CHAPTER 9

Synthesis and Summary of Research Results

In this study, three research approaches were used to explore the safety performance of intersection treatment types for bicycles at intersections: 1) crash analysis; 2) video-based conflict analysis; and 3) human factors study (simulation). This research focused on five different intersection treatment types, as shown in Figure 53.

The research was designed to gain multi-pronged insights into design-level questions uncovered in the State of the Practice review. Importantly, the research was designed to provide information on relative safety performance and inform design-related thresholds and guidelines. Each method is robust and produced valuable information, however, the differing scale, focus and limitations of each approach means that synthesizing the results requires interpretations. Table 96 presents an abbreviated summary of findings by treatment type and research approach. Text following Table 96 provides a synthesized safety performance for each treatment type based on observations and insights, followed by risk factors. To clarify the source of the observation or insight, each statement is labeled with “CRASH, CONFLICT, and SIMULATOR” labels.

Table 96. Summary findings by intersection treatment type

Safety Performance

Conventional Bicycle Lane Intersection Treatment

• CRASH – In all four cities, the second highest volume-adjusted crash rate occurs at intersections with a conventional bicycle lane, outside of a shared thru/turn lane. At the New York sites, the conventional bicycle lane has similar crash rates to pocket/keyhole and the separated bicycle lane. Conventional bike lanes with extra space between the bike lane and the curb were measurably safer than a curb-tight version of this facility and treatment.
• CONFLICT - After the pocket/keyhole bike lane, the conventional bicycle lane treatment type has the second highest predicted conflict frequency for the same bicycle and motor vehicle turning volumes.
• CONFLICT – The conflict frequency models indicate that bicycle through volume is an important predictor of the number of severe conflicts. In the models, the predicted number of severe conflicts increases more per unit increase in bicycle volumes for this treatment type as compared to others.
• SIMULATOR – When the cyclist was located further from the intersection in this treatment type, drivers exhibited more conservative yielding behaviors and turning speeds. Drivers also were observed to have the lowest average stress levels (measured in PPM) when the cyclist was closer to the intersection, most likely due to their familiarity with this treatment (drivers responded to the post-drive survey that they were most familiar with this design).

Limitations

• In the CRASH analysis, non-curb-tight designs were found to be substantively different than curb-tight designs, which may be related to the presence of motorist parking or using the intersection in a way that could not be captured in the analyses.
• The observed range of bicycle volumes for the conventional bicycle lane in the CONFLICT analysis (ranging from a maximum 6 bicycles per hour to 40 bicycles per hour) limit extrapolation of the predicted conflict models to compare results to the other treatments, such as the mixing zone or offset/protected intersection at higher bicycle volumes (except for two sites, maximum hourly bicycle volumes of ranging from 91 to 702 bicycles per hour).

Summary Statement

• Both the crash analysis and conflict analysis indicate that safety performance (either crashes or severe conflicts) for this treatment type is worse than other options when controlling for exposure. The conflict frequency modeling suggests that safety performance is more sensitive to bicycle and turning volumes than other treatments. It should be noted that other research clearly shows that bicycle facilities at the intersection are preferable to no facility in terms of bicyclist comfort (Monsere 2019), and enhanced bikeways and intersection treatments with better safety performance should be used when possible.

Separated Bicycle Lane Intersection Treatment

• CRASH – This treatment type, which had a small sample size for all cities, had a higher crash rate than conventional bike lanes at the intersection (both curb-tight and not curb-tight) for all four cities. However, after controlling for other factors in the model, this treatment’s effect did not significantly differ from the effect for the base case (curb-tight bike lane outside of a shared thru/right-turn lane). Additionally, volume estimates at this treatment type were substantially higher than non-separated bike lanes and may still have underestimated actual changes in volume in response to building these new

• facilities. With high and uncertain volume estimates, the higher crash rates at a very small sample of locations may not accurately characterize the relative risk or safety of these facilities. Further, these high volumes lend yet more evidence that bicyclists value separation from traffic.
• CONFLICT – In the conflict frequency modeling, the separated bicycle lane was found to have the lowest predicted number of conflicts for comparing the same bicycle and turning volumes across treatment types. The proportion of conflicts that were categorized as severe (5.5%) is much less than for the conventional bicycle lane (10.9%) treatment.
• SIMULATOR – This treatment type was not studied in the simulator.

