The main characteristics and results of the studies are summarised in Table 2. All studies except Störr et al. (2017) were designed as before-and-after studies, two of them with reference sites for comparison (Egeler et al., 2015; Eriksson, 2016). All studies collected video data to examine changes in cyclist behaviour and conflicts, except Egeler et al. (2015) who did manual on-site observations. In addition, accident data were analysed (Egeler et al., 2015) for three months to two years, which is generally too short, particularly for analysing accidents involving only one user group (cyclists) and direction of movement (right turn). Furthermore, most of the studies did not specify which accidents they considered to be related to the RTOR regulation.

The small number of study sites as well as cyclists and conflicts per site makes it difficult to draw reliable conclusions from these analyses, but the studies do give some indication of the determinants and effects of RTOR, which can be summarised as follows: The proportion of cyclists turning right on red tends to increase with the introduction of RTOR, but was already high at the study sites previously. With RTOR, more cyclists stay on the carriageway rather than cutting across the pavement to avoid the red light. Cyclists wriggling through others when approaching the intersection on the carriageway and conflicts between cyclists and pedestrians on the crossings might be negative effects, while less cyclists and conflicts on the pavement might be a positive effect of RTOR. No spill-over effects were found, the introduction of RTOR does not seem to have significant effects on cyclists going straight at RTOR intersections or on cyclists' right turn behaviour at other intersections.

In addition to the few identified applied pilot studies that focus specifically on RTOR for cyclists, the literature on cyclist compliance with the design and operation of signalised intersections and related implications on cyclist safety is relevant to this study. While red light running is of main interest for compliance in terms of operation, behaviour of cyclists approaching or exiting the intersection (including changing from the carriageway to the pavement or vice versa) is of main interest for compliance with the design. The pilot studies indicate an increase in conflicts in the approach, in the exit and at the crossings, and a decrease in conflicts on the pavement. No scientific studies were identified that investigate conflicts between cyclists and other users approaching or exiting the intersection, or crossing the approach or exit legs of the intersection. Few studies were found that investigate conflicts between pedestrians and cyclists.

Cyclist red light running rates from observations at specific intersections range from 7% (Johnson et al., 2011) to 60% (Fraboni et al., 2018; Richardson & Caulfield, 2015; Wu et al., 2012). Red light running rates for right-turning cyclists are consistently higher than for cyclists turning left or going straight (Alrutz et al., 2009; Johnson et al., 2011; Twaddle & Busch, 2019). This is particularly the case for cyclists turning right on a cycle track off the carriageway compared to cycle lanes in the carriageway or mixed traffic provisions (Schleinitz et al., 2019). Red light running rates are lower at more complex intersections (Schleinitz et al., 2019) and higher at lower carriageway widths (Twaddle & Busch, 2019; Van der Meel, 2013). In terms of signalling, red light running is consistently higher with longer red light phases and waiting times (Lv et al., 2022; Pai & Jou, 2014; Van der Meel, 2013). Red light running rates are highest at the beginning and at the end of red light phases (Bai & Sze, 2020; Twaddle & Busch, 2019; Wu et al., 2012). De Ceunynck et al. (2016), as one of the very few studies on RTOR for cyclists, found a spillover effect in their online-survey. Respondents with a high awareness of RTOR for cyclists reported that they are more likely to turn right on red at intersections without RTOR.

In terms of compliance with design, various types of changes in behaviour have been identified when cyclists do not feel safe or comfortable on their provision. These include switching to the pavement (Schleinitz et al., 2019) as well as changes of speed and trajectories (Kazemzadeh et al., 2023; Zhang et al., 2023). Ihlström et al. (2021) examined the practice and underlying motives of cycling on the pavement. They identified three main reasons for this practice: avoiding close proximity to cars, increasing smoothness of the ride and unclear infrastructure design.

Studies investigating conflicts between pedestrians and cyclists usually combine behavioural-based conflict measures such as speed, lateral overtaking distances or evasive actions with the typical SMoS that are also used for conflicts involving motorised vehicles (Beitel et al., 2018; Hosford et al., 2020; Zhang et al., 2023). For example, Liang et al. (2021) found a positive relationship between the volumes of cyclists and pedestrians and the number of conflicts (measured as combination of TTC and PET) in shared spaces, with swerving being the most common evasive action.

