Driving errors as a function of listening to music and FM radio: A simulator study

Abstract Objectives Driving is a dynamic activity that takes place in a constantly changing environment, carrying safety implications not only for the driver but also for other road users. Despite the potentially life-threatening consequences of incorrect driving behavior, drivers often engage in activities unrelated to driving. This study aims to investigate the frequency and types of errors committed by drivers when they are distracted compared to when they are not distracted. Methods A total of 64 young male participants volunteered for the study, completing four driving trials in a driving simulator. The trials consisted of different distraction conditions: listening to researcher-selected music, driver-selected music, FM radio conversation, and driving without any auditory distractions. The simulated driving scenario resembled a semi-urban environment, with a track length of 12 km. Results The findings of the study indicate that drivers are more prone to making errors when engaged in FM radio conversations compared to listening to music. Additionally, errors related to speeding were found to be more prevalent across all experimental conditions. Conclusions These results emphasize the significance of reducing distractions while driving to improve road safety. The findings add to our understanding of the particular distractions that carry higher risks and underscore the necessity for focused interventions to reduce driver errors, especially related to FM radio conversations. Future research can delve into additional factors that contribute to driving errors and develop effective strategies to promote safer driving practices.


Introduction
Injury caused by motor vehicle accidents is still a major concern on a global scale (World Health Organization 2018).Millions of people worldwide are affected due to motor vehicle accidents.With reference to India, road-traffic accidents represent one of the top four leading causal factors for deaths within the age-range of 15-49 years (Transport Research Wing Government of India 2020).Over the last decade, a number of deaths on Indian roads show an upward trend.Despite taking several corrective measures, the rate and severity of accidents are increasing constantly.In India, one death every 4 min is caused by motor vehicle accidents (Transport Research Wing Government of India 2020).The occurrence of on-road accidents and near accidents could be attributed to many factors which could be broadly categorized into factors associated with the vehicle (e.g., vehicle breakdown), the driving environment (e.g., bad weather, faulty roads), and the driver (e.g., driving under the influence of drugs, alcohol).Another important factor related to the driver is distracted driving.Distracted driving has a high potential for contributing to driving errors which eventually may lead to on-road accidents and near accidents.According to Stanton and Salmon (2009), 75% of all road accidents are caused by one or the other form of driver error.Distracted driving and driving errors are highly related to each other (Najar and Sanjram 2018;Yadav and Velaga 2020;Najar and Sanjram 2021) with overall estimates suggesting that driver distractions were causal factors in 95% of driving error related road accidents (Rumar 1990).The advancement in technology has built a bond between drivers and in-vehicle devices.These in-vehicle devices, apart from providing valuable communication and entertainment services, present additional distractions to drivers (Lee 2007).The study found that using a mobile phone had a negative impact on situation awareness (Van Dam et al. 2020).
Apart from using mobile phones during driving, listening to music and FM radio are the most common activities that drivers engage in when driving.In a survey that involved 1,780 British drivers, around 68% of them reported that they listen to music or radio while driving (Dibben and Williamson 2007).According to Wiesenthal et al. (2003), listening to music in highly demanding driving conditions results in increased stress and mild aggression.Pêcher et al. (2009) reported that listening to music resulted in poor lane-keeping ability and reduced speeds.Similarly, an increase in mean speed variability was observed when drivers were engaged in listening to music (Hughes et al. 2013).Regarding the effect of listening to the radio conversation on driving performance, the available literature is suggesting mixed results.Some studies report that listening to radio had no effect on driving performance, for example, Hatfield and Chamberlain (2008) reported that there was no effect on lateral positioning of the vehicle, speed and reactions to hazardous incidents when the drivers were driving in a simulator and listening to talk-radio fragments simultaneously.Similarly, by using dual-task paradigms in a driving simulator, Strayer and Johnston (2001) observed that driving performance remains unaffected when the drivers were listening to radio broadcasts or listened to a book on tape.However, there are studies which have reported that driving performance is negatively affected when the drivers listen to radio conversations during driving.In a 1994 study by Jäncke et al. participants were given the option of listening to or not listening to a radio program while driving a graphically represented car on straightforward and challenging simulated roads.They observed that in the listening to radio program condition, the drivers committed more lane deviations, particularly on complicated highways.Strayer et al. (2015) also reported that during driving, if drivers simultaneously converse with live radio broadcasts, it compromises their driving performance and safety.
While driving, according to mood-arousal hypothesis, drivers generally listen to music in order to get enjoyment or elevate their mood and avoid boredom during long drives.But the distraction hypothesis suggests that listening to music distracts the driver from driving because it takes some of the attention resources away from the task of driving; hence puts the driver as well as other road users' safety at risk (Brodsky 2001).
Although the impact of music on driving performance has been given some attention, there are still many aspects of it that have received meager attention from human factors researchers.The effect that listening to music has on driving performance when it is not of driver's preference in comparison with the effect of listening to music when it is of their preference has been given the least research attention.The current study attempts to investigate the frequency and type of errors when drivers listen to the music of their preference, of someone else's preference, or listen to an FM radio conversation.The rationale for choosing these kinds of distractions are that after driving a vehicle for some distance, driving becomes monotonous and a boring task.In order to make it pleasant the driver either listens to music, plays FM radio, or starts conversing with the co-passenger/s, if there is any.The type of music that drivers listen to depends on whether the driver is accompanied by other people sitting in the car or is driving alone.Sometimes the driver has to play the music that the co-passengers tell him to play and the driver may or may not like that type of music.People tend to sacrifice their interest for their significant others (McGrath and Gerber 2019).Also, in the case of a hired cab, the driver plays the music that the passengers tell him to play.In such scenarios, it makes sense to investigate how is driving performance affected when the drivers listen to favorite music tracks, listens to such music tracks that they are told to play, or listens to FM Radio conversations that play music tracks intermittently?

