Behavioural performance
A mixed analysis of variance (ANOVA) compared performance (accuracy and RT) on the cognitive task used during MDES. This analysis had two within-participant factors each with two levels: task (0-back/1-back) and session (first/second). Experimental group (goals/TV programmes) was included as a between participants variable. With respect to accuracy, the analysis revealed a main effect of task [F (1, 97) = 20.1, p < .001] with lower accuracy in the 1-back (mean = .92, SE = .01) than in the 0-back (mean = .97, SE = .01). Analysis of RT revealed a main effect of task [F (1, 97) = 110.9, p < .001] with slower responses in the 1-back (mean 1.08 s, SE = .02) than in the 0-back (mean = .98 s, SE = .02) and an effect of session [F (1,97) = 5.6, p < .05] with longer response times in the first session (mean = 1.01 s, SE = .02) than in the second (mean = 0.98 s, SE = .02). It also revealed a task by session by group interaction [F (1, 97) = 8.5, p < .01]. We calculated the change in RT from the first to the second session in each task and compared these scores across experimental groups using a MANOVA. This yielded an effect of experimental group on RT in the 0-back task [F (1, 97) = 5.42, p < .05] but no effect on RT in the 1-back task (p < .1). In the TV programmes condition, response times in the 0-back task were shorter in the second session [mean = .91 s (SE = .02)] than in the first [mean = .97 (SE = .03), t (47) = 2.36, p < .05].
We used principal components analysis (PCA) with varimax rotation in SPSS to decompose the dimensionality of both the experience-sampling data and the written descriptions.
Written descriptions
Table 2 presents the average ratings for the transcripts in each session. We calculated the change between initial and subsequent ratings for each participant for each dimension, creating a single vector that describes how different aspects of the participants’ descriptions changed over the course of the experiment. Applying PCA to this vector yielded four components that are presented as a heat map in the upper panel of Fig. 2. Factor solutions were selected with an eigenvalue >1. Component 1 describes changes in the level of concrete details of the description, accounting for approximately 23 % of the variance and with a positive weighting reflecting increases in concreteness.Footnote 1 Component 2 reflects an increased emphasis on the self versus other people in the descriptions and accounted for 21 % of the overall variance. Component 3 reflects an increased emphasis on negative rather than positive aspects of the descriptions for 18 % of the variance. Component 4 describes a change in the emphasis of the past relative to the future in the descriptions and accounts for 13 % of the overall variance. Subsequently, we projected these data back into subject space, allowing us to describe each individual in terms of how much they changed their descriptions along the dimensions that the decomposition procedure identified.
Multidimensional experience sampling
Table 3 presents the data from the experience-sampling probes separated by session. To ensure comparability with the prior investigation, we decomposed these data at the trial level, concatenating the data from each probing episode for each individual into a single matrix and applying PCA with varimax rotation. A decomposition with three orthogonal factors yielded solutions that are broadly consistent with prior investigations (Ruby et al. 2013; Engert et al. 2014) (see lower left-hand panel of Fig. 2): (a) Future focused thoughts with a high weighting on thoughts about the self in the future, accounting for 36 % of the observed variance, (b) Past focused thoughts with a high weighting on thoughts about self and others in the past, accounting for 18 % of the variance and (c) Task-related thoughts with a high weighting indicating positive task focused experiences and a low weighting indicating negative off-task experiences, accounting for 18 % of the variance.
Table 3 Mean (SD) for the multidimensional experience sampling (MDES)
Our decomposition of the form of thoughts yielded three components, which can be seen in the left panel of Fig. 2: (a) The modality of the thoughts (images or words) with a high weighting reflecting thoughts that were described as words, accounting for 37 % of the variance, (b) The level of intrusiveness in the thoughts with a high weighting indicating more intrusive thoughts, accounting for 25 % of the variance and (c) The level of detail in the thoughts, with a low weighting indicating a higher level of detail, accounting for 25 % of the overall variance. The loadings for each component describing the content and form of thought are presented as heat maps in the lower right-hand panel of Fig. 2, and these components are consistent to a pattern revealed through similar decomposition process in two different data sets (Smallwood et al. 2016). For the purpose of analysis, the six components describing the content and form of spontaneous thoughts were averaged across the two MDES sessions (i.e. recorded before and after the initial description task, respectively) for each participant. Finally, a change score was calculated by subtracting the component in the first MDES session from the component in the second MDES session. These data correspond to changes from baseline in particular aspects of spontaneous thought following the description task.
Relation between the change in descriptions and spontaneous thought during the mind-wandering task
To understand the relationship between the changes in the descriptions of the goals and the spontaneous thoughts that occurred during the mind-wandering task, we conducted a multivariate analysis of variance (MANOVA). In this analysis, changes in the six components describing the content and form of thoughts were dependent variables, and the four components describing the change in the descriptions were entered as continuous independent variables. We also included the condition (goal/TV programmes) as a categorical factor. We modelled the main effects of each factor and allowed each component to interact with the experimental condition. This MANOVA indicated a condition by concrete detail interaction [F (6, 84) = 2.7, p < 0.05] that was due to an association between thoughts about the self and future in the goal condition (r = 0.42, p < 0.005, 95 % CI lower = .10, upper = .65), but not the TV program condition (r = −0.16, p > 0.2, 95 % CI lower = −.47, upper = .16). Thus, when an opportunity for spontaneous thought led to an increase in the concreteness of personal goals, this change was related to increases in thoughts regarding the future across this period. Table 4 presents the zero-order correlations between the change in each type of spontaneous thought and each element of the written descriptions, and Fig. 3 illustrates the significant relationship between increasingly concrete descriptions of goals with future thinking. To assess whether this pattern replicated across the samples of participants who did and did not participate in the initial resting-state study, we calculated the correlation between increases in spontaneous future thinking and the changes in concrete goals separately in the behavioural and fMRI samples. This identified a similar pattern in both groups (behaviour only, r = 0.38, p = 0.06, fMRI, r = 0.44, p < 0.05).
