U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
202-366-4000
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
 |
| This report is an archived publication and may contain dated technical, contact, and link information |
|
Publication Number: FHWA-HRT-10-068
Date:November 2010 |
Crosswalk Marking Field Visibility Study
CHAPTER 6. DATA ANALYSIS
OVERVIEWCROSSWALK MARKINGS
OVERVIEWDuring data collection in the vehicle, the rear-seat experimenter recorded when crosswalks were identified by the participant. During a second lap, the participant's appearance rating for each crosswalk was recorded. Following the driving portion of the study, the participant rank-ordered photographs of crosswalks to reflect his or her preferences.
CROSSWALK MARKINGSThe prime objectives of this study were to determine the relative visibility of three crosswalk patterns through the use of detection distance and to identify the variables that affect this distance. The differences in detection distances were evaluated with consideration of variables grouped into the following classes:
Table 10 lists the variables that were originally considered. All variables were considered in the nighttime evaluations. The variable for retroreflectivity of pavement markings was not included in the daytime analysis. For some variables, using groups (categories) as well as continuous values was explored. For example, the eye height values were examined as a continuous value in some models and grouped into four ranges in others. Including a variable as a categorical variable after grouping, rather than as a continuous variable, can sometimes be advantageous in that the nonlinear relationship between the variable and detection distance can be detected (when the actual relationship is not linear). Table 10. Original list of potential variables for analysis of crosswalk detection distance.
The presence of a pedestrian or bicycle in the participant's view was found to be not significant in several preliminary models. The researchers expect this is more of a reflection of the low number of events with the pedestrian/bicycle variable rather than the presence of pedestrians or bicyclists not having an influence. Less than 4 percent of all events for both study sites and existing sites included a pedestrian or bicycle. For study sites, less than 1 percent of the detection distances occurred when a pedestrian was in the participant's field of view. Therefore, this variable was removed from the study site evaluations. Original investigations tried to uniquely use the variables listed in table 10 . The limited range in some of the values (maximum crossing width difference of only 10 ft between sites) and the overlap between values (e.g., the 45 mi/h sites always had no sidewalks) created concerns about the modeling process. Therefore, a new variable–street group–was developed. The street group variable accounted for the following roadway characteristics:
Table 11 shows the characteristics for each street group. The vehicle type overlapped with the driver eye height groups, so those variables were combined in the models that included driver eye heights in groups. Table 12 lists the revised list of potential variables. Table 11. Street group characteristics.
Table 12. Revised list of potential variables for analysis of crosswalk detection distance.
While the preference was to identify if retroreflectivity affects detection distance, the existing sites available for this study did not permit such an evaluation. Most existing sites had retroreflectivity readings less than 38 mcd/m2/lx. Those existing sites with higher retroreflectivity values had continental markings located at intersections or had transverse markings located at midblock crossings, which created confounding issues. Both continuous readings and groups of similar readings were explored. However, the retroreflectivity variable had to be dropped in the evaluations of the existing sites. An evaluation of retroreflectivity was not considered for the study sites because similar marking material was used at all the study sites to ensure that retroreflectivity would not be a factor. The focus of this study was on different marking patterns. The evaluation of retroreflectivity would have been a side benefit that, unfortunately, was not possible. Figure 42 shows the average detection distance for each crosswalk subdivided by daytime and nighttime conditions, and figure 43 shows the data grouped by location type. As expected, the average adjusted detection distance was always longer during the day than at night. For several locations, such as transverse markings at an intersection, the detection distance was similar in both the daytime and the nighttime, with some being different by only a few feet. Initial statistical evaluations support the observation that effects of several variables on detection distances are different in the daytime as compared to the nighttime. Therefore, the evaluations were subdivided into daytime and nighttime conditions. Initial evaluations also revealed the need to evaluate the study sites separately from the markings at existing locations. Of particular interest is the detection distance to the markings installed for this study. Figure 44 shows the average detection distance for the three test markings: transverse, continental, and bar pairs. Again, there was a difference between the average adjusted detection distance for daytime and nighttime conditions, with a minimal difference for the transverse marking sites. The analyses of the adjusted detection distance data were conducted using Analysis of Covariance (ANACOVA) mixed models treating the variables in table 12 as fixed factors/covariates and drivers and crosswalks as random factors. Models were estimated by the restricted maximum likelihood method implemented in the JMP® statistical package (a SAS® product). Several models were explored to determine the best models to describe the variables that influence detection distance. In addition to the main effects, models with two–way interactions were examined. The analysis began with a model that included all main effects variables and logical two–way interaction variables called the "extended" model. Several of the interactions were not significant and were dropped from the models when the p–value was less than 0.05. This revised model was called the "reduced" model.