Objectives and hypothesis
The explanations regarding the effects of music on driving have primarily focused on the information-processing capacity of the driver (Dalton et al. 2008).Such explanations assume that listening to music would be arousing for the driver, hence would put more demand on the limited mental resources.Music would add additional irrelevant stimuli to the driving environment and would lead to increased cognitive workload and therefore would negatively influence driving performance (Recarte and Nunes 2000).According to Lavie's load theory (Lavie 1995(Lavie , 2005)), a particular task consumes cognitive resources according to its complexity, that's the more complex a task is, the more cognitive resources it would consume.Consequently, it is conceivable that the more cognitive resources a secondary task consumes, the more would be the compromise in driving performance (a primary task in this context) during the performance of the respective secondary task.Therefore, the objective of the current study is to investigate the respective demands of cognitive resources by listening to music that the driver prefers to listen (music of driver's choice; hereafter referred to as Driver Selected Music (DSM)), listening to music which is selected by the researcher (music of researcher's choice; hereafter referred to as Researcher Selected Music (RSM)), listening to FM radio conversation (hereafter referred to as FMRC), and driving without listening to any music or FM radio conversation, i.e., baseline (hereafter referred to as Driving Without Music (DWM)).It is possible that cognitive workload plays a role if there are significant differences among DSM, RSM, FMRC, and DWM in terms of their effect on driving performance measures.The more compromise in driving performance, the more demand for cognitive resources by the respective secondary task.It is pertinent to mention here that the rationale behind choosing the above mentioned four experimental conditions is that they match real-life driving scenarios.

Participants
A group of 64 young male drivers aged 18-25 years (M = 21.8,SD = 1.4) voluntarily participated in this distracted driving study conducted on a driving simulator.Out of them, eight participants were not able to complete the driving task because of simulator sickness, so their data was discarded and not considered for analysis.So, data from 56 participants were considered for final analysis.All drivers held valid driving license and drove at least 5,000 km per year.
They had normal vision or were corrected to normal vision.All the drivers claimed to have a normal hearing without any deficiencies.The participants were recruited from the campus of one of the authors' Institute and from other surrounding academic institutions.For their voluntary participation in the study, each participant received 350 Indian rupees (INR).The study's objective was made clear to the participants.They were assured that the data collected from them would only be utilized for research purposes and that strict confidentiality would be maintained.Following that, each participant's signed consent was collected.

