Age-Related Differences in Mind Wandering: The Role of Emotional Valence

Abstract

Objectives

Older adults consistently report fewer experiences of mind wandering compared to younger adults. Aging is also associated with a shift in the emotional focus of our thoughts, with older adults tending to experience an increase in attention toward positive information, or a “positivity bias,” relative to younger adults. Here, we tested if the positivity bias associated with aging can also predict age-related changes in the content of older adults’ mind wandering.

Method

Older adults and younger adults completed a go/no-go task with periodic thought probes to assess rates of emotionally valenced mind wandering.

Results

Older adults reported significantly less negatively and neutrally valenced mind wandering compared to younger adults, but there was no age difference in reports of positively valenced mind wandering. Overall rates of mind wandering predicted poorer task performance for both age groups: Individuals who mind wandered more, performed worse, but this did not differ by the emotional valence. Both older adults and younger adults showed similar in-the-moment performance deficits, with mind wandering reports being associated with worse immediate no-go accuracy and faster reaction times, consistent with mindless responding.

Discussion

Focusing on different dimensions of thought content, such as emotional valence, can provide new insight into age-related differences in mind wandering. Older adults’ mind wandering reports were less negative and neutral compared to younger adults’ reports suggesting a positivity bias for older adults. However, this positivity bias does not seem to affect task performance. We discuss the implications of the findings for mind wandering theories and the positivity bias.

Attention lapses, Positivity bias, Sustained attention, Task-unrelated thoughts

Despite our best efforts to maintain focus on current tasks and goals, sometimes our attention shifts toward internal information, an experience called mind wandering (Smallwood & Schooler, 2006). Since Smallwood and Schooler’s review, there has been substantial development in the field as to how we define and study this phenomenon. Distinguishing different types of mind wandering is important for understanding how and when it occurs and its potential consequences (Christoff et al., 2016, 2018; Seli, Kane, Metzinger, et al., 2018; Seli, Kane, Smallwood, et al., 2018). In the current study, we conceptualized mind wandering as task-unrelated thoughts, or TUTs, as we are interested in thoughts that are unrelated to the current activity participants are engaged in and specifically the emotional valence of these thoughts in relation to age-related differences in mind wandering.

Aging and Mind Wandering

The study of mind wandering, and spontaneous cognition more broadly, in healthy aging has become an area of much interest (for a review see Maillet & Schacter, 2016). It is well established that as we age, many cognitive processes decline, including attention control and processing speed (Craik & Salthouse, 2011; Hasher & Zacks, 1988; but see Verhaeghen et al., 2012). TUTs occur (at least partially) because of failures to maintain attention to ongoing task goals (e.g., McVay & Kane, 2012). It is therefore surprising that older adults consistently report fewer TUTs compared to younger adults (meta-analytic estimate g = –0.89; Jordão, Ferreira-Santos, et al., 2019).

Some have suggested that older adults report fewer TUTs because mind wandering is resource-demanding (Smallwood & Schooler, 2006) and limited cognitive resources (e.g., Craik & Byrd, 1982) make it difficult for older adults to simultaneously perform an ongoing task and mind wander. As well, McVay et al. (2013) argued that age-related differences in ongoing current concerns or goals interact with levels of executive control to predict reduced mind wandering in older adults. That is, older adults might have fewer ongoing concerns, which might trigger less mind wandering to begin with, despite their reduced resources (but see Diede et al., 2022 for evidence against this). Others have posited that older adults report fewer TUTs because they lack awareness of their mind wandering or have a bias against reporting off-task thoughts (e.g., Zavagnin et al., 2014; but see, Frank et al., 2015). Alternatively, older adults may report fewer TUTs because of dispositional factors, such as higher conscientiousness, motivation, or interest in ongoing tasks (e.g., Jackson & Balota, 2012; Nicosia & Balota, 2021). The purpose of the current study was not to adjudicate among these accounts but rather to examine a novel question regarding age-related differences in mind wandering, namely, whether the contents of mind wandering may differ between younger adults and older adults, potentially providing a more nuanced view of age-related differences in mind wandering.

In our work (e.g., Banks & Welhaf, 2022; Banks et al., 2016; Goller et al., 2020) with younger adults, we typically find that rates of negative and neutral TUTs each tend to occupy the greater proportion of younger adults’ mind wandering (~40% of all TUTs), whereas positive TUTs occur less frequently (~20% of all TUTs). These patterns are consistent with the notion that mind wandering, at least in younger adults, is primarily negatively valenced or associated with negative mood (Smallwood et al., 2009) because it is directed toward concerns or worries (Killingsworth & Gilbert, 2010; Klinger, 1978; Stawarczyk et al., 2013). Yet, little is known about the emotional valence of older adults’ TUTs. In the present study, we adopted a novel approach by examining whether the age-related difference in TUTs varies depending on the emotional valence associated with the mind wandering report.

