홈 탐색 도움말
로그인
ardo
/
Obsidian
1
0
포크 0
파일 이슈 0 풀 리퀘스트 0 위키
트리: a0e0ef701f
브랜치 태그
Obsidian
/
Zotero
/
001_artiklid
/
reimerimpactagecognitivedemand2013.md

reimerimpactagecognitivedemand2013.md 51 KB
히스토리 Raw

category: literaturenote citekey: reimerimpactagecognitivedemand2013 title: Impact of age and cognitive demand on lane choice and changing under actual highway conditions authors: "Reimer, Bryan; Donmez, Birsen; Lavallière, Martin; Mehler, Bruce; Coughlin, Joseph F.; Teasdale, Normand" year: 2013 date: 2013-00-00 2013 doi: 10.1016/j.aap.2012.12.008 publication: "Accident Analysis & Prevention" url: "https://doi.org/10.1016/j.aap.2012.12.008" zotero_key: R8BIT68G zotero_storage: 9UF6CWSE collections: transpordi_planeerimine folder: 001_artiklid firstAuthor: "Reimer, Bryan"

status: converted

Contents lists available at SciVerse ScienceDirect

Accident Analysis and Prevention

journal homepage: www.elsevier.com/locate/aap

Impact of age and cognitive demand on lane choice and changing under actual highway conditions

Bryan Reimer a,∗, Birsen Donmez a,c, Martin Lavallièrea,b, Bruce Mehler a, Joseph F. Coughlina, Normand Teasdale b

  • a MIT AgeLab & New England University Transportation Center, 77 Massachusetts Avenue, E40-279 Cambridge, MA 02139, USA
  • b Université Laval, Faculté de Médecine, & Vieillissement, Centre de recherche FRSQ du Centre hospitalier affilié universitaire de Québec, 2300, rue de la Terrasse, Québec, Québec, Canada
  • c University of Toronto, Department of Mechanical and Industrial Engineering, 5 King's College Road, Toronto, ON M5S 3G8, Canada

a r t i c l e i n f o

Article history: Received 18 January 2011 Received in revised form 24 November 2012 Accepted 7 December 2012

Keywords: Lane change behavior Lane choice Distraction Cognitive workload Passing Aging

a b s t r a c t

Previous research suggests that drivers change lanes less frequently during periods of heightened cognitive load. However, lane changing behavior of different age groups under varying levels of cognitive demand is not well understood. The majority of studies which have evaluated lane changing behavior under cognitive workload have been conducted in driving simulators. Consequently, it is unclear if the patterns observed in these simulation studies carry over to actual driving. This paper evaluates data from an on-road study to determine the effects of age and cognitive demand on lane choice and lane changing behavior. Three age groups (20–29, 40–49, and 60–69) were monitored in an instrumented vehicle. The 40's age group had 147% higher odds of exhibiting a lane change than the 60's group. In addition, drivers in their 60's were less likely to drive on the leftmost lane compared to drivers in their 20's and 40's. These results could be interpreted as evidence that older adults adopt a more conservative driving style as reflected in being less likely to choose the leftmost lane than the younger groups and less likely to change lanes than drivers in their 40's. Regardless of demand level, cognitive workload reduced the frequency of lane changes for all age groups. This suggests that in general drivers of all ages attempt to regulate their behavior in a risk reducing direction when under added cognitive demand. The extent to which such self-regulation fully compensates for the impact of added cognitive demand remains an open question.

© 2012 Elsevier Ltd. All rights reserved.

1. Introduction

(N. Teasdale).

Driving is a complex skill that can be considered as a combination of different functional and operational activities involving low level control of the vehicle guided by maneuvers and strategic decisions (Michon, 1995). Lane changing is a driving maneuver frequently associated with accidents (Pande and Abdel-Aty, 2006) and requires engagement of a coordinated combination of sensory/perceptual, cognitive processing, and manipulative actions. Humans are generally considered to have finite information processing resources (Wickens, 1984; Wickens & McCarley, 2008), and situations that make multiple calls on these resources, particularly those that require divided attention, may tax capacity to the point that performance and safety margins suffer. Possible

such as non-critical lane changes, or by actions such as slowing driving speed. In a naturalistic study of 16 commuters using either interstate or

strategies for coping with increased demand mightinclude limiting overall workload by reducing the frequency of optional maneuvers,

state highways in southwestern Virginia, Olsen et al. (2002) found thatlane changes were most often initiated due to a slow lead vehicle and occurred more frequently on the interstate. In a driving simulator study, Bar-Gera and Shinar (2005) observed that their subjects frequently passed lead vehicles that were faster than their own average speed. The authors suggested thattheir subjects chose to perform the passing maneuver to minimize the workload associated with following a lead vehicle. Thus, the speed differential between a lead vehicle and the driver's own vehicle is not the only factor that guides lane changing decisions. Aggressiveness, sensation seeking, and competitiveness have also been suggested to affect lane change behavior (Bar-Gera & Shinar, 2005; Matthews et al., 1998).

