach type using mean and median volumes, respectively. Table 29. Yearly Motor-Vehicle Crash Predictions—Analysis Model 4 Intersection approach type Yearly crash predictions per approach At mean CRT volumes At median CRT volumes Total motor-vehicle crashes—Analysis Model 4 STR 0.051 0.048 RTL 0.093 0.087 CRT 0.072 0.068 Fatal-and-injury motor-vehicle crashes— Analysis Model 4 No regression model available For Analysis Model 4, the yearly total crash predictions for CRT approaches are lower than the yearly total crash predictions for RTL approaches but higher than for STR approaches. 5.3.5 Analysis Model 5 Results Analysis Model 5 focuses on those vehicle maneuvers that could potentially conflict with the right-turning vehicle (Movement 3) as it turns from the intersection approach (i.e., the departure end). Thus, Analysis Model 5 includes sideswipe and rear-end crashes that involve one vehicle making Movement 3 and another vehicle making Movement 1, 2, or 3. The objective of this analysis was to assess whether CRT approaches experience more crashes at the departure end than RTL or STR approaches. The layout for this analysis model is presented in Figure 33. 78

Figure 33. Summary of Analysis Model 5 Inputs and Criteria 79

Volume statistics (mean and median) corresponding to Analysis Model 5 are shown in Tables 30, where Vol1 and Vol2 are defined in Figure 33. Total and FI crash statistics are shown in Table 31. The selection of possible vehicle maneuvers for Analysis Model 5 resulted in a small number of crashes. NOTE: The seven-year FI crash counts are again very low for this scenario; indeed, at most two FI crashes occurred at any single STR or RTL approach over the seven-year study period; at most one FI crashes occurred at any single CRT approach over the same period. Table 30. Mean and Median Motor-Vehicle Volumes by Intersection Type—Analysis Model 5 Intersection approach type Number of approaches Mean MV volumes (24-hr counts)a Median MV volumes (24-hr counts)a Vol1 Vol2 Vol3 Vol1 Vol2 Vol3 STR 217 1,435 10,208 1,091 9,118 RTL 95 2,274 14,507 2,169 13,845 CRT 83 1,799 12,070 1,581 12,189 a Vol1 and Vol2 are as defined in Figure 33. Table 31. Seven (7)-Year Total and FI Motor-Vehicle Crash Counts by Intersection Approach—Analysis Model 5 Intersection approach type Number of approaches 7-yr total crash counts 7-yr FI crash counts Minimum Maximum Sum Minimum Maximum Sum STR 217 0 7 74 0 2 12 RTL 95 0 9 50 0 2 12 CRT 83 0 6 49 0 1 8 The regression results for the total and FI models are summarized in Table 32. Based on this analysis model, the right-turn treatment has no statistically significant effect on total crashes (p = 0.37) or FI crashes (p = 1). Because the right-turn treatment for Analysis Model 5 was not statistically significant at the 90 percent confidence level, the NB regression analysis was not followed-up with a direct comparison of CRT approaches to either STR or RTL approaches. Using Equation (1) and the regression coefficients shown in Table 32, yearly MV crash counts were predicted for all three approach types. To account for the differences in MV volumes among the three intersection approach types (as shown in Table 30), the three models were applied to all three intersection approaches with MV volumes (Vol1 and Vol2) set at the observed mean and median volumes for CRT approaches. The predicted average crash frequencies are shown in Table 33 for each approach type using mean and median volumes, respectively. For Analysis Model 5, the right-turn treatment has no statistically significant effect on yearly total or FI crash predictions. 80