Age-Related Positivity Bias

Compared to younger adults, older adults focus on positive information more than negative information, more commonly referred to as the “positivity bias” (for reviews, see Carstensen & DeLiema, 2018; Mather & Carstensen, 2005). This bias may be due to controlled shifts of attention toward and memory for more positive information in older adults, as predicted by motivational theories of the positivity bias (Carstensen, 1993,, 2006; Carstensen et al., 1999). Support for this comes from studies showing an age-related decrease in attention toward and memory for negative stimuli (Charles et al., 2003; Shamaskin et al., 2010) and an age-related increase in the processing of positive stimuli (Isaacowitz et al., 2006; Mather & Knight, 2005). That is, evidence for a positivity bias can take multiple forms (e.g., older adults’ decreased bias toward negative information or older adults’ increased bias toward positive information).

How might the positivity bias play out in mind wandering, and specifically the emotional valence of mind wandering? The Socioemotional Selective Theory of the positivity bias (Carstensen et al., 1999) argues that older adults have a shorter time horizon, so they direct their cognitive resources and attention toward positive information. From this perspective, then, we would assume that whatever (little) cognitive resources older adults can devote to mind wandering might be directed towards thinking about positive information. Thus, the role of current goals and reduced cognitive resources may not only explain why mind wandering is reduced in older adults (McVay et al., 2013), but also how older adults’ mind wandering content might differ from that of younger adults’ content. If there is an age-related positivity bias in older adults’ mind wandering, it will likely then be reflected in the distribution of their thought content. Specifically, older adults might show lower rates of mind wandering compared to younger adults, but when they do report being off-task, older adults’ thoughts may tend to be positive in valence, whereas younger adults’ thoughts will be primarily negative.

The Impact of Mind Wandering on Task Performance

It is well established that mind wandering affects ongoing task performance both at the between-subject level (i.e., greater rates of mind wandering are associated with poor task performance) and the within-subject level (i.e., momentary fluctuations in task performance relative to mind wandering episodes). The between-subject effect appears to be driven by negative TUTs in younger adults. Specifically, younger adults who report a higher rate of negative TUTs perform worse on tasks measuring working memory and sustained attention, likely because they capture attention and distract younger adults over the duration of the task to a greater degree than other types of TUTs (e.g., Banks & Welhaf, 2022; Banks et al., 2016). However, it’s possible that the relationship between task performance and emotional valence of TUTs is different for older adults. Specifically, if there is a positivity bias in older adults’ mind wandering, then the correlation between positive TUTs and task performance should be stronger as these are the thoughts that capture older adults’ attention to a greater degree.

The impact of mind wandering is also evident at the within-subject level. When participants report they were off-task, their immediate performance (either accuracy or reaction time [RT]) suffers relative to when they report being on-task (e.g., Bastian & Sackur, 2013; Hawkins et al., 2019; McVay & Kane, 2012). That is, there is behavioral evidence in the few trials leading up to reports of mind wandering to suggest that these self-reports validly reflect participants’ experiences. Note that this approach does not attempt to identify the duration of participants’ mind wandering, rather it shows that subtle behavioral differences (which participants are likely unaware of) can manifest in the moments before they are randomly asked to report on their thoughts.

In younger adults, these within-subject effects also seem to be sensitive to the emotional valence of mind wandering. For example, Banks and Welhaf (2022) found that when participants reported a negative TUT (compared to a neutral or positive TUT) their “no-go” accuracy was lower and they were more variable in their RT on the preceding “go” trials (see also Goller et al., 2020). This unique pattern of negative TUTs relating to task performance may be because of how strongly they capture, and redirect, attention away from the current task (Banks & Welhaf, 2022). Although younger adults do report positive TUTs, there is no evidence that they hurt performance. However, this might not be the case for older adults. As mentioned earlier, Socioemotional Selective Theory argues that older adults direct cognitive resources to attend to positive information and away from negative information. If this is the case, then older adults should show poorer performance before positive TUTs compared to neutral or negative TUTs. Thus, on this view, the relationship between task performance and emotional valence of TUTs could be specific to negative TUTs in younger adults and positive TUTs in older adults as these are the thought types that likely capture attention (either passively or actively).

Current Study

The current study aims to extend our previous investigations of the frequency and impact on task performance of emotionally valenced mind wandering to older adults. Our main questions and hypotheses were as follows:

 

H1: We predict that older adults will report less mind wandering than younger adults. Further, we predict that this difference will be driven by older adults reporting more instances of positively valenced mind wandering and fewer instances of negatively (and neutral) valenced mind wandering compared to younger adults.

 

H2: We predict that mind wandering will be negatively correlated with task performance in both age groups. Further, we predict that this relationship will be specific to different emotional valences of mind wandering in each age group. Namely, younger adults who report more negative TUTs will perform worse on the SART, but this won’t be the case for neutral or positive TUTs. As for older adults, we predict that older adults who report more positive TUTs will perform worse on the SART, but this won’t be the case for neutral or negative TUTs.

 

H3: We predict that immediate performance (mean RT, RT variability, and accuracy) will be worse on trials leading up to reports of mind wandering compared to on-task reports in both age groups. Further, we predict that in instances when younger adults report negative TUTs, their performance will be especially worse compared to neutral or positive TUTs. For older adults, we predict that instances of positive TUTs will be preceded by poorer performance compared to neutral and negative TUTs.

