Since today our paper on spatial patterns and temporal dynamics of urban bicycle crashes is available online. It’s published in the Journal of Transport Geography and can be accessed via the publisher’s website .
The main argument of the paper is that spatial patterns and temporal dynamics on the city-scale level are relevant for any further analysis. An explorative analysis prior to an explanatory approach helps to focus on relevant regions and time intervals. Thus the demand for exposure data and the extent of further in-depth investigations can be cut down tremendously. In order to facilitate the environment for detecting patterns and dynamics we designed a multi-staged workflow with the delineation of spatial reference units and a measure of relative difference between regions and time intervals at its core.
Here are three aspects of the study, which might be of interest in connection with the paper:
For our study we used authoritative data from the city administration. This crash data base is fed by police reports and cleaned and managed by the responsible department. Although the quality of the data and the level of detail is comparable high, the data suffer from severe underreporting. It is known from studies (e.g. Schepers et al. (2014) ) that the share of singe-bike-crashes is much higher than reflected in common crash statistics. Additionally there is a relation between probability of reporting and crash and the severity of material or human damage, according to a recent OECD report . Consequently the conclusions drawn from any study based on incomplete data need to be handled with great care. Of course, the same holds true for this study.
2) Exposure data and risk calculation
It was quite a challenge to provide sufficient and unambiguous arguments for our idea to focus exclusively on absolute numbers of crashes. We did not aim to provide risk estimations but to develop a workflow for the detection and description of spatial and temporal variabilities of crash frequencies. Apart from conceptual reasons for this strategy, adequate exposure data were simply not available at the time of this research. Meanwhile we’ve developed first prototypes of bicycle flow models that cover a whole city on the finest scale (see Wallentin & Loidl (2015) ) and will publish first results of risk calculations on the local scale soon.
Results of this study suffer from some limitations, which can be also regarded as issues on the research agenda. From a methodologically point of view, I’d say that the yet unsolved modifiable areal unit problem (MAUP) is the most serious limitation of this and every other study that relies on fixed spatial reference units. The main argument for not defining reference units data-driven, e.g. through cluster detection analysis or kernel density estimation (KDE), is comparability over time. It would be impossible to grasp temporal effects, such as the seasonality described in the paper, without fixed spatial reference units.
In the meantime we’ve made significant steps forward to tackle aspect 2) and 3). Apart from methodological progress, I hope that our research contributes to a better understanding of where and when crashes occur. As far as I can oversee the literature on bicycle crash analysis, there is a large body of literature about causal factors (one of the most extensive study in this respect is Teschke et al. (2012) ). However, the spatial character of bicycle crashes is often neglected or only regarded as bias for statistics. As bicycling takes place in physical space and thus is a spatial phenomenon by its very nature, I’m convinced that spatial (and of course temporal) analysis can be a small, but yet important piece of information that contributes to the whole mosaic of better understanding bicycle safety. Based on this evidence, adequate, targeted measures can be taken in order to improve the situation for vulnerable road users.