The level of cognitive demand is another factor that can influence lane changing decisions and requires particular attention

∗ Corresponding author. Tel.: +1 617 452 2177. E-mail addresses: reimer@mit.edu (B. Reimer), donmez@mie.utoronto.ca (B. Donmez), martin.lavalliere.1@ulaval.ca (M. Lavallière), bmehler@mit.edu (B. Mehler), coughlin@mit.edu (J.F. Coughlin), Normand.Teasdale@kin.ulaval.ca

as technological developments create additional possibilities for drivers to engage in cognitively demanding activities. In driving simulators, participation in a cognitive secondary task has been shown to interfere with the frequency of lane changes (Beede & Kass, 2006; Cooper et al. (2009)) and to affect the degree to which drivers monitor surrounding traffic conditions (Zhou et al., 2009). In Cooper et al. (2009), the effect of a hands-free cell phone conversation on lane changing was investigated in three levels of traffic density. The results show that when drivers conversed on a cell phone, they made fewer lane changes, were more likely to remain behind a slow vehicle, and had a lower overall mean speed. Beede and Kass (2006) reported similar results in which drivers changed lanes most frequently when not engaged in any secondary activities. The changes in driving behavior observed by Cooper et al. (2009) and Beede and Kass (2006) may be viewed as compensatory actions taken by the driver to reduce the workload associated with the driving task and diverting extra capacity to other non-driving activities. Alternatively, these changes in behavior might be interpreted as an effect of saturation in cognitive capacity, which results in reduced attention to elective driving activities such as changing lanes.

Cantin et al. (2009) assessed mental workload under various driving conditions in a simulator study by measuring reaction times to periodic auditory probes and found that reaction times during the execution of a lane change is slower than during periods without lane changes. The relative increase in workload, as evidenced by increased response delay was greater for older adults. As drivers age, they self-report that they are less likely to pass another vehicle (Boyle et al., 1998). In a driving simulator, Farah and Toledo (2010) found that drivers under age 25 tend to accept smaller passing gaps and are more likely to pass a lead vehicle compared to drivers over age 25. Under actual driving conditions, older drivers are known to self-regulate workload, e.g., drive slower, travel during less congested periods, and avoid distracting technologies (D'Ambrosio et al., 2008; Langford & Koppel, 2006). Less is known, however, about the lane changing behavior of different age groups under varying levels of cognitive demand. Finally, the degree to which lane changing behavior with cognitive workload observed in previous driving simulator studies carries over to field driving is not well established.

This paper expands upon recently published work on physiological reactivity and changes in visual behavior in response to graded levels of cognitive demand across different age groups (Mehler et al., 2012; Reimer et al., 2012) by examining lane changing behavior in the same dataset. The data were captured during an extended period of driving during non-rush hour, daytime periods on a multi-lane interstate during which participants were free to maintain or adjust their lane positioning at will. The sample was limited to cases that were classified as being largely unaffected by traffic flow (e.g., cases with extended periods of stop and go traffic and cases impacted by complete gridlock were excluded) or adverse weather conditions. The dataset therefore contains behaviors that occurred during periods of continuous traffic flow where the decision to make lane changes was effectively at the drivers' discretion. These data provided an opportunity to examine the frequency of lane changing and lane selection under both single task driving and under conditions of objectively defined levels of secondary cognitive demand. The age groups studied allow for the characterization of lane changing behavior and lane choice across the lifespan. This work also provides an opportunity to examine additional aspects of the extent to which drivers of different ages do or do not compensate for the added demand of secondary cognitive workload during highway driving as well as a validation point for observations from controlled simulation studies.

2. Methods

2.1. Participants

One-hundred and sixtyfiveparticipants initially reported for the study. Participants were self-reported experienced drivers, driving more than three times a week and having held a valid driver's license for over 3 years.Inclusioncriteria required participants to be free of police reported accidents for the past year, making the group limited to a potentially safer set of drivers than would be observed in a broader community sample. The group was considered to be relatively healthier than a community sample as individuals were excluded if they reported a recent hospitalization, specific health conditions (such as a positive history for heart attack, angina, coronary heart disease, stroke, a pacemaker, or uncontrolled diabetes), recent use of medications that cause drowsiness or suggest safety concerns (e.g., anti-psychotic or anti-convulsant medications), or medications that impact heart rate (e.g., beta blockers). Additionally, participants with scores below 26 on the Mini-Mental State Examination (Folstein et al., 1975), an indication of cognitive impairment, were excluded. Compensation of \$60 was provided for the 3-h experiment.

Cases were excluded from the final dataset as follows: failure to meet requirements upon eligibility review (7), inability to perform the secondary tasks to criterion during training (8) [4 each in the 40's and 60's age categories], pilot runs (first 4), sleepy while driving (3), heavy traffic (10), adverse weather during a portion of the drive (8), protocol errors (5), equipment failures (7), poor video quality (2), withdrew from study after learning requirements (2), and extra cases dropped after design cells (age and gender) were filled (3). The sample considered here consisted of the remaining 106 individuals and was balanced by gender and across three age groups: 20–29 (n = 36), 40–49 (n = 35), and 60–69 (n = 35). The average age by group was 24.6 (SD: 2.7), 44.4 (SD: 3.0), and 63.3 (SD: 3.1). Male and female participants did not differ significantly by age within each group (F(1,34) = .86, p = .36; F(1,33) = .83, p = .37; F(1,33) = .22, p = .64).

