Response by Monash University Accident Research Centre to “Re-investigation of the effectiveness of the Victorian Transport Accident Commission's road safety campaigns” (White, Walker, Glonek and Burns, November 2000)

Monash University Accident Research Centre - Report #177 - 2000

Full report in .pdf format [158KB]

Authors: M.H. Cameron & S.V. Newstead

Abstract:

This document is the response by Monash University Accident Research Centre (MUARC) to an investigation of some specific MUARC research. The investigation has been published by the South Australian Department of Transport (White, Walker, Glonek and Burns 2000). It relates principally to MUARC's research on traffic enforcement and road safety advertising in Victoria. From re-analysis of data used in MUARC report no. 74, White et al have concluded that the estimates of crash and financial savings attributable to the TAC-funded countermeasures are not supportable. They have also concluded that the re-analysis has failed to support the claims of MUARC report no. 52 concerning the crash reductions that can be achieved through high levels of TAC-funded road safety television advertising.

In this document, MUARC responds to White et al and reaches the following conclusions:

1. Scientific evaluations conducted by MUARC have shown substantial reductions in road trauma in Victoria due to increased random breath testing using 'booze buses' and the new speed camera program, each supported by TAC advertising.

2. The statistical models of monthly casualty crashes as functions of enforcement, advertising and socio-economic factors, developed in MUARC report no. 52, are sound. They have been tested by MUARC and by White et al and have been found to be satisfactory. White et al's investigations have provided additional evidence of the relationship between the TAC speed-related advertising and crashes.

3. The estimates of the points of diminishing returns of levels of TAC drink-driving and speed-related advertising, originally provided in MUARC report no. 52 based on the statistical models in that report, are sound. The economic analysis of advertising levels, which was based on the coefficients of the advertising variables in the statistical models, has not been questioned.

4. White et al's re-analysis of the data used in MUARC report no. 74 is not relevant to report no. 52 because of important differences in their objectives, the types of crashes analysed, the time periods covered, the treatment of levels of speed-related advertising, the inclusion of car-based random breath testing, and the assumptions made and subsequently tested.

5. White et al's so-called parsimonious three-factor model of crash variations in Victoria was based on data dredging and cannot be considered to be a valid alternative to MUARC's models.

6. White et al's test of the quantitative relationships between crashes and the enforcement and advertising variables is not valid. It is not an adequate test of the presence or absence of quantitative relationships.

Executive Summary

INTRODUCTION

This document is a summary of the response by Monash University Accident Research Centre (MUARC) to an investigation of some specific MUARC research. MUARC welcomes reviews of its work because of the critical importance of many of MUARC's results and conclusions. MUARC endeavours to conduct its research to the highest possible scientific standards, while recognising the need to provide results and advice in a timely manner. When necessary, MUARC always qualifies any of its work which is less than definitive.

The investigation relates principally to MUARC's research on traffic enforcement and road safety advertising in Victoria. MUARC is independent of any policy considerations which favour or disfavour various strategic approaches to the use of these measures to achieve road safety goals. MUARC's aim, within the constraint of the resources available to it, is to provide objective information on the effectiveness and cost-effectiveness of these measures in the broader context of factors affecting road trauma. Scientific principles have always been applied by MUARC to achieve this aim.

THE RE-INVESTIGATION PROCESS

In April 1997, the South Australian Office of Road Safety wrote to MUARC requesting that the data used for MUARC report no. 52 (RN52), Evaluation of Transport Accident Commission road safety television advertising (Cameron, Haworth, Oxley, Newstead and Le 1993), be supplied to the Office so that testing of MUARC's statistical models could be carried out.

During 1998-2000, MUARC was provided with eight draft reports, totalling 889 pages, produced during the investigation by the Office of Road Safety. The first two reports focused on the data used in MUARC report RN52. The third and subsequent reports focused on the data from MUARC report RN74, Modelling of some major factors influencing road trauma trends in Victoria 1989-93 (Newstead, Cameron, Gantzer and Vulcan 1995). 

