Relationship between Demerit Points Accrual and Crash Involvement

Monash University Accident Research Centre - Report #116 - 1997

Authors: K. Diamantopoulou, M. Cameron, D. Dyte & W. Harrison

Abstract:

The objective of this study was to consider whether a driver's prior demerit points data (or their offence data used in other ways) could be used (in combination with other variables) to improve the prediction of the driver's subsequent crashes. To achieve this aim, multivariate statistical models were developed to represent drivers' crash involvements during 1993-1994 as a function of potential predictor variables measured during 1991-1992.

The initial multivariate model fitted included all available driver and licence variables (ie. driver age, driver sex, driver location and endorsement of licence) as well as the driver's reported casualty crash and serious injury involvements, and total traffic convictions during 1991-1992. Adding a driver's prior offences (whether as demerit point levels or categories of offence) into this model produced the two models with the best predictive ability in identifying drivers with future crash-involvements. For both these models, the proportion of drivers amongst the 500 highest scoring drivers who were subsequently involved in 1993-1994 crashes was 12.4%, ie. considerably greater than the 7.3% crash-involvement rate for the top 500 drivers identified by the initial model which did not use prior offence data. The corresponding crash-involvement rate for all drivers in the database was 0.76%.

The research on the relationship between demerit points accrual and subsequent crash-involvement has shown that inclusion of a driver's prior offences (whether as demerit point levels or categories of offence) in a multivariate model adds to the predictive ability of that model in identifying drivers with subsequent crash-involvements. The more efficient model uses a driver's prior offences classified into demerit point levels. Demerit points alone can be used to predict a driver's subsequent crash involvement, but an even better model can be produced by including prior casualty crash involvements as well.

Executive Summary

The Victorian Parliamentary Road Safety Committee recommended that an analysis be undertaken of the VicRoads driver and accident databases to determine the relationship between driver accident involvement, demerit points accrual and other relevant factors as a means of identifying groups that have a significantly higher than average accident potential. The Monash University Accident Research Centre (MUARC) was contracted by VicRoads to determine both the nature and strength of the relationship between prior offence history and subsequent road accident involvement of drivers in Victoria. While the focus of the Parliamentary Committee's recommendation was on the relationship between demerit points and subsequent crashes, the study considered all available potential predictor variables such as the full range of offence information available, prior crash involvements, and driver characteristics such as age, sex, residential postcode and licence type.

The main objective of the study was to consider whether a driver's prior demerit points data (or their offence data used in other ways) could be used (in combination with other variables) to improve the prediction of the driver's subsequent crashes. To achieve this aim, multivariate statistical models were developed to represent drivers' crash involvements during 1993-1994 as a function of potential predictor variables measured during 1991-1992.

The initial multivariate model fitted was the base model (Model 1). This model included all available driver and licence variables (driver age, driver sex, driver location and endorsement of licence) as well as the driver's reported casualty crash and serious injury involvements, and total traffic convictions* during 1991-1992. To this model the driver's prior offence** data (in various forms) was added with the aim of finding a model which best predicted the driver's crash involvements during 1993-1994. Addition of the offence data resulted in the estimation of the following models:

Model 2: base model + total number of offences during 1991-1992;
Model 3: base model + number of offences during 1991-1992 by category of offence;
Model 4: base model + number of offences during 1991-1992 by demerit point level of offence;
Model 5: base model + total number of demerit points incurred during 1991-1992.

The above models were all superior to the base model with high statistical significance (p<0.0001). Hence, addition of a driver's prior offence data (in some form) adds to the ability of a model in predicting subsequent casualty crash-involvements.

The two models which included offence categories that reflected their relative importance in predicting drivers' crash involvements during 1993-1994 (ie. Model 3, the "category of offence" model, and Model 4, the "demerit point level of offence" model) were the best fitting of the five models. Both these models were significantly superior to both the "total offences" and "total demerit points" models (p<0.0004).

In addition to the above models, another model was fitted to the data in which the driver's prior casualty crash and serious injury involvements were omitted. This was found to be a less informative model in terms of predicting subsequent crash-involvements. Thus, although delays occur in obtaining crash data for each driver in Victoria, the ability of the models to predict subsequent crash involvements would be enhanced if such data could be included.

For the best fitting model (ie. Model 4, the "demerit point level of offence" model), the groups with the highest estimated probability of being represented in 1993-1994 crashes were:

young drivers;
male drivers;
drivers endorsed with a truck licence;
Melbourne residents;
drivers with casualty crash and serious injury involvements in 1991-1992;
drivers who had incurred licence convictions in 1991-1992;
drivers with 'four demerit points', 'three demerit points', 'one demerit point', 'six demerit points', and/or 'two demerit points' offences (in that order) in 1991-1992.

The efficiency of the models was measured in terms of their ability to identify the drivers with the highest "accident potential". Each model's efficiency was compared by estimating the relative risk of casualty crash-involvement of the highest scoring drivers to that of all drivers in the database, and by estimating the proportion of "correct positives" (ie. drivers with high scores who were subsequently involved in crashes).

The risk of 1993-1994 crash involvement of the highest scoring 1% of drivers relative to all drivers in the database was greatest (and very similar) for the "category of offence" and "demerit point level of offence" models. For both these models, the highest scoring 1% of drivers had an estimated risk of crash-involvement 4.4 times that of all crashed drivers.

Amongst the 500 highest scoring drivers, the "category of offence" and "demerit point level of offence" models were equally efficient in identifying the crash-involved drivers. For both these models, the proportion of drivers amongst the 500 highest scoring drivers who were subsequently involved in 1993-1994 crashes was 12.4%, ie. considerably greater than the 7.3% crash-involvement rate for the top 500 drivers identified by the base model which did not use prior offence data. The corresponding crash-involvement rate for all drivers in the database was 0.76%.

This research on the relationship between demerit points accrual and subsequent crash-involvement has shown that inclusion of a driver's prior offences (whether as demerit point levels or categories of offence) in a multivariate model adds to the predictive ability of that model in identifying drivers with subsequent crash-involvements. The more efficient model uses a driver's prior offences classified into demerit point levels. Demerit points alone can be used to predict a driver's subsequent crash involvement, but an even better model can be produced by including prior casualty crash involvements as well.

* A traffic conviction is defined to be a licence cancellation, licence suspension or licence disqualification.
** Offences are only those that incur demerit points (eg. exceeding speed limit, running a red-light), and do not include other offences such as drink-driving offences.

Sponsor: VicRoads