Effects of Acute Exercise on Emotional Memory

Abstract

Research has demonstrated beneficial effects of acute exercise on memory for neutral materials, such as word lists of neutral valence/low arousal. However, the impacts of exercise on emotional memory is less understood. Across three laboratory experiments in college students, we tested if acute exercise could enhance both neutral and emotional memory performance, anticipating a greater effect for emotional memory. We examined effects of exercise at varying intensities (Experiment 1: high-intensity; Experiment 2: low- and high-intensity; Experiment 3: moderate-intensity), of diverse modalities (Experiment 1: treadmill jogging; Experiment 2: cycling; Experiment 3: open-skill (racquetball) and closed-skill (treadmill jogging) exercise), and on emotional memory performance assessed at increasing levels of hippocampal dependency (Experiment 1: Y/N recognition task; Experiment 2: paired-associative recognition task; Experiment 3: cued-recall task). We found that, in all experiments, acute exercise did not significantly influence emotional or neutral memory performance relative to sedentary control conditions. However, we observed several noteworthy outcomes indicating that acute exercise may be linked to improvements in memory confidence and accuracy for central aspects of emotional memory stimuli, and that select exercise modalities (e.g., treadmill exercise) may also be associated with increased frequency of memory intrusions.

Introduction

One type of episodic memory that has been extensively investigated is memory for information that evokes an emotional response (Maratos et al., 2001). Although the underlying mechanisms of emotional memory are complex, and likely moderated by their learning context, neuroscience has made considerable progress in understanding the neural underpinnings of emotional memory. For details on this topic, the reader is referred elsewhere (Buchanan, 2007; LaBar, 2007; Mather & Sutherland, 2011; Talmi, 2013). According to arousal-biased competition (Mather & Sutherland, 2011), arousal serves the function of biasing attention and neural resources towards goal-relevant information, rendering it more dominant, at the expense of irrelevant information, which is rendered less salient (Ásgeirsson & Nieuwenhuis, 2019; Markovic, Anderson & Todd, 2014; Mather & Sutherland, 2011; Mather et al., 2016). Further, noradrenaline released in response to emotional events may facilitate amygdala modulation of memory encoding and consolidation (McGaugh, 2015; Talmi, 2013). With arousal biasing memory towards emotional and against neutral information (i.e., giving – typically – emotional information a memory ‘boost’ over neutral information), we anticipate that this memory enhancement effect for emotional stimuli will be stronger when exercise induces arousal. Notably, although both acute exercise and acute stress can increase levels of stress hormones,

Acute exercise participation can modulate memory, however, prior findings varied in the direction and magnitude of this influence (Loprinzi et al., 2019a). For example, the effect of acute exercise on memory for neutral stimuli appears to be moderated by attributes of the exercise (e.g., intensity, duration, modality) and memory type (e.g., short- vs. long-term memory) (Loprinzi et al., 2019a). Less research, however, has evaluated how altering the exercise and memory task influences the effect of acute exercise on emotional memory. Therefore, the purpose of the present research was to systematically investigate the effects of acute exercise participation on emotional memory. We also examined whether modifying the materials, the hippocampal dependency of the task (e.g., recognition versus recall), or the acute exercise paradigm uniquely impacted the ability to remember emotional stimuli.

Exercise-Induced Effects on Emotional Memory

There are several plausible mechanisms through which acute exercise may influence emotional memory. A recent review (Keyan and Bryant, 2019) suggests that acute exercise may facilitate synaptic plasticity of the amygdala and hippocampus to specifically enhance emotional memory, acting on various factors (e.g., BDNF, noradrenaline, and glucocorticoid production). Indeed, acute exercise during the consolidation phase of emotional memory, was recently shown to enhance emotional memory via noradrenergic-glucocorticoid interactions (Jentsch and Wolf, 2020). Furthermore, Skriver et al.’s (2014) study in humans showed that, after a 20-minute bout of high-intensity, acute exercise, noradrenaline levels were increased immediately and five minutes post-exercise. These elevated levels of noradrenaline post-exercise were associated with greater [procedural] memory retention seven days later (Skriver et al., 2014).

Accumulating research has begun to investigate the effects of acute exercise on emotional memory (e.g., Smith, Watson, & Most, 2021). In a systematic review by Loprinzi et al. (2019), six studies evaluating this topic were identified, but due to the limited selection of relevant articles, the evidence was inconclusive. There were also mixed outcomes among the evaluated studies, underscoring the importance of additional research in this emerging line of inquiry. For example, Weinberg et al. (2014) had participants view a series of emotional images followed by engaging in acute knee flexion and extension exercises or a passive control group (experimenter passively moved the participant’s leg during knee extension/flexion). Although the active contraction group achieved better recognition accuracy for emotional stimuli than the passive group, recollection estimates (e.g., remember hit rate – remember false alarm rate) were not different between the groups. Follow-up analyses also demonstrated no valence by group interaction effects, which indicated that positive and negative images were not remembered differently across experimental groups. Similarly, Wade and Loprinzi (2018) observed no differences in recognition memory for emotional pictures from the International Affective Picture System (IAPS; Lang, Bradley, & Cuthbert, 2008) after a 14-day delay between individuals who exercised for 15 minutes prior to encoding, compared to controls. In contrast, Segal et al. (2012) demonstrated that a six-minute cycling exercise protocol, performed at 70% of VO₂ max, after viewing positively-valenced images, was effective in enhancing memory consolidation of mildly positive images in older adults.

Thus, even though empirical evidence has shown that moderate- and high-intensity acute exercise can enhance non-emotional episodic memory (Labban & Etnier, 2011; Loprinzi et al., 2021), evidence of effects on emotional memory specifically is scarce and inconclusive so far.

The Present Research

Evaluating the effects of acute exercise on emotional memory is a worthwhile endeavor. Although emotional material is often thought to be more accurately recognized than non-emotional material, memory for emotional material can be inaccurate, and may vary considerably, depending on the valence and arousal of the stimuli, as well as the retention interval (Bessette-Symons, 2018). As such, across levels of valence, arousal and retention intervals, we conducted three experiments to evaluate if acute exercise can improve the accuracy of emotional memory. We anticipated that acute exercise, relative to rest, would enhance memory for neutral and emotional stimuli, but that this effect would be greatest for emotional stimuli, as arousal induced through exercise would bias memory towards the emotional stimuli. Additionally, given that the effects of acute exercise on emotional memory performance may be contingent upon a variety of factors, including, for example, the modality, intensity, timing and duration of the exercise paradigm, the emotional stimuli, and the memory task, the present trio of experiments are a direct follow-up to Wade and Loprinzi (2018).

For Experiment 1, instead of implementing a moderate-intensity bout of exercise (Wade and Loprinzi, 2018), we incorporated a high-intensity bout of acute treadmill exercise. The motivation for this came from a recent systematic review (Loprinzi, 2018) and meta-analysis (Loprinzi et al., 2019a) demonstrating that high-intensity acute exercise more favorably influences episodic memory when compared to lower-intensity acute exercise. We also modified the temporal period in which the exercise bout occurred. Instead of implementing the acute bout of exercise prior to memory encoding (Wade and Loprinzi, 2018), for Experiment 1, we placed the bout of exercise (15 minutes) both before and after memory encoding. We chose this paradigm because a recent meta-analysis demonstrated that the greatest effect of acute exercise on episodic (neutral) memory occurs when the acute bout of exercise occurs during the early consolidation period (Loprinzi et al., 2019a). Importantly, however, recent work suggests that acute exercise can enhance episodic (neutral) memory when the bout of exercise occurs before or after memory encoding (Loprinzi et al., 2021). Thus, depending on the timing of the bout of exercise and memory task, acute exercise may enhance memory through encoding and consolidation-based processes. In Experiment 1, memory was tested 1- and 7-days after encoding.

For Experiment 2, we made several additional modifications. Specifically, instead of an old/new recognition task (Experiment 1), we evaluated performance on a paired-associative recognition task assessed at baseline and following acute exercise, using a task with known hippocampal involvement (Madan et al., 2017) to probe potential acute exercise-induced effects on hippocampal-dependent emotional memory. Associative memory binding is also known to be affected by the emotional nature of the stimuli (Mather, 2007). Prior work in animals suggests that hippocampal activation occurs in proportion to increases in exercise intensity (Ahmed & Mehta, 2012). This, coupled with prior work linking associative memory binding with the emotional nature of the stimuli, allows us to evaluate whether acute exercise disproportionately influences emotional memory in a more hippocampal dependent task. To modulate the strength of the exercise stimulus, we also extended the exercise duration from 15 minutes to 25 minutes and implemented both high- and low-intensity bouts of exercise in order to more directly compare whether high-intensity acute exercise increases emotional memory more compared to lower-intensity acute exercise. We also changed the modality of exercise from treadmill (Experiment 1) to cycling (Experiment 2), as recent work suggests that the modality of acute exercise may uniquely influence memory function (Loprinzi et al., 2019a). Lastly, Experiment 1 only included a female sample, and to increase generalizability of our findings, we included a mixed-sex sample in Experiment 2.

For Experiment 3, we altered the exercise modality to include both open- and closed-skill exercises. Open-skill exercises (e.g., racquetball) involve less predictable movement patterns relative to closed-skill exercise (e.g., treadmill exercise). We chose to change the exercise paradigm because recent research suggests that open-skilled (v closed-skilled) exercise has more favorable effects on both synaptic plasticity (Hung, Tseng, Chao, Hung and Wang, 2018) and episodic (neutral) memory function (Gu et al., 2019; for opposing results, see Cantrelle, Burnett, Loprinzi, 2020). Additionally, rather than utilizing images (IAPS) to test emotional memory (Lang et al., 2008; Wade and Loprinzi, 2018; Experiments 1 and 2 of the present study), Experiment 3 implemented dynamic emotional stimuli (i.e., video) in an effort to augment the emotional response.

Across these three experiments, we hypothesized that acute exercise would improve emotional memory when occurring before and after the memory task (Experiment 1), would increase emotional more than neutral memory in a hippocampal-dependent task (Experiment 2), and that open-skilled (v closed-skilled) exercise would have greater effects on emotional memory (Experiment 3).

Methods – Experiment 1

Study Design.

This study’s research design was a three-arm, parallel-group (between-subject) randomized controlled experiment. This study was approved by the University of Mississippi’s ethics committee (#19-006) and all participants provided written informed consent prior to participation. Participants were randomly assigned into one of three groups: (a) exercise prior to memory encoding [EPE], (b) exercise during consolidation [EDC]) or (c) a control (CON) group. We ensured allocation concealment by waiting to determine the group participants would be randomized into (via a computer-generated algorithm) until after they arrived at the laboratory. Both exercise groups jogged on a treadmill for 15 minutes, whereas the control group engaged in a time-matched seated task.