2.2. Apparatus and secondary task

Participants drove an instrumented mid-sized sports utility vehicle (Volvo XC 90) equipped for time synchronized data collection. Data presented here were recorded from the vehicle's CAN bus, a microphone mounted inside the vehicle, and from a camera mounted near the center of the vehicle facing forward. Three levels of an auditory presentation – verbal response, delayed digit recall task (n-back) were employed to increase drivers' workload. Each level consisted of four 30-s blocks during which 10 single digit numbers (0–9) were presented in random order at a spacing of 2.25 s. At the lowest level of demand (0-back), drivers were to repeat each digit as it was presented. At the moderate level of demand (1-back), drivers were to respond to each new presentation by recalling and saying out loud the previous number in the presentation sequence. Atthe highestlevel of demand (2-back),the number two places back in the sequence was to be repeated. The form of this task was identical to earlier studies conducted in our laboratory (Mehler et al., 2009; Reimer, 2009; Reimer & Mehler, 2011) and was developed based on recommendations by Zeitlin (1993)for secondary cognitive tasks for use in driving contexts. This form of n-back task holds the amount of auditory demand constant across levels, verbal resources required for responding are essentially the same, and only the memory demands change – increasing in a systematic manner with each level of the task. For additional details on the development of this task, training materials, and the complete stimulus sets see Mehler et al. (2011).

2.3. Procedure

Participants signed an approved informed consent and completed a questionnaire covering driving and health history. They were then extensively trained on the secondary task prior to entering the vehicle with a minimum of n + 1 practice blocks per task level(e.g., 3 practice blocks for the 2-back).Additional repetitions of the instructions and practice trials were presented for each demand level until participants demonstrated a minimum proficiency of 7 correct responses on the 0 and 1-back (out of 10 and 9 items respectively) and of at least 4 (out of 8) on the 2-back. Participants who were unable to meet the criteria for the 2-back within nine practice blocks were excluded(8 individuals). Uponbeing seatedinthe vehicle, an audio recording reviewed the secondary task and presented an additional n + 3 practice blocks for each demand level.

A research associate, seated in the back of the car, operated the data collection equipment, provided driving directions, and monitored participants to ensure that they had adequate control of the vehicle at alltimes. Approximately 30 min of driving were provided for habituation prior to the study period. The study period consisted of an initial 6 min of single task driving followed by the three levels of the n-back task. Each task was presented over a 2-min interval (four 30-s blocks) and each task was followed by 2 min of single task driving. The presentation order for the three levels of the task was counterbalanced across the sample.

The experiment was conducted on Interstate 93 starting in the vicinity of the intersection with I495. Participants were traveling north from Boston, Massachusetts to the area generally just south of Manchester, New Hampshire. The posted speed limit was 104.6 km/h (65 mph). When first entering the highway in Boston, Massachusetts, participants were prompted: "We are going to be driving north on 93 for approximately 40 min. You can continue driving in this lane or move into another lane so that you are comfortable with the traffic flow." Participants were thus allowed to select a lane of travel and pass other vehicles at will. Instances of weather or other traffic conditions that impeded normal traffic flow or otherwise constrained a driver's ability to change lanes were recorded by the research assistantin the experimental log and reviewed by a senior staff member. As noted previously, a total of 18 cases were dropped from consideration due to weather or traffic density constraints on normal traffic flow.

The distribution of the number of travel lanes across the sample was not uniform due to the nature of variations in traffic and driver speed that affected the start of the data assessment period for individual drivers. Excluding exits, on-ramps and other road transitions, the roadway began with four lanes of travel, decreased to three and, eventually in Southern New Hampshire, to two. Then, near the end of the portion of highway used in the study, the number of lanes increased back to three. Twelve participants began the experimental period while in a section of four lane highway, 90 in three lanes and 4 in two lanes. Twenty five participants completed the experimental segment while still on a portion of the highway with two travel lanes.

2.4. Data coding and data periods used in analysis

Recorded audio was used to assess participants' accuracy in responding to the n-back tasks. Task performance was scored as a percentage of the number of correct responses out of the total number of expected responses. Lane change and lane choice data were independently extracted through a manual analysis of video recordings by two research associates. Discrepancies were than reviewed and reconciled by the first author. No differentiation between critical and non-critical lane changes was made. The procedure for classification of lane changes was analogous to Olsen et al. (2002) and consistent with Cooper et al. (2009) where the

Fig. 1. Secondary task performance (%) for each age and demand level (0, 1, and 2-back task)*.

* Figures 1, 3, and 4 are boxplots which represent the five-number summary (minimum, first quartile, median, third quartile, and maximum) as well as potential outliers as indicated with circles.

onset of each lane change was classified as the point when the vehicle was observed to be first moving in a lateral direction toward the destination lane. The completion of each lane change was recorded as the point where the vehicle was fully centered in the destination lane. Time spent in the leftmost lane was computed as the time from the lane crossing entering the left lane to the lane crossing exiting the left lane. A lane crossing was classified as when the middle of the car crossed the lane marker. Only lane changes that resulted in the centerline of the vehicle crossing over the dividing line were considered; partial motions toward an adjacentlane were not coded.

