Effects of Emotional Valence of Mind Wandering on Sustained Attention Performance

Abstract

The construct of mind wandering has notoriously been characterized as heterogenous which may mean that not all types of mind wandering produce the same pattern of results. One operationalization of mind wandering, task-unrelated thoughts (TUTs), can also itself vary in many dimensions, including the emotional valence of TUTs. The current study summarizes several years of work examining the impact that the emotional valence of TUTs has on different aspects of sustained attention. Participants in several studies reported whether their TUTs were negative, neutral, or positive in emotional valence during a sustained attention-to-response task (SART). The first major focus was a meta-analysis where we examined correlations between each TUT valence and SART performance measures. For the second major focus, we tested how different TUT valences changed over the course of the task. The results suggest that negative TUTs typically show stronger associations with SART performance measures, although all TUT valences have numerically similar correlations. Regarding time-on-task effects, across the studies, there was consistent evidence for a linear increase in negative TUTs across blocks. Evidence for this linear increase was not consistent for neutral and positive TUTs. The results of the current study suggest that the relationships between TUTs and performance, and their likelihood of occurring during a task, are not necessarily the same for every type of TUT. These results highlight the importance of continuing to investigate different types of TUTs and different forms of mind wandering, in general, to better understand how this phenomenon occurs.

Mind wandering describes the phenomenon where our attention shifts from an ongoing task to thoughts unrelated to what we are currently engaged in (for a review see Smallwood & Schooler, 2006). Recent work has argued that mind wandering is a complex and multifaceted construct that ranges on several dimensions (Seli et al., 2018a, 2018b). If this is the case, not every form of mind wandering may produce the same effects or relationships with other variables. As such, examining outcomes of interest through different forms, or contents, of mind wandering might elucidate specific patterns of results that might otherwise be missed or mislabeled. In the current study, we reexamined several studies from our research group to assess how one aspect of mind wandering we have repeatedly examined, the emotional valence of mind wandering, related to different aspects of sustained attention performance.

Mind Wandering Impairs Task Performance

It is well established that mind wandering is associated with poorer task performance in several different domains. Specifically, mind wandering appears to affect academic performance (e.g., Wammes et al., 2016), driving performance (e.g., Baldwin et al., 2017; Yanko & Spalek, 2014), reading comprehension (e.g., Feng et al., 2013; McVay & Kane, 2012b; Unsworth & McMillan, 2013), memory tasks (Thomson et al., 2014), working memory tasks (e.g., Banks & Boals, 2017; Krimsky et al., 2017; Unsworth & Robison, 2016), and basic attention tasks (e.g., McVay & Kane, 2012a; Robison et al., 2020). Indeed, a recent meta-analysis found empirical support for this observation noting a modest relationship between mind wandering and task performance (ρ = −.24; Randall et al., 2014). Thus, when examining the global form of mind wandering, there appears to be very convincing evidence of its negative impact. However, many, if not all, of these studies have simply examined an overall rate of mind wandering without considering how different contents of mind wandering might differentially be related to task performance (e.g., Welhaf et al., 2020).

One specific dimension of mind wandering that we have focused on in our previous work is emotional valence. Specifically, we have asked participants to categorize their thoughts as being either negative, positive, or neutral in valence. Our previous work has pointed to a strong pattern that negative task-unrelated thoughts (TUTs) are the most detrimental of the three types of thoughts (e.g., Banks et al., 2016). Specifically, individuals who report more negatively valenced mind wandering during different tasks tend to perform worse on those tasks (see Banks & Welhaf, 2022; Banks et al., 2016; Banks, Tartar, & Tamayo, 2015; Banks, Welhaf, & Srour, 2015). However, rates of neutral and positive TUTs show inconsistent patterns in predicting task performance suggesting their impairment of cognition might depend on other factors like task demand (Banks et al., 2016).

Why do negative TUTs appear to have a greater impact on task performance than other emotional valences of mind wandering? One possibility is that negatively valenced information has preferential access to conscious awareness or more strongly captures attentional resources. Several theories of mind wandering describe how individuals’ current concerns either fully or at least partially can trigger instances of mind wandering (for reviews see Kane & McVay, 2012; Randall et al., 2014; Smallwood, 2013). For example, according to the current concerns hypothesis (Klinger, 1999; Klinger et al., 1973), mind wandering is said to occur because individuals’ attention is drawn to their most salient or pressing concerns that are deemed to be more important than the current task. It is not unrealistic that these concerns have some (negative) emotional tie to them and therefore would predict that as individuals experience more negative TUTs, their performance declines. Extending this, the Failure × Concerns Theory (McVay & Kane, 2009, 2010, 2012a) proposes that part of the reason that mind wandering occurs is that individuals fail to inhibit distracting information in the form of concerns cued by the environment or their own personal stream of conscious (stemming from Klinger, 1971, 1999). However, individuals with greater executive control (i.e., higher working memory capacity [WMC] and attentional control abilities) can maintain task goals and suppress TUTs before they enter conscious awareness. Indeed, we have found that individuals with higher executive control do tend to report fewer negative TUTs during a sustained attention-to-response task (SART) (Banks & Welhaf, 2022). Another prominent view, the decoupling hypothesis (Smallwood & Schooler, 2006) proposes that mind wandering is not simply a failure of attention control, but rather an active and ongoing shift of attention from the external task to an individuals’ internal stream of thought. Here, emotionally valenced TUTs might draw more attention inward (perhaps because they are important to process and address for current concerns), which results in a decoupling of attention from the primary task and leads to poorer performance.

