The Effects of Age, Distraction, and Simulated Central Vision Impairment on Hazard Detection in a Driving Simulator

Abstract

Significance.

Despite similar levels of visual acuity and contrast sensitivity reductions, simulated central vision impairment (sim VI) increased response times to a much greater extent in older than younger participants.

Purpose.

Driving is crucial for maintaining independence in older age, but age-related vision impairments and in-vehicle auditory distractions may impair driving safety. We investigated the effects of age, simVI (reduced visual acuity and contrast sensitivity) and auditory distraction on detection of pedestrian hazards.

Methods.

Thirty-two normally-sighted participants (16 older) completed four highway drives in a simulator and pressed the horn whenever they saw a pedestrian. Pedestrians ran toward the road on a collision course with the approaching vehicle. SimVI was achieved by attaching diffusing filters to a pair of lab goggles which reduced visual acuity to 20/80 and contrast sensitivity by 0.35 log units. For drives with distraction, subjects listened to an audiobook and repeated out loud target words.

Results.

SimVI had a greater effect on reaction times (660 ms increase) than age (350 ms increase) and distraction (160 ms increase), and had a greater effect on older than younger subjects (828 ms and 492 ms increase, respectively). SimVI decreased safe response rates from 94.7% to 78.3%. Distraction did not, however, affect safety because older subjects drove more slowly when distracted (but did not drive more slowly with simVI), suggesting that they might have perceived greater threat from the auditory distraction than the simVI.

Conclusions.

Older participants drove more slowly in response to auditory distraction. However, neither older nor younger participants adapted their speed in response to simVI, resulting in unsafe detections. These results underline the importance of evaluating safety of responses to hazards as well as reaction times in a paradigm which flexibly allows participants to modify their driving behaviors.

Driving is vital in maintaining one’s independence and quality of life in the United States, especially with increasing age.¹ However, older drivers are also more frequently involved in motor vehicle collisions.² Both age-related degradations of vision^3–5 and cognitive resources⁶ may impair the ability of older drivers to respond to hazards in a timely manner, which may put them at increased risk for crashes. In addition, auditory distractions are becoming increasingly more common in today’s driving environment⁷ and may adversely affect hazard detection. Therefore, an important question is how impairments in vision interact with auditory distraction to affect hazard detection in both older and younger drivers.

Drivers often devote a portion of their attention resources to auditory activities such as communicating with passengers, listening to an audiobook, or interacting with in-vehicle navigation and assistance systems. Mental effort is required to engage in these activities, which decreases the amount of cognitive resources available to be devoted to the driving task which could have a negative impact on driving performance.⁷ However, the nature in which vision impairment and auditory distraction interact to affect detection of hazards while driving has yet to be fully understood. Therefore, the current study is aimed at assessing the effects of age, central vision impairment and auditory distractions, and their interactions, on detection of moving pedestrian hazards.

Lee et al.⁸ examined the effects of simulated reduced visual acuity, using positive sphere lenses, and auditory distraction on hazard perception performance of older and younger drivers. The auditory distraction task involved determining whether a given navigation instruction was logical or impossible to follow according to the road layout. Their results suggested that simulated visual acuity loss and auditory distraction both led to an increase in reaction times to hazards. Older drivers were more greatly distracted and responded more slowly to hazards than younger drivers. Lee and colleagues⁸ made use of a standard video-based hazard perception paradigm, which involved watching a series of video clips but did not require subjects to be actively engaged in motor coordination tasks important for driving such as steering or operating the brake and accelerator.⁸

Wood et al.⁹ conducted a study that examined the effects of central vision impairment (simulated reduced visual acuity and simulated cataracts with reduced visual acuity and contrast sensitivity), age, and distraction on a closed-road driving course.⁹ Distraction was created using a mental arithmetic task that was either visual or auditory (reporting the sum of two numbers that were either presented on the dashboard or were played through speakers). The authors investigated the effects of these factors on road sign recognition and hazard avoidance. A key finding was that the combined adverse effects of visual distraction and simulated cataracts were significantly greater for older than younger participants. However, the hazards along the course were large pieces of low contrast foam,⁹ which were stationary, and did not pose real danger to the subjects. By comparison, in open-road driving, many hazards are moving and present real collision risks.