Participants completed three laboratory sessions. The first session involved an acute exercise bout (or control session) and a training phase in which participants viewed 50 images from the IAPS (Lang et al., 2008).

Two follow-up assessments measured exercise-induced effects on image recognition memory at follow-up periods 1-day and 7-days after the baseline memory training. See Table 1 below for a schematic of these study procedures.

Table 1.

Schematic of the Study Protocol

Group	Start	– – – – – – – – – – – – – – – – – – – – – – →					Finish
EPE	15-min exercise	5-min rest	Memory encoding (50 Images)	15-min rest	1-day memory recognition follow-up (100 Images)	7-day break	7-day memory recognition follow-up (100 Images)
EDC	20-min rest		Memory encoding (50 Images)	15-min exercise	1-day memory recognition follow-up (100 Images)	7-day break	7-day memory recognition follow-up (100 Images)
CON	20-min rest		Memory encoding (50 Images)	15-min rest	1-day memory recognition follow-up (100 Images)	7-day break	7-day memory recognition follow-up (100 Images)

Note. EPE, Exercise Prior to Encoding; EDC, Exercise During Consolidation; CON, Control.

Participants.

We recruited participants utilizing a convenience-based sampling approach, and focused recruitment exclusively on female participants because sex differences have been observed in emotional memory function (Canli, Desmond, Zhao and Gabrieli, 2002), and exercise-induced changes in cognition may be influenced by sex (Barha, Davis, and Falck, 2017). We aimed to recruit 15 to 20 female participants in each group, consistent with the sample size of Wade and Loprinzi (2018). Assuming a partial η² of .07 (Siddiqui & Loprinzi, 2018), alpha of .05, three groups, and two measurements per group, a total sample of 36 participants would be needed for a power of 0.80. In total, 18, 15, and 16 participants, respectively, completed the EPE, EDC, and CON groups. We oversampled participants to account for possible attrition. Participants included undergraduate or graduate students and were between the ages of 18 and 23 years of age.

Participants were excluded if they (a) self-reported as a daily smoker, (b) self-reported being pregnant, (c) exercised within 5 hours of testing, (d) consumed caffeine within 3 hours of testing, (e) had a concussion or head trauma within the past 30 days, (f) self-reporting taking marijuana or other, illegal drugs within the past 30 days, or (g) had been diagnosed with attention deficient disorder or a learning disability.

Exercise Protocol.

For the EPE and EDC groups, participants jogged on a treadmill for 15 minutes at 70% of their heart rate reserve (HRR). The HRR equation (ACSM, 2014) used to evaluate exercise intensity was: HRR = [(HR_max - HR_rest) * % intensity] + HR_rest. To calculate HR_rest, at the beginning of the visit, participants sat quietly for 5 minutes, and HR was recorded from a Polar HR monitor. To estimate HR_max, we calculated the participants’ estimated HR_max from the formula, [208 − (0.7*age)]. The treadmill speed and incline were manipulated to ensure that participants stayed within 10 beats per minute of their target heart rate (measured via chest-strapped F1 Polar monitor). Heart rate was recorded at rest and at the mid-point and end-point of the bout of exercise. Our selected exercise intensity (70% of HRR) constitutes vigorous-intensity, aerobic exercise (Garber et al., 2011).

Control Protocol.

Those in the control group played a medium-level, online administered Sudoku puzzle for 20 minutes. There is experimental evidence that playing this puzzle does not enhance cognitive function (Blough and Loprinzi, 2019); thus, this may be a suitable control group.

Emotional Memory Task.

Similar to Wade and Loprinzi (2018), participants completed a study phase and two follow-up assessments for long-term recognition of emotional memory. At the beginning of the study phase, participants were instructed that their memory of the study materials would be tested on two subsequent visits. For the study phase, participants viewed 50 IAPS images. Among these 50 images, 10 were selected from age- and gender-specific normative data to elicit a negative valence – high arousal state; 10 for negative valence – low arousal; 10 for neutral valence – neutral arousal; 10 for positive valence – high arousal; and 10 for positive valence – low arousal. Using the Self-Assessment Manikin (SAM) scale as described below, thresholds of < 4.2, 4.2-6.2, and > 6.2, respectively, were used to identify negative, neutral and positive images.

The study phase involved viewing the 50 IAPS images, with each image displayed on a computer monitor screen for 6-seconds. After this baseline study phase visit, participants returned for two testing visits, occurring 1-day and 7-days later.

Encoding.

During the study phase, immediately after viewing each image, participants made valence and arousal judgements using the pictorial SAM scale (Bradley and Lang, 1994). Using images anchored with numeric values ranging from 1 (low) to 9 (high), participants were asked to rate (for valence and arousal separately) how each image made them feel. Valence was assessed as follows, “At one extreme (9) of this scale you can feel happy, pleased, satisifed, contented, and/or hopeful. The other end of this scale (1) is when you feel completely unhappy, annoyed, unsatisied, melancholic, despaired, and/or bored” Arousal was assessed as follows, “At one extreme (9) of this scale you can feel stimulated, excited, frenzied, jittery, wide-awake, and/or aroused. The other end of this scale (1) is when you feel relaxed, calm, sluggish, dull, sleepy, and/or unaroused.”

Retrieval.

The procedures for the two follow-up visits were identical. For each follow-up visit, the participant returned to the laboratory and viewed 100 IAPS images. These 100 IAPS images included the 50 images from the training visit as well as 50 new (unseen) images that were matched (to the original 50 images) for valence and arousal levels from the 5-above mentioned emotional type classification. This matching occurred by using the age- and gender-specific normative data from the IAPS database (Lang et al., 2008).

During the two follow-up assessments, participants viewed 100 images (presented in a random order), with each image displayed on a computer monitor screen for 3-seconds. After each image, participants selected one of three responses, including “remember”, “know”, or “new”. Detailed explanations of these responses were provided. Briefly, “remember” responses are taken as an index of recalling the stimuli along with associated contextual details, whereas “know” responses point to familiarity with the stimuli, without true recollection. New responses indicate that the participant perceives the stimuli as completely novel (see Migo et al., 2012).

Additional Measures.

Demographic information, including age (years) and race-ethnicity were collected via a self-report questionnaire. Body mass index (kg/m²) was calculated from measured height and weight. Lastly, self-reported moderate-to-vigorous physical activity (min/week) was determined from the Physical Activity Vitals Sign questionnaire (Ball et al., 2016).

Statistical Analyses.

Frequentist and Bayesian repeated measures ANOVAs (rmANOVA) were conducted. For the physiological (heart rate) manipulation check, a 3 (Time: resting, midpoint, endpoint) x 3 (Group: EPE, EDC, CON) rmANOVA was computed. The main analyses included a rmANOVA models, including the following factors: Group (three levels: EPE, EDC, CON), Time (two levels: 1 Day, 7 Day follow-up), Emotion (five levels: Negative, High Arousal; Negative, Low Arousal; Neutral, Low Arousal; Positive, High Arousal; Positive, Low Arousal), Dimension (two levels: Valence, Arousal), Response Type (three levels: Remember, Know, New), and Item Type (two levels: Old Item, New Item). Group was a between-subject factor, whereas Time, Emotion, Response Type, and Item Type were within-subject factors. All Frequentist post-hoc testing used Holm-corrected post-hoc comparisons.

To supplement the Frequentist analyses, Bayesian analyses were utilized to test the robustness of the examined effects. Unlike Frequentist analyses, Bayesian analyses allow for the ability to obtain evidence in favor of the null hypothesis and discriminate between “absence of evidence” and “evidence of absence” The inclusion Bayes factor (BFi) is reported, which represents the change from prior to posterior inclusion odds, in a ratio reflecting support for the effect being included versus support for the effect being excluded. We follow the convention that a BF > 3 indicates moderate evidence in favor of the alternative hypothesis, whereas a BF< 1/3 indicates moderate evidence in favor of the null hypothesis (see Table 1 in Wagenmakers et al., 2018). Notably, herein, all Bayes factors > 1000 are reported as > 1000 since very large BF values should not be interpreted literally. Finally, sensitivity analyses were computed that included self-reported physical activity (minutes of moderate-to-vigorous physical activity per week) as a factor in the rmANOVA, but physical activity did not interact with group assignment to influence memory.

Results – Experiment 1

Participant Characteristics.

Table 2 displays the characteristics of the study variables. Participants, on average, were 20 years of age. The sample was predominately white (68.8% to 80.0%).

Table 2.

Characteristics of the Study Variables

Variable	EPE (N=18)	EDC (N=15)	CON (N=16)	P-value
Age, mean years	20.6 (2.0)	20.5 (1.2)	20.4 (1.6)	.92
Race-Ethnicity, %
White	72.2	80.0	68.8	.69
BMI, mean kg/m2	22.9 (4.7)	23.3 (3.1)	23.4 (7.1)	.94
MVPA, mean min/week	169.2 (97.7)	164.3 (120.5)	191.2 (127.1)	.91

Note. BMI, Body mass index; MVPA, Moderate to vigorous physical activity; EPE, Exercise Prior to Encoding; EDC, Exercise During Consolidation; CON, Control; Values in parentheses are SD estimates. MVPA was assessed at the beginning of the first visit and evaluated using the Physical Activity Vitals Sign questionnaire (Ball et al., 2016).

Manipulation Checks

Physiological Response.

Figure 1 displays the heart rate results (rest, midpoint, endpoint) across the three experimental groups. Significant main effects of Group, F(2, 46) = 164.3, p < .001, η² = .37, BF > 1000, and Time, F(2, 92) = 104.2, p < .001, η² = .36, BF > 1000, were qualified by a Group × Time interaction, F(4, 92) = 274.7, p < .001, η² = .19, BF > 1000. There were no group differences at rest, ps > .05, no midpoint and endpoint differences across the two exercise groups, ps > .05, but significant midpoint and endpoint differences were observed when comparing the exercise conditions to the control condition, ps < .001.

Figure 1. Heart Rate Responses in Experiment 1.

Note. HR was measured at rest, midpoint, and at the end of the intervention across the three experimental groups. Error bars represent 95% CI. EPE, Exercise Prior to Encoding; EDC, Exercise During Consolidation; and CON, Control.

Valence and Arousal Response.

Figure 2 displays the valence and arousal responses across the three experimental groups. Results did not differ based on group, with the dimension and emotional responses occurring as expected. For example, higher levels of reported valence occurred for pictures normed as positive valence, and similarly, higher levels of reported arousal occurred for pictures normed as being high arousal images.

