An EEG Study of Mind wandering During Decision-Making
J. N. ZadelaarUniversiteit van Amsterdam (UvA), Amsterdam Brain and Cognition (ABC )
July 2015
Mind wandering is the cognitive process of attention drifting from a task at hand towards task-unrelated thoughts or daydreams, often causing suboptimal task performance. Pre-stimulus alpha activity, brain-waves related to a task-unengaged resting-state, observed milliseconds prior to a stimulus, are similarly related to suboptimal performance. Both these phenomena would cause suboptimal performance because both lead to flawed information processing. As such, pre-stimulus alpha-activity is expected to: 1) be indicative of mind wandering and 2) negatively influence mental processing. A Random Dot Motion task is conducted. Mind wandering is measured using thought probes, and alpha-activity using EEG. Firstly, a repeated-measures ANOVA between thought-probes and alpha-activity proved non-significant, suggesting that pre-stimulus alpha-activity is not indicative of mind wandering. Secondly, a full Ratcliff diffusion model where the drift rate parameters are informed by activity did not have an overall better fit than a model not informed by alpha-activity, suggesting that pre-stimulus alpha-activity does not influence mental processing. These results may however be the result of the experimental design or analyses, rather than the absence of the expected effect. Our belief is that future research may still support our hypotheses.
Keywords: mind wandering, EEG, decision model, full Ratcliff diffusion model, pre-stimulus alpha-activity
1. Introduction
Mind wandering is the cognitive process of an individual’s attention drifting away from the task at hand, usually towards internal processes such as task-unrelated thoughts and daydreams (Giambra, 1995; Smallwood & Schooler, 2006; Smallwood, McSpadden & Schooler, 2008). This psychological phenomenon is often negatively correlated with task performance, as it impairs processing of incoming information (Smallwood, Baracaia, Lowe & Obonsawin, 2003), leaving people with incorrect or incomplete information on which to base decisions, thus leading to suboptimal behavior (Smallwood & Schooler, 2006; Barron, Riby, Greer & Smallwood, 2011). As such, mind wandering has received increased attention from the scientific community; an increased understanding of mind wandering may help prevent accidents, understand attention deficits, and improve academic, professional and personal performance (He, Becic, Lee & McCarley, 2011; Smallwood & Schooler, 2006; Smallwood, Fishman & Schooler, 2007).
Mind wandering might be related to alpha-activity: neural oscillations at 8-12Hz brain-waves that are typically associated with a resting state when individuals are awake but not actively engaged in a task. Alpha waves appear to play an inhibitory role in the brain (Jensen & Mazaheri, 2010), with the function of enhancing performance. If alpha activity occurs in task-irrelevant areas, these areas and related processes would be inhibited from interfering with the task, thus enhancing task performance.
This is supported by the finding that alpha activity increases in the visual cortex during motor-tasks, while it increases in the motor-cortex during visual tasks (Pfurtscheller, 1992) and is related to improved performance (Klimesch, Sauseng & Hanslmayr, 2007). However, alpha activity within task-relevant brain areas might inhibit these areas from functioning properly, which would explain the relationship between increased alpha activity and decreased task performance (Hanslmayr, Aslan,
Staudigl, Klimesch, Herrmann & Bäuml, 2007).
Interestingly, it may be pre-stimulus alpha activity, rather than overall alpha activity, which is related to task-performance (Babiloni, Vecchio, Bultrini, Romani & Rossini, 2006; Ergenoglu et al., 2004; Mathewson, Gratton, Fabiani, Beck & Ro, 2009; Thut, Nietzel, Brandt & Pascual-Leone, 2006). Specifically, it could be that alpha activity in task-relevant brain areas just before the appearance of a stimulus causes decreased performance for the upcoming stimulus (Romei, Gross & Thut, 2010). This hypothesis is consistent with the theoretical background of mind wandering; mind wandering is thought to lead to (i.e., cause) impairments in the processing of incoming stimuli - an impairment at an early stage of processing, as opposed to later, higher-order operations (Smallwood, Fishman & Schooler, 2007). This hypothesis corresponds to the finding that the neurophysiological state just before, rather than during, the appearance of task stimuli influences performance. This further supports our expectation of a link between mind-wandering and pre-stimulus alpha activity.
In summary, previous research suggests there may be a relationship between the theoretical construct of mind-wandering and the empirical phenomenon of pre-stimulus alpha activity. Such a relationship would attribute pre-stimulus alpha activity a theoretical, functional explanation, thus helping to understand it better. Moreover, pre-stimulus alpha activity may help explain how mind-wandering relates to decreased performance by incorporating it in a quantitative model describing the cognitive processes underlying task performance (Eggen & Sanders, 1993; Mellenbergh, 2011; Smith & Ratcliff, 2004). Quantitative modeling allows us to test whether mind-wandering, measured as pre-stimulus alpha-activity, is related to changes in the model’s parameters. If so, we may conclude that mind-wandering affects performance via changes in the psychological processes corresponding to the affected model parameters, thus providing an explanation of how mind-wandering influences performance (Smallwood & Schooler, 2006).
