Study on two basic road safety variables about persons involved via specific statistical methods Touati, Abdel ; Lebreton, Patrick ; Vervialle, Francoise
Publication details: Road safety on four continents: Warsaw, Poland 5-7 October 2005. Paper, 2005Description: 11 sSubject(s): Bibl.nr: VTI 2005.0795Location: Abstract: In 2003, road fatalities made up 1.1 % of annual deaths in France. Road insecurity is comparable to a public health problem that leads to the deaths of thousands of people and a high number of injured persons (several tens of thousands) every year. Road safety is a major human issue because of this. Perfecting means of prevention is supported by previous knowledge and various study approaches. This presentation aims to study the two basic road safety variables from the French accident file (BAAC file) and the possible contribution of new statistical methods for perfecting a more reliable file and thus enabling to produce less biased results. The BAAC file fully describes road accident circumstances and is thus an invaluable source of information for road safety surveys. We will hereby cover two methodological problems involving them. We will propose to cover the seriousness of injuries to persons involved in the accident file, as a file with different levels. Its analysis requires using appropriate statistical techniques. Indeed, our goal is to create a model of the seriousness of injuries to persons involved, in order to isolate and especially determine the influence of wearing a seat belt on the seriousness of injuries to passengers involved in a road accident. This 5-year study lasted from January 1999 to December 2003. The correlation between users of the same given vehicle must be taken into account in assessments, in order to avoid skewing the assessment of model parameters. Either multilevel logistical regression or partial least squares - "PLS" regression methods (as well as derived methods), which enable to simultaneously analyse several dependent variables at the same time, can solve these kinds of problems. We will use a derivative of this last method because of the many benefits that it provides. Results obtained are shown as odds ratios. Results are similar to those obtained from other studies of the same subject. The second problem is the lack of reliability of certain sensitive variables such as alcohol. Indeed, the alcohol variable has an unspecified rate of 20 % for non-fatalities and 39% for fatalities. In this second part, we will study the possibilities of discriminant methods of analysis, in order to either assign the "unspecified" mode of the alcohol variable to "alcohol positive" or "alcohol negative", according to accident circumstances. This method has, until now, been rarely studied, apart from a study conducted by the NHTSA (U.S Department of Transportation), when a discriminant analysis method was used to assess missing values. Logistical regression, which is the oldest method, will serve as a standard analysis. It will enable to decide if more recent methods can produce better results.Current library | Call number | Status | Date due | Barcode | |
---|---|---|---|---|---|
Statens väg- och transportforskningsinstitut | Available |
In 2003, road fatalities made up 1.1 % of annual deaths in France. Road insecurity is comparable to a public health problem that leads to the deaths of thousands of people and a high number of injured persons (several tens of thousands) every year. Road safety is a major human issue because of this. Perfecting means of prevention is supported by previous knowledge and various study approaches. This presentation aims to study the two basic road safety variables from the French accident file (BAAC file) and the possible contribution of new statistical methods for perfecting a more reliable file and thus enabling to produce less biased results. The BAAC file fully describes road accident circumstances and is thus an invaluable source of information for road safety surveys. We will hereby cover two methodological problems involving them. We will propose to cover the seriousness of injuries to persons involved in the accident file, as a file with different levels. Its analysis requires using appropriate statistical techniques. Indeed, our goal is to create a model of the seriousness of injuries to persons involved, in order to isolate and especially determine the influence of wearing a seat belt on the seriousness of injuries to passengers involved in a road accident. This 5-year study lasted from January 1999 to December 2003. The correlation between users of the same given vehicle must be taken into account in assessments, in order to avoid skewing the assessment of model parameters. Either multilevel logistical regression or partial least squares - "PLS" regression methods (as well as derived methods), which enable to simultaneously analyse several dependent variables at the same time, can solve these kinds of problems. We will use a derivative of this last method because of the many benefits that it provides. Results obtained are shown as odds ratios. Results are similar to those obtained from other studies of the same subject. The second problem is the lack of reliability of certain sensitive variables such as alcohol. Indeed, the alcohol variable has an unspecified rate of 20 % for non-fatalities and 39% for fatalities. In this second part, we will study the possibilities of discriminant methods of analysis, in order to either assign the "unspecified" mode of the alcohol variable to "alcohol positive" or "alcohol negative", according to accident circumstances. This method has, until now, been rarely studied, apart from a study conducted by the NHTSA (U.S Department of Transportation), when a discriminant analysis method was used to assess missing values. Logistical regression, which is the oldest method, will serve as a standard analysis. It will enable to decide if more recent methods can produce better results.