Why Does Mind Wandering Increase Over Time?

The ability to maintain attention over periods of time plays an important role in the completion of everyday tasks. The vigilance decrement refers to the empirical finding that task performance tends to worsen over time (for reviews see Esterman & Rothlein, 2019; Fortenbaugh et al., 2017; Thomson et al., 2015). There are several competing theories for why the vigilance decrement occurs. Proponents of the “resource depletion” hypothesis (e.g., Warm et al., 2008) argue that attention is a limited resource that is exhausted over the course of a task. This perspective, however, does not specify the role of mind wandering in explaining, at least partially, why performance declines over time. Others have proposed that mind wandering increases with time on task because sustained attention tasks are inherently boring and understimulating and suggest that this vigilance decrement is a byproduct of “mindlessness” (e.g., Manly et al., 1999). In addition to sustained attention tasks being understimulating and boring, individuals might also recognize that it is more worthwhile to switch from performing the primary task to mind wandering to satisfy opportunity costs (Kurzban et al., 2013).

Thomson et al. (2015) proposed a hybrid model of sustained attention that integrates pieces from both the resource and executive control models of mind wandering. The “resource-control” model suggests that executive control processes are needed to maintain focus on the current task and inhibit TUTs from occurring. However, as the implementation of executive control starts to become more variable over time, TUTs are more likely to occur in the later part of the task. Put differently, because executive control processes are “fresh” at the start of the task, participants can both perform the task and also inhibit potential TUTs from interfering with primary task performance. However, as time goes on, fewer executive control processes are devoted to the external task and instead are shifted toward mind wandering.

Several studies have shown that mind wandering increases with time on task (e.g., Cunningham et al., 2000; Krimsky et al., 2017, McVay & Kane, 2012a; Thomson et al., 2014). As with the majority of the mind wandering literature, studies typically investigate how overall rates of TUTs (or other forms of mind wandering like depth of mind wandering; e.g., Brosowsky et al., 2023) increase over time, outside of a few studies. For example, Martínez-Pérez et al. (2021) investigated how intentional and unintentional TUTs increased over two sustained attention tasks, a SART and a psychomotor vigilance task (PVT; Lim & Dinges, 2008). They found that both forms increased but this depended on the task. Specifically, intentional TUTs continuously increase across the PVT and unintentional TUTs continuously increase in the SART. Martínez-Pérez et al. argued that this differential effect reflects that changes in intentional mind wandering might occur in tasks where mind wandering has little impact on performance. On the other hand, the changes in unintentional mind wandering (which reflect, at least partially, failures of executive control) are more likely to occur in tasks that place a higher demand on executive processes.

Likewise, Unsworth and Robison (2016) examined how different attentional states changed over the course of a PVT. Unsworth and Robison reported that reports of mind wandering, inattentiveness, and external distraction all appeared to increase across blocks consistent with the general finding that participants become less focused as they spend more time on task (McVay & Kane, 2012a; Thomson et al., 2014). This general increase in mind wandering, inattentiveness, and external distraction appears to be consistent with the “resource-control” model, such that the increased variability in the implementation of executive control results in increases in all three of these types of reports.

The resource-control view of mind wandering and sustained attention (Thomson et al., 2015) presents an interesting perspective on making predictions about what types of TUTs might be more likely to increase over time, specifically regarding the emotional valence of TUTs. For example, negative TUTs might be lower at the beginning of the task because executive control processes are still maximal and able to control focus on the task while not allowing attention to be captured by the negative emotional content of a negative TUT. However, as the time on task continues, and executive resources become detached from the external task, negative TUTs might have an easier time entering conscious awareness and thus become more frequently reported in these later blocks. Neutral and positive TUTs, on the other hand, might not need to be actively inhibited throughout the task, or may be more easily inhibited, and instead might occur at similar rates across blocks rather than increasing. Thus, a novel question addressed by the current study is how different types of TUTs, specifically broken down by emotional valence, change in frequency over the course of a task.

Overview and Analytic Approach of the Present Study

Given the rising interest in differentiating forms or content in TUTs, we sought to examine whether different emotional valences of TUTs led to similar effects on sustained attention performance. We examined this in two ways. First, we examined how TUTs correlated with different measures of SART performance (e.g., reaction time (RT) variability, accuracy) based on each emotional valence dimension and synthesized these results using meta-analysis. Second, we examined how the emotional valence of mind wandering changes over the course of a task. While it has been previously shown that mind wandering, in general, increases with time on task, we asked if this was specific to any emotional valence. Thus, the approach here aims to summarize our work and specify if these forms of TUTs result in similar effects across studies.

To do so, we analyzed several published, submitted, or previously unanalyzed datasets collected by our research group over the last several years (including samples from Nova Southeastern University, Western Carolina University, and Prolific). These studies include Banks and Welhaf (2022), Goller et al. (2020), Welhaf, Astacio, & Banks (2024), Banks et al. (submitted) and data from one unpublished masters’ thesis completed at Nova Southeastern University, Holtzman (2022). In each of these studies, participants completed versions of the SART with thought probes to assess the emotional valence of mind wandering. These studies were chosen as they all included SART assessments with enough blocks and thought probes to reliably capture changes in reporting rates over the course of the task. Other studies that also measured the emotional valence of mind wandering were excluded as they either had too few probes or blocks in the task (e.g., complex span tasks; Banks & Boals, 2017) or had manipulations that significantly affected either overall TUT rates or their relationship with other variables (e.g., Banks et al., 2014, 2019; Banks, Tartar, & Tamayo, 2015; Banks, Welhaf, & Srour, 2015).