Figure 42. Graph. Average adjusted detection distance at each crosswalk in order of appearance.
Figure 43. Graph. Average adjusted detection distance at each crosswalk grouped by location type.
Figure 44. Graph. Average adjusted detection distance for crosswalk markings at study sites. Daytime Detection–Study Sites Evaluations began with examining which variables affected daytime detection distance at the nine sites where the markings were installed for this study. Data for both approaches were included for a total of 18 crosswalks. Table 13 lists the results for the model that includes all potential main effect variables along with two–way interactions (the analysis output for random effects is suppressed and not shown in the table for space). The variables that were statistically significant at the 0.05 level include the following:
While there were several significant variables, the model also contained several variables that were not significant. The next step in model development was to remove insignificant two–way interaction variables until an acceptable model was reached. As variables were eliminated or other variables tried, the significant variables changed. The reduced model that the researchers think provides good information on which two–way interaction variables influence the adjusted daytime detection distance at the study sites is shown in table 14. Table 13. ANACOVA findings for daytime adjusted detection distance for extended model (includes potential variables and two-way interactions) for study sites.
Table 14. ANACOVA findings for daytime adjusted detection distance for reduced model for study sites.
The two–way interaction variables that influenced daytime detection distance are as follows:
None of the main effect driver variables, which included driver eye height, gender, and age group, were significant at the 0.05 level. While gender was not surprising, a difference due to age group was expected. The selection of participants was designed to ensure adequate representation of the 55 years old and older group by having half of the participants in that age group. Figure 45 shows the average adjusted detection distance subdivided by age group and light. The graph supports the finding that age group was not a significant variable for this particular study. In each marking type and light combination, the average adjusted detection distance was similar for younger and older participants.
Figure 45. Graph. Average adjusted detection distance by age group and light at study sites.
Vehicle type and driver eye height were also not significant for the daytime condition. Figure 46 shows the average adjusted detection distance subdivided by light and vehicle type. For bar pairs, there were no differences between sedan and SUV in the average values. A small difference occurred in daytime for continental markings. Transverse markings had the most variation in detection distance by vehicle type.
Figure 46. Graph. Average adjusted detection distance by vehicle type and light at study sites. Marking Type and Traffic Presence Interaction A significant interaction between marking type and traffic presence was identified for the daytime adjusted detection distances. This significant interaction indicates that the effect of one factor (e.g., marking type) may be different for each level of the other factor (e.g., traffic presence) and may need to be assessed conditionally on each level of the other factor. Table 15 provides the results for the least square means table and the Tukey test.
Table 15. Effect details of traffic presence and marking type interaction on daytime detection distance at study sites.
The interaction plot indicating the effect of marking type on detection distance conditional on the levels of traffic present is shown in figure 47. It can be observed that the least square mean detection distances for bar pairs and continental are longer than that for transverse markings, and the difference is larger when traffic is not present compared to when traffic is present. The Tukey's test (see table 15 ) shows that when traffic is not present, the detection distances for continental and bar pairs are similar (508 and 467 ft, respectively) but significantly different from transverse markings (265 ft), whereas there is no significant difference among the three marking types when traffic is present.