Instrumentation
The experiment used a fixed base driving simulator of the Vertex TM CSV 2001 Dynamic model.The simulator consisted of a steering wheel with force feedback, a brake pedal, an accelerator pedal, and a clutch pedal.The driving scene was displayed on three 24-inch full HD monitors, each having a resolution of 1,920 × 1,080.The visuals cover a field of view of around 180° with a minimum of 120° in front and 120° in the rear.The driving simulator setup is represented in Figure 1.
The driving track consisted of a 12 km semi-urban driving environment with oncoming and ongoing traffic also present on the road.Directions for driving were presented on the screen throughout the trial.An outline of the driving track is shown in Figure 2. The first 6 km of the track (from location A to B, as shown in Figure 2) consisted of a double-lane road with a dividing line and the next 6 km (from location B to C) consisted of a single lane road.Other critical details of the driving track are listed below: 1.There were 10 intersections in total, out of which 4 were right turns, 4 were straight through, and 2 were left turns.2. When turning right, there was oncoming traffic coming straight through or oncoming traffic turning left, which has to be given the right-of-way (see Figure 3).
3. Stop signs were displayed while approaching a major road and traffic from the left has priority.4. A number of signs were used to give driving related information to the participants.The signs along with their description and number are presented in Figure 4.

Design
This experiment employed a within groups design with 4 treatment conditions (DSM vs. RSM vs. FMRC vs. DWM).
Each participant was exposed to all four treatment conditions.In order to rule out the practice effect or the fatigue effect, the treatment conditions were randomized so that the sequence of the treatment conditions is not affecting the   driving of the drivers.In other words, each participant drove the 12 km loop 4 times and the 4 treatment conditions were randomly balanced across all the 64 participants.The speed limits were displayed to the drivers at important places, for example, at curves, near to residential areas (please see Figure 2).The drivers were required to drive in a simulator under 4 conditions with one no distracting condition (DWM) and three distracting conditions.The three distracting conditions were: (i) listening to music which was selected by the researcher (RSM), (ii) listening to music which was selected by the driver (DSM), and (iii) listening to FM radio conversation (FMRC).During DWM (which served as baseline), the participants were not engaged in any activity other than driving.For the RSM condition, the researcher selected six Telugu (local regional language) songs (recent weekly top songs suggested by jiosaavn music playlist; jiosaavn is an Indian music app.), which could be played for 20 min (that is the maximum time in which the whole distance (12 km) could be covered).Whereas for the DSM condition, the participants were given the choice of selecting the music (six songs) of their preference before they actually participated in the drive.For FMRC condition, the researchers recorded a Radio Jockey's talk which played on a local popular radio channel known as Radio Mirchi 98.3 FM.Generally, in FM radio programs there would be songs playing intermittently during conversations with celebrities or the RJs talking about any daily life issues.One such conversation was recorded by using a Motorola mobile which had an inbuilt feature of audio recording.The duration of the recorded talk was 18 min.Based on the practice trials, the average time required for completing driving in any treatment condition was recorded to be 15 min.Considering that some drivers might drive slowly, an extra time of 3 min (in the case of FMRC) and 5 min (in the case of RSM and DSM) were given to participants.However, it is worth to mention that no participant needed beyond 15 min time to drive in either of the treatment conditions.Driving errors were recorded with the help of "CarnetSoft" an inbuilt software of the driving simulator used in the study (Vertex Research Centre 2015).Each instance of error was instantly recorded and stored in Student Assessment System (a data logger, which is also a part of the driving simulator).The methodology followed in the current study is shown in Figure 5.

Procedure
Before exposing the participants to experimental conditions, they were given a practice trial so that they get accustomed to all the functions of the driving simulator.During the practice session, participants followed the on-screen navigation instructions, and turn-by-turn audio navigation instructions were also given.After completing the practice session each participant was assigned to all four experimental conditions in a random order.