Figure 2. Valence and Arousal Ratings of the Studied Images.

Note. Error bars represent 95% confidence intervals. EPE, Exercise Prior to Encoding; EDC, Exercise During Consolidation; and CON, Control. For both valence and arousal, numeric values ranged from 1 (low) to 9 (high).

In a 5 (Emotion: Negative, High Arousal; Positive, High Arousal; Negative, Low Arousal; Neutral, Low Arousal; Positive, Low Arousal) × 2 (Dimension: Valence, Arousal) × 3 (Group: EPE, EDC, CON) rmANOVA, there was not a main effect for Group, F(2, 46) = .275, p = .76, η² = .002, BF = .12, but there was a main effect for Emotion, F(4, 184) = 65.52, p < .001, η² = .15, BF > 1000, and a main effect for Dimension, F(1, 46) = 7.68, p = .008, η² = .03, BF = 885.13. Group did not interact with Emotion, F(8, 184) = .81, p = .59, η² = .004, BF = .008, nor did Group interact with Dimension, F(2, 46) = .135, p = .87, η² < .001, BF = .07. Similarly, there was not a Group × Dimension × Emotion interaction, F(8, 46) = .48, p = .87, η² = .002, BF = .017. The main effects for Emotion and Dimension were qualified by an Emotion × Dimension interaction, F(4, 46) = 120.63, p < .001, η² = .26, BF > 1000.

Memory Results

See Appendix A for the full point estimate memory results, stratified by Group, Time, Emotion, Response Type and Item Type. In a 3 (Group) × 2 (Time) × 5 (Emotion) × 3 (Response Type) × 2 (Item Type) rmANOVA, there were no exercise-induced effects. There was no main effect for Group, and Group was not involved in any further high-level interactions either, all ps > .05. Although emotional memory did not vary as a function of exercise, memory was influenced by Time (greater accuracy at the earlier test interval) and Emotion (superior memory for negative, high arousal items). These results did not involve Group and are provided in Appendix B.

Methods – Experiment 2

Study Design.

A three-arm randomized controlled experiment was employed. Participants were excluded if they reported a cardiovascular or respiratory illness/disease. Eligible participants were randomized into one of three groups, including high-intensity cycling exercise, low-intensity cycling exercise, and a non-active control group. This study was approved by the University of Alberta’s ethics committee (PRO00077772) and all participants provided written informed consent prior to participation. After the participant arrived at the laboratory, the experimenter drew a folded piece of paper from one of two boxes (divided by gender to ensure equal representation of men and women in each group), which contained the participant’s group assignment (high-intensity exercise, low-intensity exercise, or control). Thus, allocation concealment was maintained by having the researcher and participant not know which group the participant was randomized into until after arriving in the laboratory.

Both exercise groups cycled on a stationary bike for 25 minutes, whereas the control group engaged in a time-matched seated task. See Table 3 below for a schematic of the study procedures.

Table 3.

Schematic of the Study Protocol

Group	Start	– – – – – – – – – – – – – – – →		Finish
LEX	List 1: Association-Memory Task (104 images)	25-min Exercise	List 2: Association-Memory Task (104 images)	5-min Debriefing Period
HEX	List 1: Association-Memory Task (104 images)	25-min Exercise	List 2: Association-Memory Task (104 images)	5-min Debriefing Period
CON	List 1: Association-Memory Task (104 images)	25-min rest	List 2: Association-Memory Task (104 images)	5-min Debriefing Period

Note. LEX, Low intensity exercise; HEX, High intensity exercise; CON, Control.

Participants.

Participants were recruited via an online portal delivered by the Department of Psychology at the University of Alberta. A total of 105 undergraduate students from Fall 2019 Introductory Psychology courses participated in exchange for partial course credit. Participants included undergraduate or graduate students, ages 16 to 35 years whom were free of cardiovascular and respiratory disease. The sample size was based on previous single-group experiments that used the same or close variants of the emotional association memory task in experiment 2, targeting a within-subject memory measure in Madan et al. (2017), Caplan et al. (2019), and Fujiwara et al. (2021). Using the original data, Cohen’s d values were derived from means (SD) of associative recognition memory accuracy for emotional and neutral pairs and the intercorrelation between the two and were as follows: Madan et al. (2017), experiment 1: d = 0.798 (based on M_emotional= 0.610 ± 0.207, M_neutral= 0.695 ± 0.161, r=0.859); experiment 2: d = 0.452 (based on M_emotional= 0.312 ± 0.220, M_neutral= 0.380 ± 0.290, r=0.903); experiment 3: d = 0.722 (based on M_emotional= 0.526 ± 0.158, M_neutral= 0.587 ± 0.16578, r=0.862); Caplan et al. (2019): d = 0.695 (based on M_emotional= 0.474 ± 0.177, M_neutral= 0.553 ± 0.204, r=0.830); Fujiwara et al. (2021); d = 0.519 (based on M_emotional= 0.467 ± 0.171, M_neutral= 0.507 ± 0.214, r=0.944). Using G*Power (v.3.1.9.7) on these effect sizes, and assuming an alpha level of 0.05 and a power level of 0.80 resulted in required sample sizes of 15 (exp.1 of Madan et al., 2017), 31 (exp.2 of Madan et al., 2017), 18 (exp.3 of Madan et al., 2017), 19 (Caplan et al., 2019), and 32 (Fujiwara et al., 2021) participants. Based on the smallest observed effect size, we conservatively oversampled to a target of 35 subjects per group to account for potential data loss. Notably, even though these previous studies employed the same or very similar task, none of them had a between-subjects design. Therefore, the projected required sample size should be treated as an estimate. Of the original 105 participants, 95 (42 males) were retained. Participants were excluded due to failure to adhere to the exercise protocol (n = 6), below-chance memory performance (n = 1), and external interruption (fire alarm; n = 1). Prior to the exercise intervention, The Godin Leisure-Time Exercise Questionnaire (LTEQ; Rowe, Mahar, Raedeke, & Lore, 2004; Shephard, 1997) was administered to evaluate the frequency of vigorous, moderate, and light-intensity exercise activity performed on a weekly basis. A weekly leisure activity score was calculated using the following formula, (times/week of vigorous exercise × 9) + (times/week of moderate exercise × 5) + (times/week of light exercise × 3). Weekly activity scores of 24+, 14-23, < 14, respectively, represent being active, moderately active and insufficiently active/sedentary. Self-reported physical fitness measured with the LTEQ is moderately correlated to objective physical fitness levels (e.g., VO_2max scores (r = .56; Copeland et al., 2005).

Emotional Memory Task.

The association memory task (Madan et al., 2017; Caplan et al., 2019; programmed in Presentation ^®, Neurobehavioral Systems) is composed of a set of 208 pictures (104 negative, arousing pictures and 104 neutral, non-arousing pictures), from the IAPS database (Lang et al., 2008) and the internet. Each picture (Madan et al., 2017) was rated for level of arousal on a 9-point modified version of the Self-Assessment-Manikin (Bradley & Lang, 1994). A score of ‘1’ indicates high arousal, and negative pictures were rated higher in arousal (M ± SD = 5.09 ± 0.85) than neutral pictures (M = 7.70 ± 0.35; t(212) = 35.74, p < .001; Madan et al., 2017). Participants performed the memory task twice, before the intervention (List 1) and after the invention (List 2). Each list was comprised of 52 same-valence pairs (26 negative-negative pairs, 26 neutral-neutral pairs). There was no repetition of images between List 1 and List 2.

Encoding.

Each list contained an encoding phase and a retrieval phase. In the encoding phase, pairs were presented one at a time, with both pictures (each picture measured 450×300 pixels) adjacent to one another on a computer screen for 2000-ms preceded by a fixation cross for 1000 msec. Pairs were always of the same valence, either two negative pairs, or two neutral pairs. Participants were explicitly instructed to study the pairings and informed that their memory for each pair would be tested later. The encoding phase concluded with a visual 2-back task to interrupt memory rehearsal (see (Madan et al., 2017) for additional details).

Retrieval.

Following the encoding phase, in the retrieval phase each pair was first probed with a judgment of memory (JoM) task and a 5-alternative forced choice (5-AFC) associative recognition task. A probe image was presented from one the 52 pairs shown in the encoding phase, followed by the JoM task intended to emulate cued recall of the pictures. During the JoM task, participants were prompted by the question: “Recall Associate?” and presented with the options “Yes” or “No”. Participants were instructed to be “conservative” with their memory judgments and to only select a ‘yes’ response if they were confident that they correctly remembered the picture that was previously associated with the probe picture. One trial of JoM lasted 4900-ms, followed by a 100-ms blank screen and 1000-ms fixation-cross. Following the JoM task, participants were presented with the 5-AFC association recognition task. The same probe picture as in the JoM task was presented in the center of the screen (225 × 150 pixels), surrounded by an array of five pictures (one correct target, four lures, all lure pictures were of the same emotional valence and selected from other pairs in the encoding phase) in fixed screen positions. Participants were given 3900-ms to select the correct target picture, followed by a 100-ms blank screen. The retrieval phase concluded with the same 3-minute 2-back task as was used after the encoding phase. Participants were debriefed after the study. This included 5-minutes of providing a list of mental health resources in case participants experienced negative aftereffects from viewing the images.

Exercise Protocol.

The high-intensity exercise group (HEX; n = 31) consisted of a 5-minute warm-up phase on a stationary spin bike, a 5-minute “ramp-up” phase to increase the participant’s HR to at least 85% of their calculated HR maximum (220 beats per min – age; Nes et al., 2013), and 15 minutes of cycling at 85-90% of their calculated HR maximum. Heart rate was recorded from an electrocardiographic chest strap heart rate monitor (Garmin ‘DUAL’ chest heart rate monitor). Participants’ HR was displayed via Bluetooth on a Garmin Forerunner 35 watch, which only the experimenter viewed throughout the exercise task (Gillinov et al., 2017; Labban & Etnier, 2011). Heart rate was manually recorded at rest and every two minutes during the exercise bout (Gillinov et al., 2017). Along with gear changes, subjects were encouraged to maintain constant revolutions per minute (RPM) around 65-75. Exercise intensity was controlled by the experimenter via the adjustments of the bike gears accordingly to the participant’s heart rate.

The low-intensity exercise group (LEX; n = 33) consisted of 25 minutes of cycling at or below a heart rate of 100 bpm. For both exercise groups, participants verbally rated their perceived level of exertion every 2-4 minutes (high-intensity exercise) or every 4-7 minutes (low-intensity exercise) on a 10-point rating scale (Williams, 2017) that described and visually illustrated increasing levels of exertion.