Table 32. Regression Results for Motor-Vehicle Crash Models—Analysis Model 5 Parametera Regression coefficientsb Type 3 p-value Parameter significant at 90% confidence level? Estimate 90% Lower confidence limit 90% Upper confidence limit Total motor-vehicle crashes—Analysis Model 5 Intercept -13.66 -16.86 -10.65 Intersection approach type STR -0.32 -0.73 0.08 0.37 No RTL -0.32 -0.76 0.13 CRT 0 Vol1 1.04 0.76 1.33 < .0001 YES Vol2 0.36 0.07 0.67 0.04 YES Dispersion 1.04 0.62 1.61 Fatal-and-injury motor-vehicle crashes—Analysis Model 5 Intercept -4.36 -5.26 -4.35 Intersection approach type STR -0.15 -0.15 0.27 1 No RTL -0.06 -0.06 -0.05 CRT 0 Vol1 0.17 0.17 0.17 < .0001 YES Vol2 -0.02 -0.02 -0.02 1 No Dispersion -0.50 -0.50 -0.46 a Vol1 and Vol2 are as defined in Figure 33. b Using the model form in Equation (1). Table 33. Yearly Motor-Vehicle Crash Predictions—Analysis Model 5 Intersection approach type Yearly crash predictions per approach At mean CRT volumes At median CRT volumes Total motor-vehicle crashes—Analysis Model 5 STR 0.060 0.053 RTL 0.061 0.053 CRT 0.084 0.073 Fatal-and-injury motor-vehicle crashes—Analysis Model 5 STR 0.035 0.034 RTL 0.039 0.038 CRT 0.041 0.040 81

5.3.6 Analysis Models 6 Results Analysis Models 6a, 6b, and 6c are similar to Analysis Model 5, except that the left-turn movement (Movement 1) is excluded from the analysis. The analysis is broken into three sub- analyses in order to hone in on certain potential problems (rear-end crashes vs. sideswipe crashes, crashes between a right-turning vehicle and a through vehicle vs. crashes between two right-turning vehicles, etc). The three sub-analyses are: • Analysis Model 6a focuses on rear-end and sideswipe crashes that involve one vehicle making Movement 3 and another vehicle making either Movement 2 or 3. • Analysis Model 6b focuses exclusively on rear-end crashes that involve one vehicle making Movement 3 and another vehicle making either Movement 2 or 3. • Analysis Model 6c focuses exclusively on rear-end crashes that involve two right- turning vehicles (Movement 3). The layout for Analysis Models 6a, 6b, and 6c is presented in Figure 34. Volume statistics (mean and median) corresponding to Analysis Model 6 are shown in Tables 34, where Vol1 and Vol2 are defined in Figure 34. Total and FI crash statistics are shown in Table 35. The selection of possible vehicle maneuvers for Analysis Model 6 resulted in a small number of crashes. NOTE: The 7 year FI crash counts are again very low for this scenario; indeed, at most two FI crashes occurred at any single STR or RTL approach over the 7 year study period; at most one FI crashes occurred at any single CRT approach over the same period. Table 34. Mean and Median Motor-Vehicle Volumes by Intersection Type—Analysis Model 6 Analysis model Intersection approach type Number of approaches Mean MV volumes (24-hr counts)a Median MV volumes (24-hr counts)a Vol1 Vol2 Vol3 Vol1 Vol2 Vol3 6a or 6b STR 217 1,435 8,841 1,091 7,965 RTL 95 2,274 12,215 2,169 11,564 CRT 83 1,799 9,407 1,581 8,691 6c STR 217 1,435 1,091 RTL 95 2,274 2,169 CRT 83 1,799 1,581 a Vol1 and Vol2 are as defined in Figure 34. 82

