The turbo roundabout was invented in the 1990s in the Netherlands (Fortuijn, 2009). There are various types of turbo roundabouts, all of which share some common features. A typical design is illustrated in Figure 1. Like conventional roundabouts, turbo roundabouts have a circulatory roadway and give right of way to traffic in the circle. The special characteristic is the lane addition inside the circulatory roadway at one or more two-lane entries, vehicles on the left lane of the entry are then directed to the added lane on the inside. Vehicles on the right lane of the entry have to drive to the continued right lane of the circulatory roadway. Unlike conventional roundabouts with a single circular lane, turbo roundabouts have non-concentric circular lanes and multi-lane sections. Every lane that is added on the inside of the circulatory roadway is subtracted on the outside of the circulatory roadway at any of the following exits.

Although turbo roundabouts are widespread in the Netherlands, they are rare in other countries, with only a few examples in Germany, Poland, the Czech Republic and Austria. The German guidelines for turbo roundabouts (FGSV, 2015) define four entry and four exit types, which are shown in Table 1. Entries are grouped by the number of upstream entry and circulatory roadway lanes. Entry type 2EN-1CR represents a lane addition and is typical for turbo roundabouts. Single-lane entries similar to those at conventional single-lane roundabouts (e.g. 1EN-1CR) can be also found. Exits are grouped by the number of downstream exit and circulatory roadway lanes. A suffix is added to distinguish combinations with the same number of entry and exit lanes. 1EX-1CR-conv has one exit and one circulatory roadway lane. This design is frequently used in single-lane roundabouts and needs to be distinguished from 1EX-1CR-split, where a two-lane roadway splits into two single-lane roadways.

The entry and exit types shown in Table 1 are also common in turbo roundabouts in other countries, with the exception of 1EX-1CR-split. Turbo roundabouts with three-lane entries and corresponding circulatory roadways do not exist in Germany but can be found in the Netherlands.

With these characteristics, turbo roundabouts should combine the advantages of single-lane roundabouts, which are known for their high safety levels (Bondzio et al., 2012; Brilon et al., 1993; Schmotz et al., 2020; Spahn & Bäumler, 2007; Vieten et al., 2010), with the high capacity of multi-lane roundabouts. Safety issues such as lane changes on the circulatory roadway, which occur at multi-lane roundabouts, should not occur at turbo roundabouts, since incoming traffic is already assigned to the correct lane upon entering the turbo roundabout. An example of the reachable destinations by lane for 2EN-1CR is shown in Figure 1. At the same time, turbo roundabouts allow entries and exits with multiple lanes and therefore provide a higher capacity than single-lane roundabouts. The capacity of turbo roundabouts is up to 4 500 vehicles per hour (Brilon & Geppert, 2014) and depends to a large extent on the distribution of traffic flows across entries and exits (Kocianova, 2016). A key aspect of ensuring high safety levels at turbo roundabouts is preventing lane changes on the partially two-lane circulatory roadway, which is why solid markings or physical lane dividers separate the parallel lanes. Speed reducing elements like perpendicular entries and exits are also possible at turbo roundabouts to increase safety levels.