Figure 47. Graph. Least square mean daytime adjusted detection distance by marking type and traffic presence at study sites.
Driver Speed and Street Group Interaction The driver speed and street group interaction was statistically significant for daytime detection distance. As an initial examination, plots of adjusted detection distance and driver speed by light level and posted speed limit were generated. Figure 48 shows the plot for daytime, and figure 49 shows the plot for nighttime. As can be seen in figure 48 , the driver speeds on the 45-mi/h road were higher than the driver speeds at the 30-mi/h sites. Another pattern revealed is that around adjusted detection distances of 600 to 700 ft, several drivers were at speeds less than 20 mi/h. A closer investigation of those data points revealed that those drivers were still accelerating after completing a turn.
Figure 48. Graph. Adjusted detection distance by driver speed, posted speed limit, and daytime at study sites.
Figure 49. Graph. Adjusted detection distance by driver speed, posted speed limit, and nighttime at study sites. Because the model indicated that the driver speed and street group interaction was significant, plots were generated of adjusted detection distance by driver speed for each of the street groups. Figure 50 shows the relationship between driver speed and detection distance for street group 1. Street group 2 is shown in figure 51 , and street group 3 is shown in figure 52 . The figures also show a plot of the regression equation line that would be generated using the coefficients from the reduced model. The following conditions were assumed when generating the plots of regression lines:
The plots for street group 2 show the relationship between low speed and long detection distances; although, closer evaluation of the data revealed that the observation should be that when drivers are at low speed because of turning or accelerating after a turn, they detect crosswalks at a longer distance. A visual review of these graphs shows that higher speeds are associated with shorter detection distances (for street group 3, only slightly shorter detection distances).
Figure 50. Graph. Daytime adjusted detection distance by driver speed for street group 1 at study sites.
Figure 51. Graph. Daytime adjusted detection distance by driver speed for street group 2 at study sites.
Figure 52. Graph. Daytime adjusted detection distance by driver speed for street group 3 at study sites. Gender and Street Group Interaction The interaction between street group and gender was statistically significant in the reduced model (see table 14 ). The interaction plot indicating the effect of street group on detection distance conditional on gender is shown in figure 53 . The least square mean detection distance for street group 1 is longer for female than for male while differences between female and male for the other street group levels is minimal. The least square means along with two difference tests (Tukey HSD and Student's t) are provided in table 16 , which supports the above observation. There is no reason to believe that women have better eyesight than men (especially for only one street group and not other street groups), and the researchers attribute the gender difference to attention differences or response bias in that women were more willing to "guess" early to identify the marking.
Figure 53. Graph. Least square mean daytime adjusted detection distance by gender and street group at study sites.
Table 16. Effect details of street group and gender interaction variable on daytime detection distance at study sites.
Nighttime Detection–Study Sites This section presents the findings on nighttime detection at the study sites. Table 17 shows the results when potential main effect variables and two–way interactions are included in the model for nighttime detection (i.e., extended model). Table 18 shows the effect details. Most of the variables were not significant; only the following variables had a p–value less than 0.05:
Table 17. ANACOVA Findings for nighttime adjusted detection distance for extended model (includes potential variables and two-way interactions) for study sites.
Table 18. Effect details for variables in table 17.
Several additional combinations of main effects and two–way interactions were explored in the modeling efforts. Again, the elimination of non–significant variables changed the p–value of other variables, resulting in some variables becoming significant. Removing non–significant two–way interaction variables for the nighttime detection data resulted in the eye height and street group interaction term becoming not significant. In addition, some two–way interaction variables that were not significant became significant (e.g., driver speed and gender as well as driver speed and street group). Table 19 shows the resulting model, and table 20 shows the effect details. Table 19. ANACOVA findings for nighttime adjusted detection distance for reduced model for study sites.
Table 20. Effect details for variables in table 19 .