A paper by White, Walker, Glonek and Burns, Re-investigation of the Effectiveness of the Victorian TAC's Road Safety Campaigns, has been published (White et al 2000a). The final report on the investigation has also been released (White et al 2000b). This document is a summary of MUARC's response to the paper and report. MUARC's response also refers to material in the earlier Office of Road Safety reports. Interested readers are strongly advised to consult MUARC's full response for complete understanding of the many issues involved.

MUARC RESEARCH ON ENFORCEMENT AND ADVERTISING

White et al (2000b) claim that “Although a number of different statistical techniques were employed by MUARC in their earlier evaluations [of enforcement and media campaigns], multiple linear regression became the favoured technique in their later work”.

MUARC has completed thirteen substantial studies involving the analysis of real crash data in relation to Police enforcement, road safety advertising and sometimes additional factors. Details are given in the full response. Five of the MUARC studies (or parts thereof) have used multiple regression time series analysis to link crashes with road safety program measures and other factors. The remaining ten studies and part-studies have all been quasi-experimental evaluations of the impacts of the programs. The quasi-experimental time series evaluations were not critically dependent on the need to develop crash models including all influential factors in order to reach their conclusions. MUARC has used a mixture of the study methods over the last decade.

Multiple regression [time series analysis] has not been the favoured statistical technique used by MUARC in this area. It has not been used in any study which MUARC has regarded as being a scientific evaluation of enforcement and/or advertising and has not needed to qualify the results. Multiple regression time series analysis has been used only in studies either attempting to represent the underlying mechanisms of road safety programs or studies attempting to consolidate previous findings. 

White et al (2000a) claim that “Much of the MUARC research was brought together in a report by Newstead, Cameron, Gantzer and Vulcan (1995)”, ie. RN74, and that this is “representative of the MUARC research that has been influential in shaping government policy on road safety television advertising levels” (White et al 2000b). 

RN74 is not representative of the MUARC evaluation research on enforcement and advertising. RN74 is not even representative of the research in RN52, which may have been influential in government policy. Details of important differences between RN52 and RN74 are given later.
MUARC has also studied the trends in the Victorian annual road toll at the macroscopic level. These studies have indicated that the decrease in the Victorian road toll during 1990 and subsequent years was not consistent with expected trends and was a greater reduction than expected from pre-existing trends. These findings do not constitute an evaluation of the 1990 initiatives, but the findings do provide evidence of a real change in road safety in Victoria during the 1990's.

MUARC CONSULTANCY ADVICE

MUARC has been commissioned to provide road safety policy advice to jurisdictions outside Victoria. On occasions, this advice has been based on MUARC's evaluations of Victorian programs and/or the expert opinion of the nominated consultants. 

In 1996, MUARC was commissioned by the South Australian Office of Road Safety to prepare the report, Possibility of adapting some road safety measures successfully applied in Victoria to South Australia (Vulcan, Cameron, Mullan and Dyte 1996), RN102. The report recommended, inter alia, “that resources be allocated to double the exposure of television advertisements which support the speed camera and random breath testing programs”.

Referring to this recommendation, White et al (2000a,b) state that “The main aim of the project reported here [the re-investigation] is to check the soundness of the advice concerning the effectiveness of high levels of television advertising, through re-analyses of the data originally analysed by Newstead et al (1995)”. 

The consultancy report, RN102, makes no reference to Newstead et al (1995), RN74. Reference is made to RN52, but this was not the only basis on which the consultants advised their recommendation. RN102 included a summary of the RN38 and RN42 evaluations which indicated the substantial reductions in severe crashes associated with the Victorian enforcement and advertising programs aimed at drink-driving and speeding. RN102 also noted that the level of road safety television advertising in Adelaide during 1993-1995 was less than half the level in Melbourne during the same years. 
In all of MUARC's consultancy advice to other jurisdictions related to road safety advertising in support of enforcement, reference has been made to the evaluation studies RN38 and RN42, as well as to RN52. This has been considered appropriate, given that RN52 includes a number of assumptions and qualifications on which it was based. 

IMPORTANT DIFFERENCES BETWEEN RN52 AND RN74

White et al (2000b) state “It is considered that two selected RN74 analyses are representative of the MUARC research that has been influential in shaping government policy on road safety television advertising levels”. It is possible that RN52 may have been influential, if its stated assumptions were accepted by its readers. However, it is not correct that the research in RN74 is representative of that in RN52.