Three data periods were analyzed: pre-task, n-back (cognitive demand tasks), and recovery. Thepre-taskperiod consisted of 6 min of single task driving prior to the initiation of the first n-back task. The n-back period consisted of the 6 min of dual task activity corresponding to the aggregate of the three separate 2-min-long secondary tasks. The 6 min of data for the recovery period were drawn from the 2 min of single task driving that followed each of the three dual task periods. Also considered was an analysis of each of the three task demand levels (0, 1, and 2-back). Since each task was 2 min long, a 2-min reference "baseline" period was used for comparison. Consistent with Mehler et al. (2012) and Reimer et al. (2012),the 2-min baseline period was selected as minutes 3.5 to 5.5 of the pre-task period. For the purpose of classifying frequencies, a lane change was assigned to the period in which it was initiated. Gender was initially included in all statistical analysis, but was later dropped from the final models, as it was not a significant predictor in any of the models.

3. Results

3.1. Secondary task performance

Fig. 1 presents the secondary task performance for each age group by the level of cognitive task difficulty. The rate of correct responses was analyzed with a Poisson model due to the high level of non-normality in the data. There were 40, 36, and 32 stimuli that required responses in the 0-back, 1-back, and 2-back conditions, respectively. The logarithm ofthe number of stimuli was used as an offset variable in the model. Because the data consisted of repeated measures, generalized estimating equations (GEE) were used for estimation. The model was fitted using PROC GENMOD in SAS 9.1, with the specifications of log link function and Poisson distribution. Significant effects were observed for cognitive task difficulty (-2(2) = 79.03, p < .0001) and its interaction with age (-2(4) = 10.18, p = .04). Follow-up contrasts revealed that, regardless of age, increasing demand resulted in degraded secondary task performance (0-back vs. 1-back: -2(1) = 37.0, p < .0001; 0-back vs. 2-back: -2(1) = 64.5, p < .0001; 1-back vs. 2-back: -2(1) = 30.4, p < .0001). This effect suggests that as in Mehler et al.(2012), Reimer (2009), and Reimer and Mehler (2011), error rates increased with higher levels of cognitive task difficulty. The differences between age groups depended on the cognitive task demand. Under the highest demand condition (2-back), the 20's age group responded correctly to a higher percentage of stimuli than both the 40's (-2(1) = 4.0, p = .045) and 60's (-2(1) = 3.11, p = .078) age groups, although the comparison to the 60's group was only marginally significant. There were no significant differences across the age groups for the 0-back or 1-back conditions (p > .1).

It is appropriate to note again that 4 potential participants from each of the 40's and the 60's groups were excluded from data collection due to difficulties with the high demand 2-back task during training. We only wanted to consider drivers who were able to engage with the task and thus actually experience increased cognitive load. Consequently, the task performance levels for the two older age groups may well have been somewhat higher than what would be observed in unscreened samples.

3.2. Number of lane changes

A negative binomial model was developed to compare the number of lane changes across different age groups for three study periods:the 6-min period prior to the admission ofthe n-back tasks (pre-task), the aggregate of the three 2-min-long n-back tasks (nback), and the combination of the three 2-min intervals of single task driving following each n-back task (recovery). Each of these three periods was 6-min in duration. The model was fitted using PROC GENMOD in SAS 9.1, with the specifications of a log link function and negative binomial distribution. Repeated measures were accounted for by using GEE. Fig. 2 shows histograms for the number of lane changes across different age groups and study periods.

Both the main and interaction effects were statistically significant (p < .05). For the 20's age group, the lowest number of lane changes occurred during the n-back period, followed by the pretask period, and the highest number of lane changes was observed during the recovery period. For this age group, the expected number of lane changes in the recovery period was estimated to be 2.39 times the expected number of lane changes in the n-back period (95% CI: 1.70, 3.35, -2(1) = 25.25, p < .0001) and 1.45 times the expected number of lane changes in the pre-task period (95% CI: 1.05, 2.00, -2(1) = 5.02, p = .03). The expected number of lane changes in the pre-task period was 1.66 times that in the n-back period (95% CI: 1.07, 2.58, -2(1) = 5.03, p = .03).

For the 40's age group, the only significant difference was between the recovery and n-back periods with a multiplicative increase of 1.34 in the expected number of lane changes in the recovery period (95% CI: 1.003, 1.80, -2(1) = 3.93, p = .048).

For the 60's age group, n-back period resulted in lower number of lane changes compared to both the pre-task and recovery periods. Compared to the n-back period, a multiplicative increase of 2.17 and 2.38 were observed in the expected number of lane changes in the pre-task and recovery periods, respectively (95% CI: 1.35, 3.50, -2(1) = 10.15, p = .001; 95% CI: 1.47, 3.84, -2(1) = 12.59, p = .0004).

Comparisons of different periods across age groups did not reveal major significant findings, likely due to the between subject nature of these comparisons. The only statistically significant

Table 1 Number of lane changes by age and demand level (the cells represent the number of drivers who performed a given number of lane changes).

Age Demand level Number of lane changes
0 1 2 3 4 5
20's (n = 36) Baseline 21 7 5 2 1 0
0-Back 25 7 2 2 0 0
1-Back 25 9 1 1 0 0
2-Back 26 7 3 0 0 0
Total 97 30 11 5 1 0
40's (n = 35) Baseline 18 8 8 0 1 0
0-Back 19 11 4 0 0 1
1-Back 16 8 10 1 0 0
2-Back 26 6 3 0 0 0
Total 79 33 25 1 1 1
60's (n = 35) Baseline 23 7 4 1 0 0
0-Back 28 5 1 1 0 0
1-Back 28 4 3 0 0 0
2-Back 27 7 1 0 0 0
Total 106 23 9 2 0 0

finding was that, during the n-back period, the 40's group had a higher number of lane changes compared to the 60's group: a multiplicative increase of 2.31 in the expected number of lane changes (95% CI: 1.24, 4.31, -2(1) = 6.94, p = .008).

