1 Executive Summary

The aim of this report is to discover how Random Breath Testing (RBT), arguably the major contributor to the decline in fatal crashes involving alcohol, has been conducted across states in Australia.Specifically, how its pattern of enforcement demonstrates effectiveness through the significant decrease in the number of road users caught with blood alcohol concentration above the legal limit and number of deaths caused by drink driving. This obesrvation was achieved from data collected by the Australian Government Database and thoroughly analysed and investigated throughout this report. This resulted in developing two research questions, which seek to explain the changes in the number of RBT conducted in the last twelve years and how those changes have affected the rate of intoxicated driver deaths across various states in Australia. The finding was remarkable, displaying clear differences in the use of RBT across Australian states, concluding that changing the enforcement level accordingly improved the effectiveness and results in NSW when compared with QLD. Knowing this would aid the authorities to develop appropriate strategies in solving the prevalent issue of drink driving.

2 Full Report

2.1 Initial Data Analysis (IDA)

Totaldata = read.csv("RBT.csv")

Importing this data set to R created three additional variables with no data, which were removed using R code in order to clean the data set.

Additionally, the format in which the data was imported contained two independent variables and was difficult to graph. In order to simplify the data frame, it was split into a separate data frame for each desired state.

NSW=read.csv("NSW.csv")
QLD=read.csv("QLD.csv")
SA=read.csv("SA.csv")
TAS=read.csv("TAS.csv")
NT=read.csv("NT.csv")
Totaldata = subset(Totaldata, select = -c(X,X.1,X.2) )

head(Totaldata)
##   Year State RBT.conducted Positive.RBT Licences
## 1 2008   NSW       4204525        27368       NA
## 2 2009   NSW       4440862        26595       NA
## 3 2010   NSW       4637033        24411  4791490
## 4 2011   NSW       4520010        22117  4893688
## 5 2012   NSW       4735462        19982  4984973
## 6 2013   NSW       5153136        20193  5060762
##   Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit
## 1                                                                                                            58
## 2                                                                                                            68
## 3                                                                                                            48
## 4                                                                                                            52
## 5                                                                                                            44
## 6                                                                                                            46
##   Number.of.deaths.from.crashes.involving.a.driver.or.motorcycle.rider.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit
## 1                                                                                                                                  78
## 2                                                                                                                                  94
## 3                                                                                                                                  74
## 4                                                                                                                                  70
## 5                                                                                                                                  56
## 6                                                                                                                                  53
dim(Totaldata)
## [1] 96  7
class(Totaldata)
## [1] "data.frame"
str(Totaldata)
## 'data.frame':    96 obs. of  7 variables:
##  $ Year                                                                                                                               : int  2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 ...
##  $ State                                                                                                                              : chr  "NSW" "NSW" "NSW" "NSW" ...
##  $ RBT.conducted                                                                                                                      : int  4204525 4440862 4637033 4520010 4735462 5153136 5916338 6117884 4908582 4898088 ...
##  $ Positive.RBT                                                                                                                       : int  27368 26595 24411 22117 19982 20193 19531 18750 17442 18022 ...
##  $ Licences                                                                                                                           : int  NA NA 4791490 4893688 4984973 5060762 5142396 5245755 5337947 5439711 ...
##  $ Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit                      : int  58 68 48 52 44 46 34 36 41 42 ...
##  $ Number.of.deaths.from.crashes.involving.a.driver.or.motorcycle.rider.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit: int  78 94 74 70 56 53 50 45 59 55 ...
#
sapply(Totaldata, class)
##                                                                                                                                Year 
##                                                                                                                           "integer" 
##                                                                                                                               State 
##                                                                                                                         "character" 
##                                                                                                                       RBT.conducted 
##                                                                                                                           "integer" 
##                                                                                                                        Positive.RBT 
##                                                                                                                           "integer" 
##                                                                                                                            Licences 
##                                                                                                                           "integer" 
##                       Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit 
##                                                                                                                           "integer" 
## Number.of.deaths.from.crashes.involving.a.driver.or.motorcycle.rider.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit 
##                                                                                                                           "integer"