Figure 34. Summary of Analysis Model 6 Inputs and Criteria 83

Table 35. Seven (7)-Year Total and FI Motor-Vehicle Crash Counts by Intersection Approach—Analysis Model 6 Analysis model Intersection approach type Number of approaches 7-yr total crash counts 7-yr FI crash counts Minimum Maximum Sum Minimum Maximum Sum 6a (rear-end and sideswipe crashes) STR 217 0 7 71 0 2 12 RTL 95 0 9 50 0 2 12 CRT 83 0 6 48 0 1 8 6b or 6c (rear-end crashes only) STR 217 0 7 51 0 2 10 RTL 95 0 7 37 0 2 12 CRT 83 0 6 42 0 1 8 The regression results for the total and FI models are summarized in Table 36. Based on this analysis model, the right-turn treatment has no statistically significant effect on total crashes (p = 0.34 for Model 6a; p = 0.15 for Model 6b; p = 0.30 for Model 6c). However, the right-turn treatment has a statistically significant effect on FI crashes for all analysis models (p < = 0.001). Table 36. Regression Results for Motor-Vehicle Crash Models—Analysis Model 6 Parametera Regression coefficientsb Type 3 p-value Parameter significant at 90% confidence level? Estimate 90% Lower confidence limit 90% Upper confidence limit Total motor-vehicle crashes—Analysis Model 6a Intercept -13.42 -16.36 -10.69 Intersection approach type STR -0.36 -0.77 0.06 0.34 No RTL -0.32 -0.77 0.14 CRT 0 Vol1 1.13 0.84 1.42 < .0001 YES Vol2 0.27 0.05 0.51 0.05 YES Dispersion 1.08 0.65 1.67 Fatal-and-injury motor-vehicle crashes—Analysis Model 6a Intercept -17.26 -23.21 -12.11 Intersection approach type STR -0.25 -1.00 0.53 < .0001 YES RTL -0.09 -0.84 0.70 CRT 0 Vol1 1.47 0.94 2.04 < .0001 YES Vol2 0.20 -0.17 0.68 0.40 No Dispersion -0.13 -0.50 1.46 84

Table 36. Regression Results for Motor-Vehicle Crash Models—Analysis Model 6 (Continued) Parametera Regression coefficientsb Type 3 p-value Parameter significant at 90% confidence level? Estimate 90% Lower confidence limit 90% Upper confidence limit Total motor-vehicle crashes—Analysis Model 6b Intercept -16.28 -19.98 -12.89 Intersection approach type STR -0.53 -1.00 -0.06 0.15 No RTL -0.49 -1.00 0.03 CRT 0 Vol1 1.35 1.02 1.71 < .0001 YES Vol2 0.38 0.11 0.67 0.02 YES Dispersion 1.30 0.75 2.09 Fatal-and-injury motor-vehicle crashes—Analysis Model 6b Intercept -17.60 -23.93 -12.16 Intersection approach type STR -0.42 -1.21 0.40 < .0001 YES RTL -0.09 -0.84 0.71 CRT 0 Vol1 1.53 0.98 2.14 < .0001 YES Vol2 0.18 -0.20 0.68 0.46 No Dispersion -0.05 -0.50 1.68 Total motor-vehicle crashes—Analysis Model 6c Intercept -13.40 -16.20 -10.81 Intersection approach type STR –0.44 -0.90 0.03 0.30 No RTL -0.35 -0.86 0.17 CRT 0 Vol1 1.41 1.08 1.77 < .0001 YES Dispersion 1.36 0.78 2.17 Fatal-and-injury motor-vehicle crashes—Level 6c Intercept -16.12 -20.94 -11.79 Intersection approach type STR -0.38 –1.17 0.43 < .0001 YES RTL -0.03 -0.76 0.76 CRT 0 Vol1 1.55 1.01 2.16 < .0001 YES Dispersion -0.08 -0.50 1.59 a Vol1 and Vol2 are as defined in Figure 34. b Using the model form in Equation 1. Because the right-turn treatment for Analysis Models 6a, 6b, and 6c for FI crashes was statistically significant at the 90 percent confidence level, the NB regression analysis was followed-up with a direct comparison of CRT approaches to either STR or RTL approaches; these comparisons are summarized in Table 37 for FI crashes and show that: 85