Previous research on the safety of turbo roundabouts mainly focuses on comparing safety levels with those of other types of intersection. This is achieved through before-and-after study designs or by making direct comparisons between intersections with different designs in real-world case studies or simulations (Arun et al., 2021). The safety levels of turbo roundabouts are consistently higher than those of conventional two-lane roundabouts, thanks to the reduction in conflict points and driving speed (Kocianova, 2016). Eliminating the dangerous conflict points at the exits from the inner circular lane and removing weaving (lane changing) from the circulatory roadway are particularly effective in ensuring high safety levels. Compared to two-lane roundabouts, crash risk at turbo roundabouts is reduced by about 70% (D. de Baan, 2009; Hansen & Fortuijn, 2006), similar reductions are found for signalised intersections (Skvain et al., 2017; Vos, 2016), conventional single-lane roundabouts and yield-controlled intersections (Vos, 2016). Studies comparing the safety levels of turbo roundabouts and single-lane roundabouts come to different conclusions. For example, D. de Baan (2009) find higher crash risks for turbo roundabouts, whereas Vasconcelos, Silva, and Seco (2013) find similar safety levels. Comparative studies analysing conflicts mainly use the time-based conflict indicators Time-To-Collision (TTC) and Post-Encroachment Time (PET), and also find advantages for turbo roundabouts (Arun et al., 2021). Bulla Cruz, Lyons, and Darghan (2021), in a simulation study, find 72% fewer conflicts at turbo roundabouts compared to conventional two-lane roundabouts, Shetty, Sauciur, and Pande (2025) find less conflicts compared to signalised intersections. Lower speed at turbo roundabouts is one key factor for their higher safety levels, compared to conventional two-lane roundabouts (Leonardi & Distefano, 2023) and to signalised intersections (Shetty et al., 2025). Macioszek and Kurek (2020) investigate the perceived risk and find lower levels at turbo roundabouts compared to conventional two-lane roundabouts.

Very few studies could be identified that analyse more detailed crash constellations and their determinants at turbo roundabouts. Early studies from Germany find that there are more crashes at the entries than at the circulatory roadway and exits, and that right-of-way and rear-end crashes are the most common crash types. However, these studies rely on very small sample sizes, with some studies examining just one turbo roundabout (Brilon & Harding, 2008; Hantschel & Maier, 2013). These findings are in line with those of studies that analyse conventional single-lane or two-lane roundabouts (Bondzio et al., 2012; Schmotz et al., 2020; Spahn & Bäumler, 2007). Křivda et al. (2020) investigate conflicts and behaviour at 100 turbo roundabouts in Europe and identify the following types of conflict as the most relevant: failure to give way, vehicles becoming intertwined, and drivers taking shortcuts.

Studies investigating the determinants of crashes at turbo roundabouts focus on the impact of various types of lane dividers on the circulating roadway. An exception is the study by Vos (2016), which analyse the influence of giving priority to cyclist at crossings and the design of the inner lane at 204 turbo roundabouts in the Netherlands. Turbo roundabouts that give priority to bicyclists are found to have a higher level of safety compared to those where motorized vehicles have right of way. Although not statistically significant, roundabouts with a rectangular inner lane addition at the circulatory roadway also demonstrate a higher level of safety than those with a smoother, curved design. Macioszek (2015) compare turbo roundabouts with physical and solid line marked lane separation, and find a low proportion of crashes with personal injury for both groups (4% and 7%). Not yielding the right-of-way (33% and 40%), rear-end (23% and 29%) and incorrect lane changes (16% and 19%) are the main cause of crashes in both groups. For turbo roundabouts with physical lane dividers, additional crashes due to inappropriate speed to traffic conditions (25%) have a notable share. By analysing the behaviour in turbo roundabouts with physical lane dividers, Chodur and Bąk (2016) conclude, that physical lane dividers are an effective measure to prevent lane changes at the circulatory roadway. Macioszek (2015) and Kieć et al. (2019) report more vehicle side impact crashes with pavement markings as lane dividers compared to physical raised dividers. Smaller radii reduce the likelihood of crashes in several studies investigating turbo roundabouts (Kieć et al., 2019; Porter et al., 2019) and conventional roundabouts (Spahn & Bäumler, 2007). Spahn and Bäumler (2007) investigate conventional roundabouts and find that those with lighting and tangential entries have lower crash rates, while raised kerbs on the inner traffic island lead to higher crash rates.

The literature review shows that there is a good level of knowledge about the general safety levels of turbo roundabouts compared to signalised intersections and conventional roundabouts. Research gaps have been identified with regard to the effects of the specific turbo roundabout characteristics on safety levels and detailed crash constellations, which have only been considered in studies with very small sample sizes thus far. Previous studies are mainly based on descriptive statistics. Model-based approaches, such as regression analysis, have rarely been applied thus far.