2.2 Summary

  • The data came from the Australian Government database, and was collected by The Bureau of Infrastructure, Transport and Regional Economics. The raw data can be found at https://data.gov.au/data/dataset/australian-random-breath-testing/resource/6c5cbea3-79dc-40b9-9775-49521a57eacb.
  • The data is valid because it is very recent and was recorded by a reputable government institution. It records every Random Breath Test conducted since 2008, collating it with the number of deaths caused by drink driving. Since this data is as valid as the Australian Police and Government records who conducted the tests and study, it can be relied on entirely for this investigation.
  • The primary issue with this data set is the missing fatality rates for Victoria, Australian Capital Territory and Western Australia, as well as a few other data points. This can be easily overcome by omitting just these two states from any research questions performed on the data.
  • Each row represents the independent variables, which in this case is specifying the year and state that the data was collected in.
  • Each column represents the dependent variables that were measured in the data study, which is the total:
    • RBT conducted
    • Positive RBT
    • Total number of licenses
    • Driver deaths with BAC above legal limit
    • Total Deaths from crashes involving a driver with BAC above legal limit.

2.3 Domain Knowledge

Random breath testing (RBT) is an efficient method of analysing a motorist’s blood alcohol concentration (BAC) by analysing a breath sample using a breathalyser (Drug and Alcohol Research and Training Australia, 2020). Police officers typically select motorists at random to undergo the roadside test (NSW Government, n.d.). The BAC analyses the quantity of alcohol (in grams) in every 100 millilitres of blood. For example, a BAC reading of 0.02 is equivalent to 0.02 grams of alcohol per 100 millilitres of blood (Drug and Alcohol Research and Training Australia, 2020). In Australia, the legal limit for driving is 0.05. However, this varies between the states, the category of license and the type of vehicle being driven. In New South Wales (NSW) specifically, there are 3 BAC levels: 0.00, less than 0.02, and less than 0.05 (NSW Government, n.d.).

The effect of alcohol on the human central nervous system has been studied extensively, through epidemiological, laboratory, on road and closed course studies, with the simplified conclusion that as BAC increases, the impairment on the brain becomes more significant (Martin et al., 2013; NSW Government, n.d.). Martin et al., (2013) define impairment as “a decreased ability to perform a task”. It was found that in comparison to a sober state, the sensory and motor functions such as complex reaction time, attention span, lane-tracking ability, and other vision related operations (e.g. saccadic vs. smooth eye movement, glare recovery time and peripheral attention) were significantly impaired, especially with a BAC of 0.05% or higher. Despite this, it is important to note that impairments can happen even at low BAC levels (Alcohol and Drug Foundation, 2020). Furthermore, the apparentness of any impairment is dependent on the complexity of manoeuvres required to operate the vehicle (Martin et al., 2013).

2.4 Research Question 1

How have the number of RBT changed over the time period? Why?