• Although the overall effect of the right-turn treatment was statistically significant (for FI crashes) in Analysis Model 6a, neither one-to-one comparison was statistically significant at the 90 percent confidence level (p = 0.59 for CRT vs. STR; p = 0.84 for CRT vs. RTL). • Similarly for Analysis Model 6b, neither one-to-one comparison was statistically significant at the 90 percent confidence level (p = 0.39 for CRT vs. STR; p = 0.85 for CRT vs. RTL). • Similarly for Analysis Model 6c, neither one-to-one comparison was statistically significant at the 90 percent confidence level (p = 0.43 for CRT vs. STR; p = 0.95 for CRT vs. RTL). Table 37. Contrast Results for Motor-Vehicle Crash Models—Analysis Models 6 Comparison Crash rate lower CRT? Chi2 p-value Contrast significant at 90% confidence level? Total motor-vehicle crashes—Analysis Model 6a Overall effect of intersection approach type is not statistically significant Fatal-and-injury motor-vehicle crashes—Analysis Model 6a CRT vs. STR No 0.59 No CRT vs. RTL No 0.84 No Total motor-vehicle crashes—Analysis Model 6b Overall effect of intersection approach type is not statistically significant Fatal-and-injury motor-vehicle crashes—Analysis Model 6b CRT vs. STR No 0.39 No CRT vs. RTL No 0.85 No Total motor-vehicle crashes—Analysis Model 6c Overall effect of intersection approach type is not statistically significant Fatal-and-injury motor-vehicle crashes—Analysis Model 6c CRT vs. STR No 0.43 No CRT vs. RTL No 0.95 No Using Equation (1) and the regression coefficients shown in Table 36, yearly MV crash counts were predicted for all three approach types for each analysis approach. To account for the differences in MV volumes among the three intersection approach types (as shown in Table 34), the three models were applied to all three intersection approaches with MV volumes (Vol1 and Vol2 or Vol1 only) set at the observed mean and median volumes for CRT approaches. The predicted average crash frequencies are shown in Table 38 for each approach type using mean and median volumes, respectively. 86

Table 38. Yearly Motor-Vehicle Crash Predictions—Analysis Models 6 Intersection approach type Yearly crash predictions per approach At mean CRT volumes At median CRT volumes Total motor-vehicle crashes—Analysis Model 6a STR 0.057 0.048 RTL 0.059 0.050 CRT 0.081 0.069 Fatal-and-injury motor-vehicle crashes—Analysis Model 6a STR 0.010 0.008 RTL 0.011 0.009 CRT 0.012 0.010 Total motor-vehicle crashes—Analysis Model 6b STR 0.039 0.032 RTL 0.041 0.033 CRT 0.066 0.054 Fatal-and-injury motor-vehicle crashes—Analysis Model 6b STR 0.008 0.006 RTL 0.011 0.009 CRT 0.012 0.010 Total motor-vehicle crashes—Analysis Model 6c STR 0.039 0.033 RTL 0.043 0.036 CRT 0.061 0.051 Fatal-and-injury motor-vehicle crashes—Analysis Model 6c STR 0.008 0.006 RTL 0.011 0.009 CRT 0.011 0.009 While the overall analysis of treatment type suggests that CRT approaches may experience more rear-end and sideswipe crashes between the right-turning vehicle and either (1) a through vehicle or (2) another right-turning vehicle than RTL or STR approaches (Analysis Model 6a), neither of the one-to-one comparisons was statistically significant. This is true for Analysis Models 6b and 6c as well. 5.3.7 Analysis Models 7 Results Analysis Model 7 focuses on crashes between right-turning vehicles (Movement 3) and pedestrians, and includes the two pedestrian maneuvers/crossings that would potentially conflict with a right-turning vehicle. Figure 35 illustrates the right-turn movement and the two conflicting 87

