Do participants show the same shift between conditions?
The table compares each network’s average expression in two conditions. Directional agreement asks how many participants shift in the more common direction, whichever direction that is. The main findings below identify every network for which all selected participants agree; the individual points show the variation behind that consensus.
Explore the figures
Choose conditions A and B, then read across the network table. Click a network row to inspect its participant plots. Average network coefficient retains the orientation of the opposing pattern; contribution magnitude measures strength regardless of orientation. Changing the boundary delay checks how sample assignment affects the comparison.
Do network contributions differ between task and fixation?
Main findings for the selected contrast
Which networks differ, and does the average describe the participants?
| Network | Mean A | Mean B | Mean paired difference | Participant SD of differences | Directional agreement |
|---|
Interpretation and timing limits
The primary contrast pools all stimulation against internal inter-block fixation. Image-category contrasts are exploratory extensions; each category occurs only twice per run. Category-versus-fixation comparisons use all internal gaps as a common reference, not just the immediately adjacent gaps. Initial and final fixation are excluded.
Each block or gap is summarized separately, then equally averaged within condition/run, runs within participant and participants equally. This period-based weighting is explicit: existing Task dynamics summaries use condition-sample means within runs, so small differences can occur if intervals contain different numbers of samples. Nothing is refitted, and no discontinuous blocks are concatenated.
The delay control moves both ends of every block and its intervening gaps by the same 0/4/6 seconds. It changes sample membership, not the expression time series, and does not estimate a hemodynamic response. Fixation gaps contain delayed task responses; neither contrast sign nor a shifted boundary identifies a neural onset. See Task dynamics for measured-boundary trajectories.
All 15 participants, including motion/coverage-flagged recordings, remain included. There are no p-values, multiplicity corrections or causal conclusions here. Robustness to motion and independent validation remain separate questions.
Exact selection and provenance
More about the measures and interpretation
All comparisons are A − B. Directional agreement is the fraction of participants with the more common sign of A − B, whichever direction that is. Ties remain in the denominator. Agreement describes consistency of direction, not effect size or statistical significance. These are paired descriptive summaries, not significance tests. Full-cohort averaging gives each participant equal weight after averaging selected runs. Both conditions must have the same contributing runs within each participant; missing comparators are not silently filled with zero.
Click a table row to select its network. Network coefficient units differ across networks; compare conditions within a network rather than ranking raw differences between networks. Magnitude is mean |expression| × the norm of the network’s unique off-diagonal coefficients. Positive and negative coefficients can cancel in an average while retaining substantial contribution magnitude.
Background: opposing patterns, direction and strength
The cohort model stays fixed. Its continuously projected expressions are summarized during measured task blocks, internal fixation gaps, or particular image categories. This compares condition-dependent expression of learned dynamics; it does not fit short blocks separately or establish that the model’s transition law changes between conditions.
Each learned network can contain two opposing spatial poles: sets of relationships among brain regions with opposite weights. When those poles emphasize different functional systems, their opposition provides a way to describe how those systems participate in the learned dynamics. A visual–default-mode opposition, for example, is an anatomical interpretation to establish from the network map, not a property assumed for every network.
A single time-varying network coefficient controls the joint expression of both poles. A positive coefficient preserves their learned signs; a negative coefficient reverses both together. The average network coefficient therefore describes the prevailing orientation of this opposing pattern during a condition. A task–fixation difference asks whether that orientation shifts consistently across participants. It does not show that one pole changes independently or that the spatial network itself reorganizes.
How to interpret direction and strength
For network k, its contribution is xₖ(t)Wₖ: a time-varying coefficient multiplied by a fixed matrix of ROI-pair weights. “Positive pole” and “negative pole” refer to weights in that stored matrix. A negative coefficient makes originally negative weights contribute positively and originally positive weights contribute negatively. Neither pole is intrinsically the active one. These are model contributions to ROI second moments, not direct measurements of regional activation, inhibition or BOLD anticorrelation.
Average network coefficient retains direction, but strong expressions in opposite directions can cancel. Mean contribution magnitude measures strength regardless of direction: mean |xₖ(t)| × the norm of the unique off-diagonal weights. Equal magnitudes can accompany opposite average coefficients. Neither measure is a probability; magnitude is not explained variance. Both are summaries of continuously projected model coordinates, calculated after fitting.
The coefficient and matrix must be interpreted together. Reversing both signs leaves their product unchanged, so a functional description should name the systems at the two poles and their condition-related relationship rather than treating a positive coefficient as inherently greater brain activity.