Therefore in the current study we used a driving simulator to provide a safe, controlled and replicable experimental environment in which to evaluate the effects of age, central vision impairment and auditory distraction on detection of moving pedestrian hazards^{10, 11} in scenarios where participants were actively engaged in controlling vehicle speed and steering. One potential way to compensate for vision impairment, and/or auditory distraction would be to drive more slowly (as has been reported for drivers with real central vision impairment¹²). Thus, an important advantage of simulated driving tasks over video-based hazard perception tasks is that subjects are able to control the speed at which they drive. If age, vision impairment, or auditory distraction increased participants’ reaction times, it might be assumed that the responses would be less safe. However, if the participants also compensated by driving more slowly, a longer reaction time would not necessarily be less safe than a shorter reaction time when driving more quickly. We, therefore, measured both reaction times to the pedestrian hazards and the safety of the response (i.e., whether or not detection was made in enough time to avoid a collision given the speed and distance of the car to the pedestrian at the time of detection). As in several prior studies,^{9, 13–15} we used a simulation of central vision impairment comprising reduced visual acuity and reduced contrast sensitivity. Simulated vision impairment ensures a relatively homogeneous degree of impairment, which can be difficult to achieve with a heterogeneous sample of people with real vision impairment. Auditory distraction was created by instructing subjects to listen to audiobook excerpts while driving, a task that was relevant to both younger and older drivers and required a continuous level of engagement throughout the drive.¹⁶

Our primary hypotheses were that age, simulated vision impairment and distraction would each adversely affect detection performance (increasing reaction times and reducing safe detections) and that the three factors would interact to further impair detection performance. Specifically, we predicted that the effects of distraction and simulated vision impairment would be greater when the two factors were combined and that both the effects of simulated vision impairment and distraction would be greater for older than younger participants. Finally, we expected that subjects might drive more slowly to compensate for the simulated vision impairment and/or auditory distraction.

METHODS

This study implemented a 2 (levels of age - younger vs. older) × 2 (levels of distraction - with vs. without distraction) × 2 (levels of simulated vision impairment - with vs. without simulated vision impairment) mixed design. The study was conducted in accordance with the tenets of the Declaration of Helsinki and was approved by the institutional review board (IRB) at the Schepens Eye Research Institute.

Participants

A total of 32 subjects (Table 1), 16 younger adults (range 22 – 37 years) and 16 older adults (range 60 – 82 years), were recruited from the New England College of Optometry and a database of subjects at the Schepens Eye Research Institute. Subjects were English-speaking, current drivers with at least two years of driving experience. They had no adverse ocular history of eye disease that might affect visual acuity or visual fields. All subjects were able to participate in a conversation without any hearing difficulties (without or with a hearing aid - as evaluated by the experimenter during the informed consent process). All subjects’ vision exceeded the requirements for an unrestricted license in Massachusetts (visual acuity of at least 20/40 and a 120° visual field extent¹⁷). The level of lens opacity among older subjects, as measured in accordance with the LOCS III grading system¹⁸ (SL990 slit lamp; SCO Ophthalmic, Scandicci, Italy), was on average trace to grade 1+ nuclear sclerotic, anterior cortical, and/or anterior subcapsular cataract. No subject had cataract grading worse than grade 2. Only one rater (the first author) graded lens opacity. Two older subjects were pseudophakes.

Table 1.

Demographics, vision measures and driving exposure for study participants.

	Younger (n=16)	Older (n=16)	P-value*
Demographics#
Age [years]	27 (6.7)	67 (4.9)	<.001
Male [n]	7	8	1
Vision measures#
Visual Acuity without goggles [logMAR]	.02 (0.05)	.06 (0.09)	.03
Visual Acuity with goggles [logMAR]	.62 (0.14)	.58 (0.14)	.73
Contrast Sensitivity without goggles [logCS]	1.78 (0.06)	1.67 (0.11)	.001
Contrast Sensitivity with goggles [logCS]	1.43 (0.08)	1.33 (0.11)	.001
Driving exposure†
Number of days driven in a week	3.5 (.8 – 7)	5 (2.4 – 7)	.97
Average travel distance per week [km]	64 (28 – 209)	89 (46 – 227)	.69

^*P-values are for the difference between older and younger subjects

^#Mean (SD);

^†Median (IQR)

Driving Simulator and Pedestrian Detection Scenarios

A high-fidelity LE-1500 driving simulator (FAAC, Inc., Ann Arbor, MI) with five 42-inch liquid crystal display monitors (LG M4212C-BA, native resolution of 1366 × 768 pixels; 225° × 32° field of view) was used. The simulator included a motion-based seat, with three degrees of freedom in its movement, standard driving controls for an automatic transmission vehicle, and an interior dashboard modeled on a Ford Crown Victoria (see Figure 1). Data from the driving simulator, including the subjects’ vehicle control inputs (steering, horn presses, etc.), and positional information of the vehicle and scripted entities in the world, were logged at 30Hz throughout the drive.

Figure 1.

LE-1500 driving simulator (FAAC, Inc., Ann Arbor, MI) used in this study. The auditory distraction task was delivered though a pair of speakers (Amazon Basics A100 Multimedia Speakers, China) positioned to the left and right of the participant’s seat (white circle). A video recorder (yellow circle – DVC, WEILIANTE Co., Ltd, China) was positioned above and behind the car seat to simultaneously record participants’ repetition of target words in the auditory distraction task along with video of the central screens of the simulator (to facilitate scoring of auditory task performance for segments without and with pedestrians on the screen).