There are important differences between RN52 and RN74 in their objectives, the type of crashes analysed, the time periods covered, the treatment of levels of speed-related advertising, and the inclusion of car-based random breath testing. The economic analysis in RN52 was based on models of all casualty crashes (not just serious casualty crashes), the period 1983-1991 for the LAH 1 casualty crashes (not 1983-1992; this period was used for the HAH crashes), the levels of speed-related advertising were measured in TARPs (not Adstock; this was used only for drink-driving advertising), and the random breath test data included car-based tests (not just the bus-based tests included in the model in RN74).
The aim of RN52 was to provide advice on the effectiveness of TAC advertising; in particular the relationship between the level of advertising investment and reductions in injury costs. It was expected that this urgent study would require assumptions to be made. The modelling of monthly casualty crash variations was based on assumptions about the functional form of the relationships, specific explanatory variables, and the correlations between these variables. RN52 clearly stated the assumptions. The reader was cautioned that assumptions had been made in a total of six places in the report. During the project there was limited diagnostic testing of the models in the time available. However, the tests done and subsequent testing by White et al have found that the models were satisfactory. 
The aim of RN74 was not to evaluate the enforcement and advertising programs, it was to combine the results of evaluations of Victorian road safety programs and information on the influence of two socio-economic factors (unemployment rates and alcohol sales) to assess the contribution of each program/factor to Victorian road trauma trends during the early 1990's. RN74 did not have a specific focus on the advertising component of the Victoria's road safety initiatives, unlike RN52.

The narrower focus of the crashes considered for the analysis in RN74, compared with RN52, is the most critical difference between these two studies. White et al (2000b) state that “It is considered unlikely that the main points to be made in this report would be different if any if any other MUARC analyses had been selected for re-investigation”. MUARC's response is that it is not unlikely, if RN52 had been the focus, contrary to White et al's (2000b) opinion.

DATA DREDGING

White et al (2000a) claim that “The methods employed for the selection of variables in the MUARC modelling process can be described as 'data dredging' ”. To support this claim, White et al (2000b) quote principally from RN29, Linking economic activity, road safety countermeasures and other factors with the Victorian road toll (July 1992).

RN29 was MUARC's initial analysis of factors linked to road trauma trends in Victoria. It examined relationships with monthly fatalities only and considered the potential explanatory variables within a conceptual framework used to minimise any spurious selection due to chance. RN29 was not considered to be a scientific evaluation of the factors. It did suggest important influential factors, such as those representing economic conditions, which needed to be taken into account in MUARC's subsequent evaluations of Victoria's road safety initiatives.

The quasi-experimental time series evaluations of the RBT 'booze bus' initiative (RN38) and the increased speed cameras (RN42), with the supporting publicity in each case, needed to take into account the different trends in vehicle travel and/or unemployment rates in Victoria and NSW to ensure the integrity of the evaluation design. Unemployment rate was chosen for non-arbitrary reasons associated with the specificity of the data and theories about a causal role. Separate rates were available for both the metropolitan and rural areas of Victoria and NSW. Unemployment rate was also considered to represent variations in discretionary, higher-risk travel. Details are given in MUARC's full response.

The multiple regression time series analyses (RN52 and RN74) continued to use unemployment rates as an explanatory factor without consideration of alternative economic variables. Initially the models used TARPs to represent road safety television advertising as an explanatory factor, but when MUARC became aware of Adstock and its conceptually better basis for a link with awareness and hence potentially with crashes, Adstock became the preferred measure.

This process of evolution of the choice of covariates and explanatory factors for inclusion in MUARC's quasi-experimental evaluations and crash modelling analyses, respectively, did not constitute 'data dredging' as implied by White et al (2000a,b). The factors were chosen with careful attention to avoiding spurious inclusion and on the basis of reasoned consideration of their possible causality and potential explanatory role.

White et al (2000b) have proposed an alternative to unemployment rate as a measure of economic activity (outlined below). They have apparently considered a range of indicators for this role, with a range of lags, during their investigation. In their 1999 re-investigation reports they proposed that monthly unemployment rate, brought forward by 12 months, would be adequate in providing the sole explanation for the trends in serious casualty crashes during 1983-92.