Overall, the applied pilot studies show tendencies for the specific effects of RTOR on cyclists, and the general literature on cyclists' compliance with operation and design provides background knowledge on behavioural mechanisms with relevance for RTOR. At the same time, the literature review highlights the need for more comprehensive empirical evidence on the effects of RTOR on cyclist behaviour and its implications for tailor-made study designs.

Cycle provisions at signalised intersections differ greatly and might be different in the approach and the exit leg of an intersection. The literature review shows that cyclist compliance varies depending on the type of cycle provision, so a detailed approach was required that distinguished between the different possible types of cycle provisions and between potential conflict points.Table 3 shows the final typification, distinguishing between nine types of study sites.

In the first step, participating cities were selected based on their willingness to cooperate as well as population size, spatial distribution in Germany, modal share of cycling and topography. The sample of nine cities (Bamberg, Darmstadt, Düsseldorf, Cologne, Leipzig, Munich, Münster, Reutlingen and Stuttgart) ranges from major cities (> 500 000 inhabitants) to large cities (> 100 000 inhabitants). No medium-sized or small cities could be convinced to participate.

In the next step, the nine city partners made suggestions for potential study sites based on pre-defined criteria which included the legal exclusion criteria for the general RTOR as introduced in section 3 and, in addition, high crash numbers, recent or planned reconstruction, shared facilities for pedestrians and cyclists on the pavement, missing stop line for right-turning cyclists and waiting pockets for two-step left-turning cyclists in the right turn path. Cyclist volumes should be significant to generate sufficient numbers of observations for on-site video recording. All potential sites were initially assessed and suitable sites were visited on-site to confirm their suitability in terms of the quality of the surface and width of facilities, sight lines, potential bicycle traffic volumes and the possibility of mounting cameras. Based on these considerations, the final set of study sites was selected.

The nine city administrations proposed a total of 112 sites to be considered for the study resulting in a sample of 48 study sites that were, based on the assessment, finally selected for the before video recordings. After the before video recordings had been completed, the signposting took place. At five of the 48 sites, the RTOR sign was not mounted (due to space or time constraints) resulting in a reduced final sample of 43 sites with the complete before and after data collection. The total of 43 study sites were classified as shown in Table 3. One site in this study is defined as one corner of an intersection, which means that 43 corners were investigated. 27 of these sites are single sites where only one corner of the intersection is regulated with RTOR, six intersections have two corners with RTOR and one intersection is fully equipped with RTOR at four corners.

The before-video recordings took place in September and October 2018. Video data was recorded at each site for at least one day with two cameras, one filming the approach and one filming the exit of the right-turn relation to make sure that the status of the traffic lights as well as interactions and compliance could be observed reliably.

The RTOR traffic sign was subsequently mounted at the sites in early 2019 based on a special legal approval for the pilot study by the local road traffic authorities. The signposting was accompanied by press releases in local and national media. There was no additional awareness campaign on the streets.

The after-video recordings followed from May to August 2019 so that a period of familiarisation with the new RTOR regulation of at least three months was guaranteed at all sites. Video data was recorded similarly to the before observation with the same camera positions and duration of recording. All videos were recorded on working days excluding local holidays.

Video data was analysed for three hours in the afternoon (including PM peak hour) for each site in both the before and after periods. Data was analysed manually by instructed surveyors with the same methodology for before and after observations. Direction of travel and arrival at green or red light was coded for all cyclists approaching the intersection. For arrivals at red light, information was added on whether cyclists turned on red without stopping, on red with stopping or waited for green. This data was used to analyse cyclists' compliance with traffic rules. In addition, the facilities used by cyclists approaching and exiting the intersection were coded, including changing from the carriageway to the pavement and situations where cyclists dismounted and pushed their bicycle.

Volumes of right-turning cyclists and conflicting traffic were counted for the before and after periods and included as exposure in the analysis.

Interactions of cyclists with other users were determined for six pre-defined conflict points as shown in Figure 2. Cyclists' behaviour and their interactions were analysed along the entire right turning movement from approaching to exiting the intersection, resulting in no or several interactions of one or more types per cyclist.

