Key findings
  • Telematics was used to provide braking scores for a small sample of young drivers
  • Likely initial braking behaviour improvements were not sustained in the 2nd month
  • This deterioration was significantly reduced by immediate telematics feedback
  • No differences in braking behaviour were detected between genders or licence types

Background

Road traffic collisions are one of the leading causes of death globally, with approximately 1.19 million people dying in road traffic collisions each year and are the leading cause of death in young people aged between 5 and 29 years of age (WHO, 2023). There was a 7.3 percent increase in the number of deaths on Australian roads in 2023 compared to 2022, and the number of deaths in the 17-25 age group increased by 7 percent (BITRE, 2023). The increases in road traffic crashes and fatalities highlight the necessity for road traffic interventions in Australia that target young drivers.

Typically, young drivers are classified as drivers between the ages of 18 and 25 years. These drivers are more susceptible to making errors and taking risks while driving, compared to older more experienced drivers. This is especially true for young male drivers, as males are typically three times more likely to be killed in road traffic collisions compared to females (WHO, 2023).

Educating drivers about the safety implications of speed and the financial penalties of illegal driving behaviours can be an effective approach in promoting safer driving in young people. Cognitive-based training interventions have been used to successfully reduce speeding behaviours in young drivers (Williams, 2023), and feedback on driving behaviour is considered a particularly useful countermeasure for risky driving behaviours (Feng & Donmez, 2013). Molloy et al. (2019) suggested that the timing for the delivery of feedback may influence its effectiveness, with immediate feedback being more effective in improving driving behaviour. In-vehicle telematics is a tool that can provide such feedback.

The use of in-vehicle telematics as a tool in measuring and monitoring driving behaviour is increasing rapidly (The Royal Society for the Prevention of Accidents, 2013). Telemetry is the process of recording and transmitting the reading of a device or instrument and is used in a vehicle context to transmit live driving data, which can then be used to draw outcomes about drivers or vehicles. Previously driving behaviour research has relied heavily on self-reported measures (Kaye et al., 2018). However, the emergence of in-vehicle telematics creates the opportunity to automate naturalistic driving data. Telematics devices installed into motor vehicles require minimal maintenance, meaning that driving behaviour can be continuously monitored in an unobtrusive way and without the presence of a supervisor (Elamrani Abou Elassad & Mousannif, 2019).

In-vehicle telematics has been a popular tool in motor vehicle insurance studies, with telematics data added to pre-existing insurance pricing models to boost the accuracy of the models (Ayuso et al., 2019) or used to create driver identification models (Azadani & Boukerche, 2022). However, there are few studies that have used in-vehicle telematics data to model driving behaviour, or to determine the effectiveness of interventions on driving behaviour, with most studies that have included telematics data being focused on improving the accuracy of insurance premium risk models (Boylan et al., 2024). Stevenson et al. (2021) investigated the effectiveness of using feedback from an in-vehicle telematics device along with financial incentives to improve driving behaviour in young people and, while observed driving behaviours trended in a positive direction, the improvements were not statistically significant. The Young Driver Telematics Trial (NSW Insurance Regulatory Authority, 2019) also found that feedback from in-vehicle telematics led to reduced rates in speeding, harsh acceleration, harsh braking and harsh turning in young drivers. The promising results from these studies warrant further investigation into the effectiveness of feedback from in-vehicle telematics on the driving behaviour of young drivers.

The aim of the analysis in this paper was to determine whether immediate feedback from in-vehicle telematics improves the braking behaviour of young drivers. The analysis focused on determining whether braking behaviour was related to gender and licence type. It was hypothesised that visual feedback administered immediately during driving would lead to improvements in the driving behaviour of young people.

Methodology

Recruitment

Participants were recruited from Victoria, Australia using online advertisements through social media. The inclusion criteria were: aged 18 to 30 years; licensed to drive solo (i.e., probationary or full licence); and owned a motor vehicle that was compatible with the GOFAR in-vehicle telematics device. Interested people were given a link to an “expression of interest” survey. Upon completion of the survey, they emailed to confirm registration in the study. Researchers then met with each participant to instal the device in their vehicle. The device was also used as an incentive with participants able to keep their GOFAR telematics device and its attachments after the study ended. While the telematics data of the participants would continue to be recorded by GOFAR after the study period had ended, GOFAR did not have access to the demographic information collected by the researchers.