Methods and Procedures

Transparency and Openness

We report our sample size justification and data exclusion criteria, as well as all measures and manipulations included in the study (Simmons et al., 2012). All deidentified data, Rmarkdown files for analyses, and the preregistration for this study are available on the Open Science Framework (https://osf.io/q6ejp/). All aspects of this study were preregistered unless otherwise specified.

Participants

We aimed to collect data from 175 younger adults (age range = 18–35 years old) and 175 older adults (age range = 60+ years old) via Prolific Academic (https://www.prolific.co) to complete a 25-min study for $5.00. These sample sizes were based on an a priori sensitivity analysis conducted in G*Power 3.1 for comparing correlations from independent conditions (i.e., older adult vs younger adult RTsd × TUT correlation comparison). We powered our study for this test (comparing correlations) as it required the largest sample to detect a difference and thus would also allow for robust power (>99%) for detecting age-related differences in mind wandering. With a sample size of 175 in each group, alpha = .05, and power = .80, we would be able to detect a correlation difference of roughly .30 between the age groups. To account for exclusions and data loss, we collected data from a total of 372 participants (n_older = 188; n_younger = 184). Based on prior meta-analytic results of age-related differences in mind wandering (Jordão, Ferreira-Santos, et al., 2019), we had ample power to detect age-related differences in overall TUT rates. The sample size also allowed us to reliably estimate and compare correlations between the age groups. The study was approved by the Washington University in St. Louis institutional review board. All tasks and questionnaires were programmed in Gorilla (https://gorilla.sc), and we required participants to complete the study on a laptop or desktop computer to ensure accurate recording of RTs (Anwyl-Irvine et al., 2020).

Semantic Sustained Attention to Response Task

Participants completed a version of this go/no-go task adapted from a previous version built for online delivery (McVay & Kane, 2012; Welhaf & Kane, 2023, see Author Note 1). Participants were instructed to press the space bar for words from a target category (animals; 90% of total trials) while withholding responses to another category (crops, i.e., fruits/vegetables; 10% of total trials). Participants first completed 10 practice trials by responding to boy’s names and withholding responses to girls’ names. The real task began with 16 unanalyzed buffer trials. Each stimulus word was presented for 272 ms, followed by a mask (XXXXXXXXXX), which was presented for 1,224 ms. Participants were instructed to press the spacebar during either the word or the mask.

After being presented with task instructions, participants answered a quiz question about the Semantic Sustained Attention to Response Task (SART): “When performing this task, which words should you NOT press any key for?” with the following response options: (1) animal names, (2) girl names, (3) crop names, and (4) boy names. Participants pressed the number key corresponding to the correct answer (Option 3). If answered incorrectly, participants were reminded of the task instructions and then were given another chance to answer the question.

Participants completed 480 trials divided into four seamless blocks of 120. Stimuli were composed of 36 animal names (i.e., “go” trials) and 4 fruit/vegetable names (i.e., “no-go” trials) with each stimulus pseudorandomly presented three times (a total of 108 “go” animal stimuli and 12 “no-go” crop stimuli). Each block contained a new set of stimulus words, which was randomly selected for each participant. The dependent measures included mean RT to correct “go” trials, the within-subject standard deviation (SD) of RTs to correct “go” (animal) trials, “go” trial accuracy, “no-go” trial accuracy, and a signal detection measure of response bias, d-prime (dʹ, see Author Note 2).

Thought-Probes

During the SART, participants were periodically interrupted by a screen asking them to report on the content of their thoughts (exact instructions participants read are in Supplementary Material). As in previous work, probes were randomly presented after 50% of no-go trials (e.g., McVay & Kane, 2009; Welhaf & Kane, 2023. Each of the 24 thought-probes presented participants with the following response options: (1) task-related (i.e., completely focused on the task); (2) task performance/evaluation (i.e., thinking about how the participant was doing on the task); (3) off-task neutral; (4) off-task negative; and (5) off-task positive. Participants were given 10 s to make their response by pressing the numeric key that matched their response. If they did not respond within the deadline, the probe was counted as missing (see Preregistered Exclusion Criteria below for handling missing thought probes).

We calculated TUT rate as the sum of response options 3–5 divided by the number of valid probes (i.e., probes answered within the 10 s window). We also calculated the rate of each specific TUT valence in the same manner (e.g., number of Neutral TUTs/total valid probes).

During the thought probe instructions, participants completed a comprehension question regarding responding to thought-probes (as done in Welhaf & Kane, 2023). Specifically, the question asked, “When responding to questions about your thoughts, what time frame should you report your thoughts from?” with the following response options: (1) Since the very beginning of the task, (2) The moment right before the thought menu appeared, (3) Over the last 30–60 s, (4) Since the last time a thought-menu appeared. (Option 2 was the correct answer.) If answered incorrectly, participants were shown a separate screen that told them they were wrong, were reminded of the appropriate time frame, and were given another chance to answer (see Author Note 3).

