The feasibility of identifying speeding-related and fatigue-related crashes in Police-reported mass crash data

Monash University Accident Research Centre - Report #197 - 2003

Authors: K. Diamantopoulou, E. Hoareau, P. Oxley and M. Cameron

Abstract:

During 1999, the Transport Accident Commission suggested that Monash University Accident Research Centre undertake a project to examine the possibility of better defining crashes involving speeding and fatigue based on an analysis of Police-reported mass accident data in Victoria. Such data would enable better tracking of the relative involvement of these problem behaviours in crashes, and better assessment of the effectiveness of countermeasures aimed at each specific behaviour in crashes.

A review of the NSW Roads and Traffic Authority's (NSW RTA) procedures to identify speed and fatigue in crashes led to the recommendation that Victoria should not adapt the same procedures. This was due to the procedures' insufficient scientific basis in the derivation of identifying speed-related and fatigue-related crashes, as well as there being too many important differences between Victorian and NSW crash reports. It was therefore proposed to obtain data from recent in-depth crash investigations in which the degree of speeding and/or fatigue was determined objectively.

An in-depth crash study by the NSW RTA was intended to be used to identify fatigue but was deemed unsuitable. Consequently, the study's aim to identify fatigue in mass crash data was not feasible. A second in-depth crash study conducted by the NHMRC Road Accident Research Unit, RARU, of University of Adelaide, used accident reconstruction techniques to determine free-travelling vehicle speeds of crashed vehicles. This study was used in the development of a procedure to identify speed in Victorian Police-reported mass crash data.

The Adelaide RARU data was analysed using General Linear Modelling techniques. Two types of models were developed; one that included only main effects and assumed no interactions between factors, and a second model that allowed for interactions and modelled them explicitly. The significant factors obtained from these models were used to develop indicator functions that estimated the travelling speed of a vehicle prior to a crash. These indicator functions were applied to a subset of the Victorian Police-reported mass crash data to identify speed-related crashes for the period 1984-2000. A subset was selected to match the situation of the RARU data, i.e. urban daytime crashes in 60 km/h speed zones.

Using these indicator functions, average vehicle speeds of crashed drivers and the proportions of crashed drivers with vehicle speeds >75 km/h and >90 km/h were estimated. The aim was to see if there had been any changes in these speed measures on urban 60 km/h Melbourne roads in day-time hours during 1984-2000. The indicator functions could then be used to assess the impact of new speed enforcement and/or publicity programs - by monitoring any changes that occurred in average speeds or in the proportion of crashed drivers with speeds exceeding the speed limit as a response to the new speed initiatives.

Executive Summary

ADOPTING NSW PROCEDURES FOR VICTORIA

The New South Wales Roads and Traffic Authority (NSW RTA) has had procedures for identifying speeding-related and fatigue-related reported crashes for many years, based on the reporting officer's opinion of the presence of the behaviour and whether the vehicle manoeuvred in a way characteristic of the behaviour. These procedures were reviewed as part of this study to determine if they were applicable to the Victorian context. The review of the NSW RTA procedures indicated that the identification of speeding-related and fatigue-related crashes was derived from Police data, rather than being tested on crash data or being independent of the Police crash reports. It was thus questionable whether these procedures had sufficient scientific basis to warrant directly adopting them for the Victorian Police-reported crash data. Therefore it was recommended that Victoria should not adapt the same procedures for use on Victorian Police crash reports because of their insufficient scientific basis, and also because there were too many important differences between Victorian and NSW crash reports.

Because the NSW procedures were determined to be unsuitable, there was a need to investigate whether the identification of speeding-related and fatigue-related crashes directly from mass reported data is feasible at all for Victoria. It was therefore proposed to obtain data from recent in-depth crash investigations in which the degree of speeding and/or fatigue was determined objectively. These data would then be matched with Police crash reports for the same crashes, if possible. This would allow for key Police report data and appropriate functions of this information to be identified that discriminates most effectively between whether the crash involves the behaviour (speeding or fatigue, respectively) or not.

FEASIBILITY OF DISCRIMINATING SPEEDING-RELATED AND FATIGUE-RELATED CRASHES

Two in-depth crash studies, in which the degree of speeding and/or fatigue was thought to have been determined objectively, were identified as being potentially appropriate. One, a study conducted by the New South Wales Roads and Traffic Authority (NSW RTA) was believed to contain crashes where the presence or absence of fatigue had been determined using pre-defined criteria. The other, a study conducted by the NHMRC Road Accident Research Unit, RARU, of the University of Adelaide, determined the free travelling speeds of approximately 150 vehicles involved in crashes.

Fatigue-related crashes

The NSW RTA in-depth crash investigation was designed to examine the relationships between vehicle defects and road safety, and as such, the data gathered was essentially based on vehicle information such as vehicle controls, suspension, steering, tyres and chassis. Behavioural causal factors such as fatigue were not identified. On this basis the NSW RTA data was deemed unsuitable for the identification of fatigue-related crashes. Consequently, the study's objective to identify fatigue in Victoria's mass crash database of Police records was not feasible at this stage.