Study design

The study followed a randomised control trial design. Before starting the study, all participants were randomly assigned to either the control group or the intervention group and completed a demographic characteristics questionnaire and a modified version of the Driving Behaviour Questionnaire (DBQ-27). The in-vehicle telematics device was then installed in their vehicle.

GOFAR devices were used in the study (Figure 1). These devices are approved for providing Australian Tax Office compliant logbooks and have been used in the New South Wales Government Young Driver Pilot (NSW Insurance Regulatory Authority, 2019), suggesting that these devices provide accurate and valid results. The devices were used in eight research trials in which five million kilometres of driving data were collected (GOFAR, 2024), indicating that the GOFAR is a valid telematics device capable of measuring driving behaviour. The GOFAR device was also used by Chen et al. (2023) to measure and monitor driving behaviour.

Figure 1
Figure 1.GOFAR in-vehicle telematics device, mobile application, and feedback ray

(GOFAR, 2023, published with permission)

Participants’ driving data were collected across two months. The first month was the baseline period, both groups received post-trip driving feedback. The second month, the intervention group received immediate driving feedback while the control group continued to receive only post-trip feedback. All participants could access their post-trip feedback through the GOFAR phone application. The post-trip feedback included:

  • a braking score and an acceleration score reflecting the braking and acceleration performance for each trip

  • fuel economy during the trip, the amount of fuel used, the amount of emissions produced

  • distance travelled

If participants had activated their Bluetooth on their mobile phones, map information was also available and a trip path was displayed as a blue line. The line turned red whenever the driver exhibited poor driving behaviour (i.e., harsh braking, harsh acceleration, speeding).

The intervention group received their immediate driving feedback from a mounted light called the GOFAR ray (Figure 1). The ray changed colour depending on the driver’s behaviour; a blue light for ideal braking/acceleration, a purple light if the driver exceeded braking/acceleration thresholds of ± 0.2g and a red light if the driver exceeded thresholds of ±0.4g. The red light was also triggered by speeding when GPS data were available. The ray was mounted on the dashboard of the participant’s vehicle in their line of sight, however, due to differences in vehicle design the exact placement on the dashboard was not the same in every vehicle. While the Young Driver Telematics Trial (NSW Insurance Regulatory Authority, 2019) suggested that the feedback ray could be distracting, in most cases this was due to the brightness of the LED lights on the ray rather than its positioning. The brightness of the ray is adjustable in the GOFAR mobile application. This was the only additional feedback that the intervention group was provided with compared to the control group.

An optional structured interview was completed online using Zoom at the end of each month of the study. The interview focused on the participants’ attitudes towards the telematics device and whether they believed it influenced their driving behaviour. The response rate for these interviews was relatively low (21 out of 37 (56.8%) for the first interview and 8 out of 37 (21.6%) for the second interview). In the first interview, 16 participants (76%) said the GOFAR device improved their driving behaviour. In the second interview where questions relating to the GOFAR ray were introduced, all participants who responded and had access to the GOFAR ray (n=5) said the immediate visual feedback from the ray had a positive influence on their driving behaviour and disagreed when asked if the ray distracted them from other driving tasks.

Measures

The total score from the modified DBQ were used as a predictor variable in all models in the study (Reason et al., 1990). The total DBQ scores range from 0 to 135, with higher scores reflecting a higher frequency of errors, lapses and violations. Total DBQ scores were used in the analysis as opposed to the four subscales of the DBQ because only overall risk taking needed to be considered in the modelling. This is consistent with the approach in previous studies that used the total DBQ scores as a measure of risky driving behaviour (Qu et al., 2020). Batool and Carsten (2016) have suggested that attitudes towards road safety are reflected in total DBQ scores. While we had access to acceleration scores, these were not investigated as Bagdadi (2013) suggested that using harsh braking events (HBEs) as opposed to harsh acceleration events results in a 1.6 times increase in near-crash detection accuracy, indicating that HBEs explain more about driving behaviour than harsh acceleration events.