In each of the analyzed studies, participants completed semantic SART with intermittent thought probes to assess TUTs. Thought probes generally included options for neutral, positive, or negative TUTs. Slight differences in the probe and SART stimuli presentation are detailed for each study; but, in general, the tasks and probe options are comparable across studies. For each study, we also briefly describe general procedures and other tasks and measures completed by participants. However, our focus is on the SART and thought probes so those will be described in detail for each study and are summarized in Table 1.

Table 1.

Wording of Thought Probe Response Options Across Studies

Study	On task	TRI	Negative	Neutral	Positive
Banks and Welhaf (2022)	Task-related thoughts	Task-related evaluative thoughts	Task-unrelated thoughts, negative content (sadness), task-unrelated thoughts, negative content (anger), task-unrelated thoughts, negative content (other)	Task-unrelated thoughts, neutral content	Task-unrelated thoughts, positive content
Goller et al. (2020)	Task-related thoughts pertaining to the current task	Task-related evaluative thoughts—positive, task-related evaluative thoughts—negative	Task-unrelated thoughts, negative content	Task-unrelated thoughts, neutral content	Task-unrelated thoughts, positive content
Welhaf, Banks, & Bugg (2024)	Task-related	Task performance/evaluation	Off-task negative	Off-task neutral	Off-task positive
Welhaf, Astacio, & Banks (2024)	Task-related thoughts	Task-related evaluative thoughts	Task-unrelated thoughts, negative content	Task-unrelated thoughts, neutral content	Task-unrelated thoughts, positive content
Holtzman (2022)	Task-related thoughts	Task-related evaluative thoughts	Task-unrelated thoughts, negative content	Task-unrelated thoughts, neutral content	Task-unrelated thoughts, positive content
Banks et al. (submitted)	Task-related thoughts	Task-related evaluative thoughts—positive, task-related evaluative thoughts—negative	Task-unrelated thoughts, negative content	Task-unrelated thoughts, neutral content	Task-unrelated thoughts, positive content

Note. TRI = task-related interference.

Our analyses of these data addressed two specific questions, namely, (a) How does each emotional valence of TUT correlate with SART performance; and (b) Are there significant differences in the time-on-task effects for different emotional valence of TUTs? To address our first question, we utilized meta-analysis to summarize the correlations between the emotional valence of mind wandering and standard measures of sustained attention performance across different datasets. This meta-analytic approach allowed us to better estimate the relationship between the variables of interest while accounting for differences in sample sizes and samples. We hypothesized that increased frequency of negative TUTs during a sustained attention task would be associated with decrements in performance (i.e., higher RT variability, poorer task accuracy). However, these associations would not exist for neutral or positive TUTs as they are less likely to capture attention and distract individuals from their ongoing task.

To test the second question, we analyzed TUT rates as a function of probe type and block using linear mixed-effect models to compare the slopes of each type of Emotional TUT across the task.¹ Based on the resource-control perspective on mind wandering (Thomson et al., 2015), we tested for a linear increase in each thought report across blocks. Our main hypothesis was that there would be a significant difference in the slope for negative TUTs compared to neutral and positive TUTs. Specifically, while all forms of TUT might increase over the task, negative TUTs would show the steepest increase indicating participants became more prone to these types of TUTs as the task went on.

Transparency and Openness

Data were analyzed using the R computing environment (R Core Team, 2023) using the lme4 (Bates et al., 2015) and lmerTest (Kuznetsova et al., 2017) packages. Data visualizations were created using ggplot2 (Wickham, 2016). Meta-analyses were conducted using the meta package (Balduzzi et al., 2019). The study and analytic plan were not preregistered. All data and analysis scripts used can be found on the Open Science Framework (https://osf.io/5twup/; Welhaf & Banks, 2024).

Method

Participants

Participants from the current studies were recruited from university subject pools at Nova Southeastern University (Banks et al., submitted; Banks & Welhaf, 2022; Holtzman, 2022; Welhaf, Astacio, & Banks, 2024), Western Carolina University (Goller et al., 2020), and online participants from Prolific academic (Welhaf, Banks, & Bugg, 2024). Note that Welhaf, Banks, and Bugg (2024) examined age-related differences in emotional valence of mind wandering but for the current analyses, we only report data from the younger adult sample. The median sample size across the studies was N = 175 (range = 78–351). All participants in the analyzed studies were between 18 and 35 years old.

Materials

Sustained Attention-to-Response Task

In all studies, participants completed a semantic SART. Participants were instructed to respond (by pressing the spacebar) to category exemplars from one group (i.e., animal “go” trials) and withhold responses to another category (i.e., vegetables/crops “no-go” trials). Depending on the study, participants completed either four or five blocks of the seamless blocks of the SART (trial range: 480–625). The go/no-go proportion in all versions was roughly the same (90% go, 10% no-go). For the meta-analytic analyses, we calculated several common indicators of sustained attention performance including, intraindividual RT standard deviation to correct “go” trials (RTsd), d’ (a signal detection measure of accuracy), and no-go accuracy.