Figure 35. Summary of Analysis Model 7 Inputs and Criteria 88

pedestrian movements of interest. Movement 13 consists of the pedestrian crossing the approach that the right-turning vehicle turns from (in this case, Approach 1); Movement 14 consists of the pedestrian crossing the approach that the right-turning vehicle turns into (in this case, Approach 2). Volume statistics (mean and median) corresponding to Analysis Model 7 are shown in Tables 39, where Vol1 and Vol3 are defined in Figure 35. Total and FI pedestrian crash statistics are shown in Table 40. NOTE: The difference between total and FI pedestrian crashes is minor, as expected, since pedestrian crashes are almost all FI crashes. Table 39. Mean and Median Motor-Vehicle and Pedestrian Volumes by Intersection Type—Analysis Model 7 Intersection approach type Number of approaches Mean volumes (24-hr counts)a Median volumes (24-hr counts)a Vol1 (motor vehicle) Vol2 Vol3 (pedestrian) Vol1 (motor vehicle) Vol2 Vol3 (pedestrian) STR 217 1,435 2,077 1,091 1,133 RTL 95 2,274 1,120 2,169 775 CRT 83 1,799 510 1,581 277 a Vol1 and Vol3 are as defined in Figure 35. Table 40. Seven (7)-Year Total and FI Pedestrian Counts by Intersection Approach—Analysis Model 7 Intersection approach type Number of approaches 7-yr total crash counts 7-yr FI crash counts Minimum Maximum Sum Minimum Maximum Sum STR 217 0 4 51 0 4 49 RTL 95 0 3 45 0 3 44 CRT 83 0 2 12 0 2 11 The regression results for the total and FI models are summarized in Table 41. Based on this analysis model, the right-turn treatment has a statistically significant effect on total crashes (p = 0.05) and on FI crashes (p = 0.04). The right-turn treatment for Analysis Model 7 was statistically significant at the 90 percent confidence level for both total and FI pedestrian crashes (p = 0.05 and 0.04, respectively). Therefore, the NB regression analysis was followed-up with a direct comparison of CRT approaches to either STR or RTL approaches; these comparisons are summarized in Table 42 for total and FI pedestrian crashes and show that although the overall effect of the right-turn treatment was statistically significant: • The comparison between CRT and STR approaches was not statistically significant at the 90 percent confidence level (p = 0.95 for total crashes; p = 0.86 for FI crashes). • The comparison between CRT and RTL approaches was statistically significant at the 90 percent confidence level (p = 0.10 for total crashes; p = 0.07 for FI crashes). 89

Table 41. Regression Results for Pedestrian Crash Models—Analysis Model 7 Parametera Regression coefficientsb Type 3 p-value Parameter significant at 90% confidence level? Estimate 90% Lower confidence limit 90% Upper confidence limit Total pedestrian crashes—Analysis Model 7 Intercept –12.13 –14.96 –9.49 Intersection approach type STR 0.02 –0.55 0.64 0.05 YES RTL 0.57 0.02 1.18 CRT 0 Vol1 0.71 0.40 1.02 0.0001 YES Vol3 0.50 0.33 0.67 < .0001 YES Dispersion 0.31 –0.06 0.86 Fatal-and-injury pedestrian crashes—Analysis Model 7 Intercept –11.94 –14.83 –9.25 Intersection approach type STR 0.07 –0.53 0.71 0.04 YES RTL 0.65 0.07 1.28 CRT 0 Vol1 0.68 0.36 0.99 0.0003 YES Vol3 0.50 0.33 0.68 < .0001 YES Dispersion 0.38 –0.03 0.98 a Vol1 and Vol3 are as defined in Figure 35. b Using the model form in Equation (2). Table 42. Contrast Results for Pedestrian Models—Analysis Model 7 Comparison Crash rate lower at channelized RTL? Chi2 p-value Contrast significant at 90% confidence level? Total pedestrian crashes—Analysis Model 7 CRT vs. STR Yes 0.95 No CRT vs. RTL Yes 0.10 YES Fatal-and-injury pedestrian crashes—Analysis Model 7 CRT vs. STR Yes 0.86 No CRT vs. RTL Yes 0.07 YES Using Equation (2) and the regression coefficients shown in Table 41, yearly pedestrian crash counts were predicted for all three approach types for each analysis approach. To account for the substantial differences in MV and pedestrian volumes, respectively, among the three intersection approach types (as shown in Table 39), the three models were applied to all three intersection approaches with MV and pedestrian volumes (Vol1 and Vol3 ) set at the observed mean and median volumes for CRT approaches. The predicted average crash frequencies are shown in Table 43 for each approach type using mean and median volumes, respectively. 90