MUARC does not consider it is appropriate to use 'data dredging' to select a factor to explain the variation in a road crash data series. If sufficient factors, with a variety of leads and lags, are considered the analyst is almost certain to find a factor which explains the series well. However this approach runs the danger that the selected factor has no causal basis for the explanation, and that the apparent relationship is spurious. The use of a conceptual or theoretical model to select the factors for consideration is important to minimise this danger.

WHITE ET AL'S ALTERNATIVE INDEX OF ECONOMIC ACTIVITY

White et al (2000a,b) have proposed the Leading Index (LI) of Economic Indicators for inclusion in the models in RN74 instead of unemployment rate. MUARC questions the process through which White et al have chosen and used this variable as well as the conclusions they reach.

White et al (2000b) present arguments for the choice of LI which include “The peak in the Leading Index occurs at much the same time as the peak in all (ie, LAH + HAH) casualty crashes”. MUARC cautions that the inferences which are made from pair-wise comparisons of time series variables can be very misleading when the relationships between crashes and other factors are truly multivariate. The peak in the crash series may change after adjustment for the influence of other factors.

White et al (2000b) state “[A] reason for selecting this measure was, admittedly, because it peaked at much the same time as the peak in crash numbers. In that respect the measure is a product of 'data dredging', ... ”

White et al (2000b) also state “Because the month-to-month variation in the Leading Index was shown ... to be unrelated to the month-to-month variation in crash numbers, a smoother version of the Leading Index was created by taking the 12-month centred-moving-average (CMA 12).$399.44 These are remarkable statements about the choice and use of an economic indicator to replace the one which MUARC has found satisfactory. MUARC questions how future LI values (up to six months ahead) could be causally related to the crashes in a specific month?

When White et al (2000b) included the CMA 12 Leading Index in the RN74 models, the magnitudes of the estimated coefficients of the enforcement and advertising variables, and their statistical significance levels, were reduced. MUARC considers this to be an inevitable outcome of a process whereby an alternative economic indicator has been chosen on the basis of its coinciding peak and general correlation with the crash series. 

Despite MUARC's concerns, in White et al's (2000b) LAH serious casualty crash model which included the CMA 12 Leading Index, both the speed-related enforcement and advertising variables were statistically significant. These findings are evidence of a link with crashes even after the explanation associated with White et al's CMA 12 Leading Index is taken into account.

White et al (2000b) also considered a model of all (LAH plus HAH) serious casualty crashes, with the CMA 12 Leading Index and the speed-related enforcement and advertising variables as explanatory factors. The speed-related advertising was found to be statistically significant, however the enforcement variable was not.

WHITE ET AL'S THREE-FACTOR MODEL

White et al (2000a,b) have proposed that a three-factor model incorporating (1) linear trend, (2) seasonality, and (3) the CMA 12 Leading Index would be adequate to explain the monthly variations in all Melbourne serious casualty crashes during 1983-1992. They discount the contribution of the speed-related advertising to explaining variations in the same crash series, saying that “speed advertising made only a weak contribution$399.44. In fact, White et al (2000b) had shown that the contribution of the speed-related advertising was highly statistically significant (p = 0.006).

White et al (2000b) have presented the results of fitting an additive model of their three factors to the monthly crashes, rather than a multiplicative model. The multiplicative model form has been used throughout MUARC's multiple regression time series analysis. The additive model cannot be claimed to be more parsimonious than a multiplicative model. White et al (2000b) appear to suggest that an additive model is simple. The multiplicative functional form used by MUARC ensures that the number of crashes predicted by the model cannot be negative. An additive model would not necessarily meet this constraint and could lead to incorrect conclusions about the significance of factors included in the model.

MUARC questions the change in functional form of the crash model for this analysis. If the multiplicative form had been retained by White et al (2000b), their own analysis shows that, had the speed-related advertising been considered for inclusion in the three-factor model, it would have been statistically significant. The three-factor model cannot be described as parsimonious, because it does not take into account at least one factor (ie. speed-related advertising) known also to be associated with crashes. 