Post-Experiment Exclusion Questions

Participants answered several questions about their experiences during the study (adapted from Welhaf & Kane, 2023). We first asked participants to describe which part of the study they found the most challenging (to check for potential “bots”). Next, participants answered two Likert-scale questions regarding their immediate environment during the experiment. We also asked participants about their media multitasking frequencies during the study. Finally, participants reported on their sleepiness at the beginning of the study.

Results

Below, we report the results of our preregistered analyses and note where, if at all, we deviated from the preregistered plan. All data aggregation and analyses were performed in R (R core team, 2021) using tidyverse (Wickham et al., 2019). ANOVAs were performed using the afex package (Singmann et al., 2020). Data visualizations were created using ggplot2 (Wickham, 2016).

Exclusions Based on Preregistered Criteria

We dropped data from 11 participants for failing to provide a reasonable response to the open-ended difficulty question. We dropped data from 14 participants for reporting taking a medication that may affect their memory. We dropped data from four participants for missing more than four thought probes. Finally, we dropped data from 13 participants for low SART “go” trial accuracy indicating a failure to understand task instructions. We also dropped data from two participants who reported an age not within our preregistered ranges (this was not a preregistered exclusion, rather these participants falsely reported their age on Prolific and completed the study based on our screeners).

Final Sample Demographics and Cognitive Task Performance Differences

Table 1 provides demographic and descriptive statistics for both groups (N = 329). Older adults were slower on SART “go” trials compared to younger adults, t(324.97) = 4.732, p < .001, d = 0.52 [0.30, 0.74]. Older adults were nominally less variable in their “go” RTs compared to younger adults, t(315.17) = −1.727, p = .085, d = −0.19 [−0.41, 0.03]. Older adults performed more accurately on both “go” trials t(302.11) = 1.996, p = .047, d = 0.22 [0.00, 0.44], “no-go” trials t(325.87) = 3.138, p = .002, d = 0.35 [0.13, 0.56], and on the signal detection measure of sensitivity (dʹ), t(326.75) = 3.289, p = .001, d = 0.36 [0.14, 0.58]. Thus, older adults, on average, performed better on the SART compared to younger adults consistent with prior work (e.g., Carriere et al., 2010; Jackson & Balota, 2012; Maillet et al., 2018; McVay et al., 2013; see Author Note 4).

Table 1.

Descriptive Statistics of Measures of Interest for Younger (N = 167) and Older Adults (N = 162)

Age group	Variable	M (SD)	Range	Skew	Kurtosis
Younger adults	Age	27.70 (4.79)	18–35
	Gender (% female)	22.16%
	Processing speed	321.46 (76.29)	210.67, 676.66	1.95	5.15
	SART RTsd	129.69 (40.96)	55.67, 296.00	0.78	0.78
	SART Mean RT	559.21 (99.99)	304.72, 964.35	0.62	1.41
	SART Go Acc	.97 (.04)	.74, 1.00	–2.72	9.30
	SART No-Go Acc	.56 (.20)	.06, 0.92	–0.41	–0.61
	SART dʹ	2.36 (0.85)	–0.22, 4.07	–0.30	0.04
	TUTs	.20 (.22)	.00, .91	1.16	0.55
	Negative TUTs	.07 (.11)	.00, .67	2.48	7.51
	Neutral TUTs	.10 (.14)	.00, .87	2.27	6.71
	Positive TUTs	.04 (.08)	.00, .50	2.84	9.72
Older adults	Age	65.60 (4.83)	60–84
	Gender (% female)	47.53%
	Processing speed	369.49 (102.53)	94.05, 856.55	1.78	4.67
	SART RTsd	119.06 (30.27)	65.84, 210.42	0.58	–0.06
	SART Mean RT	611.79 (103.74)	308.77, 962.13	0.19	0.54
	SART Go Acc	.98 (.03)	.83, 1.00	–2.94	9.55
	SART No-Go Acc	.63 (.20)	.02, .98	–0.75	0.19
	SART dʹ	2.65 (0.80)	–0.22, 4.38	–0.55	0.59
	TUTs	.12 (.18)	.00, .83	1.85	3.15
	Negative TUTs	.03 (.07)	.00, .42	2.74	8.38
	Neutral TUTs	.06 (.10)	.00, .74	3.17	14.55
	Positive TUTs	.03 (.08)	.00, .52	3.75	16.33

Note: RTsd = Intra-individual standard deviation of reaction time to correct “go” trials; SART = sustained attention to response task; TUTs = rate of task-unrelated thoughts during SART.

Age-Related Differences in Emotional Valence of Mind Wandering and SART Performance

Our first set of analyses regarding age-related differences in mind wandering tested our first hypothesis (H1), specifically, if we replicated the age-related difference in overall TUT rates and if this differed by emotional valence. As expected, we replicated the well-established age-related reduction in overall TUT rates (e.g., Jackson & Balota, 2012; McVay et al., 2013; Robison et al., 2022; see Table 1): younger adults reported being off-task roughly twice as often as older adults did during the SART, t(315.84) = −3.849, p < .001, d = −0.42 [−0.64, −0.20].