Table 43. Yearly Pedestrian Crash Predictions—Analysis Model 7 Intersection approach type Yearly crash predictions per approach At mean CRT volumes At median CRT volumes Total pedestrian crashes—Analysis Model 7 STR 0.025 0.017 RTL 0.043 0.029 CRT 0.024 0.016 Fatal-and-injury pedestrian crashes— Analysis Model 7 STR 0.024 0.016 RTL 0.043 0.029 CRT 0.023 0.015 The yearly pedestrian crash predictions for CRT and STR approaches are similar (0.024 total crashes per CRT approach vs. 0.025 total crashes per STR approach; 0.023 FI crashes per CRT approach vs. 0.024 FI crashes per STR approach). These comparisons also hold when the predictions are made based on median CRT volumes. The yearly pedestrian crash prediction for RTL approaches is approximately 70 to 80 percent higher than for the other two types of right-turn treatments. This suggests that CRT approaches do not experience a particular safety problem with respect to pedestrians and, in fact, may have fewer pedestrian safety problems than RTL approaches. Summary of Safety Analysis 5.4 A cross-sectional safety analysis was conducted to determine the safety performance of channelized right-turn lanes. An overall comparison was performed of the safety performance of the following types of intersection approaches: • Intersection approaches with channelized right-turn lanes (CRT approaches) • Intersection approaches with conventional right-turn lanes (RTL approaches) • Intersection approaches with shared through/right-turn lanes (STR approaches) A multi-tiered analysis approach was conducted in which all movements were initially included in the analysis, and then each subsequent analysis became more focused on those movements more likely to conflict with the right-turn vehicle in question. In all, nine different analysis models were investigated; four of them are of interest and are highlighted next. • Analysis Model 3 focused on those vehicle maneuvers more likely to conflict with the right-turning vehicle, either at the departure end of the right turn or as the right-turning vehicle merges into traffic on the cross street. While the right-turn treatment had no statistically significant effect on total or FI MV crashes, the yearly total crash predictions showed that CRT and STR approaches have similar safety performance 91

(0.162 total crashes per year, on average, for both approach types), while the crash frequencies for RTL approaches were slightly higher (0.2 total crashes per year, on average, per approach). However, this observation is not statistically significant. • Analysis Model 4 assessed whether CRT approaches experience more crashes than RTL or STR approaches as the right-turning vehicle merges with the cross street, particularly since the CRT approach positions the driver at a higher skew angle at the point of the merge. Based on this analysis model, the right-turn treatment was statistically significant at the 90 percent confidence level; CRT approaches had a lower estimate of total crash frequency (0.072 MVcrashes per year per approach) than RTL approaches (0.093 MV crashes per year per approach) but a higher estimate than STR approaches (0.051 MV crashes per year per approach). While the overall effect of right-turn treatments was statistically significant, the comparisons between the individual right-turn treatment types were not. This suggests that the three right-turn treatments may differ in safety performance as the right-turning vehicle merges with the cross street, but this is not conclusively established. • Analysis Model 6 focused exclusively on rear-end and sideswipe crashes between the right-turning vehicle (Movement 3) and either another right-turning vehicle or a through vehicle on the same intersection approach. Based on this analysis model, the right-turn treatment had no statistically significant effect on total crashes. For FI crashes, CRT approaches had a slightly higher estimate than RTL or STR approaches, but the differences were very small—for example, 0.012 vs. 0.011 vs. 0.010 FI MV crashes per year per CRT, RTL, and STR approach, respectively. Furthermore, when the analysis was followed-up with direct one-to-one comparison of CRT approaches to either STR or RTL approaches, neither of the comparisons was statistically significant at the 90 percent confidence level. • Analysis Model 7 focused on crashes between right-turning vehicles and pedestrians, and included the two pedestrian maneuvers/crossings that would potentially conflict with a right-turning vehicle. Based on this analysis model, the right-turn treatment had a statistically significant effect on total and FI crashes. The yearly pedestrian crash predictions for CRT and STR approaches were similar (0.024 total crashes per CRT approach vs. 0.025 total crashes per STR approach; 0.023 FI crashes per CRT approach vs. 0.024 FI crashes per STR approach). The yearly pedestrian crash prediction for RTL approaches was approximately 70 to 80 percent higher than for the other two types of right-turn treatments. This suggests that CRT and STR have similar pedestrian safety performance, while RTL approaches have more pedestrian crashes, on average, than either CRT or STR approaches. 92