Rumination burdens the updating of working memory

Abstract

Working memory is a vital, but capacity-limited, cognitive instrument that requires frequent updating as our goals and environment change. Individuals diagnosed with depression have a reduced capacity compared to the general population, as they have a propensity to fixate on negative information, even when it is not relevant for the task at hand. Here we investigated how characteristics of psychiatric illnesses, such as rumination, affect a person’s ability to efficiently update emotional information in mind. We used both neutral and negative pictures of scenes in a working memory updating task that required participants to occasionally replace items held in mind during a brief delay period. Participants were presented with a probe item at the end of each trial and asked to report whether that item was in their current memory set. Responses were slowest and least accurate for images that had been replaced (i.e., “lures”), indicating there was some difficulty in successfully updating working memory in this paradigm. Participants who have both a high propensity to ruminate and a low working memory capacity were significantly more likely to false alarm to these lures. While emotional valence did not impact accuracy for these participants, their false alarms were faster for negative stimuli compared to neutral stimuli, indicating that task-irrelevant emotional information was more difficult to remove from working memory. These results demonstrate how rumination impairs goal-directed behavior by obscuring the boundary between relevant and irrelevant information in working memory.

INTRODUCTION

Working memory refers to the temporary maintenance, protection, and manipulation of information relevant to current goals. Because of its limited capacity (Cowan, 2001; Luck & Vogel, 2013; Miller, 1994), the ability to remove outdated information from working memory is critical to maintain its efficiency (Lewis-Peacock et al., 2018). Working memory is an essential component of cognition, and deficits in this ability are central to many psychiatric disorders such as major depressive disorder and generalized anxiety disorder (Foland-Ross et al., 2013; Gotlib & Joormann, 2010; Joormann & Gotlib, 2008; Wegner & Zanakos, 1994). There are several factors that influence performance in an individual, such as their unique working memory capacity and their proclivity to perseverate on task-irrelevant information. People with low working memory capacity are generally worse at updating, maintaining, and retrieving information (Redick & Engle, 2011).

Here we focus on updating, an essential process that involves replacing outdated information in working memory with new content (Ecker et al., 2010, 2014), which is correlated with an individual’s fluid intelligence (T. Chen & Li, 2007; Ecker et al., 2010; Friedman et al., 2006; Miyake et al., 2000). Difficulties in updating, specifically with replacing negative information, can give rise to intrusive thoughts in individuals with depression, anxiety, obsessive-compulsive disorder, and posttraumatic stress disorder (Joormann, 2010). Intrusive negative thinking, moreover, can impair people’s ability to stay focused on current task goals and inhibit attention to task-irrelevant information (Eysenck et al., 2007). More broadly, understanding the influence of emotion on memory is important because emotional information is often prioritized for maintenance in working memory (Iordan et al., 2013; Kensinger & Corkin, 2003; Murphy & Isaacowitz, 2008) and it is often the target of encoding, consolidation, and retrieval of information from long-term memory (Cahill & McGaugh, 1998; Hamann, 2001; LaBar & Cabeza, 2006).

When an individual fixates on unwanted intrusive thoughts in a way that is repetitive and negatively self-focused, they are engaging in rumination (Nolen-Hoeksema et al., 2008; Trapnell & Campbell, 1999). Rumination is a process that can consume working memory resources with task-irrelevant thoughts (Whitmer & Gotlib, 2013), and it is associated with deficits in manipulating information in mind, especially in expelling negative thoughts and memories (De Lissnyder et al., 2012; Hertel, 1997; Joormann, 2005; Joormann & Gotlib, 2008; Wyer & Jr, 1996). Rumination has long been understood as an important factor in depression, but it may also be involved in nearly all emotional disorders (Pan et al., 2020).

Rumination not only interacts with negative affect to impair individuals’ ability to solve problems, but it can also exacerbate individuals’ negative mood (Curci et al., 2013; Lyubomirsky & Tkach, 2003; Madian et al., 2019). In a study examining how individuals removed irrelevant negative information from working memory, difficulties in updating were tied to both negative affect and rumination (Joormann & Gotlib, 2008). These findings suggest that rumination negatively affects working memory updating in populations with emotional disorders. However, we still do not understand how individual differences in cognitive abilities contribute to emotional processing deficits in working memory. Because many emotional disorders are often characterized by low working memory capacity and/or high rumination, we focused here on the interplay between these two factors.