A finer break-down of the n-back task period is provided in Table 1. These data are based on the 2-min windows represented by the three separate cognitive difficulty levels (0, 1 and 2-back) and where "baseline" represents 2-min prior to the first n-back task. Given the short intervals, there were only a few cells with more than one lane change. Thus,for analysis purposes, we grouped the response variable into two categories: 0 and ≥1 lane changes. A logistic regression model was built to predict the likelihood of making at least one lane change for different age groups and demand levels. Repeated measures were accounted for using GEE. The model was fitted using PROC GENMOD in SAS 9.1, with the specifications of logit link function and binomial distribution.

Wald statistics revealed that age (-2(2) = 7.81, p = .02) and demand level (-2(3) = 8.24, p = .04) were statistically significant. The interaction of age with demand level was not significant (p > .05). The 40's group exhibited 147% higher odds of making lane changes compared to the 60's group (95% CI: 29%, 372%, -2(1) = 7.44, p = .006). No other differences were found between age groups. These findings are to some extent in line with the results of the negative binomial model reported above for the different study periods. As for demand level, drivers had 112% higher odds of making lane changes without the cognitive task compared to with the 2-back task (95% CI: 24%, 262%, -2(1) = 7.63, p = .006).

3.3. Mean speed

A repeated measures ANOVA was conducted to compare mean speed across the three age groups for the three study periods previously discussed: pre-task, n-back, and recovery (Fig. 3). The main effects of age (F(2,102) = 5.53, p = .005) and study period (F(2,204) = 16.79, p = <.0001) and their interaction (F(4,204) = 3.91, p = .004) were all significant. In general, participants in the 20's age group drove significantly faster than those in the 60's group (t(102) = 3.06, p = .003). The 40's group also drove significantly faster than the 60's group, but only during the n-back task (t(169) = 3.08, p = .003) and recovery (t(169) = 3.39, p = .001) periods and not during pre-task. The 20's and 60's groups drove significantly faster prior to the n-back task period than they drove during n-back task and recovery periods (20's pre-task vs. 20's n-back:

Fig. 2. Histograms for number of lane changes across age groups and study periods.

t(204) = 4.05, p < .0001; 20's pre-task vs. 20's recovery:t(204) = 3.77, p = .0002; 60's pre-task vs. 60's n-back: t(204) = 4.72, p < .0001; 60's pre-task vs. 60's recovery: t(204) = 4.39, p < .0001). Thus, the results suggestthatin generalthe 40's group's average speed profile stayed fairly constant across the three study periods, whereas the 20's and 60's groups reduced their speeds when presented with the nback task and maintained these slower speeds during the recovery periods. These results are based on aggregate speed profiles for 6 min study periods. A more detailed analysis on speed maintenance across the three demand levels follows.

Fig. 4 presents the mean speed for each category of age and demand level. The repeated measures ANOVA yielded significant main effects of age (F(2,103) = 5.05, p = .008) and demand level (F(3,309) = 8.79, p < .0001). The interaction between age and demand level was not significant (p > .05). Compared to the 20's

Fig. 3. Mean speed across different age groups and study periods.

and 40's age groups, drivers in their 60's drove significantly slower (105.4 and 105.3 vs. 101.7 km/h, respectively) (60's vs. 20's: t(103) = −2.78, p = .006; 60's vs. 40's: t(103) = −2.73, p=.008). During single task driving (baseline), drivers of all ages drove faster than during all three demand level periods (baseline: 106.1 km/h, 0-back: 103.9 km/h, 1-back: 104 km/h, and 2-back: 102.3 km/h)(baseline vs. 0-back: t(309) = 2.87, p = .004; baseline vs. 1-back: t(309) = 2.76, p = .006; baseline vs. 2-back: t(309) = 5.12, p < .0001). The 0- and 1-back tasks both resulted in faster average speeds thanthe 2-back task (0-back vs. 2-back:t(309) = 2.26, p = .02; 1-back vs. 2-back: t(309) = 2.36, p = .02).

3.4. Lane choice

Table 2 presents the number of drivers who did or did not at all drive on the leftmost lane (high speed lane) across the three study periods. A logistic regression model was built to predict the

Fig. 4. Mean speed for each age and demand level (baseline, 0, 1, and 2-back task).

Table 2 Percentage of drivers who drove in the leftmost lane across age group and study period.

Age Pre-task n-Back Recovery Row average
20's (n = 36) 66.7 80.6 94.4 80.6
40's (n = 35) 82.9 80.0 92.3 85.7
60's (n = 35) 51.4 40.0 62.9 51.4

Table 3 Percentage of drivers who drove on the leftmost lane across age group and demand level.