Control Protocol.

The control group (CON; n = 31) included sitting while viewing the video “David Attenborough Desert Seas National Geographic” (Wild Nature TV, “David Attenborough Desert Seas National Geographic”, 2017, 0:00-25:00) on YouTube for 25 minutes (i.e., the same duration as the two exercise groups). The movie showed various marine activities and contained no graphic or violent scenes (animal death/killings, etc.) to limit the potential for physiological arousal caused by the movie. During both HEX and LEX, the same nature video was shown to match the experimental groups.

Statistical Analyses.

Frequentist and Bayesian rmANOVAs were conducted. For the physiological (heart rate) manipulation check, a 5 (time: rest, during the 5-minute warm-up, 10-minute midpoint interval, and the final 10-minutes) × 2 (group: LEX, HEX) rmANOVA was computed. The main analyses included the following factors: Group (HEX, LEX, CON), List (List 1 – pre-intervention, List 2 – post-intervention) and Emotion (negative pairs, neutral pairs). Post-hoc tests were Bonferroni-corrected across groups but not across analyses. Dependent variables were 5-AFC recognition accuracy (hits) and a d’-prime equivalent to account for guess rates in n-AFC paradigms (Hacker & Ratcliff, 1979). Higher d-prime scores indicate better memory, adjusted for the guess rate of 20% in this 5-AFC task.

Results – Experiment 2

Participant Characteristics.

Table 4 displays the characteristics of the study variables. Participants, on average, were 19 years of age. The sample was also partially white (45%-48%).

Table 4.

Characteristics of the Study Variables

Variable	LEX (N=33)	HEX (N=31)	CON (N=31)	P-value
Age, mean years	19.3 (1.9)	19.7 (1.2)	19.1 (1.4)	.41
Gender (male)	14 (42.4%)	14 (45.2%)	15 (48.4%)	.76
Race-Ethnicity, %				.93
White	16 (48.5%)	14 (45.2%)	15 (48.4%)
BMI, mean kg/m2	24.8 (9.5)	24.0 (4.5)	23.3 (3.2)	.70
Activity score, mean LTEQ	56.8 (24.9)	59.3 (22.2)	61.1 (27.7)	.79

Note. BMI, Body mass index; LTEQ: Godin-Leisure Time Exercise Questionnaire; LEX, Low Intensity Exercise; HEX, High Intensity Exercise; CON, Control; Values in parentheses are SD estimates, or percentage of the sample. BMI data was from measured height and estimated (self-reported) weight; the sample size for the BMI data across the three respective groups was 28, 27, and 23. One missing value for race-ethnicity for the CON group.

Manipulation Checks

Physiological Response.

Figure 3 displays the HR results at rest, during the 5-minute warm-up, 10-minute midpoint interval, and the final 10-minutes of the intervention across the experimental groups. Significant main effects of Group, F(1,62) = 285.29, p < .001, η² = .99, and Time, F(1.97,121.83) = 625.73, p < .001, η² = .91, were qualified by a Group × Time interaction, F(1.97,121.83)=145.70, p <.001, η² = .70. Post-hoc t-tests showed significant differences between groups in all time points after the resting phase, with higher heart rates in the high exercise group than the low-intensity exercise group (all p’s <.001). As intended, HEX yielded a significant increase in HR from rest to the midpoint of the intervention, remaining stable until the end of the exercise protocol. LEX yielded a lower, stable HR from the beginning of the warmup until the end of the exercise protocol.

Figure 3. Heart Rate Responses in Experiment 2.

Note. HR was measured at rest, 5-minute warm-up, 10-minute midpoint interval, and in the final 10 minutes across the low- and high-intensity experimental groups. Error bars represent 95% CI.

Memory Results

A 3 (Group) × 2 (List) × 2 (Emotion) rmANOVA was conducted to test whether the exercise intervention influenced associative memory. Significant main effects for List, F(1,92)=13.7, p < .001, η²=.13 and Emotion, F(1,92) = 14.2, p < .001, η²=.13, but not Group, F(2,92) = .26, p = .77) were observed. These main effects indicated better associative memory for pairs from List 2 (M= 63.0%, SD= 0.18) than List 1 (M=58.1%, SD= 0.16), and for neutral pairs (M= 62.7%, SD= 0.18) compared to negative pairs (M=58.4%, SD = 0.16). There were no significant interactions involving Group. Bayes factors for the main effect of List (BF=209.20) and Emotion (BF=34.62) were in clear support of the significant main effects. Bayes factors for the non-significant main effect of Group (BF=0.083), and the non-significant interactions involving Group (Group × Emotion: BF=0.03; Group x List: BF= 0.02; Group x Emotion x List: BF=0.000133) were all definitively supportive of null findings (see Figure 4A).

Figure 4. Associative Recognition Accuracy.

Note. A) proportion of recognition hits, B) d’-prime. Error bars are 95% confidence intervals around the mean.

These analyses were repeated with the d’-prime accuracy outcome measure, correcting for guess rates. Similar to the results with simple recognition hits, d’-prime outcomes yielded a main effect of List, F(1,92) = 19.2, p < .001, and a main effect of Emotion, F(1,92) = 12.8, p=.001. Bayesian statistics confirmed robust main effects of List (BF>1000) and Emotion (BF=34.21). Bayes factors for the non-significant main effect of Group, F(1,92) = .45, p = .64, BF=0.08), and interactions that involved the group factor, i.e., Group × Emotion (BF=0.04), Group × List (BF= 0.03), as well as the three-way interaction Group × Emotion × List (BF=3.31x10⁻⁴) were all definitively supportive of null findings (see Figure 4B). Additional sensitivity analyses were computed evaluating if biological Sex interacted with Group to influence d’-prime, with findings demonstrating that Sex did not interact with Group, p = .94, List, p = .95, Emotion, p = .76, or any higher-order three-way interactions, all ps > .05.

Methods – Experiment 3

Study Design.

A three-arm, randomized controlled experiment was employed. Participants were randomized into one of three groups, including a closed-skill exercise group, open-skill exercise group, and a control group. This study was approved by the University of Mississippi’s ethics committee (#19-046), with participant consent provided prior to participation. Randomization was performed using a computerized algorithm. Allocation concealment was maintained by having the researcher and participant not know which group the participant was randomized into until after arriving in the laboratory.

Both the closed- and open-skill exercise groups exercised at a moderate-intensity for 30 minutes. The closed-skill group exercised on a treadmill, the open-skill group played racquetball, and the active control group engaged in a series of light stretches. See Table 5 below for a schematic of the study design, including the temporal periods of assessments.

Table 5.

Overview of the Study Protocol

Group	Start	– – – – – – – – – – – – – – – – – – – – – – →					Finish
Open-Skill Exercise	Pre-Assessments of Valence and Arousal	Watch 10-min Car Crash Video	Post-Assessments of Valence and Arousal	30-min of moderate-intensity racquetball	7-min rest †	Cued Emotional Memory Recall	Surveys (e.g., Impact of Event)
Closed-Skill Exercise	Pre-Assessments of Valence and Arousal	Watch 10-min Car Crash Video	Post-Assessments of Valence and Arousal	30-min of moderate-intensity treadmill exercise	7-min rest †	Cued Emotional Memory Recall	Surveys (e.g., Impact of Event)
Control	Pre-Assessments of Valence and Arousal	Watch 10-min Car Crash Video	Post-Assessments of Valence and Arousal	30-min of stretching	7-min rest	Cued Emotional Memory Recall	Surveys (e.g., Impact of Event)

Note.

^†Involved a 2-minute recovery walk followed by 5-minutes of seated rest.

Participants.

The total sample size included 193 participants, including 65, 65, and 64 participants in the closed-skill, open-skill, and control groups, respectively. Among the 193 participants, 165 completed the exercise bout with a mask on, due to concerns with the COVID-19 pandemic. This sample size was based on an a-priori power analysis using data from previous experiments that used the same emotional video employed in the present study. Partial eta-squared estimates (derived from their ANOVA f-values, df, and using a 95% CI)¹ were as follows for these previous experiments: Keyan and Bryant (η_p² = 0.1379; from, F(2,51) = 4.08), Devilly et al. (η_p² = 0.1412; from, F(2,58) = 4.77), and Gittins et al. (η_p² = 0.0614; from, F(1,75) = 4.91). Based on these estimates, we utilized the lower (more conservative) η_p² of 0.0614 (Gittens et al.). Using G*Power (v.3.1.9.2), and assuming 3 groups, alpha level of 0.05, power level of 0.80, and an effect size f of 0.2557 (calculated from η_p² of 0.0614), a total sample size of 153 individuals would be needed. We intentionally oversampled given the following reasons: (1) to account for potential missing data (e.g., non-compliance, technical errors, etc.) and (2) although these prior experiments employed the same emotional video as used in the present experiment, notably, they were not answering the same research question as the present study. Given the latter point, and like most a-priori power analyses, the estimated effect size is a rough guess, and as such, we employed a conservative a-priori power analysis. Notably, our sample size (N = 193) is considerably higher than previous studies (Keyan and Bryant, N = 54; Devilly et al., N = 61; Gittins et al., N = 85).

The recruitment protocol was analogous to Experiment 1. Participants included undergraduate and graduate students at the University of Mississippi between the ages of 18 and 28. In addition to the exclusionary criteria used in Experiment 1 (except for ADHD diagnosis), participants were excluded if they consumed alcohol daily or were color blind. Daily alcohol use was classified as >30 drinks/month for women, >60 drinks/month for men, or consumption of alcohol within 24-hours of their laboratory visit. Color blind status was assessed objectively by showing individuals (on a computer screen) rectangular shapes in several different colors and asking them to identify the color. Participants incorrectly responding to any of the colors were excluded from the study.

Experimental Groups.

Participants were randomized into one of three experimental groups, including closed-skill exercise (treadmill), open-skill exercise (racquetball) or control (stretching). The exercise conditions involved participants exercising (either jogging on treadmill or engaging in racquetball) for 30-minutes at 60% of their HRR. Heart rate reserve was determined via the same equation used in Experiment 1. Heart rate was assessed every 5 minutes throughout the exercise bout. The treadmill speed and incline were manipulated to keep the participant’s heart rate within 5 beats per minute of their estimated 60% of HRR. The same applied for the racquetball participants, as they were asked to play harder or reduce effort in order to keep their heart rate within 5 beats per minute of their estimated 60% of HRR. Racquetball was played as a two-person game (researcher and participant). Notably, 60% of HRR represents moderate-intensity exercise (Garber et al., 2011).