Scenarios and pedestrian events were created using the Scenario Toolbox (FAAC Inc., Version 3.9.4.25873) and were based on scenarios used in prior driving simulator studies of pedestrian detection by drivers with real vision impairment.^{10, 11} In brief, four rural highway routes were used, each including oncoming traffic (average 16 vehicles, range 15–19) and 12 moving pedestrians (2.0 m tall, subtending 0.9° vertically when they first appeared, and dressed in a white shirt and blue jeans). The routes were along undivided highways (60 mph posted speed limit, equivalent to 97 kph) with a single lane in each direction, including long straight sections and long gentle curves without any sharp turns (lessens the likelihood of simulator sickness), no stop signs or traffic lights, and no lead cars. There were, however, speed limit signs and directional road signs. The background scenery contained grassy hills with some trees (see Figure 1) and buildings along the side of the road.

Pedestrians appeared at pseudorandom intervals every 15 to 60 seconds. They appeared from either the left (n = 6) or right side (n = 6) of the road when the car was approximately 5 seconds away (if travelling at the set speed limit of 97 kph). Pedestrians were programmed to run at a constant velocity toward the road on a trajectory perpendicular to the subject’s travel lane, as if to cross in front of the subject’s vehicle. If the subject’s vehicle was travelling at 97 kph, the pedestrian was on a direct collision course with the subject’s vehicle (with a constant bearing angle between the vehicle and the pedestrian) and the two road users would have reached the collision point in the middle of the driving lane at the same time. If the participant drove more slowly than the speed limit, then the pedestrian would have reached the collision point before the car, but would still have appeared as a potential hazard (unless the participant drove extremely slowly). However, to avoid upset from actual collisions, the pedestrians were programmed to stop just before entering the participant’s driving lane. We instructed subjects to drive as close to 97 kph (60 mph) as possible to ensure that the pedestrians were on a true collision course (a constant bearing angle) as well as to increase the difficulty of the driving task. The subject’s task was to press the horn as soon as they perceived a pedestrian; they were not asked to make any judgments about the level of hazard posed by the running pedestrian.

Simulated Central Vision Impairment Goggles

When visual acuity falls below the minimum requirement for an unrestricted license (e.g., 20/40 in Massachusetts), the majority of jurisdictions in the United States still permit people to drive but on a restricted license (e.g. no highway driving, or must use a bioptic telescope¹⁹). We selected to simulate a level of visual acuity (20/80) which would be within the range permitted for a restricted license in many states.¹⁷

To simulate the central vision impairment, Bangerter diffusing filters (Fresnel Prism & Lens Co., Eden Prairie, MN) were fitted to both lenses of a pair of laboratory safety goggles (Guardian Pro Over-The Glass, Safety Glasses USA Inc., Three Rivers, MI). These filters have been widely used in prior studies^{14, 15, 20} to simulate central vision impairment. The filters reduced subjects’ visual acuity and contrast sensitivity (Table 1) without reducing their binocular field of view. Based on a prior study,¹⁵ we used a combination of a 0.1 opacity filter applied on the front surface and a 0.4 opacity filter on the back surface of the safety goggles’ lenses. Subjects always wore the simulated vision impairment goggles (hereafter referred to as “goggles”) over their habitual corrections for testing of their visual acuity and contrast sensitivity, as well as for the drives in the simulator.

Our aim was to achieve a similar amount of visual acuity reduction in the older and younger age groups. Therefore, for each of the older participants, we verified that their binocular visual acuity with the goggles was within ± 0.1 logMAR of the average binocular visual acuity of the younger participants with the goggles (Table 1 - this was possible because the data for younger participants were collected before the data for older participants). There was only one older participant whose visual acuity with the goggles was outside ± 0.1 logMAR of the average of the younger participants. For this participant a different pair of goggles was used with a 0.1 opacity filter on the front surface of the lenses only, which reduced that participant’s visual acuity to 20/80.

Auditory Distraction Task

In selecting an auditory distraction task, we had three main criteria: the task would have face validity and be relevant to both older and younger participants; the task would be neither too easy for younger subjects nor too difficult for older subjects; and the task would be continuous throughout the drive encouraging a constant level of engagement from the participant. Mental arithmetic^{9, 21} and n-back²² tasks are commonly used as distraction tasks in driving research, but they lack relevance and face validity (i.e. they are not tasks one would normally perform when driving). On the other hand, listening to an audiobook is something that people do when driving and is relevant to drivers of all ages.