We will use the drift diffusion model (DDM), a commonly applied model for analyzing perceptual decision-making (Ratcliff & McKoon, 2008; Smith & Ratcliff, 2004; Vandekerckhove & Tuerlinckx, 2007; Wagenmakers, 2009). This model provides an account of the cognitive processes involved in two-choice decision tasks performed under time-pressure. The DDM assumes that decisions are made by gradually collecting information from a stimulus until the amount of accumulated information in favor of one of two choice options surpasses a threshold (see figure 1). The process of evidence accumulation is susceptible to noise, which causes random fluctuations in response time or response accuracy.
FIGURE 1: The Full Ratcliff Diffusion Model. Figure retrieved from: Vandekerckhove & Tuerlinckx (2007).
Figure 1 shows a version of the DDM with seven parameters. Information begins accumulating from the start point (𝑧), which can fall anywhere between one of the two response boundaries. The start point represents potential response biases, an a priori preference for one of the two response options. The drift rate (denoted as 𝑣) reflects the rate of information accumulation, where a higher drift rate (a steeper average line) reflects more efficient mental processing. Boundary separation (𝑎) represents response caution, the amount of information that must be accumulated before a decision is made, where a higher boundary setting reflects more cautious responding. Once a boundary is reached, the corresponding choice is made. The predicted response time is equal to the time it took for the accumulation process to reach the boundary plus non-decision time (𝑇!"), which
represents the time required for non-decision processes such as encoding the stimulus and producing a motor response. Modern versions of the DDM allow for between trial variability in three of the DDM parameters (Ratcliff & McKoon, 2008). In particular, it is assumed there is trial-to-trial variability in drift rate according to a Gaussian distribution with mean 𝑣 and standard deviation
η
, start point according to a uniform distribution with mean 𝑧 and range 𝑠!, and non-decision time according to a uniform distribution with mean 𝑇!" and range 𝑠!. The parameters are inferred from observedbehavioral variables of response time and accuracy (Ratcliff & McKoon, 2008; Vandekerckhove & Tuerlinckx, 2007).
Since mind-wandering is thought to influence performance by impairing information processing, we expect mind-wandering to influence the drift rate parameter (𝑣), corresponding to the “efficiency of mental processing”. This finding would explain the process leading from mind-wandering to suboptimal performance as an impairment in the speed of information processing.
In this research we investigated the link between mind-wandering, pre-stimulus alpha-activity and the DDM drift rate parameter. Participants completed in a Random Dot Motion (RDM) task while alpha-activity was measured with electroencephalography (EEG). Mind-wandering was measured through self-report questions that appear pseudo-randomly throughout the RDM task.
There were two primary research questions. Firstly, we tested whether pre-stimulus alpha-activity was related to mind-wandering. We did this by examining if higher levels of pre-stimulus alpha-activity correspond to higher levels of self-reported mind-wandering. We expect that trails with higher values of self-reported mind-wandering will have a significantly higher pre-stimulus alpha-activity mean.
Secondly, we will investigate if mind-wandering has a negative influence on mental processing. We will do this by fitting two DDM’s to the data: a baseline model and an alpha-informed model. The alpha-informed model contains separate drift rate parameters for trials with high and low pre-stimulus activity. The baseline model does not distinguish between high and low
alpha-activity trials. We expect that the alpha-informed model will provide a more parsimonious account of the data than the baseline model.
2. Methods 2.1 Participants
Thirty participants were recruited, a standard sample size for perceptual decision tasks that record neurophysiological measures (Babiloni, et al., 2006; Barron, Riby, Greer & Smallwood, 2011; Mittner et al., 2014). Inclusion criteria were: a) normal or corrected-to-normal vision, b) no neurological, sensory, motor, or psychiatric disorders, c) being right-handed, and d) being between ages 18 and 30. Participants were recruited through the student participation system (https://www.lab.uva.nl/spt/) and compensated with €20,- or two UvA participation credits. Participants were excluded from data analyses for near-change accuracy (50%) across the whole task. This study was approved by the ethics committee of the University of Amsterdam Psychology Department.
2.2 Stimuli
The task used was a Random Dot Motion (RDM) task. An RDM trial involves a set of dots of which a certain percentage moves coherently in one direction (to the left or right of the display) while the remaining dots move in random directions. The task is to indicate the direction of the coherently moving dots (see figure 2).
FIGURE 2. A schematic representation of the RDM for different levels of coherence. A) 10% coherence; one out of ten dots move in the same direction, the others move in a random direction. B) 50% coherence; half of the dots move in the same direction. C) 100% coherence; all dots move in the same direction.
The task was run and recorded in Presentation and presented on a computer screen. Every RDM trial was presented centered on a black background in a circle with an invisible border, with 120 dots present per frame. Each stimulus dot was white and 3x3 pixels in size. All dots were given a random location inside this circle. Per frame the dots moved one pixel in their designated direction. The lifetime of each group of dots was limited to 3 frames (150 ms), after this time the group was replaced by a new group of randomly placed dots that had the same signal-to-noise ratio or coherence percentage. When a dot moved out of the circle window, it was wrapped around to the opposite side of the circle where it reappeared.