Age Baseline 0-Back 1-Back 2-Back n-Back (mean)
20's (n = 36) 58.3 52.8 58.3 50.0 53.7
40's (n = 35)
60's (n = 35)		 60.0
28.6		 57.1
25.7		 65.7
22.9		 42.9
31.4		 55.2
26.7

likelihood of driving on the leftmost lane. Repeated measures were accountedforusing GEE. Themodel wasfittedusing PROC GENMOD in SAS 9.1, with the specifications of logit link function and binomial distribution. The 60's group had lower odds of driving on the leftmost lane than both the 20's and the 40's group (-2(2) = 21.19, p < .0001). Compared to the 60's group, the 40's group was estimated to have a 544% and the 20's group was estimated to have a 390% higher odds of driving on the leftmost lane. The highest odds of driving on the leftmost lane was observed during the recovery period (-2(2) = 12.99, p = .002). The odds of driving on the leftmost lane were 259% and 250% higher in the recovery period compared to the pre-task and n-back periods, respectively. We also analyzed time spent on the leftmost lane for non-zero observations. There were no statistically significant findings for this variable (p > .05).

Table 3 presents the number of drivers who did or did not drive in the leftmost lane across different task demand levels. Another logistic regression model was built to analyze these data. Drivers in the 60's age group had lower odds of driving in the leftmost lane than both the 20's and the 40's groups (-2(2) = 11.57, p = .003). Compared to the 60's group, the 40's group was estimated to have a 251% and the 20's group a 229% higher odds of driving in the leftmost lane regardless of task demand. The analysis of time spent on the leftmost lane for non-zero observations did not reveal any statistically significant findings (p > .05).

4. Discussion

An analysis of data collected from an on-road experiment was conducted to examine lane change and lane choice behavior of three different age groups (20's, 40's and 60's) under varying levels of cognitive load. The findings reveal that both age and demand level were associated with lane choice and lane change behaviors. Compared to periods of single task driving, fewer lane changes were observed under secondary cognitive task load. A similar effect was observed by Cooper et al. (2009) in a simulator experiment. Given that our study was an on-road assessment, these results provide additional ecological validity to this finding. An age effect was observed in the significantly lower likelihood of making a lane change in the 60's age group compared to drivers in their 40's. This result suggests that, as expected, older adults in this study adopted a generally more conservative driving style than middleaged adults. Contrary to expectations,the 20's group did not change lanes more frequently than the oldest group. This finding may in part relate to the sample, drawn from an urban population where younger drivers have generally lower and more variable levels of driving experience. It is plausible that individuals with less experience are not as willing to accept the risk associated with lane changes. In contrast, drivers in the 40's age group have more driving experience and may be more broadly willing to accept the risks associated with passing. An alternate explanation may be that the younger age group is less willing to engage in riskier behaviors such as changing lanes while driving an unfamiliar instrumented research vehicle in the presence of a research associate.

Speed selection also was influenced by age and demand level. The 60's group drove at a significantly slower mean speed than drivers in their 20's and 40's. This finding is in agreement with other on-road and simulator studies where older drivers maintained slower speeds than younger drivers (Hakamies-Blomqvist et al., 1999; Planek & Fowler, 1971; Szlyk et al., 1995). Moreover, older drivers in our study responded to the introduction of cognitive task load with the same compensatory strategy as younger drivers, that is, they reduced their speed by a similar amount. During the baseline, drivers travelled at greater speeds than they did when they had to perform each of the cognitive secondary tasks. These results are consistent with simulation studies by Cooper et al. (2009) and Horberry et al. (2006), which showed that drivers reduced their speed while conducting a conversation similar to a hands-free cell phone task.

A major concern around added cognitive demand during driving is that drivers' gaze concentrates around the road center (Harbluk et al., 2007; Reimer et al., 2012; Sodhi et al., 2002) and drivers are not always aware of the extent of the resulting attentional attenuation or any associated performance decrements (Horrey et al., 2008). Thus, they may be willing to engage in potentially risky distracting activities while driving, resulting in an overall reduction in their capacity to respond to various emerging demands in the driving environment. Concern over this pattern of behavior is one reason behind the call for the development of detection systems that provide information to drivers when they are distracted so that they can modify their behavior appropriately (Coughlin et al., 2011; Donmez et al., 2007, 2008). Donmez et al., 2010 found that young drivers who exhibit the riskiest distraction behavior among their peers benefit most from distraction-related feedback. This finding provides more evidence about drivers' lack of risk awareness when engaged in distracting activities and the potential benefits of providing feedback to guide appropriate behavior.

To our knowledge, this is the first time that driving in the leftmostlane as a functionof age andcognitive loadhas beenevaluated. The results show that the 60's age group was more likely to not drive on the leftmost lane at all compared to the 20's and 40's cohorts. This pattern remained consistent across single task driving and with the addition of the secondary cognitive task load. Selfregulation could explain why participants in the 60's age group differ fromtheir younger counterparts intheirutilizationofthe leftmost lane. Because the attentional and control demands of travel in this lane are generally greater due to a higher travel speed, avoiding the leftmost lane may be one strategy for reducing overall demand. Donorfio et al. (2009) used a survey to investigate selfregulation in older drivers. Older drivers described driving as a way to remain connected to society, and self-regulation represents one method they use to cope with changing capabilities due to declining health and cognitive abilities. In order to self-regulate appropriately, however, drivers need to be aware of their limitations and their capacity to find an effective balance. Unfortunately, older drivers are not always good at self-evaluation of their driving performance (Holland and Rabbitt, 1992, 1994), and older drivers have been found to err both on the side of driving beyond their capabilities and of sometimes curtailing their driving behavior prematurely (D'Ambrosio et al., 2008).