TIMING OF TURN-AROUND(S) IN VICTORIAN CRASHES

White et al (2000a,b) state that “Reports from some Victorian and overseas road safety agencies give the clear impression that the agencies believed that the TAC-funded enforcement and advertising campaigns were largely, if not entirely, responsible for halting the increase in crash numbers and for initiating their decline.$399.44

White et al (2000b) state that these “misunderstandings ... are in no way attributable to the work of the MUARC researchers$399.44. The most that MUARC has claimed is that a number of road safety measures and other factors have contributed to the reductions in road trauma in Victoria during 1990 and later years.

White et al (2000a,b) claim that the start of the decline in serious casualty crashes preceded the launch of the speeding and drink-driving enforcement and advertising campaigns, and that this casts doubt on their causal role. This analysis may be valid if a single factor (ie. the relevant enforcement or advertising) had been responsible for the trends in the crashes during 1983-92. However, MUARC has identified many factors which are associated with the trends in crashes during this period. The single-factor comparisons made by White et al (2000a,b) ignore the simultaneous effects the other factors have had on the observed crash series.

RN52 shows that the peak in casualty crashes appears to be in 1989 for both HAH and LAH. However the peak in serious casualty crashes appears to be in 1988. MUARC has offered an explanation for the different trends in serious casualty crashes compared with all casualty crashes (of which serious casualty crashes were about one-third); details are given in MUARC's full response. It is emphasised that MUARC's research has been confined to assessing factors which contributed to the reductions in road trauma in Victoria during the 1990's. MUARC has not evaluated the factors which may have been responsible for the turnaround in crashes of each level of severity prior to 1990.

TESTS OF QUANTITATIVE RELATIONSHIPS

White et al (2000a,b) suggest that RN52 and RN74 claimed that quantitative relationships between TAC advertising levels and crashes had been identified. They present the results of tests based on the data in RN74 which they claim fail to provide evidence of any quantitative relationships. 

The tests involved defining a dichotomous (0, 1) variable for each of the four speeding and drink-driving enforcement and advertising measures and then including each pair (dichotomous and raw variable) together in the original multiple regression analysis of the relevant serious casualty crashes. In each case the formerly statistically significant enforcement and advertising variables (in raw form) became non-significant and the dichotomous variables were also non-significant in every case. White et al (2000a,b) argued that, if the relationships had been truly quantitative, the raw measures should have retained their statistical significance. Failure to do so constituted a lack of evidence of a quantitative relationship, in White et al's opinion.

MUARC does not agree that the test performed by White et al (2000a,b) is an adequate test for the presence or absence of any quantitative relationships. It is known that if two highly correlated variables are included together in a multiple regression, then statistically meaningless results will occur. In their 1999 re-investigation reports, White et al outlined the same tests on the RN74 data. The correlations between the pairs of raw and dichotomised variables were greater than 0.99 in three out of the four cases. They stated “there is little point in even attempting such an analysis if the correlations between the dichotomous and quantitative versions of the same variable are so high that statistical problems of collinearity would necessarily be introduced into the regression analysis. This could be the case for correlations above 0.90 (Tabachnik and Fidell, p. 96) and would almost certainly be the case for correlations of 0.99 or above$399.44.

MUARC is surprised that, although they had clearly recognised the problem of high collinearity in the context of their test, White et al (2000b) decided to proceed with it. MUARC rejects the notion that the test is a valid test of the quantitative relationships. MUARC considers that the presence or absence of quantitative relationships linking crashes with speed-related advertising and with drink-driving enforcement and advertising has not been adequately tested by White et al (2000b).

However, the 1999 re-investigation reports include additional analyses which indicate the presence of quantitative effects of the speed-related advertising. In the reports, White et al drew a distinction between the macro- and micro-level effects of the advertising and the enforcement on crashes. The micro-level effects could be described as quantitative.

White et al analysed LAH serious casualty crashes during the period from August 1990 to December 1992, after the speed camera program commenced full operation. They found negative correlations with the monthly speed camera tickets (p = 0.097) and speed-related Adstock (p = 0.038), the latter being statistically significant. White et al (2000b) dismiss this earlier analysis as not being appropriate and do not report it. Nevertheless, the analysis did find statistically significant evidence of micro-level effects of the speed-related advertising.