For each of these conflict points, interactions were counted and their criticality was assessed in a two-step approach: In the first step, four levels of interaction were determined by assessing cyclist compliance with traffic rules using the decision tree shown in Figure 3. This assessment scheme has been used before in Maier et al. (2015). Interactions of level 0 (full control) describe the interaction of two users whose routes cross each other, but whose compliance with traffic regulations creates a clear traffic situation. Interactions of level 1 (unobstructed) and level 2 (obstructed) require a reaction, and, thanks to this reaction, do not lead to a collision. In interactions of level 3 (collision), there is no or an insufficient reaction, leading to a collision. Level 2 and level 3 interactions can be considered as critical interactions because users who have right of way are affected by users who do not comply with the traffic rules.

A reaction was defined as braking, stopping, accelerating, swerving or other visible changes in behaviour. The definition of priority differs between the conflict points in Figure 2: at Crossing 1, Crossing from the left and Crossing 2, right-turning cyclists must give way to crossing users. At the Approach, users standing and waiting are considered as having priority. A cyclist overtaking a cyclist on the left or a motorised vehicle on the right-hand side at a distance of at least one metre is not violating traffic rules. For Turning from the opposite, right-turning cyclists have priority over the user turning left. Pedestrians on the Pavement always have priority over cyclists riding on the Pavement.

Interaction rates were computed to allow comparisons between the study sites and periods with different traffic volumes. They were calculated as the multiplied ratio of right-turning cyclists and the conflicting flow for each conflict point and user group with the following basic equation:

where:

IR = interaction rate;V_RTC = volume of right-turning cyclists;V_CF = volume of conflicting flow.

The Pavement is the only conflict point for which the volume of the conflicting flow was not determined because the potential conflicting users are so diverse in behaviour (standing, playing, chatting on the pavement) and positioning (very close or very far to the traffic light) that no definition for the basic population of these users could be found and therefore they were not quantified.

In the second step, the Post-Encroachment-Time (PET) was computed to assess the criticality of the interactions in terms of the time elapsed between a user leaving a conflict point and the arrival of the conflict partner at that point (Zheng et al., 2021). PET is typically used in conflict studies on vulnerable road users (Liang et al., 2021) and was chosen here because it can be determined even if the conflict parties are not on a collision course. Time to Collision (TTC) was therefore not considered.

PET can only be recorded for interactions with crossing trajectories and was therefore only measured for the conflict points Crossing 1, Crossing from the left and Crossing 2. The PET values are interpreted as the cumulated frequencies, because an interpretation based on thresholds was not considered appropriate. PET values for interactions of levels 0 (full control) to 2 (obstructed) according to Figure 2 were included in the analysis because full control interactions might also have a small PET value. PET values larger than five seconds were not considered because higher values were assessed to be regular and non-critical interactions (Beitel et al., 2018; Liang et al., 2021). Interactions measured with PET were differentiated according to whether the right-turning cyclist is the first or the second to leave the conflict area.

The final set of criteria for the analysis includes the compliance with traffic rules, interaction numbers and rates, levels of interaction in terms of compliance with the traffic rules and PET as a SMoS measuring the criticality of an interaction. Each indicator is analysed and interpreted individually to create a detailed understanding of cyclists' changes in behaviour and their implications resulting from the introduction of RTOR for cyclists at the 43 study sites. The results are presented as average effects across all study sites, weighted by the number of sites per type of cycle provision as introduced in Table 3. Differences between the individual types are described in the text, if these are found to be significant.

Table 4 gives an overview of cyclists' compliance with traffic regulation in the before and after observations. 51% of the observed right-turning cyclists arrive at green in the before and 46% in the after observation and are therefore not affected by the RTOR regulation. Of the cyclists arriving at red light, 80% turn right on red before and 92% after the introduction of RTOR which corresponds to an increase of 15%. Five percent of the cyclists turning right on red in the before period and six percent in the after period stop before turning. For cyclists turning right on red, the share of cyclists using the pavement instead of the cycle provision (mixed traffic, cycle lane or cycle track) decreases from 40% to 30% with the introduction of RTOR.

