Manuals for analysis
Road safety research frequently deals with complex data structures, e.g. the analysis of linked data such as the combination of accident data with exposure data and/or safety performance indicators. The assumption of independence of observations that underlies standard statistical inference analysis, is not valid anymore when data are either hierarchically structured (e.g. with accident data, data on drivers and passengers in vehicles are nested within vehicles, vehicle data are nested within accidents, and accident data nested in regions), or time dependent (e.g. development of accidents over time). Thus complex datasets pose road safety researcher for the challenge of how to overcome problems caused by data dependencies.
When road safety data do not contain fully independent observations, special models are needed to analyse more complex data structures. One of the main objectives of SafetyNet is to provide the road safety researcher with best practice advice for the analysis of these complex data structures. The best practice for analysis of complex road safety data structures is provided in two inter-connected reports:
These two reports provide best advise on how to chose and apply statistical techniques that are part of two broad families of data analysis: Multilevel Modelling which is dedicated to the analysis of hierarchically structured data, and Time Series analysis which is dedicated to overcoming dependency-issues in time-related data.
The report Multilevel modelling and time series analysis in traffic safety research – Methodology gives the theoretical background for these two families of analysis, it describes for each technique objectives, model formulation, and assumptions, and it illustrates the technique with an empirical example relevant to road safety research. Attention is being paid to the following techniques: multilevel linear regression models, discrete response models, repeated measures data, multivariate models, factor analysis, structural equations models, linear and non-linear regression models for time series analysis, ARMA and ARIMA time series models, DRAG time series models, and state space time series models.
The report Multilevel modelling and time series analysis in traffic safety research – Manual contains the practical instructions to carry out the type of analyses that are described in the methodology report. The Manual presents instructions to fit each model on the basis of user-friendly software, as well as guidelines for interpreting the results. The Manual is not a stand- alone document, since it presupposes that the reader is cognizant of the contents of the methodology report. To allow an easy match between the Manual is the same as that of the methodology report, the two reports have the same numbering.
|