Unlike in Experiments 1 and 2, and as suggested elsewhere (Pontifex et al., 2019), an active control group (stretching) was employed in Experiment 3 in order to create a contact-control condition that more closely aligns with aspects of the exercise protocol (e.g., muscular activation, social contact). The active control stretching condition was modeled after previous work (Edwards, Rhodes, & Loprinzi, 2017). That is, for 30 minutes, participants engaged in light stretches that were guided by the researcher. This involved a series of upper and lower body stretches that were held for an approximate 15-second period, with an emphasis on light stretching (not to the point of discomfort). The stretching group included two, 12.5-minute bouts. Participants engaged in guided stretches for 12.5 minutes, rested for 5 minutes (gentle conversation with the researcher), and then re-completed the stretches for an additional 12.5 minutes (i.e., 30 minutes in total).

Emotional Memory Task.

Participants were asked to watch a 10-minute video that depicts the scene of a car crash. The following intentional encoding instructions were given to the participants before they viewed the video. “Please closely watch this 10-min video. Please be as attentive as possible during this video, as later I will be asking you to recall specific information from the video.”

Encoding.

The 10-minute video was watched in an enclosed, isolated unit, with audio played via headphones (self-selected volume). This video has been used in other studies evaluating emotional memory (Devilly, Varker, Hansen, Gist 2007; Gittens, Paterson, Sharpe, 2015; Keyan and Bryant, 2017). In these prior studies, this video has demonstrated evidence of construct validity, as noted by increased distress scores following the video viewing (Gittens, Paterson, Sharpe, 2015).

Immediately prior to and after viewing the video, participants completed two single-item assessments to evaluate perceived valence and arousal. The Feeling Scale was used to evaluate the affective response to the video, which includes the statement, “Please circle the most appropriate number to represent how you feel right now, in this very moment.” The response options ranged from −5 (very bad) to +5 (very good). The Felt Arousal Scale was used to evaluate the physiological arousal from the video. This scale includes the statement, “Estimate here how aroused you actually feel. Do this by circling the appropriate number. By “arousal” we mean how “worked-up” you feel. You might experience high arousal in one of a variety of ways, for example, as excitement or anxiety or anger. You might also experience low arousal in a number of different ways, for example, as relaxation or boredom or calmness.” The response options ranged from 1 (low arousal) to? 6 (high arousal).

Participants were also asked to complete an Impact of Event Survey (Horowitz, Wilner, Alvarez, 1979), which assesses three forms of distress that people sometimes have after stressful life events. It was used in the present study to assess memory intrusions, or how distressing the emotional video was during the 35-minute memory consolidation period (i.e., during the bout of exercise/stretching). The three items included the statements, “I thought about it when I didn’t mean to; Pictures about it popped into my mind; I had waves of strong feelings about it.” For each of these three items, a 5-point Likert scale was used (0 = not at all; 1 = a little bit; 2 = moderately; 3 = quite a bit; 4 = extremely). In the present sample, internal consistency (Cronbach’s alpha) for these three items was 0.85.

Retrieval.

For the emotional memory assessment and pilot procedure used to develop this instrument, see Appendix C. In alignment with other studies (Devilly, Varker, Hansen, Gist, 2007; Gittens, Paterson, Sharpe, 2015; Keyan and Bryant, 2017), a 25-item survey was used to evaluate the participant’s emotional memory. This 25-item assessment included 12 peripheral and 13 central questions (see Appendix C), involving questions related to the three constituents of episodic memory (i.e., “what”, “where” and “when” questions). Similar to other studies (Devilly et al., 2007), after each item, participants rated their confidence in their response, ranging from 1 (not at all confident) to 5 (extremely confident). Two composite confidence metrics were calculated, including a peripheral details confidence and central details confidence scores, involving their average confidence responses across the peripheral and central items.

Additional Assessments.

Based on the potential to influence the relationship between acute exercise and emotional memory, participants completed a series of surveys (and had measurements taken; body mass index) at the end of the experiment. In addition to the demographic parameters included in Experiments 1 and 2, we also estimated (from a formula using self-reported physical activity, age, gender and measured body mass index) cardiorespiratory fitness (Jackson et al., 1990), experience playing racquetball, chronotype (self-reported morning-type versus evening-type; Loureiro et al., 2015), and emotional reactivity (negative and positive reactivity subscales; Preece et al., 2019).

Analyses.

Frequentist and Bayesian repeated measures ANOVAs (rmANOVA) were conducted. These analyses included the following factors: Group (three levels: Closed-Skill Exercise, Open-Skill Exercise, Control), Memory Detail (two levels: Peripheral, Central), Memory Confidence (two levels: Peripheral Confidence, Central Confidence), Arousal/Valence (two levels: Pre-Video, Post-Video), Impact of Event Survey Item Type (three levels, Item 1, 2, 3), and Exercise/Control Heart Rate Response (eight levels: Pre, 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes, 30 minutes, Post). Group was a between-subject factor, whereas all other factors were within-subject factors. Violations of sphericity were corrected with Huynh-Feldt corrections. All Frequentist post-hoc testing used Holm-corrected post-hoc comparisons.

Results – Experiment 3

Participant Characteristics.

Table 6 displays the characteristics of the sample. Participants, on average, were 20 years of age. The sample was also predominately white (78-92%).

Table 6.

Participant Characteristics Across the Experiment Groups

Variable	Closed-Skill Exercise (n = 65)	Open-Skill Exercise (n = 64)	Control (n = 64)	P-Value
Demographics
Age, mean years	20.75 (1.5)	20.48 (1.1)	20.82 (1.6)	.37
Gender, % Female	66.2	73.4	59.4	.24
Race, % White	78.5	92.2	81.3	.41
BMI, mean kg/m2	25.37 (5.3)	25.83 (4.8)	26.21 (5.8)	.65
Medication/Diagnosis
Taking mood medication, %	1.5	3.1	4.7	.59
ADHD diagnosis, %	7.7	3.1	7.8	.46
Behavioral
MVPA, mean min/week	219.7 (175.5)	193.0 (175.1)	205.6 (175.8)	.69
Cardiorespiratory fitness, mean mL/kg/min	41.23 (7.6)	40.06 (7.6)	41.29 (8.4)	.59
Racquetball experience, %				.88
Never played	83.3	80.6	76.5
Previously played	16.7	19.4	23.5
Other
Chronotype, %				.19
Definitely morning-type	11.3	12.7	19.0
Rather more morning- than evening-type	27.4	12.7	25.4
Rather more evening- than morning-type	32.3	47.6	28.6
Definitely evening-type	29.0	27.0	27.0
Time of assessment (military time), mean	1457.5 (276.0)	1397.5 (256.8)	1379.4 (247.4)	.22
Emotional reactivity, mean
Negative reactivity	25.62 (8.7)	23.61 (7.2)	25.48 (7.1)	.26
Positive reactivity	35.86 (6.1)	37.45 (5.9)	35.47 (5.5)	.13

Note. ADHD, attention deficient hyperactive disorder; BF, Bayes factor (inclusion); BMI, body mass index; MVPA, moderate-to-vigorous physical activity; ANOVA was used to calculate p-values for continuous variables, whereas a chi-square analysis was employed for categorical variables.

Manipulation Checks

Physiological Response.

Figure 5 displays the physiological (heart rate) response to the exercise/control groups. In a 3 (Group: Closed-Skill, Open-Skill, Control) × 8 (Time: Pre, 5, 10, 15, 20, 25, 30, Post) rmANOVA, there was a main effect for Group, F(2, 190) = 358.4, p < .001, η² = .44, BF > 1000, and Time, F(4.96, 943.9) = 682.6, p < .001, η² = .28, BF > 1000, which was qualified by a Group × Time interaction, F(9.94, 943.9) = 112.2, p < .001, η² = .09, BF > 1000. Post-hoc tests demonstrated that heart rate was not different at any time point between the two exercise groups, all ps > .05. These results demonstrate that the two exercise manipulations were equally effective in increasing heart rate. With the exception of the resting time period, heart rate was significantly lower in the control condition when compared to the two exercise conditions for all other time periods, all ps < .05

Figure 5. HR Responses for Each Group Across Time.

Note. Error bars (minimally present) represent 95% confidence intervals.

Valence and Arousal Response.

Figure 6 displays the valence and arousal scores immediately before and after watching the emotional video. In a 3 (Group: Open, Closed, Control) × 2 (Time: Pre, Post) rmANOVA on valence, there was a main effect for Time, F(1, 190) = 299.7, p < .001, η² = .39, BF = 5.291e+44, but no main effect for Group, F(2, 190) = 2.76, p = .07, η² = .01, BF = .426, or Time × Group interaction, F(2, 190) = 2.29, p = .10, η² = .006, BF = .461.

Figure 6. Valence and Arousal Scores Pre- and Post-Emotional Video.

Note. Error bars represent 95% confidence intervals. For valence, the response scale ranged from −5 (very bad) to +5 (very good). For arousal, the response scale ranged from 1 (low arousal) to 6 (high arousal).

Similarly, for arousal, there was a main effect for Time, F(1, 190) = 307.0, p < .001, η² = .31, BF = 3.266e+39, but no main effect for Group, F(2, 190) = 1.31 ,p = .27, η² = .01, BF = .20, or Time × Group interaction, F(2, 190) = 2.77, p = .07, η² = .006, BF = .545. These results demonstrate that the emotional video was effective, in all three experimental groups, in increasing arousal and reducing valence.

Memory Results

Memory Intrusions.

Figure 7 displays the results of the Impact of Event Survey. As shown in Figure 3, across all groups, and for each of the three survey items, mean responses were between 1 and 2, suggesting that during the exercise/stretching sessions, participants thought about the video a little bit to moderately.

Figure 7. Impact of Event Survey (Memory Intrusions) Results.

Note. Point estimates (y-axis) represent the mean value, with response options ranging from 0 (not at all) to 4 (extremely). Error bars represent 95% confidence intervals.

In a 3 (Group: Open, Closed, Control) × 3 (Item: 1, 2, 3) rmANOVA, there was a main effect for Item, F(1.95, 370.6) = 9.39, p < .001, η² = .01, BF = 120.1, main effect for Group, F(2, 190) = 3.24, p = .04, η² = .03, BF = 1.27, but no Item × Group interaction, F(3.90, 370.6) = 1.24, p = .29, η² = .003, BF = .05. Holm-corrected post-hoc tests demonstrated that Item 2 (“popped into mind”) was higher than Item 1 (“thought about it”), M_diff = .20, t = 2.71, p = .01, d = .20, BF = 6.2, and Item 3 (“strong feelings about it”), M_diff = .31, t = 4.28, p < .001, d = .30, BF = 209.0, but Item 1 and 3 did not differ, p = .12, BF = 24. Regarding the main effect for Group, Closed-Skill Exercise was higher than Open-Skill Exercise, M_diff = .46, t = 2.49, p = .04, d = .18, BF = 145.6, but Control did not differ from Open-Skill, p = .19, BF = 3.20, or from Closed-Skill, p = .43, BF = .21. These results suggest that aspects of the emotional video (memory intrusions) were more likely to have come to mind during the Closed-Skill Exercise.