In summary, White et al's (2000a,b) test of the presence or absence of quantitative relationships was not considered valid because of high collinearity problems associated with three of the four enforcement and advertising measures under consideration. An alternative analysis by White et al indicated that there is a quantitative relationship between speed-related advertising and crashes.

STARTING MONTH OF DRINK-DRIVING ADVERTISING

White et al (2000b) have criticised MUARC researchers for using November 1989 as the start date of drink-driving advertising for the analysis behind RN52 and RN74, arguing that all other documents have used mid-December 1989 as the start of TAC road safety advertising.

MUARC had non-arbitrary reasons for the inclusion of levels of drink-driving advertising during November 1989 in the data used to develop the statistical models in RN52 and RN74. Details are given in the full response. MUARC's inability to include advertising levels prior to November 1989 was due to the absence of relevant information in useable form. 

MUARC disagrees with White et al (2000b) about the criticality of the November 1989 data to MUARC's modelling results. MUARC did not find this to be the case when it re-analysed the model developed in RN52, which is the only MUARC report relevant to the question of road safety advertising levels.

WHITE ET AL'S CONCLUSIONS AND RECOMMENDATIONS

White et al (2000b) conclude “From the re-analyses of the data of Newstead et al (1995) [RN74] it is concluded that the ... estimates of crash and financial savings attributable to the TAC-funded countermeasures are not supportable$399.44. 

MUARC responds that RN74 is not representative of the MUARC research on enforcement and advertising in Victoria during the 1990's. Scientific evaluation studies, especially RN38 and RN42, have shown substantial reductions in road trauma due to the RBT 'booze buses' and the new speed cameras, each supported by TAC advertising. 

White et al (2000b) also conclude “More particularly, the re-analyses [of RN74 data] have failed to support the claims of Cameron et al (1993) [RN52] concerning the crash reductions that can be achieved through high levels of TAC-funded road safety TV advertising$399.44. 

MUARC responds that the data and analysis in RN74 is not representative of that in RN52. There were important differences in the objectives, data analysed, and assumptions of these two studies. RN52 provided estimates of the point of diminishing returns for TAC advertising, subject to stated assumptions of the analysis, whereas RN74 did not. The assumptions of RN52 have been found to be satisfactory. White et al's findings from the re-analysis of the data in RN74 are not relevant to RN52.

White et al (2000b) recommend that “The developers of an innovative road crash countermeasure should always first consider the possibility of implementing the countermeasure in such a way that it can be evaluated experimentally$399.44. MUARC supports this recommendation very strongly. However this has seldom happened in Victoria and MUARC has been forced to use quasi-experimental designs in its evaluations of the enforcement and advertising programs. MUARC recognises that evaluation on this basis is not ideal, but represents the best available approach in non-experimental settings.

White et al (2000b) also recommend that “MUARC consultancy advice in favour of very high levels of road safety advertising should not be taken into consideration when determining appropriate levels of such advertising$399.44. MUARC re-iterates a number of points with respect to its consultancy advice and the link between TAC road safety advertising and crashes:

* RN38 and RN42 provided scientific evidence of reductions in severe crashes due to:

- the RBT 'booze buses', supported by TAC drink-driving advertising
- the new speed cameras, supported by TAC speed-related advertising

* MUARC's macro-level trend analysis showed that the decrease in the Victorian road toll during 1990 and subsequent years was a greater reduction than that expected from pre-existing trends

* RN52 estimated the points of diminishing returns, under stated assumptions, for levels of drink-driving and speed-related TAC television advertising, respectively

* the assumptions made by MUARC in RN52 have been tested and found to be satisfactory

* the advice provided to the South Australian Office of Road Safety in RN102 made reference to RN38, RN42 and RN52 (but not RN74) and to data which showed that the level of road safety advertising in Adelaide was less than half the level in Melbourne

* the advice was not dependent on the findings in RN52 (had the advice been based on the points of diminishing returns, the recommendation would have been to more than double the level of advertising in South Australia)

* the statistical models of monthly crash variations developed in RN52 were not based on 'data dredging' to find enforcement, advertising and socio-economic variables, which then may have had only spurious relationships with the crashes, to include in the models

* when White et al's (2000b) CMA 12 Leading Index replaced the unemployment rate in the RN74's LAH serious casualty crash model, the speed-related enforcement and advertising variables remained as statistically significant factors explaining the crash variations