In our experiment, we tested how individual differences in working memory capacity and ruminative tendency interact to influence the ability to update working memory. We hypothesized that individuals with a higher tendency for rumination would show worse updating performance relative to individuals who ruminate less frequently. Furthermore, we hypothesized that rumination would be particularly harmful for individuals with lower working memory capacity. Because rumination tends to involve negative thoughts, we predicted that negative information would be particularly vulnerable to the detrimental effects of rumination. To test these hypotheses, we used both neutral and negatively valenced pictures in a working memory updating task that required participants to occasionally replace items held in mind during a brief delay period. Participants were presented with a probe item at the end of each trial and asked to report whether that item was in their current memory set. Working memory updating performance was assessed by evaluating participants’ ability to correctly reject probes of outdated memories.

METHODS

Participants

The study was conducted in accordance with the relevant guidelines and regulations in the Declaration of Helsinki and was approved by the Institutional Review Board (IRB) for Social Behavioral research at the University of Texas at Austin (IRB protocol #2013–10-0110). A total of 73 individuals participated in the 90-minute experiment via Prolific, an online subject recruitment platform. All participants were fluent in English, provided informed consent, and received monetary compensation ($12/hr). One participant did not complete the experiment, and five participants were excluded due to poor performance (they did not surpass an accuracy of 50% on either the change detection task or the updating task), leaving a final data set of N=67. These participants were then median split on working memory capacity (WMC, estimated from performance on the change detection task) and on rumination tendency (estimated from scores on the RRS scale). All participants above the median were labeled as high, and all below were labeled as low. Participants on the median (N=3) were excluded. The participant counts in the resulting four groups were: high WMC / high RRS (18), high WMC / low RRS (13), low WMC / high RRS (14), low WMC / low RRS (19).

Surveys

At the beginning of the study participants completed four surveys: the Mood and Anxiety Symptom Questionnaire – Short Form (MASQ; 62 items) (Watson & Clark, 1991), the Rumination-Reflection Questionnaire (RRQ; 24 items) (Trapnell & Campbell, 1999), the Ruminative Response Scale (RRS from the Response Style Questionnaire; 22 items) (Nolen-Hoeksema & Morrow, 1991), and the Penn State Worry Questionnaire (PSWQ; 16 items) (Meyer et al., 1990) in the order listed. Items within each survey were shown in random order for each subject. One item was removed from the MASQ involving suicide and a catch question was added to the RRS asking participants to mark “almost always.” All surveys were shown using PsyToolkit’s survey function.

Stimuli

The experiment was conducted online and written using PsyToolkit (Stoet, 2010, 2017). Therefore, stimuli were presented on participants’ personal computers with a range of screen sizes, resolutions, and refresh rates. The screen size was set to 800 × 600 pixels regardless of screen specs, and all sizes and locations were measured in pixels. All stimuli were presented on a gray background and a white fixation point (10 pixels). For the change detection task, colored dots (radius = 20 pixels) were shown in one of six locations. All locations were equidistant (220 pixels) from the fixation point and located at 50°, 120°, 170°, 230°, 300°, and 350°. For the updating task, images were collected from three scene image sets: NAPS, OASIS, and GAPED (Dan-Glauser & Scherer, 2011; Kurdi et al., 2017; Nowicka et al., 2010). These data were previously rated using independent subject pools and different rating scales. Valence scores were converted to similar scoring ranges to pool across all images. Images with a mean valence score in the bottom quartile (<25%) were categorized as negative images. Images with a mean valence score in the middle quartile (37.5–62.5%) were categorized as neutral images. Images with a high variation in valence reports were removed (>15% of the max valence rating). Some images were removed by experimenter discretion (e.g., images of burning crosses), some were cropped to keep consistent dimensions (200 × 160 pixels) but were removed if the cropped image removed too much of the image’s contents.