Data
Generalized Estimating Equations (GEE), is a statistical technique that is used to analyze correlated data, such as repeated measurements from the same subjects, across different conditions.In this study, GEE was employed to model the relationships between all four experimental conditions and the number and types of errors committed by drivers.The recorded data from "CarnetSoft" showed that there were numerous zero error counts against many error types, which indicates that such errors were not at all committed by the drivers, whereas some error types were committed with high frequency.Four models were developed and tested.First model was developed by considering only the types and counts of driving errors (please refer Table A1, see online supplement).The remaining three models were created by using a negative binomial log link function (a statistical technique) of GEE model.For any count data, which includes data with a high frequency and zeros (as is observed in the current study), this function works well.The inter-correlation between repeated measures (observations from the same subjects under four different conditions) was specified as unstructured, which helped us in capturing the potential variations in correlation patterns.Results with p values less than .05were considered statistically significant.Furthermore, the first model examined the difference in the total number and types of errors made across all experimental conditions.This model aimed to determine if there were significant variations in error counts and types across the four experimental conditions.The second model focused on the distracted conditions when the drivers were exposed to "RSM" and compared their error rates and types when they were exposed to undistracted conditions (i.e., exposed to "DWM, " a baseline condition).This model aimed to understand how the RSM distraction condition impacted the error rates and types.Similarly, the third model analyzed the differences between the "DSM" condition and the undistracted condition, i.e., "DWM." The fourth model examined differences between the "FMRC" condition and the "DWM" condition.Statistical analyses were conducted by using the SPSS 26.0.Throughout this paper, the alpha level of 0.05 was used to indicate statistical significance.

Results
A list of driving errors, as committed by drivers in each experimental condition, was collected from the driving simulator.Table A2 (see online supplement) shows different error types, their frequency and percentage across undistracted and distracted conditions.During baseline condition, drivers committed 363 errors and other 638, 596, and 731 errors during RSM, DSM, and FMRC, respectively.Drivers committed errors in each experimental condition, with the average number of errors made per drive being highest while listening to FMRC (13.5), followed by RSM (11.8),DSM (11.0), and baseline or DWM condition (6.7).Results of the first GEE model indicated that the drivers were 190% more likely to make an error while listening to an FMRC (Exp (B) = 2.969, p = .000);120% more likely to make an error during listening to RSM (Exp (B) = 2.247, p = .000);and 89% more likely to make an error during listening to DSM (Exp (B) = 1.893, p < .0001).
A second GEE model examined the differences in the distracted (RSM) and undistracted drivers (DWM).Out of all errors, only seven errors had an odd ratio (Exp (B)) value more than 1; and five errors had significant value.In the results tables the errors which have an Exp(B) value more than 1 are reported.Exp(B) value of more than 1 means that such errors have higher probability of being committed as compared to those which have an Exp(B) value of less than 1.The results of the model are presented in Table 1.The results demonstrate that drivers were significantly more likely to drive too fast while turning off at a junction, crossing solid white left line of the driving lane (i.e., driving off the road on the left), forgetting to use indicator before changing from the right to the left lane, pressing wrong indicator (e.g., indicating left before turning to right lane), and driving at high speed when they were distracted by RSM than when they were not distracted (DWM).The odds of the errors "at intersection, didn't give way to the traffic approaching intersection from the left side of the driver" and "forgetting to press right indicator before changing lane from left to right" did not differ significantly across the distracted (RSM) and undistracted conditions.
A third GEE model was fitted to examine differences across the distracted (DSM) and undistracted (DWM) drivers.Out of all errors observed in the study, only eight errors had odds ratio (Exp(B)) value more than 1; and five errors had significant value.The results of the model are presented in Table 2 and it indicates that drivers were significantly more likely to drive too fast while turning off at junctions; pressing wrong indicator; at intersections, didn't give way to the traffic approaching intersection from the left side of the driver as well as from the right side of the driver when distracted than when they were not distracted.The odds of "crossing solid white left line of the driving lane, " "forgetting to press left indicator before changing lane from right to left, " and "forgetting to press right indicator before changing lane from left to right" did not differ significantly among distracted and undistracted conditions.
A fourth GEE model was fitted to examine differences among the distracted (FMRC) and undistracted (DWM) drivers.Out of all errors observed, only five errors had an odds ratio (Exp(B)) value more than 1; and two errors had significant value.The results demonstrated that during driving, when drivers were distracted by FMRC, they were significantly more likely to drive fast when turning off at a junction, and also drive at high speed at the straight through, as compared to when they were not distracted (see Table 3).The odds of "crossing solid white left line of the driving lane, " "pressing wrong indicator, " "didn't give way to the traffic approaching intersection from the left side of the driver" did not differ significantly across the distracted (FMRC) and undistracted (DWM) conditions.