To test for age-related differences in the emotional valence of TUTs, we conducted a 2 (Age: Young vs Old) × 3 (Emotional Valence: Negative, Neutral, Positive) mixed model ANOVA. There was a main effect of Age, F(2, 327) = 14.71, p < .001, η_p² = .04, and Emotional Valence, F(2, 654) = 17.41, p < .001, η_p² = .05. Critically, there was a significant, albeit weak, Age × Emotional Valence interaction, F(2, 654) = 3.10, p = .046, η_p² = .01. Figure 1 shows that the age-related difference in overall TUT rate appears to be driven by reductions in the frequency of Negative and Neutral TUTs in older adults. However, both age groups showed similar rates of Positive TUTs. These impressions were confirmed with follow-up comparisons. Older adults reported fewer Negative TUTs, t(916) = −3.353, p < .001, d = −0.22 [−0.35, −0.09], and Neutral TUTs, t(916) = −3.665, p < .001, d = −0.24 [−0.37, −0.11] compared to younger adults. There was no difference in Positive TUTs, t(916) = −0.774, p = .439, d = −0.05 [−0.18, 0.08].

Figure 1.

Raincloud plots (Allen et al., 2019) depicting TUT rates of emotional valence reports between Age Groups. TUT = task-unrelated thoughts. Dots represent individual subject means in each probe response. The closed black dots represent group-level mean estimates for each Age Group. Error bars are 95% confidence intervals.

Emotionally Valenced TUTs and Interindividual SART Performance Across Age Groups

Next, we addressed our second hypothesis (H2) regarding age-related differences in the relationship between TUTs and task performance. We tested this by comparing the correlations between SART performance measures (e.g., intraindividual RT variability and dʹ) and TUTs (both overall and by emotional valence) across the age groups. Table 2 presents all bivariate correlations among our dependent measures of interest within each age group (see Author Note 5).

Table 2.

Correlations Between Measures of Interest for Younger (Below Diagonal; N = 167) and Older (Above Diagonal; N = 162) Adults

Variable	1	2	3	4	5	6	7	8	9	10
1) Processing speed	—	0.27	0.27	–0.20	–0.05	–0.18	0.00	–0.03	0.00	0.02
2) SART RTsd	0.28	—	0.35	–0.27	0.02	–0.20	0.11	0.07	0.11	0.06
3) SART Mean RT	0.33	0.32	—	–0.21	0.68	0.42	–0.14	–0.04	–0.12	–0.11
4) SART Go Acc	–0.34	–0.50	–0.28	—	0.05	0.52	–0.15	–0.09	–0.10	–0.14
5) SART No-Go Acc	–0.12	–0.31	0.54	0.17	—	0.81	–0.14	–0.07	–0.08	–0.16
6) SART dʹ	–0.34	–0.54	0.14	0.66	0.78	—	–0.16	–0.08	–0.09	–0.17
7) TUTs	0.05	0.19	–0.01	–0.08	–0.21	–0.22	—	0.63	0.77	0.75
8) Negative TUTs	–0.04	0.03	–0.22	0.01	–0.17	–0.12	0.64	—	0.19	0.32
9) Neutral TUTs	0.08	0.17	0.11	–0.07	–0.15	–0.17	0.74	0.08	—	0.32
10) Positive TUTs	0.06	0.19	0.09	–0.12	–0.08	–0.15	0.60	0.23	0.22	—

Notes: Processing speed = Mean RT of Visual RT Task; RTsd = Intra-individual standard deviation of reaction time to correct “go” trials; SART = sustained attention to response task; TUTs = rate of task-unrelated thoughts during SART.

Correlations > |.15| are significant at p < .05, >|.19| significant at p < .01, and >|.25| significant at p < .001.

Younger adults who reported mind wandering frequently in the SART performed worse (i.e., more variable RTs and poorer signal detection accuracy). These correlations were not significant in older adults suggesting no relationship between mind wandering and performance for older adults. However, the correlations did not differ between the age groups, RTsd × TUT: Fisher’s Z = 0.736, p = .462; dʹ × TUT: Z = −0.560, p = .576. We next examined how SART performance was related to the emotional valence of TUTs within each age group. RT variability was positively associated with Neutral and, surprisingly, Positive TUTs in younger adults such that younger adults who reported more Neutral and Positive TUTs were more variable in their responding across the SART. RT variability was not associated with any emotional TUT category in older adults. None of these correlations were significantly different between age groups, Zs < |1.188|, ps > .235.

In younger adults, dʹ was related to Neutral TUT rate only, such that increases in Neutral TUTs were associated with poor signal detection accuracy. For older adults, dʹ was related to Positive TUT rate only, such that increases in Positive TUTs were associated with poor signal detection accuracy. Again, none of these correlations significantly differed between age groups, Zs < |.82|, ps > .411.