Speed-related crashes

The second study, conducted by RARU, examined the effect of free travelling speed on the risk of involvement in a casualty crash in a 60 km/h urban speed limit setting in Adelaide. Accident reconstruction techniques were used to determine free-travelling speeds of crashed vehicles. It was deemed feasible to use the data from this study, matched with South Australia Police crash reports, to develop a procedure to identify speed in Victoria mass crash data. The matching process of the South Australia Police crash data with the RARU case study data resulted in a database of 149 crashed vehicles for analysis.

DEVELOPMENT OF MODELS AND INDICATOR FUNCTIONS TO IDENTIFY SPEED IN MASS CRASH DATA

A statistical modelling analysis of the South Australia matched crash database identified a number of factors associated with driver speed, i.e.:

male drivers;
drivers aged 25 years or less;
fatal crashes;
extensively damaged vehicles;
morning crashes;
crashes in which the driver was performing a manoeuvre other than going straight ahead.

The statistical methodology used was General Linear Modelling (GLM). Two types of models were developed; one that included only main effects and assumed no interactions between factors, and a second model that allowed for interactions and modelled them explicitly.

The significant factors that were obtained from the 'main effects' and 'interactions' GLM models (listed above), were used to develop indicator functions that estimated the travelling speed of a vehicle prior to a crash. These indicator functions were then applied to Victorian Police-reported mass crash data to identify speed-related crashes for the period 1984-2000.

To apply the South Australia findings to the Victoria crash data, both data sets had to be compatible. For this to occur, certain conditions had to be selected from the Victoria data set to match the conditions inherent in the South Australia in-depth-crash study. These were selecting only:

crashes that occurred on roads in 60 km/h speed zones;
day-time crashes (i.e. between 08:00 - 20:00, as defined in the South Australia mass crash database);
crashes in urban areas as defined by the Local Government Area (i.e. areas in the Melbourne Statistical division);
drivers;
passenger cars or passenger car-derivatives (i.e., car, station wagon, taxis, utilities, panel vans).

One of the significant factors in identifying speed, 'vehicle damage' was not available for the years 1984-1988 in the Victoria crash data. A related variable 'towed' indicating whether or not the vehicle had been towed after a crash was used as a substitute for the 'vehicle damage' variable. Because of this, two types of indicator functions were developed:

one covering the full data period, 1984-2000, and using the 'towed' variable to estimate vehicle damage,
the other covering the period 1989-2000 in which the 'vehicle damage' variable was readily available.

The following speed measurements were estimated:

average vehicle speeds of crashed drivers per quarter for the periods January 1984-December 2000 and January 1989-December 2000;
the proportion of crashed drivers with estimated vehicle speeds >75 km/h per quarter for January 1984-December 2000 and January 1989-December 2000;
the proportion of crashed drivers with estimated vehicle speeds >90 km/h per quarter for January 1984-December 2000 and January 1989-December 2000.

The aim was to see if there had been any changes over time in average vehicle speeds of crashed drivers or in the proportion of drivers exceeding the speed limit in metropolitan (urban) areas on 60 km/h roads for crashes that occurred during daytime hours.

Initially the Victorian speed camera program occurred primarily in urban areas during low alcohol times of the week. These times of the week mainly correspond with daytime hours. Hence, trends in speed outcomes could then be compared to the Victorian speed enforcement programs to determine whether there had been a change in average speed or in the proportion of crashed drivers exceeding the speed limit as a response to these initiatives.

MAIN FINDINGS

Average vehicle speeds

The indicator functions that were based on the models that used the 'vehicle damage' variable and the 'towed' variable exhibited fairly similar trends in average vehicle speed over the time period 1989 - 2000. This was the case for both the 'main effects' and the 'interactions' models. This suggests that the variable 'towed' is a good estimate of 'vehicle damage' when determining the average vehicle speed of crashed drivers.

The 'main effects' models and the 'interactions' models showed similar trends in the estimated average vehicle speed of crashed drivers during 1984-2000 and 1989-2000. However, the actual estimates of average speed for the 'interactions' models were approximately 7 km/h lower than those estimated by the 'main effects' models. For the 'main effects' models a range of approximately 65 km/h-68 km/h was estimated compared with a range of about 57 km/h-61 km/h for the 'interactions' models.

A previous MUARC study (Rogerson et al, 1994) found that after the introduction of the speed camera program and the launch of TAC mass media speed-related publicity, there had been little change in average speeds in a 60 km/h zones during November 1989 to June 1990. This current study supports this result in the way that average vehicle speeds over time did not vary greatly, ranging from 65 km/h to 68 km/h ('main effects' models) or 57 km/h to 61 km/h ('interactions' models). However, there was a definite decline in average speed from July 1989 to July 1991.