Driver Speed and Street Group Similar to the daytime analysis, the driver speed and street group interaction term was statistically significant. The plots showing the nighttime detection data along with a plot of the line that would be generated using the regression coefficients are shown in figure 54 for street group 1, figure 55 for street group 2, and figure 56 for street group 3. The following conditions were assumed when generating the plots of regression lines:
The plots show that the relationship between driver speed and nighttime adjusted detection distance is different for the different street groups. For street group 3, the influence of driver speed on detection distance was nominal. For street group 2, the influence of driver speed was similar to the influence seen for the daytime data–longer detection distances are associated with lower speeds. The relationship for street group 1, however, was the opposite. Detection distances were longer at higher speeds.
Figure 54. Graph. Nighttime adjusted detection distance by driver speed for street group 1 at study sites.
Figure 55. Graph. Nighttime adjusted detection distance by driver speed for street group 2 at study sites.
Figure 56. Graph. Nighttime adjusted detection distance by driver speed for street group 3 at study sites. Driver Speed and Street Group The gender and driver speed interaction was statistically significant. Figure 57 shows the original data along with the estimated regression lines for male and female using the parameter estimates. The plots of the regression lines for male and female are not parallel, and they cross at about 30 mi/h, which would contribute to the finding that there is an interaction between gender and speed. Given that the difference in predictions is 14 ft at 20 mi/h and –19 ft at 45 mi/h, the finding may be statistically significant but probably not of practical difference.
Figure 57. Graph. Nighttime adjusted detection distance by gender and driver speed at study sites. Eye Height and Street Group along with Vehicle and Street Group Within the extended model, the interaction term driver eye height and street group was statistically significant along with the interaction term of vehicle and street group. The nighttime adjusted detection distance data by driver eye height and street group are shown for street group 1 in figure 58 , street group 2 in figure 59 , and street group 3 in figure 60 . The following conditions were assumed when generating plots of regression lines:
Because all driver eye heights below 48 inches were in the sedan and all driver eye heights of 50 inches and greater were in the SUV (i.e., driver eye height is nested within vehicle type), the plots also show the detection data by vehicle. As unique main effect variables, driver eye height, vehicle type, and street group were not significant. This could indicate that the effect of driver eye height or vehicle type changes depending upon the street group, which would be surprising because the expectation is that driver eye height or vehicle type would have the same effect regardless of the characteristics of the roads.
Figure 58. Graph. Nighttime adjusted detection distance by vehicle type and driver eye height for street group 1 at study sites.
Figure 59. Graph. Nighttime adjusted detection distance by vehicle type and driver eye height for street group 2 at study sites.
Figure 60. Graph. Nighttime adjusted detection distance by vehicle type and driver eye height for street group 3 at study sites.
The overall expectation for driver eye height was that as driver eye height increases, the nighttime adjusted detection distance would be longer. Taller drivers should be able to see farther. For vehicle type, the relationship between detection distance and vehicle type is not known because of the different headlamps. To account for the potential effects of headlamps, vehicle type was kept as a separate variable from driver eye height. The two–way interaction of vehicle type and street group was statistically significant in the extended model. The headlamps on the different vehicles along with the street lighting available on a street could affect the detection distances along the different streets. The Tukey test for vehicle and street group interaction (see table 20 ) showed that the least square means for the different combinations of vehicle type and street group were not significantly different. Even though the initial statistical evaluation, F–test, found these terms significant, the multiple comparison procedure, Tukey test, did not find a difference. While a difference due to vehicle type would not be unexpected, having the impact of vehicle type change because the vehicle was on a different street would not be expected. The relationship found for the sedan drivers showed longer nighttime adjusted detection distances for drivers sitting higher. An opposite relationship was found for the SUV drivers. As eye height increased, the adjusted detection distance decreased. Regression lines were generated using the coefficients from the extended model as shown in figure 58 , figure 59 , and figure 60 . The regression lines show general patterns rather than specific values, since the plot of the line includes several assumptions, such as marking type and speed. In most combinations (e.g., SUV drivers on street group 2) the predicted differences between the lowest and highest eye height drivers would be considered practically different, so driver eye height may be a variable of interest. A better understanding of why increasing eye height improves detection distance in the sedan but decreases detection distance in the SUV is needed. This question is beyond the scope of the study, especially given that the interaction terms of marking type with vehicle type or eye height were not statistically significant. The interaction with marking type would be of greater interest. The interaction terms of vehicle and street group and eye height and street group were not significant in the reduced model, which is an additional reason to set aside this finding. Marking Type Marking type is the only main effect significant variable in the models (see table 17 and table 19 ). None of the interaction terms that included marking type were statistically significant for nighttime adjusted detection distance at the study sites. As shown in table 20 , the detection distance is similar for the continental and bar pairs markings (about 328–335 ft), and the detection distance to the continental and bar pairs markings is different from the detection distance to the transverse markings (about 220 ft).