Emotionally Valenced TUTs and Intraindividual SART Performance Across Age Groups

Finally, we tested our third hypothesis (H3) regarding performance prior to thought reports. As described earlier, younger adults perform significantly worse preceding Negative TUTs, but performance preceding Neutral and Positive TUTs is relatively comparable (Banks & Welhaf, 2022; Goller et al., 2020), possibly because Negative TUTs capture attention and are harder to disengage from. From a positivity bias perspective, we tested if older adults showed performance decrements preceding Positive TUTs, as these reports may capture their attention more strongly than other TUTs.

We first conducted a logistic mixed effect model predicting “no-go” accuracy from Age Group (Young vs Old), Probe Response (on-task vs TUT), and their interaction. There was a main effect of Probe Response, χ²(1) = 101.44, p < .001, indicating that performance was worse on “no-go” trials when participants reported being off-task compared to when they reported being on-task. There was no Age Group effect, χ²(1) = 0.00, p = .995 or interaction, χ²(1) = 0.42, p = .516.

We next conducted another logistic mixed effect model predicting “no-go” accuracy from Age Group, Emotional Valence of TUTs (Negative, Neutral, of Positive), and their interaction. Figure 2 displays the results. There was no Age Group effect, χ²(1) = 0.42, p = .515 or interaction, χ²(2) = 0.60, p = .742. However, there was an effect of Emotional Valence, χ²(2) = 15.20, p < .001. Negative TUTs were associated with worse “no-go” accuracy compared to both Neutral, Z = −3.097, p = .006, and Positive TUTs, Z = −3.687, p < .001. Neutral and Positive TUTs had similar “no-go” accuracy, Z = −1.122, p = .501.

Figure 2.

Raincloud plots (Allen et al., 2019) depicting average accuracy on “no-go” trials before different emotionally valenced TUT responses between Age Groups. TUT = task-unrelated thoughts. Dots represent individual subject means in each probe response. The closed black dots represent group-level mean estimates for each Age Group. Error bars are 95% confidence intervals.

We next examined how participants’ mean RT varied by thought report. Speeding of RTs before TUTs is proposed to reflect increased “mindless” responding due to a lack of task focus (e.g., McVay & Kane, 2012; McVay et al., 2013; Smallwood et al., 2009; but see Stawarczyk et al., 2011 for evidence of slower RTs before TUTs). We conducted a mixed effect model on Mean RT of the four “go” trials preceding thought reports predicted by Age Group, and Probe Response, and their interaction. Previous research has found that behavioral effects can be found in as few as four trials before thought probes (Bastian & Sackur, 2013; Hawkins et al., 2019; McVay & Kane, 2012). As we only analyzed instances where all “go” trial were correct, we followed this approach to maximize the number of observations for each subject (see Author Note 6).

There were main effects of Age Group, F(1, 494.82) = 27.68, p < .001, and Probe Response, F(1, 494.82) = 6.43, p = .012. There was no interaction, F(1, 494.82) = 0.00, p = .973. Older adults were generally slower than younger adults consistent with processing speed deficits, but both groups were faster before TUTs compared to on-task reports. Thus, evidence of “mindless” responding preceding TUTs was evident in both age groups.

We next examined how pre-TUT speeding varied by emotional valence. Again, we conducted a mixed effects model predicting Mean RT of the four trials preceding TUT reports from Age Group, Emotional Valence of TUT, and their interaction (see Figure 3). There was a main effect of Age Group, F(1, 331.87) = 18.17, p < .001. There was no effect of Emotional Valence, F(2, 331.07) = 1.28, p = .279, nor was there an interaction, F(2, 331.07) = 0.86, p = .425.

Figure 3.

Raincloud plots (Allen et al., 2019) depicting preprobe Mean RT for emotional valence TUTs between Age Groups. TUT = task-unrelated thoughts. Dots represent trial means in each probe response. The closed black dots represent group-level mean estimates for each Age Group. Error bars are 95% confidence intervals.

Our final intraindividual analyses focused on RT variability in the trials preceding thought reports. We took the same analytic approach as described above in the mean RT analyses. The mixed effect model predicting RT variability revealed a main effect of Age Group, F(1, 462.72) = 4.40, p = .036. There was no main effect of Probe Response, F(1, 462.72) = 1.16, p = .282, nor was there an interaction, F(1, 462.72) = 1.01, p = .314. Older adults were similarly variable before their on-task and TUT reports, t(518) = −0.012, p = .991. Younger adults were also similarly variable before their on-task and TUT reports, t(485) = −1.599, p = .111, although this difference was nominally larger than it was in the older adults. Younger adults were, on average, more variable before both report types, t(503) = −2.076, p = .038.

When examining pre-TUT RT variability for each emotional valence, a mixed effect model indicated no effects of Age Group, F(1, 287.39) = 2.68, p = .103, Emotional Valence, F(2, 285.60) = 0.29, p = .746, or an interaction, F(2, 285.60) = 0.40, p = .673. Thus, unlike our previous work (e.g., Banks & Welhaf, 2022), we did not find support for Negative TUTs predicting poorer intraindividual SART performance.