Braking scores were used as an outcome variable to track two behaviours, safe braking behaviour and HBEs. Braking scores were calculated by GOFAR with a range from 0 to 100, higher scores represented safe braking behaviour. HBEs were used to validate the GOFAR braking scores and were defined using two thresholds, which allowed us to determine whether the immediate feedback had different effects on harsh braking and more severe braking. The first threshold for harsh events was set at -0.2g as suggested as the minimum threshold by previous studies (Kamla et al., 2019; Zhou et al., 2022), and the second less sensitive threshold for more severe braking was set at -0.4g.

Analyses

While previous studies have used machine learning techniques including random forests, support vector machines, and neural networks to model telematics data, Linear Mixed Models (LMMs) were used in this analysis as it is the best technique to account for the multi-level structure of the telematics data, with trips being nested within drivers. The GOFAR braking scores were used as outcome variables for a LMM and the results of these models were validated by using the number of HBEs in Gamma Generalised Linear Mixed Models (GLMMs).

To determine whether the immediate GOFAR ray feedback had a significant impact on driving behaviour, an interaction effect between group (control or intervention) and month of the study (first or second month) was included in each model. Age, gender and licence type were included in the models as control variables. Gender was reported as male or female as the participants only selected those two options. The distance travelled in each trip and total DBQ score were also considered. All trips of less than 1km were removed from the dataset because reliable braking scores could not be computed for such short trips. In the HBEs models the outcome measure was defined as the number of harsh events per 100 kilometres with the addition of 0.5 for the number of events exceeding a deceleration threshold of -0.2g and the addition of 0.1 for the number of events exceeding a deceleration threshold of -0.4g. The addition of these constants was needed to allow the use of Gamma distributions for these variables.

All analyses were conducted using IBM SPSS Statistics version 29 (IBM Corp., 2023). A two-sided p-value of <0.05 was considered significant for all tests.

Results

Descriptives

In total, 37 participants aged 18-28 years (x̄ 23.8 years) completed the study and recorded a total of 6,917 trips. The descriptive statistics for each of the demographic groups can be seen in Table 1. LMMs were fitted with only gender and then only licence type to determine whether the differences in braking scores and the number of HBEs between males and females, and fully licenced drivers and provisional drivers were statistically significant. There were no statistically significant differences between males and females or between licence types for any of the driving behaviour measures.

Table 1.Descriptive statistics
N Braking score HBEs per 100km <-0.2g HBEs per 100km <-0.4g
Mean SD p-value Mean SD p-value Mean SD p-value
Gender Female 20 76.25 8.00 .61 201.83 215.87 .48 5.69 12.91 .48
Male 17 76.98 7.22 175.44 182.06 5.33 12.71
Licence Full licence 26 76.33 7.19 .90 205.42 205.16 .07 4.89 11.39 .15
Provisional 11 77.26 8.54 151.36 184.05 6.94 15.50
Total 37 76.61 7.64 188.96 200.51 5.51 12.82

Spearman’s rho correlations were used to determine whether there was a relationship between braking scores and the number of HBEs exceeding -0.2g per 100km. A strong non-linear negative relationship (Spearman’s rho = -0.65) was observed. A weaker relationship was observed when events exceeding -0.4g were used instead (Spearman’s rho = -0.39). These relationships are illustrated in Figure 2.

The correlations between these measures and DBQ scores were also explored. DBQ scores had a statistically significant weak negative relationship with braking scores (Spearman’s rho = -0.13, p<.001), and the number of HBEs exceeding -0.4g per 100km (Spearman’s rho = 0.09, p<.01), but no statistically significant relationship with the number of HBEs exceeding -0.2g per 100km (Spearman’s rho = 0.02, p=.01). The directions of each of the statistically significant relationships followed our expectations, with better (higher) braking scores when the number of HBEs per 100km was lower and DBQ scores were lower. However, it seems that the braking scores are a more sensitive measure of braking behaviour than the number of HBEs especially when there are no (or very few) HBEs per 100km.

Figure 2
Figure 2.Relationship between braking score and HBEs per 100km

Braking behaviour models

The parameter estimates for the LMM for braking score can be seen below in Table 2.