Change Detection Task

We used a change detection task with a varying number of stimuli to calculate participants’ working memory capacity (Rouder et al., 2011). For each trial, participants were shown 3 to 6 colored dots for 300ms followed by a brief delay period of 1000ms. The delay period included a 200ms visual mask that consisted of a ring of random noise (one of 6 images) to cover the area where the dots appeared (outer diameter = 500 pixels, inner diameter = 290 pixels), with an onset jittered between 200–400ms. Then a second set of dots was presented for 300ms with the same number of dots and locations as the first set. Participants were told to indicate whether any of the dots changed color from one presentation to the next. There was a 50% probability on each trial that one of the dots would change color. Participants reported a change by pressing the number 1 and no change by pressing the number 0 on their keyboard. They could respond as soon as the second display disappeared up to 2,000ms. Participants received feedback (1,000ms) by the color of the central fixation dot, where green indicated a correct response, red indicated an incorrect response, and black indicated no response. At the end of each trial there was a 1,000ms break until the next trial began. Participants had to first answer 3 practice trials correctly, and they then completed two blocks of 50 trials each. After each block, an optional 60sec break occurred. Working memory capacity for each individual was estimated by performance on this task. The “number of items” held in working memory was calculated for each set size (3, 4, 5, and 6) as k = set size * (Hits – FAs) and averaged across all four set sizes.

Updating Task

We used an updating task to assess the ability of participants to replace the contents of working memory (Fig. 1A). Participants were tasked with keeping track of exactly two relevant images in working memory at all times. All images used in this task were trial unique. Participants first had to correctly answer 5 practice trials, and they then completed a total of 6 blocks consisting of 20 trials each. Each trial showed either negative or neutral images, and this valence remained consistent throughout the entire trial. Valence was randomized for each trial but balanced across the experiment. After each block, an optional 60sec break occurred.

Figure 1 |. Working memory updating task and participant trait scores.

A) Working memory updating task with either all neutral or all negative stimuli on each trial. B) Pairwise correlations (Pearson’s r) between the four trait scores and working memory capacity (WMC). RRQ: Rumination-Reflection Questionnaire; RRS: Ruminative Response Scale; PSWQ: Penn State Worry Questionnaire; MASQ: Mood and Anxiety Symptom Questionnaire.

At the beginning of each trial, two scene images (200 pixels) were shown to the left and right of the fixation for 1,500ms followed by a delay period of 2,000ms with only a small central fixation dot. During the delay period, participants could be instructed to replace images with new ones. A new image could appear at the center of the screen with an arrow (52 × 85 pixels) above it pointing to the left or the right (with equal probability), indicating which memory item to replace. The timing matched the initial encoding period such that the new image was on the screen for 1,500ms followed by a delay period of 2,000ms.

At the end of each trial a probe image appeared in the center of the screen for 500ms, and participants reported whether the image matched one of the two images currently held in working memory. Participants had 2,000ms to respond starting from the probe onset, reporting a match by pressing the number 1 and no match by pressing the number 0 on their keyboard. Test images for each trial had an equal probability of being a match or non-match image. Match images had an equal probability of being either the oldest or the newest image added to the memory set (both referred to as match probes). Non-match images could be either a new image that was never shown during the experiment (new probe), or one of the images that was replaced on that trial (lure probe). Non-match images had an equal probability of being either a new or lure probe when at least one replacement had occurred. If no replacements had occurred, then the no match condition could only be a new probe. Participants were explicitly told to reject items that were not one of the two current memory items, even if previously seen during the trial (i.e., for lure probes). Additionally, participants were given feedback on their response accuracy at the end of every trial in the same manner as in the change detection task. There was a 2,000ms break before the next trial began.

Occasionally before a new trial began, participants were first asked to assess the success of their replacement attempt(s) on the previous trial. Participants would indicate within 10 sec by clicking one of four options that ranged from 1: completely replaced to 4: not at all replaced. This question would only appear after 20% of the ~80 trials in which at least one replacement occurred. On average, participants completed about 16 replacement assessments throughout the entire experiment, and the mean rating was 1.643 (SEM 0.077), indicating a fairly high level of believe in their successful updating of working memory.