The audiobook distraction task involved listening to professional recordings of two short excerpts from the novel “Boy” by Roald Dahl during two of the four experimental drives. A third extract from “The Hitchhiker” by Roald Dahl was used during the practice drive. The reading difficulty of both audiobook excerpts were calculated using the “Flesch Reading Ease Score (FRES)” (www.readabilityformulas.com). Both excerpts were considered fairly easy – easy to read (Excerpt 1 – 70.2 FRES – fairly easy to read; Excerpt 2 – 82.9 FRES – easy to read). Recordings were played through computer speakers (Amazon Basics A100, China; Multimedia Speakers). During the practice drives we adjusted the volume of the audiobook task for each individual so that it could easily be heard above the simulator’s engine noise. Engagement in the audiobook task was ensured by requiring subjects to state out loud two predetermined target words whenever they were heard in the excerpts and by asking two comprehension questions at the end of each drive. Younger subjects answered the questions correctly 74.2% of the time, while older subjects were correct 62.1% of the time. Subjects’ responses on the audiobook task were video and sound recorded using a video recorder (DVC, WEILAND Co., Ltd, China), which was positioned above and behind the car seat (see Figure 1 – yellow circle).

The time taken to complete each drive differed between subjects (depending on their driving speed). As a consequence, the total number of target words presented by the end of each drive also varied across subjects. Thus, one high frequency target word (“the”) that was present throughout all three audiobook excerpts was chosen along with a one medium frequency word present throughout each excerpt (“school”, “father” or “mother”). The first excerpt (for experimental drives) was 10:48 minutes long, where “the” took up 5% (97 words) of the total word count and “father” took up 1% (21 words). The second excerpt (for experimental drives) was 10:29 minutes in length, where “the” took up 5% (93 words) and “school” 0.5% (13 words) of the total word count. The third excerpt for the practice drive was 11:47 minutes in length, where “the” took up 4% (82 words) and “mother” 2% (41 words) of the total word count.

Procedures

Subjects’ binocular distance visual acuity was measured with a logMAR chart (Test Chart 2000 PRO, Hertfordshire, UK) at 120 cm, which was similar to the distance between the car seat and the center monitor in the driving simulator. Subjects’ binocular letter contrast sensitivity was measured with the MARS letter contrast sensitivity chart at 50 cm (The Mars Perceptrix Corporation, Chappaqua, NY). Each test was performed once without the goggles (either without or with the habitual correction, as appropriate for that viewing distance) and again with the addition of goggles (over subjects’ glasses, if worn).

In the driving simulator, subjects first completed an acclimatization drive in a highway scenario, with no traffic or pedestrians, to become familiar with controlling the virtual car. A practice drive followed, which contained all the elements of the experimental drives (including pedestrian events and oncoming vehicles). Subjects began the practice drive without any distraction and without simulated vision impairment and were instructed to press the horn as soon as a pedestrian was seen. After they had driven for about two minutes, they stopped the vehicle to put on the goggles and wore them for the rest of the practice drive. After another two minutes of driving, they stopped the vehicle and were given instructions on the auditory task. Subjects practiced performing the auditory task without driving until they were sufficiently comfortable with it, after which they began driving again and completed the rest of the practice drive with the goggles and the auditory distraction task.

Once subjects were comfortable with driving while wearing the goggles and performing the auditory task, they began the experimental drives. Subjects completed a drive in each of the four conditions: 1) normal vision and no auditory distraction, 2) simulated vision impairment and no distraction, 3) normal vision and auditory distraction, and 4) simulated vision impairment and auditory distraction combined. At the end of the experimental drives with auditory distraction, subjects answered two comprehension questions based on the excerpt to which they had listened. The order of the conditions was counterbalanced across subjects using a Latin Square Design in order to counteract any learning and adaptation effects. The order of the two audiobook excerpts was also counterbalanced across the two audiobook drives. Each drive lasted approximately 8 to 10 minutes depending on the speed at which the subject drove.

Subjects were instructed to obey all road regulations, to drive as close to 97 kph as possible while maintaining safe control of the vehicle and to press the horn as soon as they saw a pedestrian. We implemented a speed cap at 97kph so that subjects were not required to monitor their exact speed. Thus, to maintain a constant 97 kph, subjects were only required to have their foot on the gas pedal and could take their foot off the gas pedal and engage the brake as needed to slow down (e.g. when going round curves). Subjects had full control of vehicle steering. During the drives with auditory distraction, subjects were told that their performance on both the driving and the auditory task were equally important and that performance on both tasks was being scored.