Table 2.Linear Mixed Model for braking score
Parameter Estimate Std. Error df t-value p-value
Gender: Female -1.22 1.41 31.02 -.87 0.39
Gender: Male 0.00 0.00 . . .
Licence: Full licence 0.04 2.22 30.82 0.02 0.99
Licence: Provisional 0.00 0.00 . . .
Group: Control -1.34 1.32 31.80 -1.02 0.32
Group: Intervention 0.00 0.00 . . .
Month: First 1.03 0.25 6888.02 4.17 0.00
Month: Second 0.00 0.00 . . .
Age -0.29 0.45 30.78 -0.65 0.52
DBQ score -0.13 0.06 30.91 -2.11 0.04
Distance 0.13 0.01 6897.93 23.84 0.00
Group*month 1.18 0.32 6885.90 3.65 0.00

Table 2 shows that month was significant in the model, with the first month having better (higher) braking scores on average compared to the second month. DBQ score was also significant, with those who had worse (higher) DBQ scores having worse (lower) braking scores on average. Trips with a longer driving distance were associated with better braking scores. The interaction effect between group and month was significant in the model. The interaction effect is displayed by the plotting of estimated marginal means in Figure 3. Both the control group and the intervention group had lower scores for the second month, however, the decline was larger for the control group than the intervention group, suggesting less deterioration in braking behaviour when immediate feedback was provided.

The Gamma GLMM results for the number of HBEs per 100km can be seen in Table 3. No significant effects were seen in the -0.2g model, however DBQ scores were significant in the -0.4g model, as drivers with worse (higher) DBQ scores had more HBEs exceeding -0.4g per 100km. The interaction effect was close to being significant in the -0.2g model (p=0.07) with a greater deterioration in braking performance again observed for the control group than the intervention group (Figure 3). The braking behaviour of the control group was worse in the second month compared to the first month while the intervention group showed little change. The Group*month interaction effect was far from statistically significant in the -0.4g model.

Table 3.Generalised Linear Mixed Model for HBEs per 100km
HBEs per 100km (<-0.2g) HBEs per 100km (<-0.4g)
Parameter Estimate Std. Error t-value p-value Estimate Std. Error t-value p-value
Gender: Female 0.24 0.18 1.32 0.19 0.16 0.18 0.91 0.36
Gender: Male 0.00 0.00 . . 0.00 0.00 . .
Licence: Full licence 0.27 0.28 0.94 0.35 -0.23 0.27 -0.82 0.41
Licence: Provisional 0.00 0.00 . . 0.00 0.00 . .
Group: Control 0.10 0.17 0.57 0.57 -0.15 0.17 -0.92 0.36
Group: Intervention 0.00 0.00 . . 0.00 0.00 . .
Month: First 0.02 0.04 0.59 0.55 -0.14 0.07 -1.85 0.06
Month: Second 0.00 0.00 . . 0.00 0.00 . .
Age 0.03 0.07 0.45 0.65 0.01 0.06 0.10 0.92
DBQ score 0.01 0.01 1.58 0.12 0.02 0.01 2.88 0.00
Group*month -0.09 0.05 -1.83 0.07 0.05 0.05 0.51 0.61
Figure 3
Figure 3.Estimated marginal means for Group by month

Discussion

We aimed to test the hypothesis that immediate feedback from in-vehicle telematics is beneficial for improving braking behaviour among young drivers. The results showed significantly less deterioration in braking behaviour for the immediate feedback group than the control group, supporting our research hypothesis. The secondary aim was to determine if braking behaviour was related to gender or licence type. Neither gender nor licence type was statistically significant in any of the models indicating that the hypotheses of consistency in braking behaviour across gender and licence type cannot be rejected. While the hypothesis that the GOFAR ray would lead to improvements in driving behaviour was not supported, given the deterioration in braking scores seen in both the control and intervention groups. The immediate feedback provided by the GOFAR ray had some positive effect, as the deterioration was not as steep for the intervention group.