Discussion

The current study replicated the typical age-related differences in TUT rates and SART performance. More importantly, the study revealed several novel findings. First, we provided evidence that the age-related reduction in TUT rates is primarily driven by reductions in older adults’ negative and neutral TUTs; we did not find age-related differences in Positive TUTs. Second, while we replicated the negative correlation between overall TUT rates and global task performance in younger adults, this correlation did not exist in older adults.

Our findings with respect to TUT rates are consistent with a “positivity bias” in older adults’ mind wandering. As previously mentioned, evidence for the “positivity bias” can present as increased attention towards or memory for positive information or a decrease in attention towards or memory for negative information, typically in the context of a memory or visual attention paradigm (for a review, see Reed & Carstensen, 2012). The motivational perspective of the positivity bias suggests that older adults shift their goals toward actions that increase the salience of positive information in the focus of attention. As such, this perspective might predict that older adults would show increased positive TUTs compared to younger adults, as positive TUTs might be more emotionally pleasant than simply completing the SART and older adults might deploy cognitive resources to engage in these types of TUTs. Although older adults did not report more positive TUTs than younger adults, the pattern of age-related differences for positively valenced TUTs (i.e., no age difference) was both distinct from that found for negatively valenced or neutral TUTs as well as the typical pattern found in the broader literature (i.e., reduced TUTs with age). The reduction in negative and neutral TUTs in older adults may be consistent with the motivational account if we assume that the reason older adults show fewer negative TUTs is because they were devoting their cognitive resources to performing the task and used whatever remaining resources to generate positive TUTs.

One possible explanation for why we did not see a stronger positivity bias (i.e., an increase in positive TUTs for older adults compared to younger adults), may be due to the nature of the task and the role that mind wandering plays in dividing attention. Previous work has found that the “positivity bias” is evident during tasks where attention is not divided and resources are fully devoted to the task but absent during divided attention or when resources are depleted (e.g., Allard & Isaacowitz, 2008; Knight et al., 2007). Mind wandering naturally divides attention between the ongoing task and participants’ stream of consciousness. As well, the SART is a demanding task. Thus, older adults’ cognitive resources may have been sufficiently recruited to not only perform well on the SART (which they did) but also regulate their attention to focus on the task (i.e., reporting fewer TUTs). Future work should consider testing the hypothesis of age-related differences in the emotional valence of mind wandering in tasks that don’t require maximal cognitive resources. If resources could be freed up in a less demanding task (e.g., a simple RT task), it is possible that older adults might mind wander more, which could increase positive TUTs and may provide stronger support for a positivity bias in older adults’ mind wandering.

We also found that older adults and younger adults showed similar behavioral lapses before their reports of mind wandering, but this did not vary by emotional valence. These intraindividual performance effects of emotional valence of TUTs speak against some perspectives of the “positivity bias,” which predict more immediate or automatic effects of positive, than negative, information for older adults (Cacioppo et al., 2011; Labouvie-Vief et al., 2010). That is, these perspectives would predict that older adults should be more affected by positive TUTs in the moment, compared to negative or neutral TUTs. However, this was not the case as we found no evidence of differential effects of emotional TUTs on within-subject RTs or accuracy between the age groups. As seen in Figure 1, both older adults and younger adults had poorer “no-go” trial accuracy before negative TUTs. For younger adults, performance was significantly worse compared to neutral and positive TUTs (ps < .01), suggesting that these negative TUTs disrupted performance by capturing attention to a greater degree (Banks & Welhaf, 2022; Goller et al., 2020). However, for older adults, these differences were not significant (ps > .06). In the moment, then, older adults’ attention is not preferentially drawn or directed to positive information to the point where it disrupts performance.

It is also possible that differences in meta-awareness might play a role in age-related differences in the emotional valence of mind wandering. The probe-caught method does not rely heavily on meta-awareness because the probes themselves signal for subjects to take stock of their immediately preceding thoughts, rather than actively monitoring their thoughts over the course of the task. One approach to assess awareness of mind wandering is to use a self-caught method wherein participants report when they notice their mind has drifted off-task. If older adults are more aware of their positive TUTs, then we might expect them to self-catch those thoughts more frequently than they are caught by the probe. This is an interesting direction for future research.

Limitations and Future Considerations

While the current study offers some methodological and theoretical insights for future research to consider there are a few limitations to mention. First, we only probed after no-go trials, which may confound task performance with TUT reports, perhaps especially for older adults. Others have probed after “go” trials (e.g., Jackson & Balota, 2012) or used different tasks, such as the psychomotor vigilance task (Robison et al., 2022) or reading comprehension (Jackson & Balota, 2012) that don’t promote performance reactivity to assess age-related differences in mind wandering. Additionally, older adults recruited from Prolific, while diverse, might not be representative of all older adults as they might be more familiar with technology, which allows for them to complete online studies (e.g., Greene & Naveh-Benjamin, 2022).