Performance Measures

Detection performance measures included detection rates (percent of total number of presented pedestrians that were detected), reaction times (from the time of pedestrian appearance to the horn press) and safe response rates (percent of total number of presented pedestrians for which there was a safe response). The safety of a response was calculated based on the deceleration (m/s²) that would have been required to stop the vehicle before reaching the collision point with the pedestrian:

$$\text{Deceleration}=\frac{V_f^2-V_h^2}{2D_h}$$

Where V_f is the final velocity that was fixed to be 0, V_h is the vehicle’s velocity at the time of horn press, and D_h is the vehicle’s distance to the collision point at the time of the horn-press. The collision point was the location where the front of the car would have intersected with the pedestrian, assuming that the pedestrian had continued on its trajectory. If the necessary deceleration to stop the car prior to the collision point was greater than −5 m/s² the detection was considered unsafe.²³ In addition, any event where there was no horn-press response (i.e., the pedestrian was not detected) was categorized as unsafe. The cut off of −5 m/s² was chosen because it is commonly considered to be an unsafe deceleration rate.²³ Decelerating at this rate would result in the driver either becoming a hazard to themselves (if the road conditions were wet or slippery) or to the driver in the car behind them.

Vehicle speed was computed for two pre-defined straight road segments in each drive.²⁴ These segments were selected to be pedestrian-free because reaching to press the horn might affect vehicle speed. As simulator data were recorded at 30 Hz, a straight segment 200 m long driven at 97 kph (26.9 m/s) would have 223 samples from which each of the measures was computed.

Performance on the audiobook task was scored from the video recordings. The number of target words correctly repeated out loud was counted for the time between pedestrian events (when no pedestrian was in the scene) and the time during pedestrian events (when the pedestrian was in the scene). In some cases, there were no target words during a pedestrian event because they were quite brief (e.g., 5 s). Therefore, the total number of words repeated across all pedestrian events was tallied for each subject for each drive and expressed as a proportion of the total number of target words presented during pedestrian events in that drive. The proportion of words correctly repeated between pedestrian events was computed in a similar manner.

Statistical Analyses

Prior to running our analyses, we excluded data where a subject had pressed the horn but it did not register due to mechanical issues. This removed a total of 12 trials, leaving a total of 1536 trials that went into our subsequent analyses.

We evaluated the effects of age (older vs. younger), distraction (with vs. without distraction), and simulated vision impairment (with vs. without goggles) on detection performance and vehicle speed. Only one older subject missed one pedestrian in the audiobook condition. Otherwise all pedestrians were detected. Therefore, detection rates for older (average: 99.8%) and younger (100%) subjects were at ceiling in all conditions and were not analyzed further. However, reaction times and safety of the detections varied sufficiently across conditions and age groups for us to conduct statistical analyses. We also evaluated subjects’ audiobook task performance by comparing the effects of age and simulated vision impairment on the rate of correctly repeated words between pedestrian and during pedestrian events.

Reaction times were analyzed with Linear Mixed Models (LMMs) constructed in the R statistical programming environment. A log10 transformation was applied to normalize the raw reaction time data. We then winsorized our data by removing all outliers greater than ± three standard deviations from the mean.²⁵ This removed only 0.3% of the total data. All analyses were performed on the transformed data; however, results are reported in milliseconds for ease of interpretation. Binomial outcome variables (safety of responses) were analyzed by means of Generalized Linear Mixed Models (GLMMs).²⁶

We assessed the overall effect of age (young or old), simulated vision impairment (with and without goggles), and distraction (with and without the audiobook task) by entering these as fixed factors in our models. In order to examine the interaction of age with both distraction and simulated vision impairment, we dummy coded simulated vision impairment (0 – without goggles; 1 – with goggles) and distraction (0 – not distracted; 1 – distracted) variables. We also constructed random effects structures for subject (to account for any variability contributed by individual differences between our subjects) and unique pedestrian events (to remove any random variability contributed by the different pedestrian events as well as to remove any variability contributed by learning / practice effects). In addition, contrast sensitivity was included as a random factor, but visual acuity was not, because prior studies found that contrast sensitivity (but not visual acuity) was a strong predictor of hazard response times,^{9, 15} and baseline contrast sensitivity differed significantly between the older and younger groups (Table 1).

P-values for main effects were estimated by means of the lmerTest package.²⁷ Any interactions between age, distraction, and simulated vision impairment were computed by means of model comparisons. For each interaction, we compared a baseline model (with all interactions between the fixed factors removed) with the same model plus the interaction of interest by means of analyses of variance (ANOVAs). The resulting χ² values represented the significance of the interaction of interest.

The effects of age, simulated vision impairment and distraction on measures that were not normally distributed (vehicle speed and word recall rates in the auditory task) were analyzed by means of non-parametric tests. The effect of age group (young vs old; a binary factorial variable) was analyzed by means of an independent-samples Mann-Whitney U test. The effect of condition (within-subjects factor with 4 levels) was analyzed by means of a Kruskal-Wallis rank sum test. The difference in audiobook recall rates between and during pedestrian events was analyzed by means of a paired-samples Wilcoxon Signed Rank Test.