Task difficulty depended on the coherence level of the moving dots; the more dots move coherently, the easier the task is. The used coherence levels were 45%, 15% and 5%, which translates
to 6, 18 or 54 dots respectively. These percentages were derived from calibration data from Van Maanen and Van Rijn (submitted). Coherence levels varied randomly across trials.
Participants were seated on an armchair with two response buttons placed on the end of each arm rest, one placed underneath the index finger and the other underneath the middle finger. To indicate the direction of the coherently moving dots, the participant pressed the index finger button of the armrest on the corresponding side.
Each RDM trial started with a black screen (500ms), followed by a fixation cross (250ms), followed by the moving dots screen that was shown for 1500ms and then removed from the display. The participant was free to respond (left or right) at any time while the moving dots were displayed. Following the moving dots display there is second black screen (200ms), followed by a feedback screen (1000ms) that displays “correct” or “incorrect”, depending on response accuracy, or “no answer” if no response was given during the 1500ms window of the moving dots display (see figure 3A).
FIGURE 3. A schematic depiction of the task. A) An example of an RDM-trial. B) An example of a thought probe trial.
“Thought probes” were used throughout the experiment to measure mind-wandering. Thought probe trials ask the participant “where was your attention during the previous trial?”. The participant was given a 5-point Likert scale (1 = off-task, 5 = on-task) to indicate their response. Prior to beginning the task, participants were informed that being more off-task indicated increased mind-wandering or less focus on the decision-making task. Thought probe trials appeared pseudo-randomly; once during every 15 trial-block, though never in the first 5 trials. The index finger buttons were used to move across the Likert scale. The probe screen disappeared when: 1) simultaneously pressing the two buttons positioned by the middle fingers, or 2) automatically after 15000ms, see figure 3B.
The entire task consisted of 700 RDM trials and 50 thought probe trials, divided into five blocks of 150 trials that each consisted of 140 RDM trials and 10 thought probe trials. The blocks were separated by a screen saying “short break”, which disappeared with the press of any button. The total duration of the experimental task was approximately 50 minutes.
2.3 EEG Recording
Pre-stimulus alpha-activity was measured using electroencephalography (EEG) using 64 scalp electrodes placed according to the International 10-20 system of electrode placement, recorded in Biosemi. Electrodes of interest are: O1, O2, Oz, PO3, PO4, PO7, PO8, and POz, as we expected early visual detection to occur in the occipital lobe (Ergenoglu et al., 2004). Pre-stimulus alpha activity was defined as the power of the alpha wave activity (8-12 Hz) occurring in the [-200ms,0ms] interval relative to the start of every moving dots screen, see figure 4.
FIGURE 4. A time-frequency representations of the power of the EEG-signal over occipital regions measured during the task. The square indicates the measures of interest.
Eye-blinks were measured using four ocular electrodes: one above and one below the left and one on the outer canthi of each eye. One electrode on the left and right mastoids served as the reference electrodes. The EEG signal was filtered using a high-pass filter with a .1Hz cutoff frequency, and a low-pass filter with a cutoff frequency of 100Hz.
2.4 EEG Preprocessing and Analysis
Pre-processing and analyses of the EEG-data were performed in Matlab. Reference-signals (from the mastoid electrodes) were subtracted from electrode-signals to correct for baseline-levels of electric activity. The data were collected at a sampling rate of 512Hz and high-pass filtered at 0.1Hz. Low-pass and notch-patch filters were deemed unnecessary as these filter out frequencies higher than those of interest (alpha-waves). Next, the data was epoched, with epochs running from -750ms to 2700ms following stimulus onset (with 0 being the start of the moving dots stimulus). Single-trial
EEG epochs were transformed with a Laplacian transformation to eliminate the contribution of deep and broadly distributed sources. This was followed by a manual data-removal using Matlab; trials showing EMG (muscle) artifacts were removed. An average of 11% (SD=6.8%) of the trials were removed as EEG artifacts (range across participants from 2.1% to 28.6%). An independent component analysis was used to remove trials containing eye-blinks. Noisy channels were interpolated from adjacent electrodes. A sufficient amount of trials remained for the intended analyses.
Convolution and FFT were used to identify the frequency of brain-activity. EEG time-frequency analysis was performed using complex Morlet wavelets, which were defined as a Gaussian-windowed complex sine wave, running from 6Hz to 60Hz in 40 logarithmic steps. The time–frequency resolution was adjusted for the lower and the higher frequencies: The wavelet parameter sigma varied as a function of frequency.
The power in each frequency band at each time point was expressed as a change from baseline power using full-epoch, single-trial correction, calculated in units of decibel (dB) change, following the methods of Grandchamp & Delorme (2011) . The time interval of interest was the pre-stimulus window [-200ms,0ms], the baseline interval was [-550ms,250ms], both relative to stimulus onset. This was done separately for each trial and each electrode, then aggregated over all electrodes of interest (see section 2.3 EEG Recording), resulting in a single alpha-value for each participant for each trial. Next, a decibel correction was performed.