Emotional Memory Accuracy.

Figure 8 displays the emotional memory accuracy results. In a 3 (Group: Closed, Open, Control) × 2 (Memory Detail: Peripheral, Central) rmANOVA, there was a main effect for Memory Detail, F(1, 190) = 9.38, p = .003, η² = .02, BF = 13.45, but not a main effect for Group, F(2, 190) = 1.29, p = .27, η² = .007, BF = .11, nor a Memory Detail × Group interaction, F(2, 190) = 1.44, p = .23, η² = .007, BF = .21. Regarding the main effect for Main Detail, post-hoc testing showed greater accuracy for Central details compared to Peripheral details, M_diff = .04, t = 3.06, p = .003, d = .22, BF = 7.18. These findings suggest that memory accuracy was greater for Central than Peripheral details, but this did not vary as a function of acute exercise occurring during the memory consolidation period.

Figure 8. Emotional Memory Accuracy Results.

Note. Emotional memory accuracy results. Estimates (y-axis) are proportions, with error bars representing 95% confidence intervals.

Emotional Memory Confidence.

Figure 9 displays the memory confidence results. In a 3 (Group: Closed, Open, Control) × 2 (Memory Confidence: Peripheral, Central) rmANOVA, there was a main effect for Memory Confidence, F(1, 190) = 25.3, p < .001, η² = .025, BF > 1000, and main effect for Group, F(2, 190) = 4.71, p = .01, η² = .04, BF = 4.27, but not a Memory Confidence × Group interaction, F(2, 190) = .36, p = .70, η² = .0001, BF = .07. Holm-corrected post-hoc tests demonstrated that participants were more confident in remembering Central details compared to Peripheral details, M_diff = .20, t = 5.03, p < .001, d = .36, BF > 1000. Regarding the main effect for Group, those in the Closed-Skill group had lower memory confidence when compared to those in the Control group, M_diff = −.29, t = 3.04, p = .008, d = .22, BF = 107.8. Open-Skill Exercise did not differ from Closed-Skill Exercise, p = .25, BF = .39, or from Control, p = .12, BF = 1.66. These findings suggest that, similar to the greater memory accuracy of central events (see Figure 5), participants had greater memory confidence for central events. However, acute exercise was not reliably associated with memory confidence.

Figure 9. Emotional Memory Confidence Results.

Note. Point estimates (y-axis) represent the mean value, with response options ranging from 1 (not at all confident) to 5 (extremely confident). Error bars represent 95% confidence intervals.

Relationship between Memory Confidence and Accuracy.

Figure 10a illustrates the relationship between confidence in memory for peripheral details and accuracy, r = .205, p = .004, BF = 5.19. Figure 10b illustrates the relationship between confidence in memory for central details and accuracy, r = .37, p < .001, BF > 1000. These results demonstrate higher perceptions of confidence were associated with higher memory accuracy.

Figure 10. Relationship Between Memory Confidence and Accuracy of Peripheral and Central Events.

Note. peripheral items (a) and central items (b). Ordinate (y-axis) represents memory accuracy (proportion) and abscissa (x-axis) represents confidence scores. Dashed lines represent 95% confidence interval. For memory confidence, response options ranged from 1 (not at all confident) to 5 (extremely confident).

The magnitude of the correlation varied somewhat by Group. The relationships regarding memory confidence and memory accuracy for peripheral details in the closed-skill, open-skill, and control groups, respectively, were r = .29 (p = .01), r = .33 (p = .008), and r = .01 (p = .92). Comparing the size of these correlations, they did not differ between closed-skill and open-skill groups (p = .80); closed-skill and control groups (p = .11); the difference in the strength of the correlation approached significance comparing open-skill and control groups (p = .06).

Regarding central items, the relationship between memory confidence and accuracy, for closed-skill, open-skill, and control, respectively, were r = .49 (p < .001), r = .39 (p = .001), and r = .23 (p = .07). The size of correlations for the closed-skill and open-skill conditions were not significantly different (p = 0.49), and neither were the correlations for open-skill vs. control conditions (p = .33). The comparison between the closed-skill and control conditions approached significance, p = .09.

Collectively, these findings suggest that higher perceptions of confidence were associated with greater memory accuracy, and this association was more pronounced among those who exercised.

Sensitivity Results.

A series of 3 (Group) x 2 (Memory Accuracy) rmANOVA analyses were computed to evaluate the interaction effects of the categorical sample characteristics (Table 6). Results showed that Gender, p = .23, and Chronotype, p = .28, did not interact with Group to influence memory accuracy. Similarly, time of day of the visit did not interact with Group to influence memory, p = .94, nor did taking medications for mood, p = .23, or having a diagnosis of ADHD, p = .96, interact with Group to influence memory.

A series of 3 (Group) x 2 (Memory Accuracy) rmANCOVA analyses were computed to evaluate the independent effects of the continuous sample characteristics on memory accuracy. There was no main effect of age, p = .37, body mass index, p = .22, self-reported moderate-to-vigorous physical activity, p = .43, estimated cardiorespiratory fitness, p = .64, time of day of the laboratory session, p = .52, negative emotional reactivity, p = .24, or positive emotional reactivity, p = .13.

We also evaluated if estimated cardiorespiratory fitness (expressed as a continuous variable) interacted with Group to influence memory accuracy, which it did not, p = .10. Further, the exercise-induced heart rate response (endpoint heart rate minus resting heart rate) did not interact with Group to influence memory, p = .29.

Discussion

The purpose of the present set of experiments was to evaluate whether acute exercise impacted emotional memory accuracy. Across three experiments, we manipulated the timing of acute exercise (i.e., before or immediately after encoding), the modality of acute exercise (i.e., jogging, cycling, open- vs. closed-skill), the test type (i.e., recognition or cued recall), the valence and arousal level of the emotional stimuli, and the retention interval (i.e., ~ 30 minutes, 24 hours, or 7 days). Our main finding was that, across the three experiments, we did not observe convincing evidence that acute exercise reliably influenced emotional memory. However, our results demonstrated that (a) acute exercise may influence memory intrusions, and (b) acute exercise induced a robust relationship between perceptions of confidence and memory accuracy.

As demonstrated in a recent review (Loprinzi et al., 2019b), the relationship between acute exercise and emotional memory is mixed. Among older adults, six minutes of cycling (70% of VO_2max) during the memory consolidation period, relative to seated rest, enhanced emotional memory (assessed via IAPS stimuli; Segal et al., 2012). Similarly, among young adults, intense stepping exercise (60-85% of heart rate max) during the memory consolidation period, relative to slow walking, enhanced emotional memory (assessed via IAPS stimuli; Keyan and Bryant, 2017a). Given the evidence highlighting an enhancing effect of higher exercise intensities on emotional memory outcomes, we implemented a high-intensity, acute jogging protocol in Experiment 1 (70% HRR). Despite our modifications of Wade and Loprinzi’s (2018) study (moderate-intensity exercise) and our inclusion of high-intensity exercise, Experiment 1 did not demonstrate an effect of high-intensity exercise on emotional memory.

It is possible that some experimental inconsistency in the literature focused on exercise and emotional memory may relate to the modality of exercise, rather than intensity alone. The studies demonstrating an effect of exercise on emotional memory included modalities of cycling or stair step exercise (Keyan and Bryant, 2017a,b,c; Segal et al., 2012), in contrast to the null effect studies, which utilized isokinetic resistance exercise (Weinberg et al., 2014), or treadmill exercise (Wade & Loprinzi, 2018). As demonstrated in a recent meta-analysis (Loprinzi et al., 2019a), regardless of whether the acute exercise occurred before or after memory encoding, cycling-based exercise, but not treadmill exercise, was effective in enhancing episodic memory. Although speculative, it is conceivable that treadmill-based exercise, relative to cycling, may engage more cognitive operations, as treadmill exercise may require greater energy expenditure (upper and lower body movements) and attentional allocation (to maintain posture and stability). Clearly, additional research is needed that investigates whether exercise modality moderates the effects of exercise on memory. In our current set of experiments, we did not observe notable effects of exercise on emotional memory when considering treadmill (Experiments 1 and 3) or cycling (Experiment 2) exercise. Relatedly, and considering other exercise modalities, recent research also shows no difference in episodic (neutral) memory after engaging in acute aerobic and resistance exercise (Loprinzi, Loenneke, Storm, 2021). However, isometric handgrip activity improves neutral memory (Loprinzi et al., 2020) and has also been shown to increase negativity bias in young women’s recall of pictures, especially among women with low estradiol and low progesterone levels (Nielson et al., 2015). Isometric handgrip increases salivary alpha-amylase (Nielsen & Mather, 2015) and pupil dilation (Mather et al., 2020), both indicators of noradrenaline activity. Further research is needed to see if isometric handgrip targets the noradrenergic system more specifically than other types of exercise and to further examine the role that sex hormones play in the interactions between noradrenaline and emotional memory (Ycaza Herrera et al., 2019).

Based on prior evidence, as well as our findings from Experiment 1, we made several modifications to the study design in Experiment 2. These included changing the exercise modality from a treadmill to a cycling paradigm, modifying the strength of the exercise stimulus from 15 minutes to 25 minutes in duration, integrating both high- and low-intensity exercise in order to more precisely capture intensity-specific effects, including both male and female participants, and increasing hippocampal dependency of the memory assessment from a (remember/know/new) recognition task to an associative recognition task.