Limitations

• There is some site-selection bias present in the observational crash and conflict analysis. Compared to the other intersection types included in this study, the separated bike lane intersection treatment was more typically used when there is an expectation of lower hourly turn volumes or bicycle volumes; therefore, it is challenging to extract these results for comparison to the other treatments.
• In the CRASH analysis, small sample sizes as well as additional context not captured by the model described above, likely influenced results. In addition, our crash exposure models used estimated bicycle volumes from network models, which may not fully capture the increased volumes that are generated when more separated bikeways are installed. This underestimation of volume increases may contribute to higher than expected crash rates at these sites.
• Similar to the conventional bicycle lane, the observed range of bicycle volumes for the separated bicycle lane in the CONFLICT analysis (ranging from a maximum 17 bicycles per hour to 198 bicycles per hour) limit extrapolation of the predicted conflict models to compare results to the higher design treatments such as the mixing zone or offset/protected intersection (with the exception of two sites, maximum hourly bicycle volumes of ranging from 91 to 702 bicycles per hour).

Summary Statement

• The design of intersection approaches with SBLs are similar to conventional bicycle lane, except for vertical separation elements in advance of the intersection, as well as horizontal offsets from the vehicle turning lane. This treatment type was associated with the second highest bicyclist volumes, which suggests preferences for greater separation from motorist traffic. The analysis found differing safety performance for this treatment type in the CRASH (i.e., similar safety performance to conventional bike lanes) and the CONFLICT analysis (i.e., lowest predicted number of conflicts compared to all other treatments). With the differing safety results, and because most of the existing applications of the treatments are in locations with lower turn volumes or bicycle volumes, the SBL treatment type is not expected to have good safety performance at the higher turning volumes typically observed at mixing zones or, protected/offset intersections. At higher volume locations, full or partial phase separation may be needed to manage the conflicts and improve safety performance.

Pocket or Keyhole Bike Lane Intersection Treatment

• CRASH – This treatment type appears to be safer than a conventional bike lane outside of a shared thru/turn lane (curb-tight or not) outside of New York but has mixed performance in New York (small sample size). In the statistical model, relative to the base case (curb-tight bike lane outside of a shared thru/right turn lane), the pocket/keyhole intersection treatments (where the bike lane shifts to be between a through lane and a dedicated turn lane) is significantly associated with fewer bicyclist crashes.
• CONFLICT – The pocket/keyhole bike lane had the largest percentage (16.1%) of its total conflicts observed in the high severity category. The highest vehicles speeds were observed at the conflicts points

• and the largest speed variance (by design, the conflict point is upstream where vehicles are traveling faster). Finally, for the same combination of bicycle and turning volumes, the conflict frequency model predicted the highest number of severe conflicts for this treatment type.
• SIMULATOR – Outlying high speed profiles were observed primarily at the pocket/keyhole bike lane and mixing zone intersections and corresponded to drivers who incorrectly maneuvered the intersection from lack of comprehension (mixing zone) and delayed recognition of the right-turn lane in the pocket/keyhole bike lane intersections.
• SIMULATOR – In comparison to the protected and conventional bike lanes, the TFD on pavement markings was notably higher for the pocket/keyhole bike lane. A portion of this increase may be attributed to its unique feature – the green painted bike lane. Interestingly, in the post-experiment survey and through verbal discussion, some participants expressed how the green pavement markings remind them to check on cyclists before merging.
• SIMULATOR – At the intersections with pocket/keyhole bike lanes, a subset of participants who were observed to approach the intersection at relatively higher speeds were unable to recognize the right-turn lane far enough in advance, hence, executed the turn from the thru lane, treating the bike lane configuration as a conventional bike lane outside of the turn lane. Moreover, in the post-experiment survey, some participants expressed how the right turn lanes in the pocket/keyhole intersections were too short.