RESULTS

Sample Characteristics

Sample characteristics are provided in Table 1. The baseline visual acuity (without goggles) of the older group was on average 0.04 logMAR (i.e., 2 letters) worse than that of younger subjects (Table 1). Although the difference was just statistically significant, it was not a clinically meaningful difference. Thus, the two groups were reasonably well matched in terms of baseline visual acuity. The difference in contrast sensitivity (without goggles) between the two groups was significant and slightly greater in magnitude, being 0.1 log units worse on average in the older group (Table 1). As such, baseline contrast sensitivity was included as a covariate in analyses (see Statistical Analysis section).

Effects of Simulated Vision Impairment on Visual Acuity and Contrast Sensitivity

The goggles reduced binocular visual acuity from about 20/20 to 20/80 in both age groups (Table 1). Visual acuity with the goggles did not differ between the older and younger subjects F(1, 26)= .31; P= .58, and the average change in visual acuity with the goggles also did not differ (older 0.52 and younger 0.60 logMAR change, F(1,26)= 2.98; P= .1). By comparison, there was a significant age group difference in contrast sensitivity with the goggles, F(1,26)= 11.9; P= .002 (Table 1). However, the average reduction in contrast sensitivity with the goggles did not differ significantly between the two age groups (older 0.35 and younger 0.35 log units change; F(1,26)= .03, P= .86).

Detection Performance Measures

Reaction Times

Age (β= .19; SE= .049, t = 3.96, P< .001), distraction (β= −.07; SE= .01, t= −6.02, P< .001), and simulated vision impairment (β= −.03; SE= .01, t= −22.1, P< .001) each significantly increased subjects’ reaction times (Figure 2). Collapsed across distraction and simulated vision impairment conditions, average reaction times of older subjects (average: 1501 ms) were 350 ms slower than those of younger subjects (average: 1151 ms). Collapsed across age and simulated vision impairment conditions, distraction produced a 161 ms delay in reaction times (average without distraction: 1327 ms, average with distraction: 1488 ms). Lastly, when collapsed across age and distraction conditions, simulated vision impairment produced a 660 ms delay in reaction times (average without goggles: 1327 ms, average with goggles: 1987 ms). We found a significant interaction between simulated vision impairment and age (χ²(1, 9)= 8.66; P= .003; Figure 2). This interaction came about as the cost of simulated vision impairment on older subjects reaction time was 828 ms (average without goggles: 1501 ms, average with goggles: 2329 ms), whereas the cost of simulated vision impairment on younger subjects’ reaction times was only 492 ms (average without goggles: 1151 ms, average with goggles: 1643 ms). Thus, older subjects incurred a cost to reaction times with the simulated vision impairment which was 336 ms greater than for younger subjects.

Figure 2.

Average reaction time for each condition plotted separately for older and younger subjects. Error bars represent the standard error of the mean (SEM).

No significant interactions were found between distraction and simulated vision impairment (χ²(1, 9)= 1.77; P= .18) or between distraction and age (χ²(1, 9)= .011; P= .92). However, we found a significant, but weak, three-way interaction between distraction, simulated vision impairment, and age (χ²(4, 12)= 10.49; P= .03). Specifically, the cost of distraction was greater for older participants with than without the simulated vision impairment (165 ms with goggles vs. 146 ms without goggles), but the opposite was true for younger participants (125 ms with goggles vs. 177 ms without goggles). Given that the difference between without and with the simulated vision impairment was only 21 ms for older subjects and 52 ms for younger subjects, this three-way interaction has little real-world significance.

Rate of Safe Detections

Simulated vision impairment had a significant detrimental effect on safe detection rates (β= 1.13, SE= .12, z= 9.45, P < .0001; Figure 3). Collapsed across age and distraction conditions, simulated vision impairment reduced safe detections by 16.4% (78.3% with, 94.7% without goggles). However, distraction (β= .18, SE= .11, z= 1.60, P= .11) and age (β= −.30, SE= .250, z= −1.19, P= .23) did not cause any significant changes in the rate of safe detections. There were no significant interactions (P > 0.60).

Figure 3.

Percent of safe detections for each condition split by older and younger subjects. Error bars represent the standard error of the mean (SEM).

Average Speed

Older subjects drove more slowly than younger (averages 91.8 km/h and 98.6 km/h, respectively; W= 4944, P< .001; Figure 4). Collapsed across age, a Kruskal-Wallis test revealed that subjects drove more slowly when they were distracted (χ²(1)= 6.5; P= .01) but not when driving with simulated vision impairment (χ²(1)= .19; P= .66). We found that younger subjects did not adjust their speed in response to either distraction (χ²(1)= 1.93; P= .16) or simulated vision impairment (χ²(1)= 37; P= .54). In contrast to this, older subjects did drive more slowly as a response to distraction (χ²(1)= 4.93; P= .03) but not in response to simulated vision impairment (χ²(1)= .91; P= .34).