Like Experiment 1, Experiment 2 also did not demonstrate an effect of high- or low-intensity acute exercise on emotional memory. Moreover, cycling did not appear to be a robust stimulus to change associative emotional memory. For our third and final experiment, we made several novel modifications to the study design. These included changing the duration of the exercise stimulus from 25 to 30 minutes, changing the exercise modality from cycling to include open- and closed-skill exercise, using emotional video stimuli, rather than images, and further increasing hippocampal dependency of the memory assessment from an associative recognition task to a cued-recall task. Results indicated that (a) central details of the emotional video (memory intrusions) were more likely to have come to mind during the closed-skill treadmill exercise and (b) higher perceptions of emotional memory confidence were associated with greater memory accuracy, which was pronounced among those who exercised. This finding supports Keyan and Bryant’s (2017b) work showing that, among young adults, 20-25 minutes of incremental cycling following a memory reactivation induction was effective in recalling more central details of a 10-minute traumatic film (same video used in Experiment 3 of the present paper). Similarly, Keyan and Bryant (2017c) showed that, among young adults, stepping exercise for 10 minutes (50-82% of heart rate max), relative to walking, did not improve cued recall of an emotional video, but did enhance intrusive memories. Thus, it is possible that exercise task complexity (e.g., open-skilled racquetball) may drive executive attention away from the external stimuli (potentially intrusive scenes from the video) towards internal feedback controls devoted towards successfully performing complex motor sequences (Frith & Loprinzi, 2018). The increased memory intrusions during closed-skill exercise, however, could simply have been a result of less distraction during this type of exercise, allowing more time to think about the video.

The present set of results should be interpreted in the context of our limitations. For example, across the studies, we did not measure various factors (e.g., sleep duration, physical fitness, hormonal levels of menstrual cycle phase of female participants, and key biomarkers (e.g., cortisol)) that could potentially influence the effects of acute exercise on emotional memory. Relatedly, Experiment 3 was conducted during the COVID-19 pandemic, which could have created potential confounds and dissimilarities across the experiments. Further, there were some discrepancies in the methodological approach employed across the three experiments (e.g., different questionnaires, no measure of heart rate in the control condition in Experiment 2, differences in gender proportion across the experiments, and variations in the time of assessments between participants and experiments). Another limitation is the use of different control scenarios across experiments and employing control tasks that have differential attentional characteristics than the experimental conditions. Additionally, to provide a more comprehensive and accurate assessment of the physiological response to the exercise and control conditions, future work should continuously monitor and report heart rate responses for the duration of the conditions, including the control. Further, future work should more carefully control the degree of social contact across the experimental and control scenarios, as this will have important implications in the attentional processes and subsequent memory modulation.

Taken together, though the extant literature is mixed and there does not appear to be a consistent pattern as to when acute exercise may or may not enhance emotional memory (for similar conclusions regarding non-emotional memory, see Roig et al., 2013), the evidence herein lend support to an executive role of exercise in memory function. A wealth of research has demonstrated favorable effects of both acute and longer-term exercise on executive function and higher-order cognition, with acute influences associated with noradrenaline and glucocorticoid production, as well as serum brain-derived neurotropic factor (BDNF) increases (Keyan and Bryant, 2019; Piepmeier & Etnier, 2015). If the modulation of executive processes is one viable mechanism underlying an exercise-emotional memory relation, it is possible that acute exercise may activate and direct executive processes which contribute to enhanced confidence-accuracy relationships (as demonstrated in Experiment 3 presented herein). Perhaps exercise before a memory test improves prefrontal memory monitoring such that individuals are more confident judges of how accurate they are. To test this in a follow-up study, emotional memory could be evaluated on a separate day either with or without preceding exercise. Continued work should also aim to translate this research to clinical or therapeutic settings, in which more complex, dynamic exercise (i.e., open-skill exercise) may be leveraged as a tool to offset rumination on negative intrusive events or distracting, maladaptive thought patterns. Additionally, participation in exercise may be used to support memory confidence and accuracy via alterations in higher-order executive error detection mechanisms. However, given that these findings were not prespecified as the chief outcomes of interest in the present set of experiments, we encourage future work to expand on our preliminary results by designing rigorous studies to specifically address these factors and elucidate whether and how acute exercise fits into an executive framework for emotional memory.

Appendix A.

Complete memory results for Experiment 1.

Time	Classification	Item Type	Response Type	group	Mean	SD	N
1 Day Positive	High Arousal	New	Know	CON	0.075	0.093	16
				EDC	0.093	0.103	15
				EPE	0.056	0.078	18
			New	CON	0.912	0.120	16
				EDC	0.893	0.096	15
				EPE	0.917	0.138	18
			Remember	CON	0.013	0.050	16
				EDC	0.013	0.035	15
				EPE	0.028	0.096	18
		Old	Know	CON	0.225	0.241	16
				EDC	0.107	0.144	15
				EPE	0.172	0.259	18
			New	CON	0.044	0.089	16
				EDC	0.047	0.106	15
				EPE	0.056	0.110	18
			Remember	CON	0.731	0.263	16
				EDC	0.847	0.239	15
				EPE	0.772	0.254	18
	Positive Low Arousal	New	Know	CON	0.113	0.120	16
				EDC	0.087	0.092	15
				EPE	0.133	0.181	18
			New	CON	0.881	0.128	16
				EDC	0.880	0.121	15
				EPE	0.822	0.213	18
			Remember	CON	0.006	0.025	16
				EDC	0.033	0.062	15
				EPE	0.044	0.062	18
		Old	Know	CON	0.294	0.313	16
				EDC	0.113	0.125	15
				EPE	0.250	0.231	18
			New	CON	0.100	0.155	16
				EDC	0.080	0.166	15
				EPE	0.089	0.128	18
			Remember	CON	0.606	0.371	16
				EDC	0.807	0.237	15
				EPE	0.661	0.255	18
	Negative High Arousal	New	Know	CON	0.100	0.141	16
				EDC	0.220	0.142	15
				EPE	0.106	0.151	18
			New	CON	0.869	0.162	16
				EDC	0.753	0.185	15
				EPE	0.839	0.200	18
			Remember	CON	0.031	0.048	16
				EDC	0.027	0.059	15
				EPE	0.056	0.092	18
		Old	Know	CON	0.087	0.109	16
				EDC	0.093	0.144	15
				EPE	0.156	0.206	18
			New	CON	0.031	0.060	16
				EDC	0.013	0.052	15
				EPE	0.050	0.092	18
			Remember	CON	0.881	0.147	16
				EDC	0.893	0.179	15
				EPE	0.794	0.229	18
	Negative Low Arousal	New	Know	CON	0.037	0.089	16
				EDC	0.080	0.132	15
				EPE	0.050	0.104	18
			New	CON	0.963	0.089	16
				EDC	0.907	0.144	15
				EPE	0.928	0.136	18
			Remember	CON	0.000	0.000	16
				EDC	0.013	0.035	15
				EPE	0.022	0.073	18
		Old	Know	CON	0.256	0.299	16
				EDC	0.113	0.130	15
				EPE	0.272	0.397	18
			New	CON	0.063	0.081	16
				EDC	0.087	0.106	15
				EPE	0.028	0.046	18
			Remember	CON	0.681	0.306	16
				EDC	0.800	0.200	15
				EPE	0.700	0.388	18
	Neutral Low Arousal	New	Know	CON	0.063	0.096	16
				EDC	0.080	0.132	15
				EPE	0.050	0.071	18
			New	CON	0.938	0.096	16
				EDC	0.913	0.155	15
				EPE	0.922	0.106	18
			Remember	CON	0.000	0.000	16
				EDC	0.007	0.026	15
				EPE	0.028	0.075	18
		Old	Know	CON	0.225	0.296	16
				EDC	0.080	0.137	15
				EPE	0.194	0.321	18
			New	CON	0.038	0.062	16
				EDC	0.053	0.155	15
				EPE	0.039	0.120	18
			Remember	CON	0.738	0.299	16
				EDC	0.867	0.247	15
				EPE	0.767	0.334	18
7 Day	Positive High Arousal	New	Know	CON	0.188	0.219	16
				EDC	0.253	0.192	15
				EPE	0.100	0.128	18
			New	CON	0.637	0.275	16
				EDC	0.633	0.266	15
				EPE	0.800	0.228	18
			Remember	CON	0.175	0.184	16
				EDC	0.113	0.119	15
				EPE	0.100	0.161	18
		Old	Know	CON	0.313	0.236	16
				EDC	0.180	0.201	15
				EPE	0.172	0.252	18
			New	CON	0.113	0.163	16
				EDC	0.073	0.122	15
				EPE	0.089	0.123	18
			Remember	CON	0.575	0.302	16
				EDC	0.747	0.247	15
				EPE	0.739	0.295	18
	Positive Low Arousal	New	Know	CON	0.263	0.253	16
				EDC	0.220	0.193	15
				EPE	0.211	0.181	18
			New	CON	0.644	0.280	16
				EDC	0.713	0.220	15
				EPE	0.656	0.248	18
			Remember	CON	0.088	0.081	16
				EDC	0.067	0.090	15
				EPE	0.133	0.168	18
		Old	Know	CON	0.306	0.252	16
				EDC	0.320	0.326	15
				EPE	0.250	0.248	18
			New	CON	0.188	0.213	16
				EDC	0.153	0.203	15
				EPE	0.144	0.129	18
			Remember	CON	0.506	0.364	16
				EDC	0.527	0.322	15
				EPE	0.606	0.284	18
	Negative High Arousal	New	Know	CON	0.200	0.256	16
				EDC	0.207	0.222	15
				EPE	0.161	0.197	18
			New	CON	0.606	0.332	16
				EDC	0.613	0.264	15
				EPE	0.639	0.299	18
			Remember	CON	0.194	0.211	16
				EDC	0.180	0.227	15
				EPE	0.200	0.185	18
		Old	Know	CON	0.119	0.098	16
				EDC	0.213	0.196	15
				EPE	0.161	0.223	18
			New	CON	0.119	0.183	16
				EDC	0.127	0.144	15
				EPE	0.122	0.131	18
			Remember	CON	0.763	0.225	16
				EDC	0.660	0.282	15
				EPE	0.717	0.255	18
	Negative Low Arousal	New	Know	CON	0.200	0.273	16
				EDC	0.173	0.246	15
				EPE	0.122	0.131	18
			New	CON	0.669	0.334	16
				EDC	0.713	0.309	15
				EPE	0.783	0.207	18
			Remember	CON	0.131	0.189	16
				EDC	0.113	0.160	15
				EPE	0.094	0.130	18
		Old	Know	CON	0.319	0.229	16
				EDC	0.220	0.211	15
				EPE	0.256	0.324	18
			New	CON	0.125	0.106	16
				EDC	0.140	0.130	15
				EPE	0.100	0.114	18
			Remember	CON	0.556	0.278	16
				EDC	0.640	0.261	15
				EPE	0.644	0.357	18
	Neutral Low Arousal	New	Know	CON	0.231	0.185	16
				EDC	0.207	0.153	15
				EPE	0.089	0.118	18
			New	CON	0.644	0.248	16
				EDC	0.700	0.262	15
				EPE	0.811	0.211	18
			Remember	CON	0.125	0.148	16
				EDC	0.093	0.149	15
				EPE	0.100	0.141	18
		Old	Know	CON	0.269	0.275	16
				EDC	0.200	0.196	15
				EPE	0.222	0.317	18
			New	CON	0.144	0.182	16
				EDC	0.120	0.157	15
				EPE	0.111	0.157	18
			Remember	CON	0.588	0.275	16
				EDC	0.680	0.281	15
				EPE	0.667	0.376	18

Appendix B.