Although we defined mind wandering as TUTs, it is possible that different results may have been found had we defined mind wandering as a different type of spontaneous process (i.e., one that is dynamic or freely moving; Christoff et al., 2016, 2018). The thought probe responses we included did not sample the entire range of attentional states, but rather focused on TUTs. Previous research has also included options for “no thoughts” or “blank minds” and has dissociated intentional from unintentional mind wandering. Regarding the current hypotheses, the age-related positivity effect may differ for intentional and unintentional mind wandering reports. For example, intentional mind wandering is often seen as a more controlled process, which might boost the positivity effect for older adults based on socioemotional selectivity theory (see Reed & Carstensen, 2012). On the other hand, it’s possible that differences in the current results might be driven by younger adults’ increased likelihood of engaging in intentional mind wandering, which is often associated with more negative affective states like depression, anxiety, and stress (Seli et al., 2019). Future work should consider how these probe options affect the reporting of emotional valence of TUTs and if emotional TUTs arise more deliberately or spontaneously (see Banks & Welhaf, 2022; but see Kane et al., 2021; Murray & Krasich, 2022 for concerns about intentionality of mind wandering being a coherent category of TUTs).

Finally, although we excluded participants who reported taking medications that might affect memory or attention, we did not screen for possible psychiatric or depression diagnoses. Younger adults in the current study reported higher levels of perceived stress compared to older adults (see Supplementary Table 1), and perceived stress was associated with more negative TUTs in younger adults but not in older adults (see Supplementary Table 2). Thus, it’s possible that in addition to differences in perceived stress between older adults and younger adults, they may differ in psychiatric diagnoses, including depression or anxiety, which could contribute to age-related differences in mind wandering, especially the emotional valence of participants’ mind wandering.

Practical Implications and Conclusions

The current study highlights the theoretical importance of investigating different contents of TUTs to further explore age-related differences in mind wandering, an aspect that is often overlooked in mind wandering research. Although older adults show an overall reduced TUT rate compared to younger adults, the distributions of older adults’ TUT experiences appear to support a positivity bias through a reduction in negative and neutral TUTs. However, these positive TUTs do not seem to affect their overall performance on the current task as older adults performed a demanding sustained attention task quite well. The current work suggests some practical implications for understanding age-related changes in cognition and the effects of mind wandering on ongoing performance. Given that mind wandering, and perhaps negative TUTs specifically, is associated with performance costs in younger adults, activating goals, through priming manipulations that are associated with emotional satisfaction or meaning could minimize these performance impairments (e.g., Jordão, Pinho, et al., 2019). This would align nicely with the rationale of the socioemotional theory of the positivity bias for older adults and might be a useful mechanism for reducing the impact of mind wandering for younger adults.

Supplementary Material

Supplementary data are available at The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences online.

Author Notes

1. Participants first completed a processing speed task (Hintz et al., 2020) to assess cognition independently from the SART. As expected, older adults were significantly slower in the processing speed task compared to younger adults, t(297.73) = 4.809, p < .001, Cohen’s d [95% CI] = 0.53 [0.31, 0.75], suggesting we captured two cognitively different groups during recruitment.

2. In our preregistration, we specified a group-level outlier censoring procedure for RTsd and dʹ as they were our primary measures of interest. However, as an oversight, we didn’t include this procedure for SART Mean RT. To be consistent, we implemented this censoring to account for possible outliers, deviating from our preregistration.

3. After completing the SART, participants answered questions regarding their motivation and alertness during the SART, and their conscientiousness and perceived stress. We preregistered looking at these measures, but for simplicity report these results in Supplementary Materials.

4. Although not preregistered, we also calculated the coefficient of variation (CoV) for SART “go” RTs, which is a measure of variability that accounts for mean RT and is commonly used in studies of aging (e.g., Carriere et al., 2010; Jackson & Balota, 2012; Nicosia & Balota, 2021). Older adults had a significantly lower CoV in the SART compared to younger adults, t(307.46) = −4.517, p < .001, d = −0.50 [−0.72, −0.28]. Thus, even when accounting for mean RT, older adults had more consistent performance compared to younger adults.

5. As seen in Table 1, emotionally valenced TUTs had high levels of skewness and kurtosis suggesting non-normal distributions. Although not preregistered, we applied a square root transformation to these data, which corrected these issues. Many of the correlations were similar to those with the raw data, so we report those for ease of interpretation. Correlations with the transformed data are reported in Supplementary Material for transparency.

6. An anonymous reviewer suggested looking at different lengths of trials (e.g., two or three trials) before thought probes to see if the immediate effects of emotional valence of TUTs were still apparent. Although the four-trial approach used in the current study is common, we examined these effects using a three-trial window. The effects for both on-task versus TUT reports and the emotional valence of TUTs largely mirror the four-trial window results reported in the main text.

Funding

This work was supported by the National Institute on Aging of the National Institutes of Health (T32 AG000030-47).

Conflict of Interest

None.

Data Availability

All deidentified data, Rmarkdown files for analyses, and the preregistration for this study are available on the Open Science Framework (https://osf.io/q6ejp/). All aspects of this study were preregistered unless otherwise specified.

Acknowledgments

The authors would like to thank members of the Cognitive Control and Aging Lab at Washington University in St. Louis for providing feedback on a version of the manuscript.