Trials within each coherence condition were split into low-alpha activity and high-alpha activity trials, calculated separately for each participant. This was achieved by first calculating the median pre-stimulus alpha activity for each coherence level and splitting the data into the low-alpha and high-alpha activity halves for each coherence level.
2.5 Model fitting
To test the second expectation, two DDM’s (more specifically: two full Ratcliff diffusion models) were fitted independently to each participant’s data. We expected that the alpha-informed model (the model with different drift rate parameters for high and low pre-stimulus alpha trials) would provide a more parsimonious account of each participant’s data than the baseline model (the model without separate drift rate parameters for high and low alpha trials). The baseline model contained three drift rate parameters, one for each coherence level (𝑣!, 𝑣!", 𝑣!"). The alpha-informed model contained six drift rate parameters, one for each coherence level, separately for trials with high and low pre-stimulus alpha-activity (𝑣!"!"!, 𝑣!"#!"!, 𝑣!"#!"!, 𝑣!!"#, 𝑣!"!"#, 𝑣!"!"#). Posterior distributions of the DDM parameters were estimated using differential evolution Markov chain Monte Carlo (DE-MCMC), using the default settings (see Turner, Sederberg, Brown, & Steyvers, 2013). Prior distributions of each of the model parameters are displayed table 1.
Table 1. Descriptions of the prior distribution of every model parameter.
Parameter Distribution Mean Sd Lower limit Upper limit Interval
a Gaussian (truncated) 1 1 0 Inf
v Gaussian 0 3
eta Gaussian (truncated) 1 1 0 Inf
z Uniform [0, a]
sz Uniform [0, a]
ter Uniform [0, 1]
The same prior distributions were used for both models to ensure that any differences in parameter estimates were driven by the data. We ran each of 36 chains for 2000 iterations of the DE-MCMC sampling routine. We discarded the first 1000 samples as burn-in, for a total of 36,000 samples from the posterior distribution of the parameters. Convergence was checked by visual inspection and the multivariate potential scale reduction factor statistic (cutoff <1.15; Brooks & Gelman, 1998).
The deviance information criterion (DIC)1 of both these models was calculated separately for
each participant. The DIC is an information criterion that estimates model fit while penalizing model complexity. DIC is a standard and well-developed model selection criterion in Bayesian statistics, and it possesses computational advantages over other measures for our models and research goal (Spiegelhalter, Best, Carlin & Van Der Linde, 2002). Lower DIC-values indicate a better model-fit.
2.6 Procedure
The participant was led into a quiet, dimly lit room and seated in front of a computer screen. The researcher described the general procedure and provided a research information brochure. After the informed-consent form was signed, an EEG-cap was placed on the participant’s head. Before the task started, the participant practiced the RDM-task and was shown their EEG-signal (as viewed in BioSemi). Participants were shown the effects of smiling, frowning, gritting teeth and excessive blinking on the EEG signal, and were discouraged from these actions throughout the task. The researcher left the room and the test started. All measures, as well as the participants themselves, were monitored from an adjacent room. The entire process took approximately 120 minutes. Prior to the research participants were informed to arrive with washed, dry hair containing no hair products (e.g.,
hair gel).
1 For a clear summary of the DIC and its use, see also: DIC: Deviance Information Criterion, retrieved from:
3. Results
3.1 Trail exclusion
Firstly, 348 RDM-trials with response times equal to or below zero, and 58 RDM-trials containing middle-finger button responses were omitted, assuming that these responses were unintentional. The number of omitted trials was significantly larger for trials with lower coherence levels, 𝑋! (2) = 32.961, p < 0.0012. Secondly, 131 RDM-trials with very fast responses (<300ms) were
excluded from analysis, as these trials showed below-chance mean accuracy (<0.5). The number of trials removed did not differ significantly across coherence levels, 𝑋! (2) = 2.580, p = 0.275. Note that
84.7% of these trials belonged to two participants who would later be excluded based on performance accuracy. No thought probe trials were removed.
3.2 Participant exclusion
Mean accuracy for each participant at each coherence level was calculated. For seven participants, the mean accuracy rates approximated chance performance (0.5), with mean accuracies falling into the range [.468, .568]. This indicates an overall inability to perform the task. Even for the 45% coherence level (the easiest trials) the highest mean accuracy was .546 compared to .986 in the group of remaining participants. As such, these seven female participants were excluded from following analyses. The remaining participants were 10 men and 13 women, with a mean age of 22.7 years.
3.3 Behavioral Analysis
Response times and accuracy were examined. Given that only females were selected for participant exclusion, we decided to inspect these variables on possible gender differences first; two subsequent pairwise t-tests3 showed that females had a significantly higher response times, p < .001, while males had significantly higher accuracy levels, p < .05. This suggests that men had an overall better task performance.