The LMM for braking scores showed that both the control group and the intervention group experienced declines in braking behaviour in the second month of driving with an in-vehicle telematics device compared to the first month. The driving behaviour in the first month may have been improved by participants subconsciously changing their behaviour because they knew their driving was being monitored. This is known as a Hawthorne effect. Cooper et al. (2023) acknowledged that the Hawthorne effect may have been present in their study, in which participant awareness of the monitoring of their naturalistic driving may have influenced their driving behaviour. Ziakopoulos et al. (2020) suggest that while participants may know that their driving behaviour is being monitored initially, they tend to forget after some time. The deterioration of braking scores in the control group during the second month seems to support this assumption. However, it is also possible that participants were trying to improve their post-trip driving scores during the first month, resulting in better braking scores than would have been the case prior to the installation of their telematics device. The GOFAR mobile application also includes a leaderboard which ranks drivers based on their driving scores (braking scores, acceleration scores, and overall scores). This may have influenced the driving of competitive participants who wanted to achieve higher rankings on the leaderboard.

The deterioration in braking scores in the second month was steeper for the control group than the intervention group. This indicated that the instantaneous visual feedback provided in the form of a colour changing light by the GOFAR ray had served to better sustain any improvement that was achieved in the first month. This supports the hypothesis that the immediate visual feedback afforded by the GOFAR ray had a positive influence on driving behaviour.

However, the results were not as positive as anticipated given the lack of improvement in the braking scores of the intervention group in the second month compared to the first month of the study. While the results were unexpected, they were not unprecedented, as a study conducted by Choudary et al. (2021) also found that driving performance was nearly 14.9 percent worse for participants who receive detailed immediate feedback from in-vehicle telematics compared to those who did not receive such feedback. However, these authors note that strong negative feedback could lead to short-term improvements in driving behaviour.

The GOFAR braking score could not be ascertained as the calculations are proprietary. For this reason, the relationships of the braking score (0-100) with the DBQ and the number of HBEs per 100km were used to validate this measure. For the -0.2g HBE measure, the expected relationship was particularly strong with the GOFAR braking scores. However, the interaction effect for group with month was not quite significant for the -0.2g HBEs measure and not significant for the -0.4g measure, perhaps suggesting that the GOFAR braking score provides a more sensitive measure of braking behaviour than HBEs. This was confirmed in the plots of the GOFAR braking scores against the number of HBEs.

A previous trial has shown statistically significant interaction effects between group and month for an HBEs model. The Young Driver Telematics Trial (NSW Insurance Regulatory Authority, 2019) found that real-time feedback from a LED light ray had an overall positive impact on the behaviour of young drivers, with HBEs per 1,000km significantly reduced. The trial of 717 participants aged 17-24 years had a younger mean age (20.4 years) compared to the present study (23.8 years) and a longer study duration, with three months for each study period (baseline and intervention). The trial also defined harsh events using higher thresholds compared to the present study, with their minimum threshold for a HBEs being set at -0.45g. These factors may explain the differences in results between the trial and the present study. Molloy et al. (2019) found that immediate auditory feedback was ineffective in improving the speed management behaviours of young drivers. The present study focused on immediate visual feedback, finding that this was beneficial. However, more research is necessary to determine the longer term effectiveness of immediate visual feedback.

The lack of significance for the interaction effect in the HBEs models despite significant results for the braking score model may be explained by the more sensitive nature of the braking score measure as noted above. Also, the model for the more sensitive number of HBEs measure (exceeding -0.2g) was nearly statistically significant (p=.07), suggesting that a larger sample size might have shown significance for this model.

The findings regarding the relationship between DBQ score and braking behaviour was promising, with DBQ scores being significant in the braking score model and the number of HBEs exceeding -0.4g model. While some studies have found that self-reported driving behaviours are unreliable (Af Wåhlberg & Dorn, 2015), Zhao et al. (2012) found that those with higher violation scores on the DBQ carried out more HBEs compared to those who had lower violation scores, as found for -0.4g HBEs in this study. Using a driving simulator and a vehicle fitted with an in-vehicle monitoring system to assess the validity of the DBQ, Helman and Reed (2015) also found that the DBQ was a valid measure of driving behaviour. The significance of DBQ scores in the braking score model therefore helps to confirm that the GOFAR braking scores is a reliable measure of braking behaviour.

The lack of statistically significant findings regarding demographic variables such as gender, licence type and age was unexpected, considering that the link between demographics and driving behaviour outcomes has been previously established. Singh and Kathuria (2021) suggested that age, gender and driving experience have a major influence on the number of driving errors and violations in their review of naturalistic driving studies. Simons-Mortens et al. (2019) confirmed that males exhibited a greater frequency of harsh driving incidents than females, and that drivers aged between 18 and 20 years reported more harsh driving events than older counterparts. However, Bagdadi (2013) also conducted a naturalistic driving study and found that gender and age had no significant influence on crash risk and that there were no significant differences between males and females in harsh acceleration.