The following aims are formulated for this study in order to address these research gaps: (1) to provide an overview of turbo roundabouts in Germany with their specific characteristics, (2) to investigate how different characteristics affect safety levels, and (3) to identify detailed crash constellations and their determinants.

To achieve these aims, we conduct a thorough research on existing turbo roundabouts in Germany, thereby obtaining a largely complete sample. Data on traffic volumes, infrastructure design and operation, and crashes are collected for all sites in the sample. Next, turbo roundabouts are categorised according to their various characteristics. Crash rates are calculated for each category and detailed crash constellations are identified, based on the standardised crash types and detailed narrative crash reports. Finally, crash prediction models are developed to take into account the interdependencies between the various factors that influence crash occurrence. The detailed data enable the analysis at two levels: the turbo roundabout as a whole, and individual elements, including entries, exits and circulatory roadway.

A three-step approach involving several methods was chosen for the analysis. In the first step, crash rates were calculated for each roundabout and its entries and exits, based on the volume of motorised vehicles at each site. To avoid potential biases caused by lower exposure during the COVID-19 pandemic, only crashes from 2016 to 2019 were considered in this step, resulting in 842 crashes. Pearson correlation matrices were computed for the characteristics of each roundabout and its entries to determine interdependencies between the various characteristics and to assess whether the observed effects of individual variables could possibly be attributed to their relationship with other features. Only significant correlations with coefficients above 0.3 or below -0.3 were taken into account.

In the second step, detailed crash constellations as defined in Table 3 were computed to gain a deeper understanding of crash patterns at the different elements of the turbo roundabouts and the factors that determine them in terms of infrastructure and operation. We tested the relevance of the COVID-19 pandemic for this analysis by comparing the proportions of crash types for all roundabouts in Saxony, Saxony-Anhalt and Hesse in the years 2016 to 2020. We found no significant effect of the lower exposure in 2020 on the distribution of crash types and therefore included crashes from 2020 in this second step, resulting in a total of 1 013 crashes.

In the third step, crash prediction models were developed for entries, as this element got identified as the most relevant in terms of crash numbers. Models were computed for all crashes, as well as for the two most common crash constellations, right-of-way and rear-end, as defined in Table 3. The modelling procedure followed a stepwise approach, starting with an intercept-only model, followed by the addition of exposure variables, and concluding with a full model incorporating all relevant parameters. Initially, poisson regression was applied to assess the individual explanatory power and statistical significance of each variable. To address overdispersion commonly observed in crash count data, final parameter estimation was carried out using negative binomial regression. Only variables with coefficients statistically significant at the 10% level and not highly correlated with each other were included in the final model specification. Correlations between the variables were assessed using the Pearson correlation matrix modelled in step one.

The analysis shows a high diversity in the German turbo roundabouts in our sample. The categories are evenly distributed for most characteristics of the roundabouts, and few correlations could be found between the variables. It seems like each roundabout is a unique one, and that design characteristics are tailored to the specific local requirements. This diversity is probably due to the fact that, prior to the publication of the official guidelines in 2015 (FGSV, 2015), there were no clear recommendations for the design and scope of application of turbo roundabouts in Germany. The picture becomes more nuanced when the individual entries and exits are examined.

As a general tendency, crash rates increase with increasing complexity of the turbo roundabouts and their elements. Traffic volumes are particularly relevant, crash rates at turbo roundabouts with an ADT of >= 26 000 are significantly higher than with lower ADT values. The same holds for pedestrian and cyclist volumes which are considered in this study via the proxy variable “factor for environmental use”. Significantly higher crash rates are found for sites with a higher factor for environmental use of >= 0.9. Crash rates for the “smallest” entry type 1EN-1CR are lowest, they increase with each added lane in the entry and the circulatory roadway, and are highest for the most complex entry type 2EN-2CR. This negative effect of entry complexity on safety is consistent with previous studies (see e.g. Novák et al., 2018). Crash rates at the exits are generally low and all in the same order of magnitude.