The effects of coherence level on mean response time and mean accuracy were examined using one-way repeated-measures ANOVA, see figure 5A-B. There was a significant main effect for coherence level on mean response time, F(2, 44) = 84.18, p < .001. Post-hoc follow-up tests were conducted with three pairwise t-tests using Bonferroni’s adjustment for multiple comparisons, which indicated significantly different response times for each coherence level, with p <.001 for all comparisons, see figure 5A. This indicates that higher coherence levels (easier trials) led to significantly faster responses. Applying the same tests for mean accuracy, we found a significant main
2 Unless specified otherwise, analyses in the result section were performed in R.
3 With pooled standard deviations to compensate for the differences in sample sizes caused by trial exclusion. While not
effect for coherence level on mean accuracy F(2, 44) = 318.1, p < .001, and significantly different mean accuracies for each coherence level, again with p <.001 for all comparisons, see figure 5B. This shows that trials with higher coherence levels (easier trials) were answered correctly significantly more frequently. These analyses indicate that the coherence manipulation in the RDM task had the expected effect on task performance.
FIGURE 5. A) Mean response time in seconds per coherence level, ranging from 5% (hard) to 45% (easy). The error bars are the standard errors of the mean. The stars indicate the significance level of the p-value of a paired t-test (*** equals p<0.001). B) Mean accuracy (0 = 0% correct, 1 = 100% correct) per coherence level. For other specifications, see description of figure 6A. C) Frequency distribution of thought probe responses in percentages. The error bars indicate the within-subjects standard errors of the mean.
Finally, the thought probe responses were examined, see figure 6C. The distribution of thought probe responses shows that all responses were selected sufficiently often, indicating that the experimental task induced both on-task and off-task behavior in more or lesser degrees.
High and low alpha trials were compared on mean response time and mean accuracy, see figure 6. When comparing high and low alpha, none of these figures displays a clear pattern, indicating no systematic difference in mean response time or mean accuracy for high and low alpha trials. Two separate within-subjects ANOVA’s showed non-significant main effects alpha-activity on mean response time, F(1, 22) = .757, p = .394, and accuracy, F(1, 22) = 1.556, p = .224. These results
suggest that pre-stimulus alpha-activity did not affect task performance.
FIGURE 6. A) Mean response time for high and low alpha trials per coherence level. B) Mean accuracy for high and low alpha trials per coherence level.
3.4 Expectation 01: Thought Probe Responses and Pre-stimulus Alpha-activity
Our first expectation was that higher values of self-reported mind-wandering would have significantly higher pre-stimulus alpha-activity. Every thought probe response (e.g., 5 “on-task”) was
matched to the pre-stimulus alpha-activity measure of the RDM-trial directly preceding the probe. These were aggregated across though probe responses, resulting in one mean alpha estimate for every probe response, for every participant (i.e., every participant has one alpha-estimate estimate for probe response “1”, one for “2”, and so on). A one-way repeated-measures ANOVA, where thought probe responses predicted alpha-activity, was performed. A Shapiro-Wilk test showed that the assumption of normality was not violated, p = .074. There was no significant main effect of thought probe response on mean pre-stimulus alpha power, F(4, 72) = 1.692, p = .161, which is inconsistent with our first expectation.
One problem with this analysis was that not every participant selected every thought probe response at least once, meaning that the (dependent) variable of thought probe response had five levels for some participants, but four or three levels for others; only 9 out of 23participants had selected response 1 at least once, and 21 out of 23 participants had selected response 2 at least once. This unintendedly led R to apply a mixed-design, rather than a repeated-measures ANOVA. Time-constraints did not allow for a solution to be found for this complication so instead an alternative analysis was added; for every participant, thought probe responses were split up into high (“on-task”) and low (“off-task”), based on each participant’s mean (e.g. for a participant with a mean probe response of 3.2, responses higher than 3.2 were considered high, the others were considered low probe responses). Mean pre-stimulus alpha-activity was calculated separately for high and low thought probe responses. We expected that high thought probe trials would have significantly lower pre-stimulus alpha activity. A paired t-test showed then showed there was no significant difference in pre-stimulus alpha-activity between high and low thought probe trials, t(22) = -1.4682, p = .1562. Like the first analysis, this result contradicts the expectation that higher measures of self-reported mind wandering are related to higher alpha-measures.
3.5 Expectation 02: Alpha-informed model versus baseline model
The deviance information criterion (DIC) of both the bassline model and the alpha-informed model was calculated separately for each participant. Then the difference in DIC-values between these two models was calculated, see figure 7. Negative DIC-differences mean that the alpha-informed model provided a more parsimonious account of the data than the baseline model; more negative values indicate a better fit. This figure shows that only 9 out of 23 participants (39.1%) had below-zero DIC-differences, and most were found within the [-5,0] range, indicating a weak to moderate difference. We could not conclude that a model informed by alpha-activity accounted for the data better.
FIGURE 7. The differences in DIC-values between the DDM with separate drift rate parameters for high and low alpha trials and the DDM without separate parameters for high and low alpha trials. The grey line is set at y=0. The red lines indicate the interval y=[-5,5].
