The detailed methods section.
Method
Participants
Sample size
A sample size calculation was run based on the largest anticipated test with an estimable effect size: a two-way Mixed ANOVA on the effect of the between subject variable age (young vs old) and within subject variable block (first vs last) on omission error. We expected the power (1 – β) of 0.80, α = 0.05 and a medium effect size of Cohen’s f 0.25, according to a recent meta-analysis (Vallesi et al., 2021) as well as previous findings (Hanzal et al., 2021), where results were highlighted suggesting this strength of an effect (Brache, Scialfa & Hudson, 2010, Staub et al., 2014). The power analysis was conducted using the “Superpower” package in R (Lakens & Caldwell, 2021), determining that 34 participants (17 per age group) should be used for data analysis, ignoring power analysis for post-hoc tests (Althouse, 2021). Given the possibility that 10% of participants may be excluded from data analysis, the total number of participants was raised to 38 (19 per age group), also matching the sample size of the mentioned papers.
Recruitment
The participants were recruited out of the student population (age 20-26) through the University of Glasgow subject pool. The older adult group was recruited from the general population in the Glasgow area with age cut-off beginning at 55 (Vallesi et al., 2021).
Exclusion criteria
There were no recruiting pre-requisites based on handedness due to no expected impact, but participants were balanced for gender. They were also asked to report any existing medical conditions, eye-sight correction and medication which could impact the experiment and lead them to be considered for exclusion. They were likewise screened for dementia and visual field deficiencies during the experiment.
Sample
A total of 41 healthy adults aged between 18 and 87 years were recruited and scheduled for a single recording session. The data collection ran in the period of 6 months between March to August 2022. Of the 41 participants recruited, 34 provided usable EEG data. Data from 7 participants were excluded in later pre-processing due to excessive noise and signal artifact using later specified criteria. The final sample consisted of 34 participants (F = 16) in two age groups: young (n = 18, F = 9, mean age = 22.61, SD = 1.85, range = 20 - 26) and older adults (n = 16, F = 7, mean age = 66.50, SD = 8.45 years, range = 55 - 87). Two participants reported left-handedness. There was one smoker in the sample. All participants reported up to moderate caffeine consumption (mean units = 1.31, SD = 1.12, range = 0-4), corresponding to the recommended daily dose of 400mg of caffeine (Mitchell et al., 2014). A unit of caffeine consumption was defined as one coffee or two teas, corresponding to 100mg caffeine intake. They also showed a typical (Hirshkowitz et al., 2015) pattern of average sleeping hours citation. (mean hours = 7.34, SD = 0.85, range = 6-9). 20 participants started the recording session in the morning with the reminder doing so in the afternoon (starting hours 8:45 – 16:30). One participant reported a borderline underactive thyroid but was still included in the sample. The participants were also screened for cognitive difficulties using the Montreal Cognitive Assessment test (MoCA) (Nesreddine et al., 2005), showing that all included participants met the standard of two standard deviations from age-specific group mean on the examination: both young (mean score = 28.28, SD = 1.49, range = 26 - 30) and older (mean = 25.81, SD = 2.74, range = 22 - 30) adults. A regression revealed a significant difference between the two groups based on their MOCCA scores F(1, 32) = 10.98, p = .002, but scores still representative of a healthy aged population (Engedal et al., 2022). No participant had been administered the MoCA less than 6 months ago.
Ethics
The current study was conducted in line with BPS guidance and approved by the University of Glasgow College of Science and Engineering Ethics committee. The participation was also undertaken with a required window of 14 days since any incidences of testing positive for Covid-19 infection or were required to re-test prior to the experiment. Participants received a minimum payment of £13.50 for their participation. As overall participation lasted between 130 to 165 minutes, the final payment was adjusted to reflect the base rate of £6 per hour.
Measures
The Visual Analogue Scale for Fatigue (VAS-F) was used as the state fatigue measure. It captures changes in subjective state fatigue through 18 items divided into two subscales, one for fatigue (13 items) and one for energy (5 items) with scores 0-100. It has shown excellent reliability of α = .93 and α = .91 for the two scales, respectively (Lee et al., 1991). Two of the items on the scale were altered: “worn out” to “drained”, and “bushed” to “run down” to avoid repetitiveness and a poor understanding of the items. The spontaneous subscale from the Mind Wandering measure (Carriere et al., 2013), comprised of 4 items on a seven-point Likert scale was administered to measure mind-wandering level change in the experiment.
Visual screening test
A short (3 minute) computerised visual screening assessment was administered at the beginning of the session to ensure that the participants were suitable for subsequent perceptual testing and exclude potential visual pathology. The task was adapted from a similar experiment investigating lateralised visual attention in young and older participants (Learmonth et al., 2017). Small black 0°23’ x 0°23’ (0.5 x 0.5cm) squares were briefly presented individually at 32 possible locations extending to 12° 55’ above and below fixation, and 22° 37’ to the left and right (34 x 60cm) for 150ms. There were 32 trials in total interspaced by irregular fixation periods to measure adherence with the task. Participants were requested to press the space bar if a stimulus had been detected and to withhold their response when undetected. None of the participants (all having reported normal or corrected-to-normal vision during recruitment) had to be excluded based on performance in this visual screening test.