Thought Probes

During the SART, participants were randomly interrupted and asked to report on the content of their immediately preceding thoughts. Participants were instructed to press a key on their keyboard that indicated the content of their thoughts. These thought probes varied slightly across the studies, but for all studies, the off-task response options focused on the emotional valence of participants’ mind wandering. Table 1 lists the thought probe options used for each study. For the meta-analyses, we calculated mean TUT rates for each valence across the SART in each study (i.e., not by block as described below).² Each study also varied slightly in how many probes were presented. Specifically, Banks and Welhaf (2022), Banks et al. (submitted), Holtzman (2022), and Welhaf, Astacio, and Banks (2024) all presented 30 probes to participants, Goller et al. (2020) presented 36 probes during their SART and Welhaf, Banks, and Bugg (2024) presented 24 probes to participants.

For the time-on-task analyses, rates for each emotional valence TUT were calculated as the number of each response divided by the number of probes in each block. Note that Welhaf, Banks, and Bugg (2024) implemented a 10-s time limit on thought probes because this study was completed online, thus if a participant did not respond during that period of time, that probe was set as missing. So, participants in Welhaf, Banks, and Bugg (2024) were still presented with six probes, but the denominator in each block (i.e., the number of valid probes) might have varied.

In two studies, participants were also presented with a second screen to categorize the intentionality of their thoughts (Banks & Welhaf, 2022) or the awareness of their thoughts Welhaf, Astacio, & Banks (2024). Details of these secondary probes are reported in each study. For the current study, we only focus on responses to the initial screen regarding emotional valence.

Procedures

Participants completed the semantic SART with thought probes either in a lab or online setting. In many of the studies, participants also completed several cognitive tasks including measures of WMC and attention control, and questionnaires to assess various constructs. In each study, informed consent was acquired at the start of the study and the SART was given within the first 20 min. Participants in Banks et al. (submitted) completed the SART following an affective writing manipulation. No differences were found between the conditions in overall TUTs (ps > .05) as such we collapsed across conditions for the current analyses of time-on-task effects. Participants in Welhaf, Astacio, and Banks (2024) completed a SART before and after an affective music manipulation (Ns ~ 60 per condition). We only present data from the premanipulation SART for that study.

Results

Data Cleaning

Cleaning of SART performance data was handled using the same procedure for each study (note this might differ from data cleaning procedures either reported or unreported in previous papers). First, to ensure participants engaged with the task sufficiently and understood instructions, we removed anyone who had <70% accuracy on the frequent “go” trials (Welhaf & Kane, 2023a, 2023b).

RT outliers were cleaned using the same process in each study. First, we removed error trials, posterror, and postprobe trials. Next, RTs < 200 ms were removed. Finally, we censored individual RTs that were further than median + 3 × interquartile range (IQR) for each individual subjects’ remaining RTs. From here, we calculated RTsd. Univariate outliers were identified via box-plot for each of the SART performance measures. If a subject was outside the median + 3 × IQR of the sample mean, they were censored to a value equal to the median + 3 × IQR. Table 2 presents the number of cases censored for each SART performance measure, in each study.

Table 2.

Descriptive Statistics for Emotional Valence of TUTs and SART Performance Across Studies

Study	M	SD	Min	Max	Skew	Kurtosis	Reliability	Cases censored
Banks and Welhaf (2022) N = 119
Neutral	0.26	0.18	0.00	0.87	0.82	0.42	.839
Positive	0.08	0.12	0.00	0.87	3.60	18.37	.758
Negative	0.18	0.19	0.00	0.87	1.62	2.83	.895
RTsd	178.03	74.59	36.68	415.47	1.29	2.09	.973	4
d’	0.88	0.87	−0.68	4.18	0.69	0.98	.945	0
No-go accuracy	0.25	0.15	0.04	0.72	1.25	1.49	.919	3
Goller et al. (2020) N = 351
Neutral	0.19	0.17	0.00	0.75	1.08	0.78	.797
Positive	0.07	0.10	0.00	0.72	2.59	8.85	.846
Negative	0.08	0.15	0.00	0.83	2.78	8.30	.887
RTsd	141.22	49.63	42.07	299.33	0.80	0.91	.971	7
d’	2.00	1.15	−0.69	4.70	−0.10	−0.88	.890	0
No-go accuracy	0.53	0.25	0.05	0.97	−0.08	−1.15	.937	0
Welhaf, Banks, and Bugg (2024) N = 167
Neutral	0.10	0.14	0.00	0.87	2.48	6.71	.857
Positive	0.07	0.11	0.00	0.67	2.27	7.51	.710
Negative	0.04	0.08	0.00	0.50	2.84	9.72	.868
RTsd	125.59	38.00	55.67	265.29	0.71	0.54	.926	1
d’	2.36	0.85	−0.22	4.07	−0.30	0.04	.842	0
No-go accuracy	0.56	0.20	0.06	0.92	−0.41	−0.61	.896	0
Welhaf, Astacio, and Banks (2024) N = 233
Neutral	0.32	0.26	0.00	1.00	0.84	0.03	.920
Positive	0.10	0.16	0.00	0.97	2.80	9.02	.906
Negative	0.18	0.24	0.00	1.00	1.81	2.75	.945
RTsd	225.30	129.07	57.48	595.66	1.31	0.96	.979	5
d’	0.89	0.94	−0.90	3.83	0.72	0.39	.823	0
No-go accuracy	0.31	0.20	0.01	0.93	1.16	0.67	.949	0
Holtzman (2022) N = 78
Neutral	0.26	0.16	0.00	0.70	0.61	−0.01	.781
Positive	0.08	0.09	0.00	0.50	1.67	4.26	.767
Negative	0.09	0.10	0.00	0.37	0.87	−0.24	.719
RTsd	140.18	51.96	34.62	260.65	0.40	−0.50	.965	0
d’	1.43	1.01	−0.27	4.19	0.63	−0.13	.913	0
No-go accuracy	0.47	0.20	0.05	0.95	0.30	−0.63	.915	0
Banks et al. (submitted) N = 183
Neutral	0.24	0.18	0.00	0.87	1.15	1.35	.844
Positive	0.09	0.11	0.00	0.70	2.45	7.94	.832
Negative	0.12	0.14	0.00	0.67	1.83	3.11	.845
RTsd	172.56	82.34	46.99	412.85	1.32	1.57	.981	8
d’	0.70	0.75	−0.87	2.43	0.03	−0.66	.913	0
No-go accuracy	0.19	0.08	0.01	0.39	0.10	−0.67	.824	0