To examine this finding more closely, we compared the parameter estimates of the two most extreme participants; the participant with the most negative DIC-difference (best relative fit of the alpha-informed model) and the participant with the most positive DIC-difference (worst relative fit of the alpha-informed mode), see figure 8. For both participants, the drift rates followed expected patterns with respect to coherence levels; they increased as the coherence level increased, indicating faster information accumulation during easier trials. The difference between these participants became apparent upon comparing their respective drift rate difference between high and low alpha trials. For the participant with the most negative DIC-difference, drift rates were lower for high alpha trials than for low alpha trials across all coherence levels. This suggests that information accumulation was slower as pre-stimulus alpha got higher, as expected. The drift rates of the most positive DIC-difference participant showed little DIC-difference for high and low alpha trials however; only at the 45% coherence level did a difference even become visible, and still it remained minute. This indicates that pre-stimulus alpha-activity had close to no effect on information accumulation. This shows that the alpha-informed model did reflect the internal processes of some, but not all, participants.
FIGURE 8. Drift rate estimates per coherence level of the alpha-informed model for the two most extreme participants
4. Conclusions & Discussion 4.1 Summary
The goal of this research was to examine the relationship between pre-stimulus alpha-activity and wandering. The first hypothesis was that pre-stimulus alpha-activity is indicative of mind-wandering. This was tested by examining if self-reported mind-wandering could predict pre-stimulus alpha-activity, using a repeated-measures ANOVA and a pairwise t-test. No significant relationship was found. The second hypothesis was that mind-wandering negatively influences mental processing. This was examined by fitting two drift diffusion models to every participant’s data, one where the drift rate parameters were informed by pre-stimulus alpha-activity measures and one where they were not. We expected that the alpha-informed model would provide a more parsimonious account of the data, as it could distinguish between mind-wandering (high pre-stimulus alpha) and a normal state of mind (low pre-stimulus alpha). The alpha-informed model provided a better account of less than half of the participants’ data. In summary, neither hypothesis was confirmed.
4.2 Discussion
In testing our first hypothesis, we failed to observe a relationship between self-reported mind-wandering and pre-stimulus alpha-activity. This may be explained by the experimental design. Recall that participants received performance feedback after every trial (e.g. “correct”). Possibly this
feedback, rather than a reflection on one’s own mental state, steered the self-reports of mind-wandering. This means that what was actually tested may have been the relationship between alpha-activity and self-perceived performance, rather than self-perceived mind-wandering. We did consider this eventually beforehand, but still chose to incorporate feedback to prevent the task from becoming too tedious, which could lead to motivational problems influencing task performance. Another explanations of this result is that the small sample size (23 after participant exclusion) may have resulted in a statistical power too small to detect the expected effect. This seems unlikely though, as this sample size is standard and has proven effective for this type of neuropsychological research (Bompas, Sumner, Muthumumaraswamy, Singh & Gilchrist, 2015; Lou, Li,, Philiastides & Sajda, 2014; Sherwin, Muraskin & Sajda, 2015).
In testing our second hypothesis, we did not find evidence that the alpha-informed decision-making model accounted for data better than the baseline model, suggesting that pre-stimulus alpha-activity does not influence mental processing. This might be because the applied model did not fully capture the decision process; our model assumed that response caution is stable over time, while it seems likely that in a time-limited task such as ours, the pressure to respond increases over time, causing response caution to gradually decline. So while our model assumed fixed boundaries, a dynamic model incorporating collapsing boundaries may have been more appropriate. The difference between fixed and collapsing boundaries is displayed in figure 9A-B.
FIGURE 9. A) Diffusion Decision Model with fixed and collapsing boundaries. B) Response time distributions of models with fixed and collapsing boundaries. Figure retrieved from: Hawkins, Forstmann, Wagenmakers, Ratcliff & Brown (2015).) C) Response time distribution of the dataset used in this research. The red line indicates the density.
When comparing the response time distributions of the fixed and collapsing boundaries models (figure 9B) with the response time distribution of our dataset (9C) our distribution clearly resembles the latter; both distributions seem less peaked and have a relatively stumped right tail compared to the fixed boundaries model. As such, the finding that the alpha-informed model explained the data better for only 39% the participants may have been an underestimation caused by model assumptions. A collapsing boundaries model was not our first choice because these dynamic decision models have been researched insufficiently to make any definite claims about their efficacy (Hawkins, Forstmann, Wagenmakers, Ratcliff & Brown, 2015). In retrospect, both logic and our data suggest a
collapsing boundaries model would have been appropriate though, and may have led to more favorable results.
Another matter of interest lies with participant exclusion; seven participants had near-chance accuracy levels. This suggests an inability to perform the task, but the cause of this is unknown. Yet more surprising was that all these participants were female. On top of that, females had a higher mean response time, and a lower mean accuracy. This would suggest that the task-inability may have been gender-related, but the sample is too small to substantiate such a claim, and there is no literature on inabilities to perform RDM-tasks to shed light on these findings. Potential explanations for the low accuracy levels, regardless of gender, might lie with motivation or fatigue. Closer attention to this problem is advised in the future.