5. Conclusion

As noted above, this is, to our knowledge, the first attempt to assess if older drivers differ in their lane changing and lane choice behaviors compared to younger drivers in an on-road study employing a relatively large sample size. We also investigated the effects of varying levels of secondary task cognitive load on these behaviors. During periods of heightened cognitive load, lane change frequency was found to decrease across all age groups, and drivers in their 60's were found to make fewer lane changes than those in their 40's. In general, older drivers were less likely to make a lane change than drivers in their 40's and were less likely to drive on the left hand travel lane compared to drivers in their 20's and 40's. More research is required to determine if the observed reduction in lane changes represents a conservative, compensatory approach to driving under dual task load or simply results from a basic saturation of the drivers' cognitive capacities that limits engagement in other activities. In the latter case, the act of attending to the secondary task may by default reduce attention to the driving task, resulting in less active lane changing behavior and reduced pressure on the throttle resulting in speed reduction. In either case, a reduced frequency of higher risk behaviors, such as lane changing and higher speed, during periods of added cognitive demand may be one of many factors that explains why accidents are not increasing while drivers are being confronted with more visual, manipulative and cognitive distractions. Nonetheless, more research is needed to quantify the risks associated with non-driving related in-vehicle activities.

6. Limitations

The results presented here are limited to the extent that they focus on one type of cognitive demand and a single driving environment. The changing number of travel lanes across the drive introduced some forced lane changes. While forced lane changes were not included in the analysis, the driving environment may have introduced variability that impacted the results in subtle ways. Follow-on work considering a highway segment with a fixed number of lanes is warranted. The age effects observed in this study may be understated in that individuals were excluded due to reported health conditions, infrequent driving, recent police reported accidents and an inability to perform the n-back task to criterion during training, most likely resulting in a somewhat more frequent driving and higher functioning sample than would have been obtained without exclusions. Finally, the effect of driving a highly instrumented vehicle with an observer on lane changing and other driving behaviors is unknown. It might be that some drivers were less likely to make lane changes compared to what they would do alone in their own vehicles. In spite of these limitations, these results provide important external validity to earlier simulation findings.

Acknowledgements

We gratefully acknowledge the support ofthe US Department of Transportation's Region I New England University Transportation Center at MIT and The Santos Family Foundation during the analysis of these data and writing of this paper, and Ford Motor Company for the development of the original dataset and the instrumented vehicle. In addition, Martin Lavallière was supported by a doctoral scholarship and a FRSQ-RQRV International Internship Grant from the Fonds de la Recherche en Santé du Québec. An earlier version of this manuscript appeared in the Proceedings of the Transportation Research Board 90th Annual Meeting (Lavallière et al., 2010).

References

Bar-Gera, H., Shinar, D., 2005. The tendency of drivers to pass other vehicles. Transportation Research Part F: Traffic Psychology and Behaviour 8 (6), 429–439.