Note. Reliability estimates are Spearman–Brown corrected correlations of odd and even blocks. TUT = task-unrelated thought; SART = sustained attention-to-response task; Min = minimum; Max = maximum; RTsd = intraindividual RT standard deviation to correct “go” trials in the SART; RT = reaction time.

Descriptive Statistics for Each Study

Before summarizing the correlations in each study, it is worth noting that rates of each TUT valence were relatively similar across studies (after removing participants who either did not understand the task instructions or did not engage enough with the task). Mean rates for each TUT valence and the SART performance measures are provided in Table 2. Neutral and negative TUTs were consistently reported more frequently than positive TUTs.

Emotional Valence of TUTs-by-SART Performance Meta-Analysis

We conducted a meta-analysis on the correlations between each TUT valence and three common measures of sustained attention derived from the SART: RTsd, d’, and no-go accuracy. Figure 1 displays the results of the meta-analysis between each emotional valence and RTsd across the studies. There are a few notable results of this analysis. First, overall, TUTs were modestly related to RTsd (r = .17 [0.14, 0.21]) consistent with previous research (e.g., Unsworth et al., 2021; Welhaf & Kane, 2023a, 2023b). People who reported being off-task more frequently in the SART also showed more inconsistent responding. When examining the individual relationships for each TUT valence, each TUT valence was significantly related to RTsd. The strongest relationship was with positive TUTs (r = .21 [0.15, 0.27]), followed by negative TUTs (r = .17 [0.11, 0.23]) and then neutral TUTs (r = .14 [0.08, 0.19]). It should be noted that the 95% confidence intervals of these estimates highly overlapped with one another suggesting these correlations are rather similar.

Figure 1. Forest Plot of Correlations Between Emotional Valence of TUTs and Intraindividual Reaction Time Variability.

Note. Gray boxes represent study-level correlations. Blue (dark gray) diamonds represent meta-analytic estimate across studies. TUT = task-unrelated thought; RT = reaction time; COR = correlation; CI = confidence interval.

Figure 2 displays the results of the meta-analysis for d’ and emotional valence of TUTs. Again, consistent with previous research, d’ was weakly correlated with overall TUT rates in the task (r = −.12 [−0.15, −0.08]). All TUT valences were significantly related to d’, with the strongest relationship occurring for negative TUTs (r = −.17 [−0.23, −0.11]), followed by positive TUTs (r = −.12 [−0.18, −0.06]) and neutral TUTs (r = −.05 [−0.12, −0.00]).

Figure 2. Forest Plot of Correlations Between Emotional Valence of TUTs and D’. Gray Boxes Represent Study-Level Correlations.

Note. Blue (dark gray) diamonds represent meta-analytic estimates across studies. TUT = task-unrelated thought; D’ = D-prime; COR = correlation; CI = confidence interval.

Figure 3 shows the results of the final meta-analysis between emotional valence of TUTs and no-go accuracy. TUTs overall were weakly associated with no-go accuracy (r = −.07 [−0.11, −0.04]). Thus, no-go errors, which are proposed to be a deeper form of task disengagement (e.g., Cheyne et al., 2009) do not appear to be strongly related to participants’ subjective experience of mind wandering during a task. This relationship seems to be unique to negative TUTs as analysis of each TUT valence revealed no significant correlation between neutral or positive TUTs and no-go accuracy across the studies.

Figure 3. Forest Plot of Correlations Between Emotional Valence of TUTs and No-Go Accuracy.

Note. Gray boxes represent study-level correlations. Blue (dark gray) diamonds represent meta-analytic estimate across studies. TUT = task-unrelated thought; COR = correlation; CI = confidence interval.

Time-on-Task Effects of Emotional Valence of TUTs

Our next set of analyses examined the within-subject changes in each TUT valence across the task. Table 3 displays the results of the linear mixed-effect models for each study. For each study, we conducted the mixed-effect models twice, once with negative TUTs as the reference group and once with neutral TUTs as the reference group. In doing so, we could examine the differences in slopes across each probe type. Table 3 provides the model output. Negative and neutral TUTs consistently showed time-on-task effects across all the studies (i.e., main effects of block). Positive TUTs, on the other hand, only showed time-on-task effects in two of the studies (Goller et al. and Holtzman). Thus, all forms of emotional mind wandering do not appear to show the same pattern across our studies (see Figure 4 for the block-by-block changes for each study). The interaction effects in each model speak to the differences in slopes among the probe responses. The slope for negative TUTs did not significantly differ from that of neutral TUTs in all but one study Welhaf, Astacio, and Banks (2024). Thus, despite neutral TUTs occurring more frequently (relative to negative TUTs) their within-person changes across the task were largely similar. Relative to positive TUTs, both neutral and negative TUTs significantly increased across the task and largely did so at a different rate.