Procedure
The study hypotheses, design and analysis plan were pre-registered prior to data collection and can be accessed via the Open Science Framework: https://osf.io/y2vgc/. A custom E-Prime 2.0 script was used to run the SART. Figure 1 outlines each trial. A trial consisted of a number between 0-9 presented centrally for 250ms. The number disapeared for 3000-4000ms before the next one is presented. The response window size varied to prevent learning effects and was be extended relative to standard SART protocols (Robertson et al., 1997) to facilitate capture of oscillatory patterns. Participants were instructed to fixate centrally on the presented fixation cross and attend to the stimuli presented at an angular distance of 1°. The fixation was introduced to reduce task-irrelevant eye-movements and avoid visual flicker. No colour was used in the task and the stimuli were presented black on a white background using a 21-inch CRT monitor (Samsung, SyncMaster 1100MB) with at a screen resolution of 1,024 x 768 pixels and a refresh rate of 100 Hz. The monitor was calibrated for luminance and chromaticity to ensure accurate and consistent stimulus presentation throughout the experiment. Participants were seated at a comfortable viewing distance of 60 cm from the screen, maintaining a horizontal eye level with the center of the display. A chin rest was employed to minimize head movement and maintain a stable viewing angle. The task was to respond to any number apart from the numbers 3 or 6 by left click (go), whilst withholding their response when seeing numbers 3 or 6 (no-go). The block did not include any feedback about response time or accuracy. The stimuli were pre-generated to be distributed randomly and represented in equal frequency across all block and then showed in the same order to all participants.
The experimental procedure is outlined in the attached figure (Figure 2). The participants were firstly provided with an information sheet and asked for their consent. They were then asked to provide basic demographic information and report visual deficiencies as well as other impeding conditions. A 64 Ag/AgCl BrainCap (BrainProducts, Gilching, Germany) including two electrooculographs were fitted according to the international 10/20 system (American Electroencephalograpic Society, 1991) and impedance reduced to <25KΩ using Signa gel, before they completed the VAS-F and MW-S. The procedure of impedance calibration ranged between 30 – 60 minutes per participant. Participants then complete a brief visual field test to map peripheral vision. They then underwent a practice session for the SART and were given general feedback about their accuracy on the practice block without specified desirable outcomes. Those who failed to satisfy minimum performance standards as outlined in the exclusion section were informed of this and repeated the practice block (n = 3). No participant had to repeat the practice block more than once. Then, they completed a session of eight blocks of the SART each lasting five minutes and 20 seconds with self-paced breaks between them, all recorded at 1000Hz sampling rate. Each block was initiated by the experimenter based on a cue from the participant. The break message included the information which of 8 blocks the participant just completed. The total duration of breaks between the blocks were estimated from recording lengths (mean break length = 49.57 seconds, SD = 16.78 seconds). This was followed by administering VAS-F and MW-S to record post-task subjective states. The initial eight blocks of SART were followed by one further unannounced EEG-recorded block of 90 trials. Half of the participants were allocated by single-blind randomisation into a motivational initiative group. They were instructed to try and maximise their performance on the last block. They were likewise informed that the participant with top performance on the last block will receive a monetary reward of £50. The other half were only informed that the experiment includes an additional block and asked to undertake the block with the same instructions as the previous blocks. No participant refused to partake in the additional block. After the motivational manipulation, the participants were screened for mild cognitive decline using the MoCA. Finally, the EEG cap was taken off and the participant aided with the removal of Signa gel and access to hair washing facilities, proceeded by full debrief. Behavioural analyses Data for behavioural analysis was chosen based on the following steps:
- Participants with insufficient neural data as outlined in data pre-processing were not included in behavioural analysis.
- Trials with trigger information missing due to failure of transition of an event signal were identified and removed.
- Commission trials were removed based on deviant reaction time. Firstly, block and age group mean and standard deviation in reaction time was computed. Blocks were considered separately due to expected time effect and age groups were used to effect of cognitive strategy. Then trials rising above two standard deviations beyond the mean were removed as attention lapses and trials two standard deviations bellow the mean as anticipation error (Kiesel et al., 2008).
- Participants were also removed if they exhibited any of two identified erroneous strategies in any of the 8 main experimental blocks: 1) responding to all trials at chance level (> 80% go stimuli correct and < 20% no-go stimuli correct), or 2) withdrawing the response for all trials at chance level (< 20 % go stimuli correct and >80% no-go stimuli correct) in any of the eight experimental blocks.
To ensure the suitability of reaction time data for linear modeling, transformations to the data were considered. These transformations will be implemented following assumption checks on the skewness and kurtosis of the distributions. In cases of non-normality, a remedy will be sought in the form of log transformation. Logarithms are an established method for addressing non-normality in reaction time data which typically involve positive skew, with a longer tail extending to the right (Ratcliff, 1993). They in turn help to mee the assumptions of parametric statistical tests, leading to more accurate and reliable results.