* speed-related advertising was a statistically significant factor in White et al's (2000b) CMA 12 Leading Index model of monthly serious casualty crashes during all times of the week 

* White et al's (2000b) test of whether the enforcement and advertising variables in the RN74 models were quantitatively related to the crash variations is not valid and does not establish the absence of quantitative relationships

* White et al's (1999) earlier investigation of the micro-level effects on LAH serious casualty crashes found that the speed-related advertising was statistically significant 

These points indicate that the basis for MUARC's consultancy advice is sound. Based on White et al's investigations, the evidence in favour of the effectiveness of TAC speed-related advertising supporting the Victorian speed camera program has been strengthened. The evidence for the effects of the drink-driving advertising supporting Victoria's RBT program relies on RN38 and on relevant parts of RN52 (which White et al have tested and found satisfactory).

MUARC'S CONCLUSIONS

1. Scientific evaluations conducted by MUARC have shown substantial reductions in road trauma in Victoria due to increased random breath testing using 'booze buses' and the new speed camera program, each supported by TAC advertising. 

2. The statistical models of monthly casualty crashes as functions of enforcement, advertising and socio-economic factors, developed in RN52, are sound. They have been tested by MUARC and by White et al and have been found to be satisfactory. White et al's investigations have provided additional evidence of the relationship between the TAC speed-related advertising and crashes.

3. The estimates of the points of diminishing returns of levels of TAC drink-driving and speed-related advertising, originally provided in RN52 based on the statistical models in that report, are sound. The economic analysis of advertising levels, which was based on the coefficients of the advertising variables in the statistical models, has not been questioned.

4. White et al's re-analysis of the data used in RN74 is not relevant to RN52 because of important differences in their objectives, the types of crashes analysed, the time periods covered, the treatment of levels of speed-related advertising, the inclusion of car-based random breath testing, and the assumptions made and subsequently tested.

5. White et al's so-called parsimonious three-factor model of crash variations in Victoria was based on data dredging and cannot be considered to be a valid alternative to MUARC's models.

6. White et al's test of the quantitative relationships between crashes and the enforcement and advertising variables is not valid. It is not an adequate test of the presence or absence of quantitative relationships.

REFERENCES

CAMERON, MH, HAWORTH, N, OXLEY, J, NEWSTEAD, SV, and LE, T (1993), 'Evaluation of Transport Accident Commission road safety television advertising'. Report No. 52, Monash University Accident Research Centre.

HARRISON, WA (1990), 'Update of alcohol times as a surrogate measure of alcohol-involvement in accidents'. Research Note, Monash University Accident Research Centre.

NEWSTEAD, SV, CAMERON, MH, GANTZER, S, and VULCAN, P (1995), 'Modelling of some major factors influencing road trauma trends in Victoria 1989-93'. Report No. 74, Monash University Accident Research Centre.

VULCAN, AP, CAMERON, MH, MULLAN, N, and DYTE, D (1996), 'Possibility of adapting some road safety measures successfully applied in Victoria to South Australia'. Report No. 102, Monash University Accident Research Centre.

WHITE, M, WALKER, J, GLONEK, G, and BURNS, N (2000a), 'Re-investigation of the Effectiveness of the Victorian TAC's Road Safety Campaigns'. Paper submitted to Road Safety Research, Policing and Education Conference, Brisbane, November 2000.

WHITE, M, WALKER, J, GLONEK, G, and BURNS, N (2000b), 'Re-investigation of the Effectiveness of the Victorian Transport Accident Commission's Road Safety Campaigns'. Report No. 4/2000, Safety Strategy, Transport South Australia, November 2000.

1 LAH, the 'low alcohol hours' of the week (ie. Monday-Thursday 6am to 6pm, Friday 6am to 4pm, Saturday 8am to 2pm, Sunday 10am to 4pm), are those periods when the percentage of drivers killed or admitted to hospital with a blood alcohol content exceeding 0.05%, was below 4%. HAH, the 'high alcohol hours', are the converse of these periods, during which about 38% of driver serious casualties had blood alcohol content exceeding 0.05% (Harrison 1990).

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