Each trial had an equal probability of having 0, 1, or 2 replacements, and thus the total amount of time varied for each trial, as did the precise distribution of each type of trial across participants. Five predetermined image orders were created for the experiment and one was randomly assigned to each participant. Image order only determined the order in which each image would be used in the experiment, but not how each would be used. That is, the number of replacements and the probe types were randomized across trials, and this ensured that images would not be assigned to the same experimental conditions across participants, even for those assigned to the same image order.

LTM Recognition Task

Immediately following the working memory updating task, participants were given a surprise recognition memory test for all of the scene images they had just encountered. Because the number of replacements varied randomly across trials, not all participants viewed the exact same number of images during the study phase. There were about 420 images (min 410, max 436) viewed by each participant. Approximately half as many new images (~211) were added to the memory test, for a total of 631 total images tested for each participant, with a ratio of approximately 2:1 old to new images. Images were shown in random order, one at a time at the center of the screen with 4 response options below it (sure old, unsure old, unsure new, and sure new). Participants indicated their response by clicking one of the 4 options within 10 sec. The test advanced as soon as a response was recorded, and there was a 500ms delay before the next image appeared. Of the ~420 old images, exactly 120 of them were presented as probes during the updating task. Recognition memory sensitivity for these images was quite low (d-prime: M = 0.194, SEM = 0.057; Hit rate: M = 0.526, SEM = 0.019; FA rate: M = 0.457, SEM = 0.021), even though these images had been seen twice in the updating task (once at encoding and once as the probe). After excluding these images, memory sensitivity was also quite poor for the remaining ~300 images (d-prime: M = 0.173, SEM = 0.055; Hit rate: M = 0.518, SEM = 0.018). With such a large stimulus set, it is possible that participants prioritized speed over accuracy in their responses. These exploratory data will not be considered further here.

Analyses

All statistical analyses were done using standard parametric tests with the Pingouin statistics package in Python (Vallat, 2018). All confidence intervals shown in plots were calculated using bootstrapping methods where subjects were randomly sampled with replacement and iterated 1,000 times. All t-tests were two-sided and all post-hoc comparisons were FDR (Benjamini/Hochberg) corrected. A repeated measures ANOVA was used to compare probe type differences. A two-way between-subject ANOVA was used to compare group differences, and a mixed ANOVA was used to compare valence and RRS group differences.

Open Practices Statement

The data and materials for all experiments will be available on Open Science Framework.

RESULTS

Working memory capacity (WMC) for each individual was estimated by their performance on the change detection task (see Methods). The mean WMC for all participants was M = 2.39, SEM = 0.095. After a median split, the high WMC group had a mean capacity of M = 3.05, SEM = 0.063, and the low WMC group had a mean capacity of M = 1.73, SEM = 0.082. None of the symptom surveys were correlated with WMC estimates (all p’s > 0.22, Fig. 1B). All pairwise correlations between the four symptom surveys were significant (MASQ × RRQ: p = 0.046, R²_(adj) = 0.044; MASQ × RRS: p < 0.001, R²_(adj) = 0.37; MASQ × PSWQ: p < 0.001, R²_(adj) = 0.205; RRQ × RRS: p < 0.001, R²_(adj) = 0.444; RRQ × PSWQ: p < 0.001, R²_(adj) = 0.54; RRS × PSWQ: p < 0.001, R²_(adj) = 0.264). We did not collapse across these scores to preserve the unique qualities each survey reflects. Here, we focus all further analyses on groups of participants split by WMC and RRS rumination scores (see Methods).

Lure trials were more difficult to reject

Each probe type in the working memory updating task (Fig. 1A) may have required different cognitive processes to answer correctly. Therefore, we examined performance differences between the three probe types (match, new, and lure; Fig. 2A) for both accuracy (F(2, 132) = 58.16, p < 0.001, η_p² = 0.47) and reaction times (F(2, 132) = 24.25, p < 0.001, η_p² = 0.27). Lure trials had both the worst accuracy (vs. match: t(66) = 5.07, p < 0.001, d = 0.55; vs. new: t(66) = 9.87, p < 0.001, d = 1.34) and the slowest reaction times (vs. match: t(66) = 6.59, p < 0.001, d = 0.61; vs. new: t(66) = 3.74, p < 0.001, d = 0.32). Trials with no replacements were excluded from these comparisons because lure trials could only occur on trials with at least one replacement. Lure trials are the key for evaluating working memory updating performance because they probe the item that should have been replaced from working memory. We focus on these trials for all remaining analyses.