Discussion
This study was conceptualized to investigate the frequency and types of errors that drivers commit when they are distracted in comparison with when they are not distracted.While driving in a simulator, drivers were also engaged in listening to music selected by researcher, music selected by the driver, and listening to FM radio conversation.The results demonstrate that not only did the drivers commit driving errors during distracted driving but they also committed errors when they were not distracted.This indicates that in order to understand the mechanism underlying the commission of driving errors, the researchers need to employ systemic approach, i.e., taking into consideration all the factors involved in driver-vehicle interaction.In the context of current experiment, the higher number of driving errors across all the three distracted conditions as compared to that of non-distracted driving condition ascertains that distracted driving is certainly one of the factors responsible for driving errors.Drivers need to drive within legal speed limits, particularly in urban traffic conditions, any deviation from the speed limits would bring negative consequences, and for example, accidents would take place if the driver drives at high speeds and traffic jams would happen if the driver drives at slow speeds.The most common error that drivers committed was "speed too high" (if the driver drives at a speed which is more than the specified speed limit, then it is counted as a speed too high error), with 183 instances during FMRC condition, followed by 156 instances during RSM condition, 133 occurrences during DSM condition, and 41 occurrences during baseline condition (i.e., DWM).Similarly, another common error observed was also related to speeding.In 178 instances, "driving too fast while turning off at junction" error occurred during FMRC condition, followed by 136 occurrences during RSM condition, 131 instances during DSM condition, and only 43 instances during DWM condition.This is in congruence with Catalina et al. (2020), who also reported that irrespective of the type of the music that drivers listen to while driving, the probability of committing speeding violations increase by 3% as compared to when they are not listening to any kind of music.Furthermore, they also argued that novice drivers tend to drive at an adequate speed when they are not listening to any music but when they listen to music while they drive, they are the ones who are mostly affected in terms of committing speeding errors.
As mentioned above in objectives and hypothesis section, people have limited information processing capacities and if the demands of the task are such that it exceeds the available resources, then a performance compromise ought to happen.Allocation of cognitive resources toward one task (e.g., listening to music or FM radio conversation, in the context of current study) limits the resources available for other tasks (e.g., performance on driving task) (Moreno and Park 2010).The results of the current study have demonstrated that the drivers committed a greater number of errors when they listened to FMRC, followed by RSM and DSM.This indicates that listening to FMRC demands more cognitive resources as compared to the other three experimental conditions.There are similar results reported by Jäncke et al. (1994) who found that driving performance is compromised when the drivers listen to radio conversations during driving.They reported that drivers committed more lane excursions.
In addition to frequency of errors, another aim of the current study was to delve deeper in understanding the type of errors that drivers commit while driving and listening to music simultaneously.Pertaining to this, three models were fitted: differences in the distracted (listening to RSM) and undistracted drivers; differences in the distracted (listening to DSM) and undistracted drivers; and differences in the distracted (listening to FMRC) and undistracted drivers were analyzed in the second, third and fourth model, respectively.It is observed that errors related to speeding (i.e., driving at a very high speed both on the urban roads as well as while turning at intersections) are common across all the three  2009;Stanton and Salmon 2009;Hughes et al. 2013;Wang et al. 2015).
Another line of thought for more compromise in driving performance during listening to FMRC could be that as compared to listening to music (both RSM and DSM), listening to the conversation is more cognitively demanding as well as engaging.The successful performance of driving (a complex task) is dependent upon the amount of attentional resources demanded and the availability of attentional resources to accomplish those demands (Van Vuuren 1987).Given the fact that we have limited attentional resources (Kahneman 1973) any substantial increase in the cognitive workload associated with the performance of a concurrent task (in this context, listening to FMRC) would reduce the available attentional resources that otherwise could have been allocated to driving.In the context of driving, a number of studies have shown that the more demanding the secondary task is the more compromised is the driving performance (e.g., Jäncke et al. 1994;Briem and Hedman 1995;Patten et al. 2004;Horberry et al. 2006).During DSM condition, the participants listened to music that they were already familiar with.Familiarity with any task reduces the demands on cognitive processes and the same principle holds true for listening to music (Peelle 2018).Greater number of driving errors during listening to FMRC indicates that as compared to listening to music (both RSM and DSM), listening to FMRC consumed more cognitive resources and left relatively lesser cognitive resources available for driving task (Lavie 2001(Lavie , 2005) ) whereas relatively lesser number of driving errors during listening to music (both researcher selected and driver selected) condition indicates that listening to music consumed less cognitive resources and left more cognitive resources to be used in driving task.
The current study was conducted with the intention of comparing the effects of listening to music, listening to FM radio conversation, and driving without any distraction on driving performance.Listening to music was manipulated in terms of researcher selected music, driver selected music, and FM radio conversation.This study was conducted in an artificial driving environment (consisting of 12 km semi-urban driving track) by using a fixed base driving simulator.The results show that as compared to researcher selected music and driver selected music, there is more compromise in driving performance while the drivers listen to FM radio conversations.Furthermore, with respect to types of errors, the results demonstrate that drivers are more likely to commit more speeding errors and traffic violations when they listen to music or listen to FM radio conversations than when they are not.