  • Beede, K.E., Kass, S.J., 2006. Engrossed in conversation: the impact of cell phones on simulated driving performance. Accident Analysis and Prevention 38 (2), 415–421.
  • Boyle, J., Dienstfrey, S., Sothoron, A., 1998. National Survey of Speeding and Other Unsafe Driving Actions. National Highway Traffic Safety Administration, Washington DC.
  • Cantin, V., Lavallière, M., Simoneau, M., Teasdale, N., 2009. Mental workload when driving in a simulator: effects of age and driving complexity. Accident Analysis and Prevention 41 (4), 763–771.
  • Cooper, J.M., Vladisavljevic, I., Medeiros-Ward, N., Martin, P.T., Strayer, D.L., 2009. An investigation of driver distraction near the tipping point of traffic flow stability. Human Factors 51 (2), 261–268.
  • Coughlin, J.F., Reimer, B., Mehler, B., 2011. Monitoring, managing, and motivating driver safety and well-being. IEEE Pervasive Computing 10 (3), 14–21.
  • D'Ambrosio, L.A., Donorfio, L.K.M., Coughlin, J.F., Mohyde, M., Meyer, J., 2008. Gender differences in self-regulation patterns and attitudes toward driving among older adults. Journal of Women and Aging 20 (3–4), 265–282.
  • Donmez, B., Boyle, L., Lee, J.D., 2008. Mitigating driver distraction with retrospective and concurrent feedback. Accident Analysis and Prevention 40, 776–786.
  • Donmez, B., Boyle, L., Lee, J.D., 2010. Differences in off-road glances: the effects on young drivers' performance. ASCE Journal of Transportation Engineering, Special Issue: Applications of Advanced Technologies in Transportation 136 (5), 403–409.
  • Donmez, B., Boyle, L.N., Lee, J.D., 2007. Safety implications of providing realtime feedback to distracted drivers. Accident Analysis and Prevention 39 (3), 581–590.
  • Donorfio, L.K.M., D'Ambrosio, L.A., Coughlin, J.F., Mohyde, M., 2009. To drive or not to drive, that isn't the question – the meaning of self-regulation among older drivers. Journal of Safety Research 40, 221–226.
  • Farah, H., Toledo, T., 2010. Passing behavior on two-lane highways. Transportation Research Part F: Traffic Psychology and Behaviour 13 (6), 355–364.
  • Folstein, M.F., Folstein, S.E., McHugh, P.R., 1975. Mini mental state: a practical method for grading the cognitive state for the clinician. Journal of Psychiatry Research 12 (3), 189–198.
  • Hakamies-Blomqvist, L., Mynttinen, S., Backman, M., Mikkonen, V., 1999. Agerelated differences in driving: are older drivers more serial? International Journal of Behavioral Development 23 (3), 575–589.
  • Harbluk, J.L., Noy, Y.I., Trbovich, P.L., Eizenman, M., 2007. An on-road assessment of cognitive distraction: impacts on drivers' visual behavior and braking performance. Accident Analysis and Prevention 39 (2), 372–379.
  • Holland, C.A., Rabbitt, P.M., 1992. People's awareness of their age-related sensory and cognitive deficits and the implications for road safety. Applied Cognitive Psychology 6 (3), 217–231.
  • Holland, C.A., Rabbitt, P.M., 1994. The problems of being an older driver: comparing the perceptions of an expert group and older drivers. Applied Ergonomics 25 (1), 17–27.
  • Horberry, T., Anderson, J., Regan, M.A., Triggs, T.J., Brown, J., 2006. Driver distraction: the effects of concurrent in-vehicle tasks, road environment complexity and age on driving performance. Accident Analysis and Prevention 38 (1), 185–191.
  • Horrey, W.J., Lesch, M.F., Gabaret, A., 2008. Assessing the awareness of performance decrements in distracted drivers. Accident Analysis and Prevention 40 (2), 675–682.
  • Langford, J., Koppel, S., 2006. Epidemiology of older driver crashes identifying older driver risk factors and exposure patterns. Transportation Research, Part F: Traffic Psychology and Behaviour 9 (5), 309–321.
  • Lavallière, M., Donmez, B., Reimer, B., Mehler, B., Klauber, K., Orszulak, J., et al., 2010. A field study assessing lane changing and lane choice across age and multiple levels of cognitive demand. In: Paper Presented at the Transportation Research Board 90th Annual Meeting, Washington, DC.
  • Matthews, G., Dorn, L., Hoyes, T.W., Davies, D.R., Glendon, A.I., Taylor, R.G., 1998. Driver stress and performance on a driving simulator. Human Factors 40 (1),
  • 136–149. Mehler, B., Reimer, B., Coughlin, J.F., 2012. Sensitivity of physiological measures for detecting systematic variations in cognitive demand from a working memory task: an on-road study across three age groups. Human Factors 54 (3), 396–412.
  • Mehler, B., Reimer, B., Coughlin, J.F., Dusek, J.A., 2009. The impact of incremental increases in cognitive workload on physiological arousal and performance in young adult drivers. Transportation Research Record: Journal of the Transportation Research Board 2138, 6–12.
  • Mehler, B., Reimer, B., Dusek, J., 2011. MIT AgeLab Delayed Digit Recall Task (nback), MIT AgeLab White Paper Number 2011-3B. Massachusetts Institute of Technology, Cambridge, MA.
  • Michon, J.A., 1995. A critical view of driver behavior models: what do we know, what should we do? In: Evans, L., Schwing, R. (Eds.), Human Behavior and Traffic Safety. Plenum Press, New York, p. 520.
  • Olsen, E.C.B., Lee, S.E.,Wierwille,W.W., Goodman,M.J., 2002.Analysis ofdistribution, frequency, and duration of naturalistic lane changes. In: Paper Presented at the Human Factors and Ergonomics Society 46th Annual Meeting, Baltimore, MD.
  • Pande, A., Abdel-Aty, M., 2006. Assessment of freeway traffic parameters leading to lane-change related collisions. Accident Analysis and Prevention 38 (5), 936–948.
  • Planek, T.W., Fowler, R.C., 1971. Traffic accident problems and exposure characteristics of the aging driver. Journal of Gerontology 26 (2), 224–230.

  • Reimer, B., 2009. Cognitive task complexity and the impact on drivers' visual tunneling. Transportation Research Record: Journal of the Transportation Research Board 2138, 13–19.

  • Reimer, B., Mehler, B., 2011. The impact of cognitive workload on physiological arousal in young adult drivers: a field study and simulation validation. Ergonomics 54 (10), 932–942.

  • Reimer, B., Mehler, B., Wang, Y., Coughlin, J.F., 2012. A field study on the impact of variations in short-term memory demands on drivers' visual attention and driving performance across three age groups. Human Factors 54 (3), 454–468.

  • Sodhi, M., Reimer, B., Llamazares, I., 2002. Glance analysis of driver eye movements to evaluate distraction. Behavior Research Methods Instruments and Computers 34 (4), 529–538.

  • Szlyk, J.P., Seiple, W., Viana, M., 1995. Relative effects of age and compromised vision on driving performance. Human Factors 37 (2), 430–436.

  • Wickens, C.D., 1984. Processing resources in attention. In: Davies, D.R., Parasuraman, R. (Eds.), Varieties of Attention. Academic Press, New York, NY.

  • Wickens, C.D., McCarley, J.S., 2008. Applied Attention Theory. CRC Press, New York. Zeitlin, L.R., 1993. Subsidiary task measures of driver mental workload: a long-term field study. Transportation Research Record 1403, 23–27.

  • Zhou, H., Itoh, M., Inagaki, T., 2009. Effects of cognitive distraction on checking traffic conditions for changing lanes. In: Paper Presented at the Proceedings of the Human Factors and Ergonomics Society 53rd Annual Meeting, San Antonio, TX.