Figure 2 |. Working memory performance.

A) Reaction time (ms) shown on the x-axis and accuracy shown on the y-axis for each probe type: match, new, and lure. B) Reaction time (ms) shown on the x-axis and accuracy shown on the y-axis for each subject group: high and low groups for both working memory capacity (WMC) and rumination (RRS). C) Valence shown on the x-axis and reaction time (ms) shown on the y-axis for low WMC participants split by high and low RRS. Solid circles indicate group mean, transparent circles represent individual subjects, bars indicate bootstrapped 95% CIs. (~p < 0.05 uncorrected, * p < 0.05, ** p < 0.01, *** p < 0.001).

Lure trials reveal individual differences in updating performance

Working memory capacity estimates were positively correlated with overall accuracy on the updating task (β = 0.015, p = 0.0041, R²_(adj) = 0.106) but not with reaction times. None of the symptom surveys correlated with overall performance on the updating task. However, rumination scores from the RRS survey correlated with a difference in accuracy between probe types (lure - new: β = −0.0041, p = 0.0011, R²_(adj) = 0.139; lure - match: β = −0.0032, p = 0.0073, R²_(adj) = 0.092) and a difference in reaction times between valence types (negative - neutral: β = −1.07, p = 0.0056, R²_(adj) = 0.099). To better evaluate the influence of these factors, participants were median split on both working memory capacity and rumination scores. No accuracy differences between groups were present for match or new trials. For lure trials, there were main effects of capacity (F(1, 62) = 8.37, p = 0.0053, η_p² = 0.12) and rumination (F(1, 62) = 6.63, p = 0.012, η_p² = 0.10) on accuracy, and an interaction between these factors (F(1, 62) = 3.99, p = 0.05, η_p² = 0.06). Low-capacity individuals with high rumination scores performed worse than all other groups (vs. high WMC / high RRS: t(30) = 3.01, p = 0.025, d = 1.18; vs. high WMC / low RRS: t(25) = 2.98, p = 0.025, d = 1.13; vs. low WMC / low RRS: t(31) = 2.52, p = 0.039, d = 0.94; Fig. 2B). Despite the modest sample size (N=14) for the low WMC / high RRS group, the calculated power achieved for these statistical comparisons with the other groups was 89%, 80%, and 73%, respectively.

For reaction times, there was also an interaction between working memory capacity and rumination on lure trials (F(1, 62) = 7.18, p = 0.009, η_p² = 0.11) and match trials (F(1, 62) = 6.40, p = 0.014, η_p² = 0.10). No reaction time differences passed significance after correction for multiple comparisons, but two groups were marginally slower than the high WMC / low RRS group (vs. high WMC / high RRS: t(30) = 2.33, p = 0.09, p_(unc) = 0.03, d = 0.89; vs. low WMC / low RRS: t(31) = 2.62, p = 0.086, p_(unc) = 0.014, d = 0.94). Within each participant group, emotional valence alone did not show an effect on either accuracy or reaction time. However, for low WMC individuals, emotional valence had a marginal interaction with rumination on reaction times (F(1, 30) = 3.68, p = 0.06, η_p² = 0.11). While there was no difference for neutral lures, high ruminators with low WMC responded significantly faster to negative lures than did low ruminators (t(31) = 3.37, p = 0.004, d = 1.17; Fig. 2C).