Daytime Detection–Existing SitesIssues arose in the analysis of the data at the existing sites due to potential confounding between marking type and street group. The limited number of sites was a major obstacle in the evaluation. There was only one set of midblock continental crossing data because the data for the opposite approach (with an available viewing distance of only 339 ft) were removed from the data set. There was also only one midblock transverse marking crossing with data available for both approaches. Because of this limited number of sites, the midblock sites always had the same street groups (only street group 2 for the continental site and street group 4 for the transverse site). Only continental and transverse markings were used at the existing sites; there were no bar pair markings. Reviewing patterns of the data also indicated that a square root transformation of detection distance was needed to satisfy an underlying assumption for ANACOVA. The data were transformed back to original form (i.e., the results were squared) in the graphs developed to illustrate findings. Table 21 and table 22 show the summary of fit findings and fixed effect results for the daytime detection of existing markings when using all main effect variables along with reasonable two–way interaction variables. The response variable was transformed using a square root. The significant variables for daytime adjusted detection distance at existing sites include the following:
Table 21. ANACOVA summary of fit findings for daytime adjusted detection distance for extended model (includes potential variables and two–way interactions) for existing sites.
Table 22. ANACOVA fixed effects findings for daytime adjusted detection distance for extended model (includes potential variables and two–way interactions) for existing sites.
Table 23 provides the model that reflects only significant interaction terms and main effect variables that are part of a statistically significant two-way interaction term. Table 24 provides the effects details for this model. Statistically significant interaction terms are as follows:
Table 23. ANACOVA findings for daytime adjusted detection distance for reduced model for existing sites.
Table 24. Effects details for variables in table 23.
Marking Type and Location The interaction between marking type and location was significant (see figure 61 ). The continental markings were always detected at a greater distance as compared to the transverse markings. The Tukey results (see table 24 ) show that the detection distances to the continental or transverse markings at intersections are not significantly different. The detection distance to midblock continental is statistically different from the detection distance to midblock transverse markings. The difference between the continental and the transverse markings is more apparent at the midblock locations, as illustrated in figure 61 . Both transverse and continental marking midblock sites had pedestrian warning signs (W11–2). These sites are located on either side of a basketball arena near large parking lots used by TAMU students. Therefore, both sites have heavy pedestrian traffic associated with students going to classes during the day. One site was a two-lane street with bike lanes, and the other was a four–lane divided roadway, so the roadway width may be a factor. If one assumes that the roadway width is not a factor, a general observation could be that at a midblock location the continental markings were detected at about twice the distance upstream as the transverse markings. Another interpretation of the finding is that the additional 350 ft of detection distance between transverse and continental markings reflects 8 s of increased awareness of the presence of the markings at 30–mi/h operating speeds.
Figure 61. Graph. Least square mean daytime adjusted detection distance by marking type and location at existing sites.