The diameter of the roundabout and related variables are particularly interesting. Crash rates are lowest for turbo roundabouts with a diameter of >= 65m, for entries with a deflection by the central island of >= 20m, and for higher distances between entry and exit of >= 10m. Higher diameters of roundabouts tend to increase the speed in the circulatory roadway with adverse effects for safety (Spahn & Bäumler, 2007). At the same time, higher diameters mean a higher deflection by the the central island for approaching vehicles. This supports a reduction of the approaching and entering speed (Hydén & Várhelyi, 2000) and therefore can result in a higher safety level (Al-Marafi et al., 2020; Kim & Choi, 2013). In the balance of these two contrary effects, the positive effects of higher dimensions and deflection outweigh the negative effects in our sample.

Crash rates are higher for turbo roundabouts equipped with lighting, despite the generally well-established positive effect of lighting on safety levels (Elvik, 1995). These results may be attributed to complex interactions between various design features, although few significant correlations could be identified in the data. Furthermore, the lack of detailed information on lighting, such as whether it is in operation throughout the night, its intensity, or whether it covers the approaches, represents a limitation of the analysis.

By definition, right-of-way crashes can only occur at the entries, but they are most relevant for all different types of turbo roundabouts in our sample, regardless of which variable is chosen to categorise the sites. The high share of crashes at the entries is the main reason for this, it has also been found in previous studies investigating roundabouts in general (Mandavilli et al., 2009; Spacek, 2004). The distribution of crash constellations at turbo roundabouts with solid line markings differ from the finding by (Macioszek, 2015); the share of right-of-way crashes are 16% higher, whereas the share of rear-end crashes are 19% lower.

Differences in the proportions of right-of-way crashes at the different entry types show that their design matters, the share is lowest at entries of type 1EN-1CR, which is the “smallest design” and highest for 1EN-2CR, which is more complex. Drivers need to cross the outer lane to get to the inner lane at this entry type.

Lane changing and cutting are the most relevant constellations in the circulatory roadway, independent from the type of lane separation. We need to consider that all of the roundabouts in our sample have marked lane dividers in the form of dashed, single solid and double solid lines; there are no raised dividers. The results of our analysis suggest that drivers do not respect single or double solid lines, resulting in similar crash constellations and rates for all types of markings.

The proportions of crash constellations at the exits are evenly distributed, but different patterns are found for the exit types. The proportion of lane changing crashes is particularly high at exits of type 2EX-1CR and lower at exits of type 1EX-1CR-split. Complexity of the infrastructure seems not only to have an effect on crash rates but also on the distribution of the different crash constellations. No clear patterns could be found for the proportion of crashes with crossing VRUs, this crash constellation just seems to “appear” when VRUs are present. Rear-end crashes have also high proportions, which are highest at the entry type 1EN-1CR. This suggests that crashes at this smallest entry type are being displaced from the roundabout onto the access road. This effect may be more pronounced during periods of high traffic volume.

The crash prediction models for entries confirm the findings from the descriptive analysis. Traffic volumes have a significant effect on crash numbers for all crashes and also the two constellations right-of-way and rear-end. Coefficients are slightly below or, for rear-end crashes above one, which means that there is no safety-in-numbers effect in our sample. Greater distances between the entry and exit reduce the likelihood of crashes, which confirms the descriptive analysis; deflecting drivers seems to be more important than the size of a roundabout for achieving high safety levels. Similarly, larger right corner radii reduce the likelihood of crashes which comes somewhat unexpected, but has also been observed at roundabouts in general (Al-Marafi et al., 2020). Larger corner radii typically reduce vehicle deflection by the central island, potentially allowing higher speed in the approach and entry and, as the consequence, might increase crash risks. One possible explanation for our counter-intuitive result could be that a larger corner radius improves the visibility of oncoming traffic on the circulatory roadway, which may help drivers to better assess gaps in traffic and enter more safely. The other significant variables (one lane with no bypass, number of entries at the roundabout, entry type) are linked to the overall complexity of the roundabout and confirm the descriptive findings that higher complexity of a roundabout increases the likelihood of crashes.