DISCUSSION

In this study, we showed that participants had difficulty correctly rejecting memory probes for items that were recently removed from working memory, especially when those items had a negative emotional valence. Participants with low working memory capacity and high ruminative tendencies had the most difficulty in rejecting these lures. This subset of participants not only had worse accuracy for lures, but they also had marginally faster response times, indicating a deficit in removing outdated information from working memory. These participants also responded significantly faster to negative lures, likely reflecting a bias to fixate on negative information. Importantly, these participants showed no differences in accuracy or response times on non-lure trials, compared to participants with larger working memory capacity and/or less rumination. Thus, the tendency to false alarm to lures did not come at the expense of correctly identifying items that should be in working memory. This suggests that when a new item was presented during the delay period, it did not displace the prior item in working memory, but rather it was added to the existing memory set with the now-irrelevant item. Together, these results demonstrate that individuals with low working memory capacity and high rumination tendencies are particularly vulnerable to memory errors arising from the perseveration of outdated, negative information.

These findings are consistent with results observed in Curci et al. (2013), in which rumination was found to mediate negative emotional states and working memory performance. Through experimental induction of rumination in participants, researchers found that rumination depletes working memory resources and thus impairs performance. When ruminating individuals perform an updating task, the two processes of ruminating and updating compete for the limited storage and control resources available for working memory (Curci et al., 2013). Our results extend these findings by showing that for individuals with less working memory capacity, a proclivity for ruminative thoughts can make them particularly vulnerable to updating errors.

In an updating task, participants with sufficient working memory capacity may opt to maintain all items in working memory, along with the appropriate response for each item, rather than removing any outdated items. At encoding, a working memory representation becomes temporarily bound to a representation of the context in which it was perceived, such as its serial position in a list or its location in space (Barak & Tsodyks, 2014; Ecker et al., 2013, 2014; LaRocque et al., 2013; Oberauer, 2005; van Loon et al., 2018; Wilhelm et al., 2013). It can also become bound to a prospective representation of a future response, such as whether it should be accepted or rejected in a memory test (Barak & Tsodyks, 2014; LaRocque et al., 2014; van Loon et al., 2018). The removal of an irrelevant item from working memory can occur temporarily, with its context bindings remaining intact thus allowing it to be easily retrieved, or it can be removed permanently, with its context bindings being severed (Lewis-Peacock et al., 2018). Accordingly, poor performance on an updating task – as demonstrated by the low capacity / high rumination participants in our study – could either be due to a failure to permanently remove old items, or a failure (or insufficient capacity) to properly categorize items that remain in working memory as either relevant or irrelevant.

Low-capacity participants with higher rumination also responded faster to memory probes for negative stimuli compared to participants with lower rumination. This could be explained by differences in stimulus processing, or perhaps by the evolutionary value of negative stimuli. In an electrophysiological study where cognitive load of participants was controlled, participants were found to update negatively valenced faces more quickly in working memory (Kessel et al., 2016). In more automatic stages of stimulus processing, positive stimuli are preferentially updated. However, in numerous ERP studies, negative faces were better updated at higher-order stages of stimulus processing (S. Chen et al., 2012; Lang et al., 1990; Nakashima et al., 2008; Schupp et al., 2004; Smith et al., 2013; Tang et al., 2009; Weinberg & Hajcak, 2010). These results agree with the negativity bias hypothesis, which posits that animals and humans give more weight to negative entities (Rozin & Royzman, 2001). There is adaptive value to a negativity bias. For example, negative events can present threats to our evolutionary success, and these events tend to develop more rapidly than positive events (in the case of predator threats). As a result, negative events may elicit a quicker reaction, as we found in our study for responses to lure probes of negative stimuli.

We did not see a significant difference in accuracy in responses to negative and neutral stimuli in our study. Previous literature has shown that participants more accurately updated negative stimuli, whereas processing of positive faces did not differ from neutral ones (Kessel et al., 2016). We may have obtained different results with accuracy because we used scenes as stimuli, instead of faces or words, which are more commonly used in these types of paradigms. Scenes are rich in information and may externalize emotions differently than faces and words (Belham et al., 2017). Additionally, participants in our study may not have been engaging with negative stimuli in the same way that they engaged with neutral stimuli. This is consistent with comments that nine of our participants made at the end of the experiment, stating that they found the negative stimuli to be aversive and were “more keen to forget them…as fast as possible.” As a result, they may have been personally incentivized to respond faster (without sacrificing accuracy) to the memory probe in order to advance to the next trial, which could contain neutral stimuli rather than aversive stimuli.

