The present study analyzes data from a larger research project on multitasking in older age. One subcomponent of this project is multitasking during simulated car driving. This research led to several methodological and data-based publications, but those publications have not addressed the effects of additional tasks on the reaction time of braking. The larger research project has been approved by the ethics committees of the institutions involved, and was carried out in accordance with the declaration of Helsinki. Informed consent was given by all participants before testing began.
Participants
Simulated-driving data from 120 older participants were registered for the purposes of the larger research project, and therefore were available for the present analysis. None of those persons was excluded from analysis. Post-hoc power analysis for the effect of the factor Condition, which is the main effect of interest (see below), was carried out with G*power (Faul et al., 2007). Using f = 0.25, α = 0.05, 1 - ß = 0.95, 1 group, 2 measurements, r = 0.5, ε = 1 yielded 54 as required sample size.
All analyzed participants were community dwelling, healthy adults between 64 and 79 years of age. Their demographic characteristics are shown in Table 1. We used this Participants were recruited by public advertising, flyers, and oral presentations to seniors. Interested persons were screened by a structured phone interview for the following inclusion criteria: age between 65 and 75 years, BMI < 30, absence of physical, neurocognitive, or psychiatric medical conditions, at least one active driving trip per week over the last six months, and ability to walk for at least 30 min without assistance. If eligible persons so wished, we also included spouses in our sample even if they did not fully meet our age criterion: five spouses were 64 years old, one was 78, and three were 79 years old. Participants received a nominal compensation for their contribution to our work.
Table 1
Participant characteristicsa.
|
males (n = 60)
|
females (n = 60)
|
age (years)
|
71.34 ± 2.54
|
72.22 ± 2.12
|
education (years)
|
13.32 ± 1.20
|
12.76 ± 1.34
|
height (m)
|
1.76 ± 0.10
|
1.74 ± 0.08
|
weight (kg)
|
70.10 ± 5.67
|
72.33 ± 4.20
|
BMI (kg/m²)
|
26.22 ± 2.31
|
27.52 ± 3.00
|
MMSE (0–30)
|
28.78 ± 1.20
|
28.22 ± 0.98
|
aTable entries are means ± standard deviations |
Eight participants were left-handers, as assessed by the Edinburgh Handedness Inventory (Oldfield, 1971). All stated normal or corrected to normal vision and hearing, and all passed the Freiburg Visual Acuity Test v. 3.9.0 (cut-off: 20/60; Keeffe et al., 2002). Their normal cognitive status was confirmed by the Mini-Mental-State-Examination Test (cut-off: 27/30; Creavin et al., 2016), and their years of formal education were assessed by self-report (cf. Table 1). All participants underwent electrocardiography registration and obtained a medical clearance before they were enrolled in the project, since another subcomponent of the project involved a graded exercise test. Participants received a nominal compensation of 15 € per testing day. The project was approved by the ethics committees of the participating institutions, and it was carried out in accordance with the declaration of Helsinki (World Medical Association, 2013). Informed consent was given by all participants before testing began.
Simulated driving
Participants were seated in a research-grade driving simulator (Carnetsoft® version 8.0, NL) with three 48” monitors (195°horizontal field of view), a Logitech G27 steering wheel & pedal assembly that was placed to ensure each participant’s comfort, a numeric key pad with keys 1 to 6 that was placed next to the steering wheel on the side of hand dominance, and a head set that presented situation-matched driving sounds as well as auditory task stimuli (cf. below). To minimize perceptual conflicts between the dynamic driving display on the monitors and the static laboratory surrounds, the laboratory was screened off by black cloth and the ambient light was dimmed.
The monitors displayed a driver’s view of the cockpit with front and side windows, a dashboard with typical car instruments, a rear-mirror and two side-view mirrors. Inside the simulated windows, the monitors showed a softly winding two-lane road which passed through a rural environment without intersections. The scene was populated by trees, buildings, mountains, clouds, etc., and by motor vehicles travelling in both directions. The driving course had a length of 25.7 km, and was completed within about 25 min.
Participants were instructed to follow a lead car that usually drove at a constant speed of 70 km/h. If inter-vehicle distance exceeded 50 m, the lead car slowed down to 70% of the driver’s current speed, and returned to its usual speed once the inter-vehicle distance decreased to less than 50 m; this ensured a relatively stable inter-vehicle distance throughout the experiment. Braking tasks were administered at irregular locations along the driving course: when reaching a construction site or a posted speed limit, the brake lights of the lead car lit up and its speed dropped within 7 s to 40 km/h. After another 6 s, the lead car sped up again, returning to 70 km/h within 9 s. Participants had to respond by slowing down as well, as they otherwise would crash into the lead car. Crashes were rare; they were simulated by displaying a shattered windshield and a characteristic crashing sound. The few braking trials with crashes were not repeated, and were excluded from analyses.