Limitations

• In practice, it is likely that more risk-tolerant types of bicyclists use this intersection treatment type (i.e., those comfortable enough to be in the middle of motor vehicle traffic), which may lead and/or be related to substantive differences in behavior and experience in the observational methods.
• Observed bicycle volumes in the CONFLICT analysis did not exceed maximum of 30 bicycles per hour, which limits the ability to compare how the safety performance of this intersection treatment type would compare to the treatments that were observed to have higher bicycle volumes.
• There is more variation in the designs of these facilities, especially in the placement of the merge area and the length and marking of the merge/conflict point (including the use of green color pavement marking).

Summary Statement

• Crash data suggest that this treatment type is relatively safer than a conventional bicycle lane (outside of the NYC data), but the conflict data shows higher-severity conflicts and higher speeds. Further, the simulator data shows motorists traveling at a high speed and making the right turn from the through lane rather than the right-turn lane, which negate the intention the pocket bike lane (i.e. to have the motorist cross the bike path prior to the intersection at a defined point). Taken together, these findings indicate that this treatment type may not be appropriate for an all ages and abilities network, due to the need to consistently watch for and negotiate with motorists for safe passage. Observed conflicts had the highest percentage of high severity interactions, despite having the fewest number of conflicts in the study. The pocket/keyhole bicycle lane may be an acceptable design in limited situations to accommodate a right-turn lane as long as the risk factors identified (e.g., vehicle speed) are addressed, but it is not highly recommended in other scenarios.

Mixing Zone Intersection Treatment

• CRASH – In Austin, Minneapolis and Seattle, the mixing zone treatments have the lowest crash rates per 1,000 daily bicyclists. In New York, the crash rate is second lowest only to the offset/protected

• intersections. In the statistical model, which does not include New York, the treatment effect is not significantly different from the base case (curb-tight bike lane outside of a shared thru/right-turn lane).
• CONFLICT – In the conflict frequency modeling, the mixing zone had the lowest number of predicted conflicts. The mean observed vehicle speed at the conflict zone was also the lowest (7.27 mi/h) and the proportion of severe conflicts was also low (6.1%).
• SIMULATOR – Similar to keyhole bike lane, the subjects had relatively higher total fixation on the pavement markings. When cyclists were closer to the intersection, the transition and turn zone speeds were slightly higher when parallel parking was present. Driver response at mixing zones became more correct across repeated exposure to mixing zone configurations (for participants who were previously unfamiliar with mixing zones).

Limitations

• Intersections with mixing zones may not be as comfortable for bicyclists, particularly those who are less experienced. If more risk-tolerant types of bicyclists (i.e., those comfortable enough to be in the mix with motor vehicle traffic) use this treatment type in practice, that may lead and/or be related to substantive differences in behavior and experience.
• In the video-based conflict analysis, the sites selected for conflict analysis were well designed and followed most of the existing guidance. The results from this study should be interpreted within this context— if sites with less quality designs were selected (such as longer mixing areas), it is likely that additional or different conflicts would be observed.

Summary Statement

• Crash and conflict data indicate good safety performance relative to a conventional bicycle lane in all cities. The results from the simulator experiment also indicate high levels of motorist attention, which can contribute to safe operation. However, it is likely that more risk-tolerant bicyclists are using bikeways with mixing zones because they require more mixing of bicyclists and motorists. In design and operation of mixing zones, care should be taken to maximize bicyclist comfort and comprehension of expected behavior (particularly speed) for both bicyclists and motorists.