Finally, it is important to examine our results in the light of the current literature. Recall that alpha-activity was found to have an inhibitory effect on the brain region where it occurred (Jensen & Mazaheri, 2010; Rihs, Michel & Thut, 2007), which explains the association between pre-stimulus alpha-activity and diminished task performance (Mathewson, Gratton, Fabiani, Beck & Ro, 2009; Mazaheri, DiQuattro, Bengson & Geng, 2011; Smallwood, Beach, Schooler & Handy, 2008); if a brain region is task-relevant, inhibition in this region – as caused by alpha-activity – hinders task performance. The link between alpha-activity and mind wandering, in turn, seemed likely as mind wandering is associated with the same diminished performance as pre-stimulus alpha activity (Smallwood, Fishman & Schooler, 2007). Our results however did not indicate a relationship between mind wandering and alpha-activity in the task-relevant brain region. This does not necessarily that mind wandering is not indicated by pre-stimulus alpha-activity, but possibly that it cannot be measured solely in the task-relevant brain region. This makes sense, as mind wandering is not just decreased attention for the task at hand, but also the occurrence of task-unrelated thoughts and daydreams. As such, other brain-regions or paths may also be involved in the mind wandering process, for example those related to imagination and memory. In short, pre-stimulus alpha-activity may be indicative of mind-wandering, but it is likely to be a more global neural phenomenon that currently believed – not restricted to a single brain region and possibly not even powerful enough in a single brain region to be detected as such. Our second hypothesis already unconsciously hinted at this; we expected mind-wandering to have a negative influence on the (information-accumulation of the) decision-making process. As such, it seems likely that mind wandering does not restrict itself to brain regions related to basic perception, but is also affects brain regions related to attention and decision-making, such as the frontal or partietal lobe (Andersen & Cui, 2009; Fellows & Farah, 2005). This idea is supported by a study by Braboszcz & Delorme (2011), who found a relationship between pre-stimulus alpha and self-reported mind wandering, but while observing the entire brain rather than a specific task-related region. In short, we can conclude that mind wandering may well be indicated by pre-stimulus alpha activity, but that it is a global, rather than a region-specific neural phenomenon.
4.3 Conclusions
As the results do not conclusively support our hypotheses, we may not conclude that pre-stimulus alpha-activity is indicative of mind-wandering, and decreases task performance by negatively influencing mental processing. This outcome was unexpected, but did lead to an alternative conclusion that mind wandering is a global neural phenomenon – not restricted to a single task-relevant brain region and possibly only measurable globally. As such, we encourage future research to investigate this topic further, keeping in mind the points mentioned in the discussion section.
4.4 Acknowledgments
First of all, I would like to thank Guy Hawkins for being my main supervisor during this research internship, as well as Quirine Tordoir for guiding me through the practical part of the research and all EEG-related matters. I would also like to thank Birte Forstmann for being my Research Master supervisor. Finally, a thanks goes out to Wibout van Woerkom for his professional and helpful attitude as a fellow-intern.
Literature
Andersen, R. A., & Cui, H. (2009). Intention, action planning, and decision making in parietal-frontal circuits. Neuron, 63(5), 568-583.
Babiloni, C., Vecchio, F., Bultrini, A., Romani, G. L., & Rossini, P. M. (2006). Pre-and
poststimulus alpha rhythms are related to conscious visual perception: a high-resolution EEG study. Cerebral cortex, 16(12), 1690-1700.
Barron, E., Riby, L. M., Greer, J., & Smallwood, J. (2011). Absorbed in thought the effect of mind wandering on the processing of relevant and irrelevant events. Psychological science. Bompas, A., Sumner, P., Muthumumaraswamy, S. D., Singh, K. D., & Gilchrist, I. D. (2015). The
contribution of pre-stimulus neural oscillatory activity to spontaneous response time variability. NeuroImage, 107, 34-45.
Braboszcz, C., & Delorme, A. (2011). Lost in thoughts: Neural markers of low alertness during mind wandering. Neuroimage, 54(4), 3040-3047.
Brooks, SP. and Gelman, A. (1998) General methods for monitoring convergence of iterative simulations. Journal of Computational and Graphical Statistics, 7, 434-455
Eggen, T. J. H., M., & Sanders, P., F. (1993). Psychometrie in de Praktijk. Cito: Arnhem.
Ergenoglu, T., Demiralp, T., Bayraktaroglu, Z., Ergen, M., Beydagi, H., & Uresin, Y. (2004). Alpha rhythm of the EEG modulates visual detection performance in humans. Cognitive Brain Research, 20(3), 376-383.
Fellows, L. K., & Farah, M. J. (2005). Different underlying impairments in decision-making following ventromedial and dorsolateral frontal lobe damage in humans. Cerebral cortex, 15(1), 58-63. Giambra, L. M. (1995). A laboratory method for investigating influences on switching
attention to task-unrelated imagery and thought. Consciousness and cognition, 4(1), 1-21. Hanslmayr, S., Aslan, A., Staudigl, T., Klimesch, W., Herrmann, C. S., & Bäuml, K. H.