Table 3.

Summary of Models Across Studies Varying the Reference Probe Type

Study and Model	Effect	b	SE	t	p
Banks and Welhaf (2022)
Negative reference	Intercept	0.052	0.021	2.486	.013
	Block	0.041	0.006	6.359	<.001
	Neutral TUT	0.135	0.030	4.545	<.001
	Positive TUT	0.022	0.030	0.747	.455
	Block × Neutral TUT	−0.017	0.009	−1.917	.055
	Block × Positive TUT	−0.040	0.009	−4.443	<.001
Neutral reference	Intercept	0.188	0.021	8.913	<.001
	Block	0.024	0.006	3.699	<.001
	Negative TUT	−0.135	0.030	−4.545	<.001
	Positive TUT	−0.113	0.030	−3.798	<.001
	Block × Negative TUT	0.017	0.009	1.917	.055
	Block × Positive TUT	−0.023	0.009	−2.525	.012
Goller et al. (2020)
Negative reference	Intercept	−0.001	0.011	−0.126	.899
	Block	0.033	0.004	7.671	<.001
	Neutral TUT	0.118	0.016	7.376	<.001
	Positive TUT	0.044	0.016	2.740	.006
	Block × Neutral TUT	−0.005	0.006	−0.832	.405
	Block × Positive TUT	−0.022	0.006	−3.709	<.001
Neutral reference	Intercept	0.116	0.011	10.306	<.001
	Block	0.028	0.004	6.549	<.001
	Neutral TUT	−0.118	0.016	−7.376	<.001
	Positive TUT	−0.074	0.016	−4.636	<.001
	Block × Negative TUT	0.005	0.006	0.832	.405
	Block × Positive TUT	−0.017	0.006	−2.877	.004
Welhaf, Astacio, and Banks (2024)
Negative reference	Intercept	0.061	0.018	3.337	<.001
	Block	0.038	0.006	6.754	<.001
	Neutral TUT	0.255	0.026	9.831	<.001
	Positive TUT	0.014	0.026	0.520	.603
	Block × Neutral TUT	−0.035	0.008	−4.516	<.001
	Block × Positive TUT	−0.029	0.008	−3.747	<.001
Neutral reference	Intercept	0.317	0.018	17.241	<.001
	Block	0.003	0.006	0.476	.634
	Negative TUT	−0.255	0.026	−9.831	<.001
	Positive TUT	−0.241	0.026	−9.311	<.001
	Block × Negative TUT	0.035	0.008	4.516	<.001
	Block × Positive TUT	0.001	0.008	0.769	.442
Welhaf, Banks, and Bugg (2024)
Negative reference	Intercept	0.022	0.012	1.706	.088
	Block	0.018	0.005	3.522	<.001
	Neutral TUT	0.048	0.018	2.659	.008
	Positive TUT	0.007	0.018	0.400	.689
	Block × Neutral TUT	−0.007	0.007	−1.102	.271
	Block × Positive TUT	−0.013	0.007	−2.027	.043
Neutral reference	Intercept	0.070	0.013	5.466	<.001
	Block	0.010	0.005	2.061	.039
	Neutral TUT	−0.048	0.018	−2.659	.008
	Positive TUT	−0.041	0.018	−2.259	.024
	Block × Negative TUT	0.007	0.007	1.102	.271
	Block × Positive TUT	−0.006	0.007	−0.925	.355
Banks et al. (submitted)
Negative reference	Intercept	0.035	0.016	2.225	.024
	Block	0.027	0.005	5.478	<.001
	Neutral TUT	0.132	0.022	5.944	<.001
	Positive TUT	0.044	0.022	1.980	.048
	Block × Neutral TUT	−0.003	0.007	−0.517	.605
	Block × Positive TUT	−0.024	0.007	−3.657	<.001
Neutral reference	Intercept	0.167	0.016	10.661	<.001
	Block	0.023	0.005	4.768	<.001
	Neutral TUT	−0.132	0.022	−5.944	<.001
	Positive TUT	−0.088	0.022	−3.964	<.001
	Block × Negative TUT	0.003	0.007	0.517	.605
	Block × Positive TUT	−0.021	0.007	−3.140	.002
Holtzman (2022)
Negative reference	Intercept	−0.011	0.021	−0.520	.603
	Block	0.035	0.006	5.376	<.001
	Neutral TUT	0.171	0.029	5.956	<.001
	Positive TUT	0.040	0.029	1.382	.167
	Block × Neutral TUT	−0.001	0.009	−0.146	.884
	Block × Positive TUT	−0.019	0.009	−2.146	.032
Neutral reference	Intercept	0.162	0.021	7.903	<.001
	Block	0.033	0.006	5.176	<.001
	Neutral TUT	−0.173	0.029	−5.956	<.001
	Positive TUT	−0.132	0.029	−4.574	<.001
	Block × Negative TUT	0.001	0.009	0.146	.884
	Block × Positive TUT	−0.018	0.009	−2.000	.046

Note. TUT = task-unrelated thought.

Figure 4. Time-on-Task Effects for Each Study by TUT Valence.

Note. Individual dots represent the block mean and error bars are ± 1 SE. TUT = task-unrelated thought.

