WP 7 – Data Analysis and Synthesis
Deliverables D7.2 & D7.3
These two deliverables give an introduction to Multilevel and Time Series analyses. They have been further developed and updated to form D7.4, the best-practice advice for the analysis of complex data structures.
Multilevel modelling and time series analysis in traffic safety research
The data assembled or gathered for the observatory consist of accident data, road safety risk indicators, data on road safety performance indicators and in-depth accident data. Potential users will consequently have to link data from different data-sets, to consider different levels of aggregation jointly, and to analyse the development over time. Together, deliverables D7.4 and D7.5 form a best practice advice for the analysis linked data, dealing with statistical and conceptual issues that come into play when analysing such complex data structures.
The main goal is to enable the reader to deal with data that show dependencies in space (nested data) or in time (time series data). At first, it is demonstrated how such dependencies can compromise the applicability of standard methods of statistical inferences, because they can lead to an underestimation of the standard errors and consequently of the error in statistical tests.
As a solution to this problem, two families of statistical techniques are presented that enable dealing with these dependencies. Multilevel Modelling is dedicated to the analysis of data that are structured hierarchically. It offers the possibility to include hierarchical structures into the model of analysis. In road-safety research, multilevel analyses allow for the introduction of exposure data and of safety performance indicators, even if those are not specified at the same level of disaggregation as the accident data themselves. In this way, multilevel analyses allow a global and detailed approach simultaneously. Time series analyses are employed to overcome dependency issues in time-related data. They allow describing the development over time, relating the accident-occurrences to explanatory factors such as exposure measures or safety-performance indicators (e.g., speeding, seatbelt-use, alcohol, etc), and forecasting the development into the near future.
D7.4 Methodology
This deliverable gives the theoretical background for the two families of analyses, multilevel and time series analysis. For each technique the objectives, detailed model formulation, and assumptions are described and subsequently the technique is illustrated with an empirical example relevant to traffic safety research.
Multilevel modelling and time series analysis in traffic safety research – Methodology
D7.5 Manual
This deliverable contains the manual to support the methodology report in D7.4, where the theoretical background for multilevel and time series analyses is given. For each technique described in the methodology report, this manual presents the instructions to fit the models on the basis of user friendly software, as well as guidelines for interpreting the results. The aim of the manual is to enable the reader to conduct all analyses described in the methodology and this way to get hands on experience in the analysis of road safety data. To enable the reader to track every step presented, the data sets discussed in the various sections are available.
Data files used in the manual
D7.6 Analysis of the Fatal Accident Investigation Database
This deliverable is dedicated to the analysis of a preliminary version (December 2006) of the Fatal Accident Investigation Database. As the database contains fatal accidents only, the analyses are focused on accident severity. In particular, the accident-size, the fatality risk and the reliability of the injury reporting were modelled.
Analysis of the Fatal Accident Investigation Database
D7.7 Multivariate time series analysis of SafetyNet data
This deliverable demonstrates the use of time series analysis techniques. In particular, structural time series models are developed and demonstrated for France and the Netherlands, as well as disaggregated models for two types of networks in France, and disaggregated models for several accident types in the Netherlands. It is demonstrated how road safety developments of the traffic volume, the number of accidents and the number of fatalities can be linked to the developments of exposure, accident risk and accident severity, estimated through their unobserved components: their trend (level and slope) and their seasonals. Some interpretations are given. In addition, the performance of the time series model is compared to the performance of one classical alternative: the vectorial regression model.
Multivariate time series analysis of SafetyNet data
| 