(2007). Prestimulus oscillations predict visual perception performance between and within subjects. Neuroimage, 37(4), 1465-1473.
Hawkins, G. E., Forstmann, B. U., Wagenmakers, E. J., Ratcliff, R., & Brown, S. D. (2015).
Revisiting the evidence for collapsing boundaries and urgency signals in perceptual decision-making. The Journal of Neuroscience, 35(6), 2476-2484.
He, J., Becic, E., Lee, Y. C., & McCarley, J. S. (2011). Mind wandering behind the wheel
performance and oculomotor correlates. Human Factors: The Journal of the Human Factors and Ergonomics Society, 53(1), 13-21.
activity: gating by inhibition. Frontiers in human neuroscience,4.
Klimesch, W., Sauseng, P., & Hanslmayr, S. (2007). EEG alpha oscillations: the inhibition– timing hypothesis. Brain research reviews, 53(1), 63-88.
Lou, B., Li, Y., Philiastides, M. G., & Sajda, P. (2014). Prestimulus alpha power predicts fidelity of sensory encoding in perceptual decision making. NeuroImage, 87, 242-251.
Mathewson, K. E., Gratton, G., Fabiani, M., Beck, D. M., and Ro, T. (2009). To see or not to see: Prestimulus alpha phase predicts visual awareness. Journal of Neuroscience, 29, 2725– 2732.
Mazaheri, A., DiQuattro, N. E., Bengson, J., & Geng, J. J. (2011). Pre-stimulus activity predicts the winner of top-down vs. bottom-up attentional selection. PLoS One, 6(2), e16243-e16243. Mellenbergh, G., J. (2011). A Conceptual Introduction to Psychometrics. Eleven International
Publishing: Den Haag.
Mittner, M., Boekel, W., Tucker, A. M., Turner, B. M., Heathcote, A., & Forstmann, B. U. (2014). When the brain takes a break: A model-based analysis of mind wandering. The Journal of Neuroscience, 34(49), 16286-16295.
Pfurtscheller, G. (1992). Event-related synchronization (ERS): an electrophysiological correlate of cortical areas at rest. Electroencephalography and clinical
neurophysiology, 83(1), 62-69.
Ratcliff, R., & McKoon, G. (2008). The diffusion decision model: theory and data for two- choice decision tasks. Neural computation, 20(4), 873-922.
Rihs, T. A., Michel, C. M., & Thut, G. (2007). Mechanisms of selective inhibition in visual spatial attention are indexed by α-‐band EEG synchronization. European Journal of
Neuroscience, 25(2), 603-610.
Romei, V., Gross, J., & Thut, G. (2010). On the role of prestimulus alpha rhythms over
occipito-parietal areas in visual input regulation: correlation or causation?. The Journal of neuroscience, 30(25), 8692-8697.
Sherwin, J. S., Muraskin, J., & Sajda, P. (2015). Pre-stimulus functional networks modulate task performance in time-pressured evidence gathering and decision-making. NeuroImage, 111, 513-525.
Smallwood, J. M., Baracaia, S. F., Lowe, M., & Obonsawin, M. (2003). Task unrelated thought whilst encoding information. Consciousness and cognition,12(3), 452-484
Smallwood, J., Beach, E., Schooler, J. W., & Handy, T. C. (2008). Going AWOL in the brain: Mind wandering reduces cortical analysis of external events. Journal of cognitive
Smallwood, J., Fishman, D. J., & Schooler, J. W. (2007). Counting the cost of an absent mind: Mind wandering as an underrecognized influence on educational
performance. Psychonomic Bulletin & Review, 14(2), 230-236.
Smallwood, J., McSpadden, M., & Schooler, J. W. (2008). When attention matters: The curious incident of the wandering mind. Memory & Cognition,36(6), 1144-1150. Smallwood, J., & Schooler, J. W. (2006). The restless mind. Psychological Bulletin, 132,
946–958.
Smith, P. L., & Ratcliff, R. (2004). Psychology and neurobiology of simple decisions. Trends in neurosciences, 27(3), 161-168.
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A. (2002). Bayesian measures of model complexity and fit. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 64(4), 583-639
Thut, G., Nietzel, A., Brandt, S. A., & Pascual-Leone, A. (2006). α-Band electroencephalographic activity over occipital cortex indexes visuospatial attention bias and predicts visual target detection. The Journal of Neuroscience, 26(37), 9494-9502.
Turner, B. M., Sederberg, P. B., Brown, S. D., and Steyvers, M. (2013). A method for efficiently sampling from distributions with correlated dimensions. Psychological Methods. 18. 368-384. Van Maanen, L. & Van Rijn, H. (submitted). The preparatory pupillary response predicts
trial-by-trial variation in motor-related components of response time.
Vandekerckhove, J., & Tuerlinckx, F. (2007). Fitting the Ratcliff diffusion model to experimental data. Psychonomic bulletin & review, 14(6), 1011-1026.
Wagenmakers, E. J. (2009). Methodological and empirical developments for the Ratcliff diffusion model of response times and accuracy. European Journal of Cognitive Psychology, 21(5), 641-671.