This study uses a largely complete sample of turbo roundabouts in Germany to investigate detailed crash patterns in terms of location at the different roundabout elements and crash constellation. The analysis shows that driving properly through a turbo roundabout is a complex task for drivers. Ideally, turbo roundabouts should be simple in terms of their size, entry types, additional infrastructure characteristics and low traffic volumes. For higher traffic volumes, which need larger and more complex designs, drivers need clear guidance and support to achieve high safety levels. The current German guidelines on turbo roundabouts (FGSV, 2015) allow for multiple and very different individual turbo roundabout designs, a higher degree of standardisation might support higher safety levels. Similar to findings from previous studies investigating roundabouts (Mandavilli et al., 2009; Spacek, 2004), the entries are most critical in terms of crashes also in this study. The early and intuitive guidance of drivers to the correct lane when entering the turbo roundabout should get highest emphasis to reduce all types of crash constellations, and specifically right-of-way, lane changing and cutting crashes. Fish hook pavement markings and corresponding arrows on the signposts in the approach can support drivers, these markings are not common in Germany thus far. Reduced maximum allowed speed and clear sight lines can be also assumed to increase safety levels. VRUs increase complexity, locations with low VRUs volumes are preferable for turbo roundabouts in general, and particularly when the motorised traffic volumes are high. The effect of VRUs volumes was assessed in this study based on the proxy variable “factor for environmental use”, future studies with detailed data on the volumes of pedestrians, cyclists and micro-mobility users would be helpful to better understand these effects.

In terms of infrastructure characteristics, previous studies have paid particular attention to lane dividers. This study finds no significant differences between the various types of pavement marking on the circulatory roadway. Future studies comparing marked and physical dividers would be helpful. Deflection increases safety levels and can be measured using various variables. Both should definitely be considered in the future design guidelines for roundabouts. Standardisation can be also assumed to increase safety levels, when drivers use the turbo roundabout more intuitively, just because of their experience from driving through other turbo roundabouts with similar designs. While standardisation at the level of roundabouts as a whole may be difficult, it is feasible and necessary at the level of individual elements, including entries, exits, and the circulatory roadway. Our sample points to this direction: there is far greater diversity at the level of roundabouts than at the level of individual elements.

Future international studies investigating detailed crash patterns at the level of individual turbo roundabout elements, as examined in this study, would allow for comparisons to be made and demonstrate the transferability of this study’s results. A deeper understanding of the mechanisms that influence safety could be achieved by supplementing crash research with studies on conflicts and safety perceptions, at the level of individual turbo roundabout elements and involving large samples. Our sample is with 25 turbo roundabouts small but largely complete for Germany. We could only compute crash prediction models for the entries, too many exits had no crashes at all. However, the three regression models computed in this study provide valuable insights into the effect of traffic volumes and infrastructure characteristics on crash numbers. To the best of our knowledge, this is the first study to go beyond descriptive statistics and focus on entries rather than turbo roundabouts as a whole. Future studies that compute similar crash prediction models would be valuable in helping to deepen our knowledge of the significance of the different effects. Including additional characteristics such as sight distances, surveyability, lane arrow pavement markings, and directional signage could help identify further parameters affecting safety levels. Further analyses of crashes with crossing pedestrians or cyclists would be also valuable but would need higher sample sizes and detailed exposure data.

Overall, this study confirms earlier research findings that turbo roundabouts effectively combine high capacity with high safety levels. The detailed analysis of crash patterns in this study enables specific recommendations to be made for improving safety even further and exploiting this potential more effectively.