In Queensland (QLD), the number of RBT’s conducted rose linearly from 2008, peaking at 2014 before decreasing exponentially until 2018, increasing only slightly in 2019. The 2014 peak for RBT’s conducted occurred 4 years after the peak in positive RBT’s. Furthermore, the positive RBT’s also decreased in a mostly linear trend between 2010-2014. This 4-year delay could indicate that the number of RBT’s conducted increased in response to increasing positive RBT cases. Hence as positive RBT’s decreased over time, testing also decreased. Between 2008-2015, New South Wales (NSW) experienced an increasing linear trend. However for the period of 2011-2013, the tests conducted dropped below the expected line of best fit between 2008-2015, yet still increased linearly. After an approximately 20% decrease from 2015-2016, the number of tests remained steady until 2017, before increasing between 2017-2019 in a steeper linear trend when compared to that of 2008-2015. Like QLD, the RBT’s conducted appear to increase in response to an increase of positive RBT’s. However, from 2017-2019, this correlation does not hold, because as positive RBT’s continue to decrease, the RBT’s conducted increased rapidly. There is an array of factors that contributed to the trends seen. However, the greatest contributor to the decreasing trend of positive RBT’s in QLD and NSW has been attributed to methods of deterrence enforced by the police force in conjunction with effective road safety campaigns. The NSW and QLD plateau in 2012-2013 which was followed by the steady decline in positive RBT’s coincided with NSW’s initiation of the ‘Plan B’ and QLD’s ‘Think before you drink. Think twice before you drive.’ campaign in, 2012 (NSW Government, 2017; Queensland Government, 2012). These campaigns were heavily televised during their launch and provided a range of transport options for people to contemplate, in order to avoid drink driving. Moreover, over 63% and 81% of the target audience in NSW and QLD respectively, concurred that drink driving is socially intolerable, thus responding positively to the campaign. Along with the fear of hefty fines, license disqualification and enforcement of alcohol ignition interlock orders in 2015, these deterrent approaches proved to be effective in further pushing the number of positive RBT’s down within a year of its launch, with a slight lapse between 2017-2018 (NSW Government, 2017; Queensland Government, 2012). Since 2007, youth education programs like ‘You’re no dummy’ and ‘Keys2Drive’ have been more present, informing young drivers on drink driving ramifications. Along with this, the QLD introduced the 3-year provisional licensing system that NSW had adopted in 2000. As such, QLD’s positive RBT’s was seen to steadily decline in 2010, after they graduated to full licenses. Furthermore, QLD’s enforcement of a 0.00 BAC for provisional drivers in 2010, accounts for the almost 20% decrease in positive RBT’s between 2010-2011, despite more testing (CARRS-Q, 2015; Queensland Government, 2010). In brief, multitudes of factors contribute to the RBT conducted and positive RBT trends in QLD and NSW and a steadily declining number of positive RBT’s in both states can be observed, indicating effective approaches.

plot(QLD$Year,QLD$RBT.conducted,xlab="Year",ylab="RBT conducted",main="RBT Conducted in QLD from 2008-2019")

plot(QLD$Year,QLD$Positive.RBT,xlab="Year",ylab="Positive RBT",main="Year vs Positive RBT's in QLD 2008-2019")

plot(NSW$Year,NSW$RBT.conducted,xlab="Year",ylab="RBT conducted",main="RBT Conducted in NSW from 2008-2019")

plot(NSW$Year,NSW$Positive.RBT,xlab="Year",ylab="Positive RBT",main="Years vs Positive RBT's in NSW 2008-2019")

Summary:

2.5 Research Question 2

How and why have the number of RBT conducted affected the rate of intoxicated driver deaths differently between states in Australia?

It is likely that enforcing RBT would reduce the number of alcohol induced road death tolls. However, considering the tight funding and resources, the effectiveness of this method should be critically examined. Thus, it would be of most value to analyse two similar states and compare the nature and results of their approaches to limiting drink driving accidents. According to the analysis from Research Question 1, the number of RBT’s conducted in both NSW and QLD dropped around 2014. However, NSW gradually lifted these rates back to original levels, whereas QLD continued to drop them. Through the process of constructing regression models comparing the annual RBT test rate with the drunk driver deaths of each state, assumptions about the effectiveness of these strategies can begin to be drawn.

QLD

L=lm(QLD$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit~QLD$RBT.conducted)
plot(QLD$RBT.conducted,QLD$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit, ylab="Motorists killed with illegal BAC",xlab="RBT Conducted", main="Effectiveness of RBT in NSW")
abline(L,col="red")

The regression line above shows an obvious correlation between rising RBT rates and decreasing deaths caused by drink driving in QLD yet several assumptions must be satisfied to prove its validity. The graph shows that the relationship is linear and randomly sampled as this government data records all cases of drink driving accidents and RBT conducted each year. Furthermore, there is sufficient sample variance in the explanatory variables in this case and both residuals plots indicate no trend in the data points, and therefore the classic linear regression model assumptions are satisfied and by extension, the OLS estimators are BLUE.

model=lm(QLD$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit~QLD$RBT.conducted)
plot(model$residuals, main="Residuals Plot")
abline(h=0, col="red")

model=lm(QLD$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit~QLD$RBT.conducted)
plot(predict(model),residuals(model),xlab="Predicted Y",ylab= "residuals", main="Predicted Y vs Residuals")

With the results of the regression model satisfied, it can be used to accurately estimate specific conditions and scenarios. This model is based off the linear equation y=-6.911✕10-6x+63.91, where x is the number of RBT’s conducted and y is the number of drunk driver deaths.