Various sensitivity analyses were computed for Experiment 1. With corrected recognition (hit rate – false alarm rate) as the outcome, engagement in moderate-to-vigorous physical activity did not interact with group assignment to influence memory, F(5, 21) = 1.85, p = .15, and similarly, there was not a significant three-way interaction between group, physical activity and time, F(5, 21) = 1.28, p = .31.

For Experiment 1, we observed a significant Time × Response Type × Item Type interaction, F(2, 368) = 4.03, p < .001, η² = .02, BF > 1000. Supplementary Figure 1 displays these results. The three-way interaction was decomposed by comparing recognition rates for response type (remember, now, new) and item type (old, new) across the two time periods (1-and 7-day follow-up periods). Overall, results showed that hit rates (responded “remember” for old items) decreased across time, whereas false alarms (responded “new” for old items”) increased over time. These specific paired t-test post-hoc results are as follows.

Remember. There was a greater proportion of remember responses for old items at the 1-day follow-up compared to the 7-day follow-up period, M_diff= .13, p < .001, BF = 700.1, d = .66. Similarly, there was a smaller proportion of remember responses for new items at the 1-day follow-up compared to the 7-day follow-up period, M_diff= .11, p < .001, BF = 100, d = .83.

Know. The know responses for old items was higher at the 7-day follow-up period when compared to the 1-day follow-up, M_diff = .05, p = .008, BF = 4.7, d = .40. There was a greater proportion of know responses for new items at the 7-day follow-up period compared to the 1-day follow-up period, M_diff = .10, p = .001, BF = 154.4, d = .59.

New. The new responses for old items was higher at the 7-day follow-up period compared to the 1-day follow-up, M_diff = .07, p < .001, BF = 36.6, d = .51. There was a greater proportion of new responses for new items at the 1-day follow-up period compared to the 7-day follow-up period, M_diff = .20, p < .001, BF > 1000, d = .88.

We observed a significant Emotion × Response Type × Item Type interaction, F(8, 368) 8.16, p < .001, η² = .004, BF > 1000. Supplementary Figure 2 displays these results. Overall, and generally, these findings demonstrate higher hit rates (responded “old” for old items) and lower correct rejection (responded “new” for new items) for negative, high arousal items. Paired t-test post-hoc results are as follows.

Remember. There was a greater proportion of remember responses for old items for Negative High Arousal compared to Negative Low Arousal, M_diff= .11, p < .001, BF = 21.1, d = .48, and Positive Low Arousal, M_diff = .16, p < .001, BF > 1000, d = .68. The remember responses for new items did not differ across Emotion, all ps > .05.

Know. There was a lower proportion of know responses for old items for Negative High Arousal compared to Negative Low Arousal, M_diff = −.10, p < .001, and Positive Low Arousal, M_diff = −.12, p < .001, BF = 75.3, d = .55. The know responses for new items did not differ across Emotion, all ps > .05.

New. The new responses for old items did not differ across Emotion, all ps > .05. However, there was a lower proportion of new responses for new items in the Negative High Arousal compared to Negative Low Arousal, M_diff = −.11, p < .001, BF > 1000, d = .76, Neutral Low Arousal, M_diff = −.10, p < .001, BF = 420.7, d = .64, and Positive High Arousal, M_diff = −.08, p < .001, BF = 33.5, d = .51.

Supplementary Figure 1. Proportion of Remember, Know and New Responses Across Item Type and Time.

Note. Item type = old, new; time period = 1- and 7-day follow-up periods. Error bars represent 95% confidence intervals.

Supplementary Figure 2. Proportion of Remember, Know and New Responses.

Note. Responses are displayed across item type (old, new) and classification. Error bars represent 95% confidence intervals.

Appendix C. Memory assessment for Experiment 3.

Participants were instructed to answer the following questions to the best of their ability. Then, rate their confidence using the following scale.

1 = not at all confident, 2 = somewhat confident, 3 = neutral, 4 = confident, 5 = extremely confident

				Confidence Rating
#	Ouestion	Correct Answer	Time Period (in video) Showing Correct Answer	1	2	3	4	5
1	How many injured (not dead) victims were there?	4	0:16, 0:32, 1:16, 6:42
2	How many police motorcycles were present?	1	0:01, 3:33
3	How many of the injured victims had dark skin?	3 – dark skin	0:16, 1:16, 6:42
4	Where there any noticeable houses in the background?	Yes	0:11
5	What color were the backboards that the victims were placed on?	Wooden/orange/brown	0:46, 0:55
6	What was the color of the deceased (person who has died) victim’s sneakers?	White	9:03
7	What was the color of the helmets that the paramedics were wearing?	Light blue/blue	1:28
8	What was the color of the blanket placed on the light skinned victim?	Yellow	5:37-5:45
9	What time did the car accident occur?	2:20 pm	0:03-0:06
10	What was the color of the tarp placed on the deceased victim?	Yellow	5:59
11	What color was the shirt of the victim, who was sitting on the curb?	White	6:42
12	What was the name of the county that was on the back of the first responders’ (paramedics/firefighters/police officers) jackets?	Contra Costa Country	1:28
13	What was the color of the overturned vehicle?	Taupe/Brown	2:06
14	What color pants was the deceased victim wearing?	Navy Blue	8:59
15	What were the two different color helmets that the firefighters wore?	Yellow and Red	0:16, 0:55
16	What color were the light skinned (not dead) victims’ pants?	Light blue/Blue	0:32 – 0:37
17	What was the color of the shoe placed on top of the deceased victim’s tarp?	Brown/Taupe	5:59
18	What color shirt was the helpful bystander (an individual not involved in the accident, nor a part of emergency services; a stranger who witnessed the accident and wished to help) wearing?	Purple/plaid	0:32-0:55
19	How many of the injured victims were placed into C-Collars (known as a cervical – collar, or neck brace)?	2	5:36, 7:46
20	What was the color of the deceased victim’s shirt/sweatshirt?	White	8:59
21	Was the car upside down or on its side?	Upside Down	2:06
22	What was the color of the deceased victim’s hair?	Brown	9:18
23	Where did the wreck occur (e.g., in a parking lot, country road, etc.)?	Highway/Interstate	3:33 – 4:05
24	How many female paramedics/firefighters/police officers were there?	0	*not in the video*
25	How many of the injured victims had light skin? (not including the deceased victim)	1	0:32 – 0:27

Key: Peripheral Central

Scoring: Among the 25 questions, 1-point is given for each correct response. In total, 26 points were possible, as one question (#15) was worth 2 points.

Notes for Scoring:

• A 30-min grace period was given for the time answer (2:20 pm) on question #9, “What time did the car accident occur?” That is, they were given a point if their answer fell between 1:50 PM to 2:50 PM.

• For question #12, “What was the name of the county that was on the back of the first responders’ jackets?” – The answer was deemed correct (1-point given) if the participant got at least one of the words (i.e, ‘contra’ or ‘costa’)

• For Question #15, “What were the two different color helmets that the firefighters wore?” - Each correct color was given 1-point. Thus, for this question, a total of 2-points is possible.

Survey Development

A series of steps, in alignment with the Delphi-method, were followed in developing this survey. First, among the 25-items, 3-items (#1, #2, #3) were identical to previous studies utilizing this same emotional video. For the remaining 22-items that we created, we took a four-step approach in developing these items.

Step 1

Step 1 involved the student (A.L.) and senior author (P.L.) watching the 10-min emotional video and generating 22 questions. After fully discussing and developing these 22 questions, Step 2 commenced.

Step 2

Six individuals (2 undergraduates, 2 graduates, 2 faculty) were asked to participate in the pilot session. Each pilot participant viewed the 10-min video and then completed the cued recall test. The specific instructions given to the pilot participant prior to viewing the video was as follows.

“Please closely watch this 10-min video. Please be as attentive as possible during this video, as later I will be asking you to recall specific information from the video.”

Immediately after viewing the video, the pilot participants were given the cued recall assessment. They were also asked if they had any comments about the questions. Questions that were ambiguous were modified accordingly. After completing the cued recall assessment, they were asked to re-read each question and indicate if the question was related to a central or peripheral detail of the video. They were informed that “Central details are any facts or elements directed related to the victims in the video and are not background details, whereas peripheral details are any information associated with the event that is not directly related to the victim. For each item, please provide a response ranging between 1 (central) to 6 (peripheral).”

In the initial pilot assessment, 29 items comprised the memory instrument. However, 4 items were ultimately removed due to item ambiguity and/or floor effect concerns. As such, the memory instrument is comprised of 25 items. Among the 25 items, 12 included peripheral items and 13 are central items. The mean (SD) response for the peripheral items was 4.42 (SD = 0.46) and the mean (SD) response for the central items was 1.55 (SD = 0.56). The mean (SD) valence response pre- and post-video, respectively, were 2.67 (2.52) and −3.0 (2.65). The mean (SD) arousal response pre- and post-video, respectively, were 1.67 (1.15) and 5.0 (1.0).

Step 3

In addition to the above pilot testing to ensure that there were no item ambiguity issues and that our assigned peripheral and central items measured these categories, 5 additional undergraduate/graduate participants (different from those who completed the first pilot assessment) completed the cued recall assessment 35-min after viewing the video; during this delay period, they watched a video clip from The Big Bang Theory. The purpose of this assessment was to evaluate whether there were any potential floor or ceiling effects after taking into consideration the delay period used in our experiment. The mean (SD) response for the 12 peripheral items was 6.0 (SD = 2.34; range = 4-9), which corresponded to an average % correct of 46.15 (range = 30.7%-69.2%). The mean (SD) response for the 13 central items was 4.67 (SD = 2.07; range = 3-8), which corresponded to an average % correct of 35.92 (range = 23.1%-61.5%). Given these results (% correct ranging from 23.1% to 69.2%), we felt confident that there were no floor/ceiling effect concerns.

Step 4

Step 4 involved having two faculty members, with extensive research experience, view the video and provide feedback on the 25-item memory survey. After watching the video and reviewing the process of our survey building, the two-faculty had minimal comments about the survey and the ambiguity in the questions. A few statements were made about the use of words such as ‘deceased’ and ‘first responders’, so short definitions were added after these words for clarity.