General Discussion

The current study examined the role and experience of emotional valence of mind wandering in the context of sustained attention. We sought to address two main questions. First, what emotional valence(s) of mind wandering are associated with sustained attention performance? Using meta-analysis, we examined the correlations between negative, neutral, and positive TUTs with three frequently used measures of sustained attention performance in the SART. We found that all negative TUTs were consistently associated with poorer SART performance. Correlations with neutral and positive TUTs were less consistent. Second, we also examined how these reports of different emotional valences of TUTs changed within subjects over the course of a task. While we found general increases in each TUT valence with time on task, the most evident, and consistent, increases were specific to negative TUTs. Thus, the collective results suggest that not all forms of mind wandering behave similarly.

The meta-analytic results of our previous work highlight that the emotional valence of mind wandering might be an important moderator in the relationship with task performance. While negative TUTs were consistently, albeit weakly, related to worse performance (e.g., more variable responding and worse accuracy), there appears to be some evidence that positive TUTs are also associated with poor sustained attention. Thus, emotional thoughts, whether negative or positive, might both capture attention enough to impair task performance (Strauss & Allen, 2009). As we have previously hypothesized, negative TUTs likely impact performance because they are strongly captured in attention leading to distraction. Given the preferential processing of negative emotional stimuli, it is not surprising that these types of thoughts capture our attention.

Negative TUTs were more strongly associated with accuracy measures in the SARTs (d’ and no-go accuracy). It is possible that the salient “Oops!” response triggered by erroneously responding to rare no-go trials could generate feedback to participants that leads to negatively valenced mind wandering. Although we focused on interindividual differences in negative TUTs in the current study, we have previously shown that negative TUTs are associated with poorer in-the-moment no-go trial performance (e.g., Banks & Welhaf, 2022; Goller et al., 2020). Thus, negative TUTs might correspond to increased salient task errors while positive TUTs reflect a broadening of attention that manifests in subtle fluctuations of attention (i.e., increased intraindividual RT variability). Our hypothesis that negative TUTs would be the strongest correlate with sustained attention performance was only partially supported. We discuss how the experience of positive TUTs, if only infrequent might also contribute to poorer sustained attention performance below.

Positive emotions are often associated with increased cognitive processing in several different tasks including the SART (e.g., Smallwood et al., 2009) and visual attention paradigms (e.g., Raila et al., 2015). Fredrickson’s (2004) Broaden-and-Build theory suggests that positive emotions can be used to broaden a persons’ scope of attention, cognition, and action allowing for the construction of resources. While this may seem important for everyday functioning and survival, when these positive emotions are linked to TUTs, a broadening of attention may be problematic. That is, even though positive emotions might broaden the scope of attention and interest in the environment and encourage exploration (Wadlinger & Isaacowitz, 2006), this inherently pushes attention away from the current goal of the task. Thus, positive emotions (and positive TUTs) might increase cognitive flexibility at the cost of goal maintenance (Paul et al., 2021). As such, experiencing positive TUTs might lead to similar levels of distraction and impairments of performance by tipping focus away from the task (e.g., Pourtois et al., 2017). In fact, the strongest association was between positive TUTs and intraindividual RT variability suggesting that people who reported more positive TUTs experienced greater inconsistency in their responding during the SART signaling increased fluctuations in sustained attention across the SART. Positive TUTs were not related to no-go errors which typically reflect the deepest form of mind wandering and are the most overt error types that can be made during the SART. Thus, it is possible that the relationship between no-go errors and mind wandering occurs when attention is captured to a greater degree or greater depth. However, intraindividual RT variability may be more sensitive to mind wandering that does not capture attention to as great a degree (i.e., less depth). Supporting this view, negatively valenced TUTs do appear to be reported to have greater depth than neutral or positively valenced TUTs (Goller et al., 2020). Alternatively, if positive TUTs act to shift focus away from the task and negative task events, like no-go errors (Pourtois et al., 2017), then it is not surprising to find a null association.

Theoretical Implications and Alternative Explanations

A novel aspect of this study was examining the time-on-task effect of each TUT valence. According to the resource-control perspective (Thomson et al., 2014), executive resources are fixed throughout the task, but over time, participants might actively reallocate their resources away from the primary task toward mind wandering. As executive resources are reallocated, we would expect an increase in all forms of mind wandering, but this was generally not the case. In fact, negative and neutral TUTs showed a consistent time-on-task effect, while positive TUTs remained relatively consistent (and low) throughout. Negative TUTs might have increased with time on task through an interaction between two factors. Specifically, consistent with the resource-control model (Thomson et al., 2014), participants likely reallocated executive resources toward mind wandering later in the task. Negative TUTs were likely more able to capture attention (as we have previously hypothesized), and this is apparent especially later in the task. Thus, the reallocation of resources, along with the strong attention capture, might lead to an isolated increase in negative TUTs with time on task.

These time-on-task findings provide insight into the within-person dynamics of mind wandering. It is well established that individual differences in executive attention predict the frequency of mind wandering during the task, and from the resource-control perspective, should also predict changes in mind wandering over the course of the task. We have previously shown that individual differences in executive attention abilities are negatively associated with negative TUTs (see Banks & Welhaf, 2022; Welhaf, Astacio, & Banks, 2024). If individual differences in executive attention are important for preventing negative off-task thoughts from reaching conscious awareness, this might be more prominent early in the task. However, as executive resources shift, these negative TUTs become more prominent (as indicated by the main effect of block in the studies analyzed here). Future work should consider how individual differences in executive attention interact with these time-on-task effects (e.g., do individuals with better executive attention still show stark increases in negative TUTs over the course of a task?).