Most research on rumination discusses it in the context of depression and other internalizing disorders. People who ruminate may have difficulty inhibiting irrelevant information and experience worsened depression or other psychopathologies, such as OCD, anxiety, binge-eating, or binge-drinking (Nolen-Hoeksema et al., 2008). However, there is a special population, people with highly superior autobiographical memory (HSAM), that despite increased propensity to ruminate yet a normal working memory capacity has exceptional memory abilities (LePort et al., 2016). Individuals with HSAM can recall autobiographical memories with a high degree of accuracy and vividness, regardless of the emotional salience of the event (LePort et al., 2012, 2017). Prior research suggests that people with HSAM experience normal encoding of memories but have enhanced consolidation and long-term recall of autobiographical events due to obsessive-compulsive tendencies and a disrupted ability to determine and filter important information (LePort et al., 2016; Daviddi et al., 2022). HSAM characteristics run counter to our findings that increased propensity for rumination coupled with a low working memory capacity can hinder the ability to update negative information in mind. Thus, our findings may not be applicable to certain groups with ruminative tendencies.

Rumination can occur temporarily as a state or chronically as a trait (Whiteman & Mangels, 2020). People with ruminative tendencies may be engaging in automatic thoughts about a negative event or stimulus. This is akin to mind-wandering, which can occur intentionally or unintentionally. However, the main difference between the two is that mind-wandering is a non-pathological form of repetitive thinking (McVay & Kane, 2010). While we measured rumination and self-reflection, we did not measure mind wandering in this study. Future research could explore how mind wandering impacts working memory updating, or perhaps seek to differentiate the influences of rumination and mind wandering.

Additionally, there is currently a gap in understanding how other clinical populations, such as individuals with attention disorders, update information in mind (Snyder et al., 2015). Future work should investigate how inattention impacts working memory updating. Finally, recent research has already found that working memory training with emotional stimuli can help reduce rumination (Pan et al., 2020). This is a deliberate exercise specifically targeting the improvement of updating ability. Our study suggests that such an intervention may be particularly beneficial for individuals with low working memory capacity.

There are two major limitations to this study that could be addressed in follow-up experiments. One limitation is the use of a change detection task for measurement of working memory capacity. In a paper from Gold et al., researchers found that asking whether the test array was identical to the prior sample array may affect measurement sensitivity of WMC (2006). A single correct or incorrect guess could have large proportional effects on the measured number of items stored especially when array sizes are much larger than a participant’s working memory capacity. Our arrays consisted of 3 to 6 items, which may have exceeded the working memory capacity of some participants. To prevent lucky guesses from skewing the results, a better way to measure working memory capacity may be to ask participants to point to the specific location of the color change (Gold et al., 2006). Another limitation is sample size. The data for this study were collected using online, paid participants. The university grant funding this project allowed for the collection of N=73 participants, of which 67 were retained after exclusions. These participants were split into four subgroups for further analysis, with N=14 in the low WMC / high RRS group. Our central findings rely on statistical comparisons of lure trial performance for this group, which had significantly lower accuracy on lure trials compared to the other three groups, and the calculated power achieved for these tests was 89%, 80%, and 73%. These tests are sufficiently powered, though of course these results would benefit from replication with a larger sample.”

In conclusion, we identified characteristics – low working memory capacity coupled with a tendency to ruminate – that can make an individual particularly vulnerable to working memory errors arising from the inability to discard outdated, negative information. These findings help us to better understand the cognitive and affective factors in individuals that can conspire to degrade goal-directed behavior.

SIGNIFICANCE.

Difficulties in updating working memory, specifically with replacing negative information, can give rise to intrusive thoughts in individuals with depression, anxiety, obsessive-compulsive disorder, and posttraumatic stress disorder. Intrusive negative thinking, moreover, can impair people’s ability to stay focused on current task goals and inhibit attention to task-irrelevant information. In this study we identified characteristics of healthy young adults – low working memory capacity coupled with a tendency to ruminate – that make an individual particularly vulnerable to working memory errors. These findings help us to better understand the cognitive and affective factors in individuals that can conspire to degrade goal-directed behavior.