Proportion of vehicle speeds exceeding 75 km/h and 90 km/h

When considering the proportion of crashed drivers with estimated vehicle speeds exceeding 75 km/h, the 'main effects' models produced higher estimated proportions than the 'interactions' models. For the 'main effects' models a range of approximately 15%-25% was estimated compared with a range of 9%-17% for the 'interactions' models. However, for the proportion of crashed drivers with estimated vehicle speeds exceeding 90 km/h, the 'interactions' models gave larger estimates than the 'main effects' models by about 1%.

The 'main effects' model that included the 'towed' variable showed consistently smaller estimates per quarter in the proportion of crashed drivers with vehicle speeds >90 km/h than the corresponding model that used the 'vehicle damage' variable. The proportion of speeds >90 km/h was approximately 0.2%-0.6% less in magnitude for the first model. Also, the latter model was more variable than the former model but fluctuated steadily around approximately 0.6%.

Thus the models that used the 'vehicle damage' variable and the 'towed' variable were not consistent estimators of the proportion of crashed drivers with vehicles speeds >90 km/h. However they do seem to be consistent when estimating the proportion of vehicle speeds exceeding 75 km/h in crashes.

Rogerson et al (1994) found that in response to the introduction of the speed camera program and the launch of the TAC publicity, there was a significant decrease in both the percentage of vehicle speeds exceeding 75 km/h and 90 km/h, in 60 km/h zones from November 1989 to June 1990. This decrease remained through to November 1991. These reductions were also apparent when the speed measure considered in this current study was the proportion of crashed drivers with speeds exceeding 75 km/h, but were less obvious for the corresponding proportions exceeding 90 km/h, possibly because there were too few observations. Thus the indicator functions that were developed to identify speed in mass crash data have been able to detect some changes in the speed of crashed drivers over time, particularly when considering the proportion of crashed drivers with vehicle speeds exceeding 75km/h.

Overall, both the 'main effects' and 'interactions' models exhibited similar trends during the time periods considered in the study, 1984-2000 and 1989-2000. However, the 'main effects' models produced larger estimates of the average vehicle speed of crashed drivers and the proportion of crashed drivers with vehicle speeds exceeding 75 km/h than did the 'interactions' models. Conversely, lower estimates of the proportion of crashed drivers with vehicle speeds exceeding 90 km/h were produced by the 'main effects' models compared with the 'interactions' models.

USEFULNESS OF THE INDICATOR FUNCTIONS

The indicator functions developed as part of this study can be used:

to estimate the pre-crash travelling speeds of vehicles crashing on Melbourne roads in 60 km/h speed zones;
as monitoring tools to assess and/or monitor changes in the average vehicle speed of crashed drivers as well as the proportion of crashed drivers exceeding the speed limit at various levels over a specified time period;
to assess the impact of new speed enforcement programs or new publicity and education campaigns targeted at speeding drivers - this could be achieved by monitoring any changes that occurred in these speed measures during the times when these new speed initiatives were implemented.

LIMITATIONS OF THE STUDY

The indicator functions of speed that were developed are only applicable for crashes that occurred on Melbourne roads in 60 km/h speed zones, predominantly during day time hours. These constraints were placed on the analysis because it was necessary to match the Victorian crash data with the criteria used in the RARU in-depth study. The RARU study examined the effect of free travelling speeds on the risk of casualty crash involvement in 60 km/h urban speed limit setting in Adelaide.

All crashed vehicles that met the above criteria were selected from the VicRoads crash database, not just those with assumed free travelling speeds. Thus stationary vehicles involved in a crash could also potentially have been selected. A refinement of the indicator function would be to exclude vehicles which were stationary at a crash, if possible, and re-estimate the speed of drivers prior to a crash.

RECOMMENDATIONS

It is clear that the planned procedures for this study - particularly obtaining the relevant data to allow the identification of fatigue-related crashes - were not as simple and straightforward as anticipated. It is suggested that the plan to identify fatigue-related crashes (and the expansion of speed-related crashes) be re-examined.

To develop higher quality procedures for defining/indicating the presence of key behaviours in crashes, it may be necessary for MUARC to undertake special investigations of behaviours which were the precursors or precipitators of an adequate sample of crashes or at least to add such behavioural investigations to MUARC's existing crash studies. The data would be matched with Police accident reports and the correlation of the existing data with the presence of the behaviour examined. It may be necessary to recommend that additional data be collected by the Police to more accurately identify whether certain safety-related behaviours are likely to have been present in the crash.

It is also recommended that the indicator functions of speed developed as part of this study be refined to exclude crashed vehicles that may have been stationary at the time of impact.

Sponsoring Organisation: Baseline Research Program - Department of Justice, Transport Accident Commission, Royal Automobile Club of Victoria (RACV) Ltd, VicRoads