It is possible that the relationship observed between mind wandering and sustained attention performance is driven by an unobserved construct. Specifically, prior work has suggested that both neuroticism (Crow, 2019; Robison et al., 2017) and depression (Maalouf et al., 2010; Seli et al., 2019) are related to sustained attention performance and mind wandering. Thus, it is important to examine how these constructs may impact both sustained attention and mind wandering. In one of the current studies (Banks et al., submitted), participants self-reported levels of neuroticism. Counter to prior findings (Crow, 2019; Robison et al., 2017), we did not find any relationships between neuroticism and overall TUT rate, positive, negative, or neutrally valenced TUTs, d’ or reaction time variability on the SART, or a composite WMC measure, all ps > .05. This suggests that the relationship between mind wandering and sustained attention is less likely to be due to neuroticism. However, future work should examine the role of depression in this relationship.

An alternative possibility is that the increase in negative TUTs with time on task is a reflection of emotional or motivational changes over the course of the task. Previous research has found that motivation tends to decrease from pre- to posttask (e.g., Seli et al., 2019) and even over the course of the task (Brosowsky et al., 2023). This change in motivation is often tightly coupled with mind wandering during the task (e.g., Thomson et al., 2015). Several studies have shown that motivation manipulations can reduce overall rates of mind wandering (e.g., Seli et al., 2019) and alleviate vigilance decrements in performance and mind wandering (see Neigel et al., 2020 for a review; Robison et al., 2021 for an excellent empirical test of this claim). However, it is not clear if changes in motivation are associated with changes in the specific emotional valence of mind wandering. Future research could test such a question that changes in motivation predict increases in the emotional valence of mind wandering to better understand the link between motivational state and specific forms of off-task thought.

Changes in motivation can also result in the feeling of boredom (e.g., Struk et al., 2020). This increased feeling of boredom (along with other negative emotions like frustration and underarousal; Danckert et al., 2018; Raffaelli et al., 2018) might be one explanation for why TUTs occur (Critcher & Gilovich, 2010; Eastwood et al., 2012) and for sustained attention failures, more generally (Hunter & Eastwood, 2018). Future research should aim to incorporate dispositional measures of boredom proneness or continuously track boredom over time (using experience sampling methods) to understand how changes in boredom might be related to changes in mind wandering and specifically the emotional valence of TUTs.

Limitations

While the current study has several strengths including analyzing data from several independent datasets across several years and multiple samples, it is worth noting some possible limitations and future directions. First, while the SART is frequently used in the literature to assess mind wandering and sustained attention failures, other tasks might produce different patterns. For example, the PVT (Lim & Dinges, 2008) and metronome response task (MRT; Seli et al. 2013) have also been used in studies of mind wandering and appear sensitive to manipulations that should affect sustained attention like sleep deprivation and motivation manipulations. As discussed earlier, Martínez-Pérez et al. (2021) showed different changes in intentional versus unintentional mind wandering depending on whether mind wandering was assessed in the SART or a PVT. Thus, it is possible that within-subject changes in the emotional valence of mind wandering might also vary by the executive demands imposed by the task. Second, the SART tasks in the current study were relatively short compared to more traditional sustained attention tasks that last several tens of minutes. Although we clearly showed block effects for thought reports, it would be interesting to investigate these within-subject changes over longer time periods in line with much of the vigilance paradigms. Finally, most of the data reported here relied on university student samples (outside of Welhaf, Banks, & Bugg, 2024 which used participants paid through Prolific). Thus, our conclusions should be viewed considering this specific sample. Recruiting more diverse and representative samples would provide a more holistic picture of how the emotional valence of mind wandering varies over time and impacts sustained attention performance.

Conclusion

Assessing the emotional valence of mind wandering shows that not all experiences of mind wandering behave the same way. While studies of mind wandering tend to lump different reports of off-task thoughts together, the present study shows the importance of assessing specific contents of individuals’ stream of consciousness. In doing so, we found that individuals who report more emotionally valenced mind wandering (both positive and negative reports) also perform worse on a sustained attention task. The differential development of emotionally valenced mind wandering over the course of the task suggests that not all types of mind wandering increase (or change at all) over the course of the task. Collectively, these findings highlight the heterogeneous nature of mind wandering and speak to the importance of assessing dimensions of mind wandering.

Appendix Meta-Analyses Using the Square Root Transformed Emotional Valence Reports to Correct for Nonnormal Distributions

Figure A1. Forest Plot of Correlations Between Emotional Valence of TUTs (Square Root Transformed) and Intraindividual Reaction Time Variability.

Note. Gray boxes represent study-level correlations. Blue diamonds represent meta-analytic estimates across studies. TUTs = task-unrelated thoughts; RT = reaction time; COR = correlation; CI = confidence interval.

Figure A2. Forest Plot of Correlations Between Emotional Valence of TUTs (Square Root Transformed) and No-Go Accuracy.

Note. Gray boxes represent study-level correlations. Blue diamonds represent meta-analytic estimates across studies. TUTs = task-unrelated thoughts; COR = correlation; CI = confidence interval.

Figure A3. Forest Plot of Correlations Between Emotional Valence of TUTs (Square Root Transformed) and d’.

Note. Gray boxes represent study-level correlations. Blue diamonds represent meta-analytic estimates across studies. TUTs = task-unrelated thoughts; D’ = D-prime; COR = correlation; CI = confidence interval.