Overview of data collection.

(Left) We selected 14 tasks to capture states linked to a range of cognitive domains. (Right) To represent the brain activity and internal experience associated with each task, we obtained unthresholded task maps from existing fMRI data and administered multidimensional experience-sampling (mDES) thought probes while participants completed each task.

Component extraction and distribution of tasks in the ‘thought-space’.

To represent participants’ reported thoughts in fewer dimensions, we decomposed the set of mDES probes using PCA. (Top Left) Based on examination of the Scree plot and parallel analysis we extracted 4 components. (Top Right) These components comprise the dimensions of a continuous ‘thought-space’ where we can compare reported thought across contexts in a data-driven manner. This 3D scatterplot depicts the average scores for each task on the first 3 Varimax-rotated components, “2B-Face/2B-Scene” = 2-back task with faces/scenes, “0B/1B” = 0-back/1-back task, “EasyMath/HardMath” = easy and hard math tasks, “Read” = reading task, “Memory” = memory task, “Friend” = social appraisal task, “You” = self-appraisal task, “FingerTap” = finger-tapping task, “GoNoGo” = go/no-go task, “Documentary/SciFi” = movie-viewing of documentary/sci-fi clips. (Bottom) We named the Varimax-rotated components according to their loading structure, which we visualize using Word Clouds, where word size represents loading strength (larger = stronger loading) and color represents directionality (red = positive, blue = negative).

Distribution of tasks on each component.

(Top) We named the Varimax-rotated components according to their loading structure, which we visualize using Word Clouds, word size represents loading strength (larger = stronger loading) and colour represents directionality (white = positive, black = negative). (Bottom) The distribution of tasks on each component based on average component score, error bars represent mean±SEM, “2B-Face/2B-Scene” = 2-back task with faces/scenes, “0B/1B” = 0-back/1-back task, “EasyMath/HardMath” = easy and hard math tasks, “Read” = reading task, “Memory” = memory task, “Friend” = social appraisal task, “ You” = self-appraisal task, “FingerTap” = finger-tapping task, “GoNoGo” = go/no-go task, “Documentary/SciFi” = movie-viewing of documentary/sci-fi clips.

Connectivity gradients and distribution of tasks in “brain-space”.

(Left) The five principal connectivity gradients. Regions in blue depict brain areas linked to the low end of the gradient while red regions fall to the high end. (Right) Adapted from 30; network loadings of each of the five gradients on 7 canonical resting-state networks from 34.

Gradient coordinates of each task

. To represent task-related whole-brain activation in fewer dimensions we projected unthresholded group-level task maps onto the first five cortical gradients generated from resting-state fMRI data from the Human Connectome Project. See here the distribut ion of the 14 tasks on each of the 5 gradients. “Uni.-Hetero.” = “Unimodal – Heteromodal”; “Mot.-Vis.” = “Motor – Visual”; “DMN-Cont.” = “DMN – Control”; “DAN-S.Att.” = “DAN – Sensory Attention”; “S.Mot.-Lim.VAN.” = “Sensory-Motor – Limbic-VAN”.

ICCs for each thought-pattern nested within-subject, within-task, or cross-factored subject:task.

ICCs are presented with bootstrapped 95% confidence intervals.

The influence of context versus subject on reported thought per task.

(Top) Word clouds depict the distribution of mean-centred ICCs across tasks, word size indicates distance from average (larger = further from average), word color represents direction (white = above average, black = below average). (Bottom) While ICCs varied across tasks depending on the thought pattern in question, more challenging tasks achieved consistently higher stability across components, whereas stability on some tasks depended on the component in question. The bar plots here center around 0.50, representing the threshold beyond which subject-level dependency is apparent (i.e., subjects respond more similarly to themselves than others).

Between- versus within-individual variation across tasks.

(Top Left) Adapted from 36; Xu’s Theoretical Variation Field Map visualizes how the ICC increases as individuals’ respond more consistently within themselves (i.e., lower 𝜎w2) and more uniquely from others (i.e., higher 𝜎b2), such as for the point in the top left. On the other hand, a low ICC could result from either: 1) generally unstable responses (e.g., lower right point), or 2) individuals responding too similarly to each other, becoming indistinguishable (e.g., lower left point). (Right) Variation field maps for each task on each thought-pattern. While passive, unengaging tasks typically demonstrated the lowest ICC values across components, field maps reveal that in some cases this was due to high similarity between subjects (e.g., the memory task) while in others it was a result of poor stabilit y overall (e.g., the reading task).

Relationship between within-subject stability and component score.

To examine how within-subject stability on a given component related to its appearance, we regressed ICC per task against task-averaged component score in 1000 bootstrap resamples. Bootstrapped coefficients indicated that during tasks with high levels of task-focus, subjects’ reported thought tended to be more consistent between probes. On the other hand, tasks high in Intrusive Distraction or Sensory Engagement were associated with probe responses becoming less consistent.

Association between stability in reported thought and cortical gradients.

To examine how stability in individuals’ reported thought related to brain activation during each task, we computed bootstrapped regression coefficients representing the relationship between each gradient and stability in reported thought while holding other gradients at a constant, * = 𝐶𝐼𝑎𝑑𝑗. exclude 0. “Uni.-Hetero.” = “Unimodal – Heteromodal”; “Mot.-Vis.” = “Motor – Visual”; “DMN-Cont.” = “DMN – Control”; “DAN-S.Att.” = “DAN – Sensory Attention”; “S.Mot.-Lim.VAN” = “Sensory-Motor – Limbic-VAN”.

Regions related to stable Deliberate Task-Focus significantly overlap with the Multiple Demand Network (MDN).

Participants’ reports of Deliberate Task-Focus were linked to stable task-related engagement and activation of brain regions implicated in executive functioning. As such, we sought to test the validity of our generated stability map by comparing it to the MDN, an established network with functional associations similar to Deliberate Task-Focus. Spin-testing found that the stability map for the component indeed significantly overlapped with the MDN, whereas the same effect was not present for any other thought pattern. This not only provides novel evidence supporting the association between the MDN and engagement with complex tasks but provides further nuance suggesting that as MDN activation grows, individual differences in Deliberate Task-Focus become more apparent.

Multidimensional Experience Sampling (mDES) Items.