Strengths and Limitations

In addition to the new insights gained about young drivers and their braking behaviour, the main strengths of this study are the new evidence on the benefits of using in-vehicle telematics to record naturalistic driving behaviour. The device minimises the Hawthorne effect, a known limitation of naturalistic studies with an observer, as well as providing a gamification of the driving experience that may have promise as a behaviour change approach. Providing real-time feedback on driving performance and gaming the device so that to ‘win’ drivers need to drive safely have the potential to reach young drivers in situ and change behaviour on the road. Additionally, this study was able to account for the multi-levelled nature of in-vehicle telematics by using LMMs, this is an important strength not often seen in the analysis of telematics data.

However, there are also noted limitations. First, the technology. Both participant groups could access post-trip driving feedback through the GOFAR mobile phone application (i.e., overall driving score, braking score, acceleration score). However, it was not possible for the researchers to determine how often this information was accessed. This information may have helped to provide a more detailed explanation of why braking scores deteriorated in the second month. A longer study duration is needed to determine whether the effects of the GOFAR ray feedback are sustained in the longer term. Although the GOFAR braking score measure has been validated, it is not known exactly how this measure is calculated, making it impossible to reproduce these results more generally. In addition, it would have been useful if other measures of driving performance, such as speeding and acceleration, were studied in addition to these braking scores.

Recruitment was another major limitation as perceived privacy issues related to the collection of data caused unexpected delays. This resulted in staggered participation as groups did not complete the study at the same time. Timing was also impacted by COVID-19 as the study was conducted during the lockdown period which potentially impacted the driving conditions experienced by some participants. It is unknown what influence this had on the results, with driving conditions likely to have varied across holiday periods and the seasons. Also, participants were aware of their group allocation with the intervention group told at the start that GOFAR ray installation would occur at the end of the first month. This may have led to an unknown level of bias in the results. The random allocation of participants to the intervention and control groups was to minimise the potential biases of the limitations related to recruitment and participants.

Conclusion

The findings of the study suggest that there are beneficial effects of telematics monitoring on the driving behaviour of young people in the first month. There are limitations that need to be addressed to determine if those improvements can be sustained beyond a month. However, with immediate driving feedback it seems that the decline in these beneficial effects is compromised. The GOFAR braking score measure used in this study has been validated using a self-rated measure of driving behaviour (DBQ) and the number of harsh braking events (HBEs) recorded per 100km. However, a larger sample size and a larger number of appropriate driving behaviour measures would have served to provide more information about the efficacy of immediate feedback as a means for improving driving behaviour. The paper ultimately determined that while vehicle telematics is a powerful driving behaviour data collection tool, its application for improving the driving behaviour of young drivers requires additional study.

Acknowledgements

The authors would like to thank all the young drivers who participated in this study. In addition, they would like to thank Swinburne University of Technology for their contribution to the funding of this research. The authors acknowledge that ChatGPT 3.5 was used to assist in the choice of analysis options.

Author contributions

James Boylan: Conceptualisation, Data collection, Formal analysis, Methodology, Writing – original draft, Writing – review & editing. Denny Meyer: Conceptualisation, Project administration, Resources, Methodology, Supervision, Writing – review & editing. Won Sun Chen: Conceptualisation, Funding acquisition, Project administration, Supervision, Writing – review & editing. All authors have read and agreed to the published version of the manuscript

Funding

This work was supported by the Australian Government Department of Infrastructure, Transport, Regional Development and Communications (RSIF2-59). However, the funder has had no direct contribution to the study design, data analysis, or the final composition of the paper.

Human Research Ethics Review

The study was conducted in accordance with the protocol approved by the Swinburne University Human Research Ethics Committee SUHREC 22/6551 22/4/2022.

Data statement

Due to the private nature of the telematics data the researchers are unable to share the dataset used for analysis. Participants were assured upon registration into the study that their data would not be shared.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have influenced the work reported in this paper.