Offset or Protected Intersection Treatment

• CRASH – There was a small sample size for this treatment type, but also a lower crash rate than observed for the separated bike lane outside of the shared thru/turn lane in all four case cities and the lowest crash rate observed in New York. This design had by far the highest volume of bicyclists among all the treatment options studied, showing a preference for greater separation from traffic.
• CONFLICT – The design had the lowest percentage (4.9%) of the conflicts that were classified into the high severity (PET less than 1.5 seconds and high vehicle conflict speed) despite having the largest number of observed bicycle volumes of any of the treatment types.
• CONFLICT – The conflict frequency models estimated the second lowest predicted number of conflicts (after the mixing zone) for the same combination of through bicycle volumes and turning vehicles. The model also showed the predicted number of conflicts is very low despite having much larger exposures.
• SIMULATOR – The minimum average TFD, 0.24 seconds, was observed in the intersection with the separated bike lane offset by 6 feet, with parallel parking. When the offset was the highest (18 feet) a significantly greater percentage of drivers looked at the cyclist and when they did, their fixation times were longer. Moreover, this was the only intersection design where all participants had a non-zero fixation time on the cyclist. The lowest averaged minimum speed (speed = 5.2 mph, SD = 4.5 mph) was observed in the scenario with parallel parking and an offset of 6 feet.

Limitations

• Small sample sizes precluded the inclusion of this treatment type in the multivariate regression model of bicyclist-motorist crashes.
• Most of the intersection approaches in the conflict study were in New York and the level of bicycle volumes were much larger than any other treatment type observed.

Summary Statement

• This treatment type was associated with the largest bicyclist volumes by far, underscoring preferences for greater separation from motorist traffic. Despite these higher bicycle and motor vehicle turning volumes, the crash and conflict results indicate that offset or protected intersections perform better (in terms of crash rates and high-severity conflicts) than conventional bicycle lanes in all cities and better than all other treatment types in New York City. This design appears preferable to conventional bicycle lanes and pocket/keyhole bicycle lane and comparable to mixing zones from a safety performance perspective. However, it is likely that more risk-tolerant bicyclists (i.e., those comfortable enough to be in the mix with motor vehicle traffic) are the majority of users at mixing zones, which likely contributes to mixing zones appearing to have comparable safety performance to offset/protected intersection.

Risk Factors

Bicycle Volume

• CRASH – The statistical model indicates that bicycle volumes are significantly and positively associated with bicyclist-motorist crashes on approach. Separate city-specific analyses of crash rates and numbers per treatment type underscore the impact that higher volumes have on crashes.
• CONFLICT – The models that were used to analyze conflict frequency confirmed that both bicycle volume and motor vehicle turning volume are significant predictors in the number of conflicts observed at each treatment type. The models of severe conflicts (low PET and high vehicle speed) predicted higher conflicts for a given combination bicycle and motor vehicle turning volume for the conventional bike lane and the pocket/keyhole bike lane. For all five treatment types the contribution of the same direction motor vehicle turning volume was relatively consistent; however, for the conventional bike lane and pocket/keyhole bike lane, the bicycle volume was a significantly more important predictor of the total severe conflicts (as measured by the incidence rate ratio, or IRR) than same direction turning volume. This indicates that bicycle volume is a significant consideration when deciding to move to a higher treatment type design such as mixing zones, protected intersections, or bicycle signals.

Motor Vehicle Though Volumes

• CRASH – Vehicle volume was examined both directly and indirectly via the crash analysis. While the direct variable (motorist AADT) was not found to be significant in the multivariate model, the number of vehicle lanes variable, a proxy for vehicle volumes, is significant. More information about the effects of multi-lane streets is presented in the “Number of Lanes” section below.

Left Turns (Opposing Direction Turns)

• CRASH – The presence of an opposing left turn lane is significantly associated with more bicyclist crashes. Because opposing left turn lanes are installed where motorist left turning volumes are higher, this relationship is likely due to a proxy relationship with turning volumes or other activity at the

• intersection rather than presence of the opposing left turn lane itself. Signals with a green ball only (implying permissive left turns with no protected left turn phasing) have a stronger association with bicyclist crashes (higher coefficient; lower p-value) than signals with an arrow (implying the possibility of some protected left turn phasing and/or flashing yellow arrow operations).