In the control condition, participants followed the lead car and responded in the braking task as described above. In the multitasking condition, the same participants responded in the braking task and also in a range of additional tasks. In a typing task, they had to enter a three-digit number into the numeric key pad; this task simulated driver interaction with in-vehicle devices such as a radio or a navigation system. Numbers were presented visually for 5 s, or auditorily for about 3 s. In a reasoning task, participants had to verbally argue for or against an issue of general interest (e.g., “state an argument against vacationing on a cruise ship”); this task simulated an engaging conversation with car passengers or telephone partners. We selected topics that could not adequately be answered by “yes” or “no”, to ensure that participants contemplated their response rather than issuing a brief verbal statement without much prior thinking. Again, topics were presented visually for 5 s, or auditorily for about 3 s. In a memory task, participants had to recall the prices posted at gas stations along the road, or recall the information provided by traffic announcements during the trip (highway number, position and length of a traffic jam). The time length of gas price visibility is difficult to quantify, but that of traffic announcements was about 4.5 s.
In the multitasking condition, braking tasks and additional tasks were presented in a mixed order at irregular locations along the driving course, under the constraint that no task was presented twice in a row. Task locations were identical for all participants and both conditions. SOA between a braking task and the last preceding additional task varied in dependence on task locations and momentary driving speed, within the range SOA = 11.49 ± 1.99 s (mean ± standard deviation).
Procedures
Data were registered in two distinct project phases with different research objectives, over a period of about four years. In phase I, the braking task was presented ten times per trip; the additional typing, memorizing and reasoning tasks were presented ten times per trip visually, and ten times per trip auditorily. In phase II, the braking task was presented eleven times per trip; the additional typing and reasoning tasks were presented fifteen times per trip visually, and fifteen times per trip auditorily. Thus, there were 70 task presentation per trip in phase I, and 71 in phase II. Sixty-one participants for the present study came from phase I, and 59 from phase II.
Before the first day of testing, participants received a general overview of the project, and they completed a written informed consent form as well as a questionnaire on demographics, health status, driving behavior, physical and social activities, and handedness. The actual testing was distributed across two to four days, and assessed physical fitness, cognitive fitness, and multi-tasking in a car driving and in a walking scenario. Testing order was pseudorandomized to control for serial-order effects. Specifically, simulated car driving in the multitasking condition and in the control condition were administered in pseudorandomized order on separate days. Before registration in the chronologically first driving condition, participants practiced driving in the control condition for 2 to 3 minutes.
Calculation of outcome variables
Raw data were the time series of gas pedal positions, brake pedal positions and inter-vehicle distances, as registered by the driving simulator software. We analyzed those data with R Studio v1.1.463 (R Core Team, 2020), to determine the following variables for each braking trial:
-
Gas-off reaction time (RT): Time interval between the illumination of brake lights on the car ahead, and the onset of gas pedal release. An algorithm searched for that onset in a time window from − 3 s to + 15 s about the instant of brake light illumination, rejecting small fluctuations that changed gas pedal position by less than 10%. The time window started at -3 s to capture anticipations, which are common in car driving.
-
Foot movement time (MT): Time interval between the onset of gas pedal release and the onset of brake pedal actuation. To determine the latter onset, an algorithm searched for the first instance at which brake pedal depression was greater than 0, again in the above time window.
-
Inter-vehicle distance (Distance): distance between the front bumper of the participant’s car and the rear bumper of the lead car, at the time when the brake lights on the lead car came on.
On some braking trials, participants released the gas pedal but didn’t depress the brake pedal within the search window interval. On some other trials, they kept their foot on the gas pedal throughout the search window interval. We therefore could analyze 1680 RT scores and 1563 MT scores.
Statistical analysis
We used linear mixed models, which offer a regression approach for repeated-measures designs even when data of some individuals are missing on some repetitions. It therefore was not necessary to impute or otherwise deal with missing data. Model parameters were fitted by the ‘lme4’ package (Bates et al., 2015) in R Studio v1.1.463 (R Core Team, 2020), based on maximum likelihood estimation (Field et al., 2012). To address our primary research hypothesis, which stipulates that the reaction time of braking responses will be higher in the multitasking condition than in the control condition, we used RT as dependent variable and Condition (multitasking, control) as a fixed effect of interest. To address the secondary hypothesis according to which the difference between conditions only emerges at long inter-vehicle distances, we also included Distance and Condition x Distance as fixed effects of interest. Age, Gender and Education were entered as fixed effects to control for potential confounds and provide more accurate estimates of the fixed effects of interest. The intercepts and slopes of participant ID in Condition were added as random effects, to account for non-independence in the data. To address the secondary research hypothesis, according to which higher RT in the multitasking condition is compensated by lower MT, we calculated a second linear mixed model in which MT was used as dependent variable, fixed and random effects were the same as above, but RT and its interactions with Condition and Distance were also included as fixed effects. All continuous independent variables were mean centered before they were entered into the models.
Fixed effects were evaluated with the t-statistic obtained from the ‘lmerTest’ package (Kuznetsova et al., 2017) that applies Satterthwaite’s approximation. Wald intervals were used for confidence interval estimation. Partial eta-squared (η²) from the ‘effectsize’ package was calculated as effect size, and was interpreted as small (0.01), moderate (0.06) or large (0.14) in accordance with Cohen (1988).