NSW

Line=lm(NSW$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit~NSW$RBT.conducted)
plot(NSW$RBT.conducted,NSW$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit, xlab="RBT conducted", ylab="Motorists killed with illegal BAC", main="Effectiveness of RBT in NSW" )
abline(Line,col="blue")

summary(Line)
## 
## Call:
## lm(formula = NSW$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit ~ 
##     NSW$RBT.conducted)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -7.0101 -4.7197 -0.7456  2.0686 14.2898 
## 
## Coefficients:
##                     Estimate Std. Error t value Pr(>|t|)    
## (Intercept)        1.078e+02  1.693e+01   6.370  0.00013 ***
## NSW$RBT.conducted -1.219e-05  3.358e-06  -3.629  0.00549 ** 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 6.545 on 9 degrees of freedom
##   (1 observation deleted due to missingness)
## Multiple R-squared:  0.5941, Adjusted R-squared:  0.549 
## F-statistic: 13.17 on 1 and 9 DF,  p-value: 0.005492

Similar to the case of Queensland, the linear regression model shows that the more RBT conducted the less deaths caused by BAC above the legal limit. In checking its validity, we can see that the three assumptions regarding linear parameters, random sampling, and sample variation in the explanatory variable are all satisfied. The residuals plot is then used to test zero conditional mean and homoscedasticity. As there’s no pattern can be found in the residual plot, we can conclude that the means of errors for every value of x equals zero and the variance of the error is a constant value. As a result, all Classic Simple Linear Assumptions have been satisfied and thus our Simple Regression Outcomes would be accurate.

model1=lm(NSW$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit~NSW$RBT.conducted)
plot(model1$residuals, main="Residuals Plot")
abline(h=0, col="blue")

model1=lm(NSW$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit~NSW$RBT.conducted)
plot(predict(model1),residuals(model),xlab="Predicted Y",ylab= "Residuals", main="Predicted Y vs Residuals")

summary(model1)
## 
## Call:
## lm(formula = NSW$Number.of.drivers.and.motorcycle.riders.killed.with.a.blood.alcohol.concentration..BAC..above.the.legal.limit ~ 
##     NSW$RBT.conducted)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -7.0101 -4.7197 -0.7456  2.0686 14.2898 
## 
## Coefficients:
##                     Estimate Std. Error t value Pr(>|t|)    
## (Intercept)        1.078e+02  1.693e+01   6.370  0.00013 ***
## NSW$RBT.conducted -1.219e-05  3.358e-06  -3.629  0.00549 ** 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 6.545 on 9 degrees of freedom
##   (1 observation deleted due to missingness)
## Multiple R-squared:  0.5941, Adjusted R-squared:  0.549 
## F-statistic: 13.17 on 1 and 9 DF,  p-value: 0.005492

The Linear Regression Model found is y= -1.219✕10-5x+107.8, where x is the number of RBT conducted and y is the number of drunk drivers’ deaths. By comparing this to Queensland’s model, we can conclude that the number of deaths caused by drink driving declines at a higher rate in NSW than in QLD. In other words, NSW’s strategy of increasing and decreasing the number of RBT flexibly according to the situation is more effective than that of QLD. This finding is also backed up by the concept of deterrence mechanisms: offending behaviour declines as perceived risk increases, which can be achieved by enforcing RBT in a ‘unpredictable, unavoidable and ubiquitous’ manner (Homel, 1993, as cited in Terer & Brown, 2014).

3 References