Right Turns (Same Direction Turns)

• CONFLICT – Conflict frequency models indicate that for the conventional bicycle lane and the pocket/keyhole bicycle lane, increases in the same direction turning volume cause larger increases in the predicted number of conflicts than the other treatment types.

Pedestrian Volume and Activity

• This risk factor was not directly studied by any of the methods.
• CRASH – The multivariate analysis found that bus stop presence, a potential proxy for pedestrian volumes as well as traffic complexity, is significantly associated with bicyclist crashes. This finding is likely an indicator of activity level at the intersection. There may also be transit-specific factors in effect, such as obstructed views or drivers passing transit vehicles inappropriately, though this is impossible to discern from the data collected.

Vehicle Speed and Arrivals

• CRASH – The coefficient for each speed limit category above 25 mph is positive suggesting that there are more crashes at higher speed approaches. In particular, the coefficient for speed limits 45mph or higher is significant. An approach with a posted speed limit of 45 mph or higher has over 13 times as many bicyclist-motorist crashes than one with a 25 mph or lower speed limit.
• CONFLICT – Vehicle speeds at the conflict point were highest for the pocket/keyhole bicycle lane (given the design places the conflict point prior to the turn and deceleration area). The mixing zone designs had the lowest speeds at the conflict point, followed by the offset or protected treatment.
• CONFLICT – The CS models identified that when the motor vehicle arrived first, this operation was a significant predictor of CS for all treatment types. The treatment types had varying ranges of observed proportions of vehicles arriving first. For the conventional bike lane, the pocket/keyhole bike lane and the separated bike lane, the vehicle arrived first in approximately 40 percent of the conflict observations. For the mixing zone, 60 percent of the vehicles arrived first, and for the protected intersection, it was 50 percent.
• SIMULATOR – For the mixing zone intersections, when the cyclist was located closer to the intersection, drivers exhibited similar vehicle speed patterns across zones, with and without parallel parking conditions. However, when the cyclist was located further from the intersection, there was an observable difference in vehicle speeds between parallel parking conditions – when parallel parking was present, the average transition and turn zone speed was 5.8 and 3.7 mph lower, respectively, than the corresponding average zone speed when there was no parallel parking. This may because drivers in the simulator are expecting to see a bicyclist (despite the random presentation of sequences), and when the bicyclist is behind parallel parking the drivers need to slow in these zones to search or scan for a bicyclist.
• SIMULATOR – The outlying high speed profiles were observed primarily at the pocket/keyhole bike lane and mixing zone intersections and corresponded to drivers who incorrectly maneuvered the intersection from lack of comprehension (in the mixing zone) and delayed recognition of the right-turn lane in the pocket/keyhole bike lane intersections.

• SIMULATOR – The average minimum turning speeds were observed to be lowest in both protected intersection (6 feet offset) scenarios: without parallel parking had an average minimum speed of 5.5 mph, and with parallel parking had an average minimum speed of 5.2 mph. However, these two scenarios exhibited the highest (with parallel parking) and third highest (no parallel parking) minimum speeds and distributions across the preceding transition zone, indicating the greatest rates of deceleration close to the conflict point.

Vehicle Size and Type

• CONFLICT – Large vehicles were significantly associated with increased CS for conventional bike lanes, mixing zones and offset/protected intersections.

Number of Lanes (Cross Street)

• CRASH- Cross streets with more than two lanes are associated with more crashes than cross streets with one to two lanes. The coefficient for each category of cross street lanes above 1-2 lanes is positive and significant, suggesting that there are more crashes when the bicyclist has to cross more lanes of traffic. While the coefficient for the highest bracket (7-12 lanes) is smaller than the two in the middle (3-4 lanes and 5-6 lanes), the p-value is also higher for the top bracket. This finding does not indicate that these very wide roads are somehow safer, rather, it is likely that this result is related to having relatively few approaches with 7-12 lanes in our dataset compared to other categories. Reducing lanes and roadway widths can further enhance bicyclist safety at signalized intersections.