Location and Driver Speed Driver speed was not significant. However, the interaction between driver speed and location was significant. Figure 62 shows the individual data points along with the regression line that would be generated using the coefficients from the reduced model (see table 23 ). The lower speeds on the approaches to the stop– or signal–controlled intersections can easily be seen. The effects of driver speed are statistically different for the midblock locations and the intersection locations. At the intersection locations, shorter adjusted detection distances were associated with lower speeds while the opposite occurred at the midblock locations. Faster drivers at the midblock locations had slightly shorter detection distances. The low detection distance and low speed at the intersections is related to the drivers coming to a complete stop at the intersection. Several drivers focused more on the stopping maneuver than on the task of identifying the crosswalk. They would make comments such as "oh yes, crosswalk," indicating that they only recalled the crosswalk identification task after initiating the stopping maneuver.
Figure 62. Graph. Daytime adjusted detection distance by driver speed and location at existing sites.
Age Group and Location Age group was not significant. However, the interaction between age group and location was significant. As shown in figure 63, younger drivers had slightly shorter detection distances than older drivers at the intersections (180 ft compared to 194 ft). For the midblock sites, the pattern was reversed; younger drivers had greater detection distances (576 ft compared to 525 ft). The Tukey results shown in table 24 reveal that detection distances are not statistically different for the two age groups at the midblock locations or at the intersections. Stated in another manner, the detection distance is different for midblock and intersection locations. Detection distance is not different for older and younger drivers at the midblock locations or at the intersections.
Figure 63. Graph. Least square mean daytime adjusted detection distance by driver age and location at existing sites.
Nighttime Detection–Existing SitesSimilar to the analysis of the existing site daytime data, the analysis of the existing site nighttime data had potential confounding issues. In addition to the overlaps between marking type and street group was an overlap between marking type and retroreflectivity group. At the stop-controlled intersections, all of the transverse markings had retroreflectivity readings less than 38 mcd/m2/lx and the continental markings had readings in the 500s (524 or 570 mcd/m2/lx). On the other hand, at the midblock locations, all of the continental markings had retroreflectivity readings of 81 mcd/m2/lx and the continental markings had readings in the 300s (309 or 331 mcd/m2/lx). Therefore, retroreflectivity was not included in the model. Similar to the daytime analysis, the nighttime data indicated that a square root transformation of detection distance was needed to satisfy an underlying assumption for ANACOVA. The data were transformed back to original form (i.e., the results were squared) in the graphs developed to illustrate findings. Table 25 (summary of fit) and table 26 (fixed effect tests) show the results for the nighttime detection of existing markings when using all main effect variables along with reasonable two–way interaction variables. The response variable was transformed using a square root. The significant two–way interaction variables for nighttime adjusted detection distance at existing sites include the following:
Table 25. ANACOVA summary of fit findings for nighttime adjusted detection distance for extended model (includes potential variables and two–way interactions) for existing sites.
Table 26. ANACOVA fixed effect tests findings for nighttime adjusted detection distance for extended model (includes potential variables and two–way interactions) for existing sites.
The reduced model that only includes significant two–way interaction terms along with main effects variables is shown in table 27 . The least square means and Student's t test results are provided in table 28. Although marking type was not significant either as a main effect or part of a two–way interaction in the extended model, it was significant under the reduced model for the nighttime existing site data. Table 27. ANACOVA findings for nighttime adjusted detection distance for reduced model for existing sites.
Table 28. Effects details for variables in table 27.
Driver Speed and Location Similar to the findings for the daytime, driver speed was significant along with the interaction term of driver speed and location. Figure 64 illustrates the findings for driver speed and location. For existing crosswalks at stop– or signal–controlled intersections, lower speeds are associated with shorter detection distances. The statistical evaluation found a similar trend for midblock locations; longer detection distances are associated with higher speeds (see plot of regression equation). Gender and Location The interaction between gender and location was significant, with women seeing the crosswalk markings at a greater distance upstream for the midblock locations. A similar difference between male and female detection distance was found for the study sites. As previously noted, there is no reason to believe that women have better eyesight than men. This gender difference might be attributable to attention differences or response bias in that women were more willing to "guess" early to identify the marking.Figure 65 illustrates the trend. The difference for gender was over 136 ft at the midblock locations, which was statistically significant based on Tukey's test (see table 28 ). The difference was minimal for the intersections and was not statistically significant.