Table 4 Zero-order correlations between the change in self-generated thought and in descriptions of goals and TV programmes
Finally, we explored whether the pattern linking change in the concreteness of goals was related to low levels of future thought in the baseline session, in the post-description session or both. We conducted two separate linear regressions with the change in concreteness as the dependent variable and the loading on future thinking in sessions 1 and 2 as independent variables. This model was run separately on the participants who participated in the goal and in the TV programme conditions. This revealed a significant model for the goals condition [F (2, 48) = 5.02, p < .01] but not in the TV programmes condition [F (2, 45) = .761, p = .47]. In the goal condition, the largest changes in the concreteness were associated with lower levels of future thinking in session 1 and higher levels in session 2 (see Table 5). This pattern indicates that increasingly concrete goals were produced by individuals who produced low levels of future-related thoughts but subsequently increased these reports in the second session.
Table 5 Results of linear regressions exploring the relationship between the change in concreteness of descriptions and future thinking in sessions 1 and 2
Relation to functional connectivity
Having identified that spontaneous thoughts about the future increase when an opportunity to engage in spontaneous thought leads to changes in the concreteness of personal goals, we next assessed whether these changes in concreteness were predicted by individual differences in the pattern of hippocampal functional connectivity assessed during the resting state. We first verified that the overall patterns of hippocampal connectivity were consistent across the two groups. The upper panel of Fig. 4 presents unthresholded connectivity maps for the left hippocampus seed separated into the participants who were subsequently allocated to the goals or the TV programme condition, and it is apparent that there were no gross differences in functional connectivity across the participants assigned to the two conditions. We compared these groups using FMRIB’s Local Analysis of Mixed Effects (FLAME) with cluster-forming threshold of Z = 2.3 and corrected for multiple comparisons with an FWE level of p < .05. This analysis found no differences in the group maps and revealed clusters of significant functional connectivity within subcortical regions, in medial pre-frontal and posterior cingulate cortex, lateral regions of the temporal lobe and elements of anterior lateral occipital cortex (see Fig. 4, lower panel). The significant clusters are presented in Table 6.
Table 6 Spatial cluster identified in the group-level analysis of hippocampal connectivity
Having demonstrated that the experimental groups did not differ in their general patterns of hippocampal connectivity, we conducted a group-level regression examining how these connectivity patterns predicted changes in concreteness that were common to the goals and TV condition. In this model, the independent variable was the components describing the change in concreteness. We used a cluster-forming threshold of Z = 2.3, and the subsequent spatial maps were corrected for multiple comparisons with an FWE level of p < .05. We first explored individual differences in functional connectivity that varied with increases in the concreteness of descriptions in both conditions (i.e. an association between changes in concreteness and functional connectivity that was common to both the goals and the TV programme conditions). This identified a cluster in the ventromedial pre-frontal cortex, centred on MNI coordinates: 8, 35, −14, which was 5880 mm3 in volume (see Fig. 5, upper panel). Next, we examined individual differences in connectivity that were associated with greater changes in concreteness for descriptions of goals rather than TV programmes (i.e. specific patterns of connectivity that were greater for participants who increased the concreteness in their descriptions of goals more than for TV programmes). This contrast identified a single cluster in the dorsal medial pre-frontal cortex extending into the pre-SMA: this was centred on MNI coordinates 5, 3, 64, with an approximate volume of 4880 mm3 (see Fig. 5, upper panel).Footnote 2
To facilitate the interpretation of these patterns of connectivity, we extracted the β values from both spatial clusters for each participant and plotted them against their change in concreteness, separately for each condition. It can be seen in Fig. 5 that while increases in concreteness in both conditions were associated with greater coupling between the hippocampus and the cluster in the ventromedial pre-frontal cortex, the coupling with the pre-SMA cluster was associated with more concrete descriptions in the goal condition (r = .60, p < .001 95 % CI lower = .315, upper = .799] but less concrete descriptions of television programs (r = −.68, p < .001 95 % CI lower = −.88, upper = −.22). Given the relatively small sample size, we performed a spilt-half reliability check in both cases this revealed acceptable levels of consistency (Spearman–Brown: goals = .75, TV programs = .80). Thus, our data show that coupling between the hippocampus and the ventromedial pre-frontal cortex is predictive of increased concreteness for both descriptions of goals and TV programmes, while an extensive cluster in the dorsomedial frontal cortex was uniquely associated with the tendency to increase the concreteness of personal goals.
Finally, to better understand the psychological significance of this pattern of functional coupling, we conducted a post hoc analysis targeted at determining whether the pattern of functional coupling was related to less concrete descriptions of goals in the initial session, or higher levels in the second session. We conducted a multiple regression in which the dependent variable was the hippocampal–pre-SMA coupling and the independent variables were the raw ratings of abstract and concreteness in the descriptions gained in the first and second sessions. We conducted these separately for the goal and TV programs conditions. In both cases, this produced a model which was significant [goals, F (4, 22) = 4.41, p < .01; TV program, F (4, 18) = 5.2, p < .01, see Table 7]. In the case of the description of goals, we found a reduction in connectivity between the hippocampus and the pre-SMA associated with more abstract goals at in session 1 and a reduction in abstraction in session 2. In the TV programme condition, we observed that the reverse connectivity between the hippocampus and the pre-SMA was associated with less abstraction in session 1 and more abstraction in session 2.
Table 7 Relation between hippocampal–pre-SMA coupling and ratings of descriptions as concrete or abstract in each session