Figure 4.

Boxplots for vehicle speed of older and younger subjects for each condition. Thick black line represents group median; box extent represents the inter quartile range (IQR); plot whiskers represent 1.5x IQR; and filled circles represent outliers beyond 1.5x IQR. (Note: the y-axis has been truncated and runs from 40–100 km/h)

Word Recall Rates

When collapsing across age, word recall rates were significantly lower during pedestrian events (when the pedestrian was in the scene) as compared to the time between events when no pedestrian was on the screen (V= 162; P< .001; see Figure 5, left panel compared to right panel). Older subjects recalled significantly fewer words than younger subjects both during pedestrian events (24.9% vs. 38.9%; W= 221.5, P= .007; Figure 5 left) and between pedestrian events (34.7% vs. 52.9%; W= 148, P< .0001; Figure 5 right). Finally, the simulated vision impairment did not significantly alter recall rates either during (W=.69; P= .41) or between pedestrian events (W= .95; P= .33).

Figure 5.

Boxplots of average percentage of words recalled both during pedestrian events (left) and between pedestrian events (right) plotted separately for older and younger subjects. Thick black line represents group median; box extent represents the inter quartile range (IQR); and plot whiskers represent 1.5x IQR.

DISCUSSION

Effects of Age, Simulated Vision Impairment and Distraction on Detection Rates

The finding that neither age, nor simulated vision impairment, nor distraction affected detection rates is not surprising given that the pedestrian targets were highly salient and the detection task was relatively straightforward (press the horn when you see a pedestrian). High detection rates were also found in prior driving simulator studies for both normally sighted participants with simulated central vision impairment²⁸ and individuals with real vision impairment (mostly age-related macular degeneration)^{10, 29} when performing a similar pedestrian detection task.

Effects of Age and Simulated Vision Impairment on Reaction Times and Safe Detection Rates

In contrast to the detection rate results, age and simulated vision impairment both significantly affected reaction times. The age-related increase in reaction times was most likely a result of age-related reductions in psychomotor abilities^{30, 31} (though we did not include a separate measure of psychomotor abilities). The increase in reaction times with simulated vision impairment was consistent with the effects reported in prior driving simulator studies of normally-sighted participants with similar simulations of central vision impairment²⁸ and individuals with real central vision impairment (compared to controls without vision impairment).^10,29

We found a significant interaction between age and simulated vision impairment. The increase in reaction times with simulated vision impairment was much greater for older than younger subjects even though the change in visual acuity and contrast sensitivity caused by the simulated vision impairment did not differ for the two age groups. This is an important finding because it suggests that reactions to hazards by older drivers with recent onset central vision impairment may be delayed to a much greater extent than the reactions of younger drivers with recent onset central vision impairment.

The large increase in reaction times with the simulated vision impairment resulted in a significant decrease in safe detection rates (from 95% to 78%). Interestingly, however, we found no significant effect of age on safe detection rates despite the longer reaction times in the older group. The safety calculation took account of the speed and distance of both road users at the time of the detection, so even though older subjects had significantly longer reaction times, they did not make more unsafe detections than younger subjects because they drove more slowly. However, contrary to expectations, participants did not drive more slowly with than without simulated vision impairment; the longer reaction times with the simulated vision impairment were not offset by slower vehicle speeds.

Effects of Distraction on Reaction Times and Safe Detection Rates

Auditory distraction had less of an effect on detection measures than either age or simulated vision impairment. Distraction increased reaction times slightly (160 ms overall) but had no significant effect on safe detection rates. Younger drivers did not drive more slowly either with simulated vision impairment or with distraction. Older drivers, however, drove more slowly when distracted but not with simulated vision impairment. These results suggest that older drivers might have been more aware of the detrimental effects of the auditory distraction on their driving performance than the detrimental effects of the simulated vision impairment. There is, however, evidence from survey studies that older drivers do self-restrict their driving after the onset of vision impairment by avoiding difficult driving situations.^32–34

Contrary to expectations we did not find a two-way interaction between distraction and either age or simulated vision impairment. In contrast, Lee et al,⁸ found that distraction had a greater effect on hazard detection of older than younger subjects in the presence of simulated vision impairment. This might suggest that our distraction task was not as difficult or as distracting as the task used by Lee et al.⁸ However, the lack of an age-by-distraction interaction could be explained by the fact that older participants were able to compensate for the distraction by driving more slowly in our driving simulator detection task, but they could not do so in the video-based hazard perception task in the Lee et al.⁸ study.