Limitations
The present study was conducted by using a driving simulator popular in distraction-related research as it provides a safe and controlled environment that is difficult to achieve in field studies.However, there is a possibility of exaggerated rate of committing driving errors.The drivers generally have a low-risk perception about meeting an accident when they drive in a simulator, so they tend to take more risks.Future researches could focus on investigating the relationship between listening to different types of music (example, rock music, pop music, or classical music), physiological arousal and driving performance by taking into consideration different age groups, driving experience, and personality factors of drivers.Moreover, in order to have a better understanding about driving errors, the future work could consider a comparison between elderly drivers and younger drivers.

Figure 1 .
Figure 1.representative image of a participant driving in the simulator.

Figure 2 .
Figure 2. an overview of experimental track used in the present study.

Figure 3 .
Figure 3. at an intersection, left traffic had right of way, but failed yielding behavior.Similarly, right traffic had right of way, but failed yielding behavior.

Figure 5 .
Figure 5. Methodology followed in the present study.

Figure 4 .
Figure 4.The symbols, along with their description and number, used to give driving related information to the participants.

Table 1 .
results of Gee model (baseline vs listening to researcher selected music).
Note: Significant results p < .05 in bold.

Table 3 .
results of Gee model (baseline vs listening to fM radio conversation).
Note: Significant results p < .05 in bold.

Table 2 .
results of Gee model (baseline vs listening to driver selected music).Significant results p < .05 in bold.differentexperimental conditions.Furthermore, other types of errors that are committed by drivers are categorized as violations (example, crossing solid white line, at intersection didn't give way to traffic approaching intersection either from the left side or the right side of the driver); wrong action (example, pressing wrong indicator); and inattention (example, forgetting to press the indicator before changing the lane).With respect to the types of errors committed by drivers because of distractions, the findings of the current study are also in line with previous researches (e.g., Sandin