Looking in at the changes in compliance for the different cycle provisions in more detail, the share of cyclists turning right on red is highest at 97% at sites with a cycle track in the approach and any cycle facility (either track or lane; CT–CL and CT–CT in Table 3) in the exit with hardly any change in the after period (99%). The largest change in the proportion of cyclists running red lights, from 49% in the before to 86% in the after period, is observed at sites with mixed traffic provision in the approach and the exit (MT–MT in Table 3). No differences for the share of cyclists who stop before turning right on red are identified for the different types of cycle provisions, these shares are consistently low at all sites and in both periods.

For cyclists turning right on red, the proportion of cyclists using the pavement is highest at around 80% (both before and after) for sites with a cycle track in the approach and any type of cycle facility in the exit (CT–CL, CT–CT in Table 3), when cyclists are already at pavement level and do not have to climb a kerb to access the pavement. The share of cyclists riding on the pavement in the before period is lowest for mixed traffic provisions in the approach at 29% (MT–MT, MT–CL, MT–CP in Table 3). The largest change in behaviour was found for sites with mixed traffic both in the approach and exit (MT–MT in Table 3) where the share of cyclists on the pavement decreased from 27% in the before to eight percent in the after observation.

In addition to the cyclists as listed in Table 4, we observed persons that reach the traffic light at red as a cyclist, dismount before the stop line, step onto the pavement and push their bicycle to continue cycling after having left the intersection. This behaviour occurred 26 times in the before and 10 times in the after period which are low numbers but still a substantial decrease. The impact of this specific user behaviour on interactions is assumed to be low due to the small numbers. However, the comfort might improve for cyclists when they stop changing to the pavement with the new RTOR regulation and instead complete their right-turn manoeuvre as cyclists on the carriageway. Persons dismounting and pushing their bicycle are not considered in further analysis.

About 90% of cyclists arriving on red light and cycling straight ride on green light with no significant change between the before and after periods. Differences between the different types of cycle provisions are generally low for this group and are therefore not interpreted.

Concerning interactions involving cyclists turning right on red, the total number of observed interactions is generally low. For 543 cyclists turning right on red before introduction of RTOR, 411 interactions are recorded. For 881 cyclists turning right on red with RTOR, 811 interactions are identified. This is on average less than one interaction per cyclist (0.75 interactions per cyclist before; 0.92 interactions per cyclist after) and less than 20 interactions per site across all 43 study sites.

Table 5 shows the total numbers of interactions by conflict points and levels. In both periods, interactions in the Approach have the highest share followed by interactions with users on Crossing 1 and Crossing from the left. Interactions on Crossing 2 and on the Pavement are relatively rare. The observation of interactions with users Turning from the opposite is with 6 (before) resp. 12 (after) so low in numbers, that they do not allow comparisons of behaviour in the before and after observations. Interactions at this conflict point are therefore not considered further.

Table 5

Observation		Before		After
		abs.	rel. [%]	abs.	rel. [%]
Conflict points	TOTAL	411	100	811	100
	① Approach	126	31	360	44
	② Crossing 1	95	23	140	17
	③ Crossing from the left	99	24	209	26
	④ Turning from the opposite	6	1	12	1
	⑤ Crossing 2	30	7	55	7
	⑥ Pavement	55	13	35	4
Level of interaction	TOTAL (④ excluded)	405	100	799	100
	Level 0 (full control)	80	20	180	23
	Level 1 (unobstructed)	289	71	561	70
	Level 2 (obstructed)	36	9	58	7
	Level 3 (collision)	0	0	0	0

The total number of interactions is divided by the volumes of the different conflicting flows to calculate the interaction rates which are shown in Figure 4 by conflict point and user group. Unlike the interaction numbers, the interaction rates are approximately in the same magnitude for all conflict points, showing the high relevance of the exposure for interaction numbers. The interaction rates increase between the before and after observations for each conflict point except the Pavement where only the volumes of right-turning cyclists are considered.