Effects of Age, Simulated Vision Impairment and Pedestrian Hazards on Audiobook Performance

Older subjects recalled significantly fewer words on the audiobook task (both between and during pedestrian events) than younger subjects. This is most likely due to older subjects having a more limited cognitive resource pool than younger participants^{6, 35} and attention resources being finite.³⁶ Indeed, older drivers have been reported to multitask less well than younger drivers in many studies, both in the driving simulator^{37, 38} and on the road.^39–42 The fact that older subjects recalled fewer words on the audiobook task, yet did not have significantly fewer safe detections, suggests that they were able to maintain satisfactory performance on the detection task by prioritizing that task over the audiobook task. In addition, they compensated by driving more slowly with auditory distraction. However, the combination of the increase in the reaction times of older subjects and their recall of fewer words than younger subjects suggests that they were not processing either task as well as their younger counterparts.

Interestingly, during drives with the audiobook task, we found that audiobook word recall rates were higher when no pedestrian was on the screen than when a pedestrian was on screen. During periods where no pedestrian was visible, subjects did not have to press the horn and were not distracted by pedestrians, allowing more attention resources to be allocated to the auditory task without compromising their detection performance. This suggests that subjects were flexibly reallocating their attention resources depending on moment-to-moment demands of the driving task. Therefore, it seems that when a pedestrian event was on screen, subjects allocated more attention to the driving task (i.e. by processing the pedestrian hazard and formulating an appropriate motor response) at the cost of correctly performing the audiobook task.

Limitations

When interpreting the results of the current study, some methodological limitations need to be considered. The average age of participants in the older group was 67 years, which is a relatively “young” older group. Declines in cognitive capacities and crash risk both increase markedly in the 70+ age group.^{43, 44} Furthermore we did not quantify cognitive capacities of younger and older subjects (e.g., by conducting tests of working memory⁴⁵ and cognitive capacity / load⁴⁶). This means we could not directly test the hypothesis that the greater effects of simulated vision impairment in the older participants were associated with them having a more limited cognitive resource pool, though it is well established that cognitive resources decrease with increasing age.^6,47,48

Simulating central vision impairment in people with normal vision is useful in achieving reasonably uniform levels of visual acuity and contrast sensitivity reductions. However, individuals with real central vision impairment may behave differently to subjects with simulated vision impairment as they will have had much longer to adapt to their vision impairment. Simulated vision impairment also fails to capture all aspects of real vision impairment. For example, in the current study we simulated reductions in visual acuity and contrast sensitivity, but did not simulate central visual field loss (scotomas), which is an important characteristic of age-related macular degeneration.

We also need to consider the limitations of the driving simulator task. We extended prior research^{3, 8} by using a driving simulator paradigm with moving pedestrian hazards in which participants had to control vehicle speed and position. However, the pedestrian figures were highly salient, participants did not have to make judgments about whether pedestrians were hazards, and pedestrians appeared more frequently than would be the case when driving on rural highways in the real world. Nevertheless, an important advantage of driving simulator studies is that participants are never at risk for real collisions, which is a major concern for on-road studies. Furthermore, frequent presentation of hazards enables reliable measurement of reaction times. In the real world, pedestrian hazards may be less frequent and unexpected and therefore we would probably expect to find greater effects of central vision impairment and distraction on response times and safety. Finally, our detection task required only a horn-press response. We did not measure braking, which would be a common response to a pedestrian hazard in on-road driving. However, the safety calculation did take account of the speed of the participant’s vehicle at the time of the horn-press. The relatively straightforward detection task may explain why distraction did not have greater adverse effects on subjects’ detection performance.

To ensure the pedestrians were truly on a collision course (constant bearing angle), we instructed subjects to drive as close to 97 kph as possible whenever they thought it safe to do so. This likely had two consequences. Firstly, it increased the difficulty level of the drives somewhat and secondly it may have reduced the amount by which participants might have otherwise compensated for either distraction or vision impairment by driving slowly. However, the instructions were the same for all conditions and participants were able to use the brake to reduce speed whenever they felt unsafe. It is also relevant to note that the sensation of speed in a driving simulator is not as strong as in a real car⁴⁹ which may have contributed to subjects not slowing down as expected when driving with simulated vision impairment.

CONCLUSIONS

Simulated central vision impairment increased response times to a much greater extent in older than younger participants, despite similar levels of visual acuity and contrast sensitivity reductions. Neither older nor younger participants adapted their speed in response to simulated vision impairment, resulting in unsafe detections. However, older subjects appeared to compensate for the auditory distraction by driving more slowly, possibly perceiving a greater threat to their driving safety from the auditory distraction than from the simulated central vision impairment. This interesting and unexpected finding warrants further investigation. Our results underline the importance of evaluating safety of responses to hazards as well as reaction times in a paradigm which flexibly allows participants to modify their driving behaviors.