Figure 64. Graph. Nighttime adjusted detection distance by driver speed and location at existing sites.
Figure 65. Graph. Least square mean nighttime adjusted detection distance by gender and location at existing sites.
Marking Type Marking type was statistically significant in the reduced model (see table 27 and table 28 ). Similar to the study site evaluation, none of the interaction terms that included marking type were statistically significant for nighttime adjusted detection distance at the existing sites. The least square mean detection distance to the continental markings was 186 ft (after applying a squaring transformation) as compared to 160 ft to the transverse markings. The Tukey test (see table 28 ) did identify these distances as being significantly different.
Comparison of FindingsTable 29 summarizes the findings from the statistical evaluations of the adjusted detection distances. The results are subdivided by the site type (study or existing) and light (day or night). Preliminary evaluations demonstrated that evaluations needed to be conducted separately for the study sites (where the markings were installed at midblock locations) and the existing sites (where the markings were already present at midblock locations with pedestrian warning signs or at intersections). The preliminary evaluations also clearly showed a difference in detection distance for day and night. Since the nighttime condition had an additional variable, retroreflectivity, to consider and some of the variables were believed to have different effects during the night (such as vehicle type and driver eye height), separate analyses were done for daytime and nighttime conditions. The average detection distances by location, day or night, and marking type is shown in table 30. In all combinations, daytime detection distances are longer than nighttime detection distances. As shown in table 29, the marking type (bar pair, continental, or transverse) for the study sites was statistically significant. The detection distances to bar pairs and continental markings were similar, and they were statistically different from the detection distance to the transverse markings. For the study sites, the presence of traffic had an impact on detection distance, in most cases limiting the ability to see the markings farther upstream, as expected. The impact of traffic on the transverse markings was minimal as the detection distance to these markings was already small compared to the detection distances for bar pairs or continental. Overall, shorter detection distances were associated with higher speeds. However, in most cases, it was only slightly shorter detection distances. The characteristics of the streets also influenced the detection of the crosswalk markings. An unexpected result was that the street group with 45–mi/h posted speed limit had longer nighttime adjusted detection distances for the higher speeds. This was opposite the finding for daytime conditions; daytime adjusted detection distances were (slightly) shorter for the higher speeds. Variables that included gender, driver eye height, and vehicle type as part of an interaction term were found to be statistically significant; however, closer examination found them to not be of practical significance. For the existing sites, marking type had a significant effect on detection distance. During the day, the detection distances to the continental or transverse markings at intersections were not significantly different. The detection distance to midblock continental was statistically different (longer) from the detection distance to midblock transverse markings. During nighttime conditions, variables in addition to marking type, such as location (midblock or intersection) and driver speed, had an effect on detection distances at the existing sites. Driver speeds had mixed effects on detection distance depending upon location (intersection or midblock) and light level (day or night). For intersections, an increase in driver speed was associated with longer detection distances for both the daytime and nighttime conditions. All of the intersections included in this project were either stop–controlled or signal–controlled. Several drivers appeared to be more focused on the stop maneuver than the detection task and would not call out the recognition of a crosswalk until close to the stop bar. For midblock (or uncontrolled approaches), the finding was dependent on light level. Nighttime detection distance at midblock was similar to intersections; longer detection distances were associated with the higher speeds. For daytime, the opposite occurred; higher driver speeds were associated with shorter detection distances to the midblock crosswalk. While the higher driver speeds were associated with shorter detection distances, the differences were small and would not be considered of practical difference. Table 29. Summary of adjusted detection distance findings.
Table 30. Average adjusted detection distances.
FHWA-HRT-10-068
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