Sight Distance and Parking

• SIMULATOR – In the offset/protected scenarios, drivers looked at the cyclist less when parallel parking was present. However, in the mixing zone, when parallel parking was present, drivers spent more time looking at the cyclist.
• SIMULATOR – From the post-experiment survey, many participants explicitly stated that the parallel parking made it harder for them to see the cyclist, with some participants expressing increased levels of stress when parallel parking was present. Interestingly, the average stress indicators (PPM) in the scenarios where cyclist was located closer to the intersection for all three treatments (mixing zone, pocket/keyhole and offset/protected) were higher when there was no parallel parking.

Design / Geometry

• CRASH – Skewed approaches at the intersection were associated with more crashes than non-skewed approaches in Austin, Minneapolis and Seattle. The relationship for mild skew (near 90-degree angle) is small and non-significant, whereas it is larger and significant for approaches with more substantial skew. Skewed approaches have 1.5 times as many bicycle crashes as non-skewed approaches, which may indicate visibility challenges and/or issues with driver and bicyclist communication, behavior and expectations.
• CRASH – Bicycle facilities on the left side of the street continued to be associated with more crashes than those on the right side of the street. While the number of left-side bikeway approaches in our sample is small (n=17), the effect is statistically significant. Since left-side facilities are only installed for one-way approaches, we tested an approach directionality variable (one-way vs. two-way), but it was not significant.
• SIMULATOR – Regarding the protected intersection offset, some participants expressed in the post-experiment survey that the lateral separation (in combination with the divergence in the bike lane

• trajectory from the vehicle trajectory) was confusing because the cyclist’s trajectory was not as “predictable,” since it was not parallel with the thru and right turn lane upon approaching the intersection.
• SIMULATOR – At the intersections with pocket/keyhole bike lanes, a subset of participants who were observed to approach the intersection at relatively higher speeds were unable to recognize the right-turn lane far enough in advance, a hence, executed the turn from the thru lane, treating the bike lane configuration as a conventional bike lane outside of the turn lane. Moreover, in the post-experiment survey, some participants expressed how the right turn lanes in the pocket/keyhole intersections were too short.

Conclusions

The objective of this research was to develop guidelines and tools to provide practitioners a better understanding of the safety performance of design treatments to use to reduce the frequency and severity of turning conflicts between motor vehicles and bicycles at controlled intersections. The research was framed by a state of the practice review that included the existing literature, a summary of current design guidance, and a practitioner interview process. A macro-level crash analysis examined bicycle crashes and injuries at a broad scale (e.g., using multiple state databases) to contribute to a thorough understanding of bicycle crashes at intersections, including type, frequency, and severity, in varying land use contexts (urban, suburban, rural) and was used to provide additional context to the research design. Three research approaches were used to explore the safety performance of intersection treatment types for bicycles at intersections: 1) crash analysis (both a macro crash analysis and micro crash analysis); 2) video-based conflict analysis; and 3) human factors study (simulation).

• Micro-level crash analysis examined bicycle crashes, injuries, roadway design, operational variables, and related risk (e.g., accounting for exposure) at a smaller scale to clarify how exposure and various roadway design factors influence risk across a variety of facility types and site-specific contexts.
• Video-based surrogate safety analysis examined conflicts via surrogate safety measures at a further targeted scale to provide additional insight into the dynamics between bicycle safety and known risk factors like turning volumes, speed, and design features.
• Human factors study explored and validated some design assumptions about bicycle-vehicle interactions, reaction times, cognitive behaviors, and mechanisms of a driving/bicycle simulator.

Using the synthesized results of this research and existing knowledge, a decision tool was designed to provide information on relative safety performance and inform design-related thresholds and guidelines. The findings from this study can help practitioners gain a better understanding of the safety performance of design treatments in different contexts to inform design decisions.

Limitations and Future Research

To promote the safety of all roadway users, the limitations of the study must be explicitly defined, such that findings are accurately interpreted and appropriately applied. This will better allow for implementations of these vehicle-bicycle lane configurations in relation to site specific conditions, and more finely tuned to the users’ existing perspectives, attitudes, and behaviors.