In the Approach, the majority of interactions occur with other cyclists in both periods and, there is also a substantial increase in cyclist-cyclist interaction rates from before to after observations. A qualitative investigation of selected single sites with particularly high or increased interaction rates with the introduction of RTOR shows that the interactions increase most at sites with mixed traffic provision for cyclists, bike boxes and when traffic volumes increase between before and after periods. Interaction rates with users Crossing from the left look similar in magnitudes and involved user groups to interaction rates in the Approach. They are also observed mainly with other cyclists and also increase from the before to the after period. Interaction rates cyclist-motorist hardly change at both conflict points (Approach, Crossing from the left). The qualitative investigation of selected sites shows the high relevance of sufficiently long sight lines and traffic volumes particularly for interactions with users Crossing from the left.

Interaction rates of cyclists with pedestrians at the crossings (Crossing 1 and Crossing 2) are approximately the same in the before and after periods, but cyclist-cyclist interaction rates increase substantially in the after period. Interaction rates are higher at Crossing 1 than on Crossing 2 which is plausible seeing that crossing road users at Crossing 2 usually have a red light when cyclists turning right on red arrive at Crossing 2. Interactions at this conflict point can only occur when users cross at red or in the time when one phase changes to the next.

Interactions on the Pavement occur less frequently after the introduction of RTOR because fewer cyclists ride on the pavement.

Figure 4

The shares of levels of interaction in Table 5 only change marginally between the before and after observations. Interactions of level 1 (unobstructed) have the highest shares at around 70%, followed by interactions of level 0 (full control) at around 20% in before and after periods. Interactions of level 0 (full control) are not possible for cyclists turning on red light in the before period at all crossing interactions because cyclists who ride on red automatically violate traffic rules. The share of level 2 (obstructed) interactions is low in the before and after periods and no interactions of level 3 (collision) are identified.

Figure 5 presents the shares of the level of interactions for each conflict point. In the Approach, the share of level 0 (full control) interactions is at 65% in the before period and therefore by far the highest compared to all other conflict points. This is because passing other users that wait at the traffic light is the only possible movement without violating traffic rules before the introduction of RTOR because users at that stage (in the Approach) did not run a red light. Additionally, cyclists seem to pass other users waiting at the traffic light mainly without violating traffic rules (e.g., overtaking on the wrong side and/or with insufficient distance). The share of level 0 (full control) interactions decreases substantially to 27% in the after period. This reduction shifts to a share of 61% of level 1 (unobstructed) interactions in the after period. The share of level 2 (obstructed) interactions is generally highest in the Approach and increases slightly from the before to the after period.

The proportion of levels are similar in magnitudes at Crossing 1 and Crossing from the left in both the before and after period with increasing proportions of interactions at level 0 (full control) from before to after observations. There are higher interaction rates in the after period at these conflict points but a higher share of those interactions happen in a controlled way (level 0). The increasing shares of level 2 (obstructed) interactions at Crossing 1 point to a different direction and show more critical interactions in the after period but the overall shares of level 2 (obstructed) interactions at this conflict point are generally low with seven (before) and nine (after) percent.

The major share of interactions is at level 1 (unobstructed) on Crossing 2 and on the Pavement. Level 2 (obstructed) interactions decrease at both these conflict points from the before to the after periods so that overall, RTOR seems to lead to an improvement concerning levels of interaction at these points.

The presentation of the cumulative frequencies of the PET values in Figure 6 shows the total number of observations, which is 200 in the before period and 329 in the after period. The number of PET values is smaller than the total number of interactions, because only PET values of less than five seconds are considered and because PET could not be recorded for all interactions (e.g. due to obstructed vision).

The overall variance of PET values is higher and differences between the before and after observations are more distinct in the left panel of Figure 6. Variance is mainly skewed to the left. This means that PET values without RTOR are significantly lower than with RTOR when right-turning cyclists are the first to leave the conflict area. This holds particularly for smaller PET values below two seconds and for the conflict point Crossing from the left. For Crossing 2, the low number of cases needs to be considered and only allows cautious interpretation. One possible reason for the higher PET values with RTOR might be that cyclists feel more comfortable and confident and therefore complete their manoeuvre to turn right on red in a less risky way. The low PET values at conflict point Crossing from the left are plausible as the speed of users Crossing from the left (motorists and cyclists) can be assumed to be higher than on Crossing 1 and Crossing 2 (only pedestrians and cyclists).