• Micro-Crash Analysis - Crashes are the most direct and measurable safety outcome, but all crash data analysis is retrospective and depends on the accuracy and quality of reported crash data. Bicycle crash analysis is particularly challenging because these crashes are very rare events and sufficient samples are not easily obtained for statistical analysis. Typically, the precise details of crash-level

• actions and events are not known, and the analyst is reliant instead on coded fields in the reported crash data. Crash reports are likely to contain missing or inaccurate information mainly pertaining to crash location and time, severity, participants’ characteristics and contributing factors, which can impact the analysis. Importantly, exposure data, both of bicycles and vehicles, is needed to allow comparison of counts but is often not available or only available at limited frequency and for short duration (e.g., for single day of the month). Lastly, because not all persons will choose to bicycle on all facilities, there is a potential for omitted bias in the crash data. In other words, less experienced or risk adverse bicyclists may avoid higher speed roadways while a wider array of people may choose to bicycle on lower speed facilities which may impact safety analysis results in ways that are difficult to measure.
• Video-Based Conflict Analysis - Conflicts, or interactions between vehicles, occur much more frequently than crashes. Using automated video-based conflict analysis gathers robust detail about each bicycle-motor vehicle interaction including speeds, trajectories and information about all vehicle paths. There is a reasonable body of research that correlates conflicts and other surrogate measures of safety with actual crashes (i.e., that observing conflicts is useful in understanding the actual reported crash performance). The research, however, is less developed for bicycle-motor vehicle crashes. It also is reasonable to assume that fewer conflicts would be associated with higher levels of comfort for road users, including people on bikes, and it is useful to know the rate and severity of conflicts, though this has not yet been validated by research. Data for the video-based conflict analysis was obtained from 28 sites across four cities. While geographical diversity in the sites was chosen to represent varying land use contexts and behaviors, the use of a larger sample of sites in future studies could be beneficial in further exploring user behaviors and safety across various treatment types. Lastly, the comment on omitted bias also applies to the conflict analysis.
• Human Factors Study (Simulator) - Both the crash and video-based conflict studies are observational – only what has been designed and built and how existing traffic and drivers use the roadway can be observed. In the driving simulator, the experimental environment can be completely controlled and the variation in outcome is in the driver’s performance within the same situation. In addition, significant and detailed information about each driver’s performance can be collected, however, there is a limitation on the number of people that can be included in the experiment (typical experimental cohorts are between 30 and 40 people). While many measures have been validated to real world conditions and is directly related to safety (e.g., driver speed selection), not all data maps directly to safety outcomes. The first limitation of this research is the focus on sampling passenger car drivers, resulting in design guidelines that emphasizes this user type. To some degree the safety of cyclists was accounted for by the level of separation variable, yet what remains unknown is cyclists’ perceived safety of and preferences for such configurations. Under a wider scope of research, larger budget, and extended timeline, this multifaceted limitation could be addressed. Future work could take a similar approach and study cyclist and heavy vehicle driver performance measures, behaviors, and preferences. The second limitation of this study is the exclusion of network-level confounding variables such as vehicle density and presence and density of vulnerable road users. This was a result of project-related constraints (scope, budget, time) and the factorial design of the experiment. Ultimately, the research team identified driver behavior to be the most critical contributing factor to right-hook bicycle crashes. While confounding network-level variables influence driver behavior independent of cyclists, right-hook bicycle crashes are ultimately a result of failure of situational awareness. This is derived in cyclist visibility, which is not only a function of the driver position relative to the cyclist, but perhaps most affected by the geometric design conditions (e.g., infrastructure elements, guidance, right-of-way configuration) along the approach. In this lies the rationale for prioritizing the independent variables studied over network-level variables.

These limitations have implications on user safety and network operations and thus, when the study’s findings are being considered, discussed, or applied, these should also be discussed. To alleviate some of