Classical Music, Educational Learning, and Slow Wave Sleep: A Targeted Memory Reactivation Experiment

Abstract

Poor sleep in college students compromises the memory consolidation processes necessary to retain course materials. A solution may lie in targeting reactivation of memories during sleep (TMR). Fifty undergraduate students completed a college-level microeconomics lecture (mathematics-based) while listening to distinctive classical music (Chopin, Beethoven, and Vivaldi). After they fell asleep, we re-played the classical music songs (TMR) or a control noise during slow wave sleep. Relative to the control condition, the TMR condition showed an 18% improvement for knowledge transfer items that measured concept integration (d=0.63), increasing the probability of “passing” the test with a grade of 70 or above (OR=4.68, 95%CI:1.21,18.04). The benefits of TMR did not extend to a 9-month follow-up test when performance dropped to floor levels, demonstrating that long-term-forgetting curves are largely resistant to experimentally-consolidated memories. Spectral analyses revealed greater frontal theta activity during slow wave sleep in the TMR condition than the control condition (d=.87), and greater frontal theta activity across conditions was associated with protection against long-term-forgetting at the next-day and 9-month follow-up tests (rs=.42), at least in female students. Thus, students can leverage instrumental music—which they already commonly pair with studying—to help prepare for academic tests, an approach that may promote course success and persistence.

1. Introduction

The pervasiveness of poor sleep in students is well chronicled. Sixty percent of college students are habitually poor sleepers, and students sleep fewer than the recommended 7 hours on 50–65% of nights (Hirshkowitz et al. 2015; Lund et al. 2010; Scullin 2019). Such statistics should be of wide concern to students, teachers, and administrators because insufficient sleep compromises immune functioning, exacerbates stress reactivity, and impairs numerous cognitive functions. Without sleep, students suffer from impairments to attention (Doran et al. 2001), creativity (King et al. 2017), and memory consolidation (Rasch and Born 2013).

Sleep restriction is detrimental to people of all demographic groups. However, there is longstanding evidence that women are more likely to suffer from insomnia than men (Zhang and Wing 2006). More importantly, there is emerging evidence that females are more vulnerable than males to the consequences of sleep loss (Goldstein-Piekarski et al. 2018; Gao et al. 2019; Prather et al. 2013; but see Okano et al. 2019). In rats, 72 hours of paradoxical sleep deprivation significantly impaired spatial learning and short-term memory in female rats, but not male rats (Hajali et al. 2012). In humans, Rångtell and colleagues (2019) found that one night of total sleep deprivation impaired working memory performance in women, but not in men. Gender-related sleep disparities are particularly provocative when considered within the broader educational context of achievement gaps for females in science, technology, engineering, and mathematics (STEM) disciplines (especially prevalent in mathematics-based disciplines; Ballard and Johnson 2005; Fryer and Levitt 2010).

To combat sleep-loss-related cognitive impairments for all students, some educators have implemented sleep education programs (Hershner and O-Brien 2018). Unfortunately, education programs tend to only impact students’ knowledge/awareness of sleep, rather than motivate them to change their sleep habits/behaviors (van Rijn et al. 2019; Wing et al. 2015). Another approach is to take a system-wide effort to delay school start times (e.g., Wahlstrom 2000). Doing so improves sleep duration and academic outcomes, but system-wide changes are often resisted by politicians, administrators, and other stakeholders (American Academy of Pediatrics 2014). A third approach is to incentivize individual students to go to bed earlier (Scullin 2019), but this approach is also limited because sleeping longer does not necessarily change sleep microarchitecture or increase the probability that classroom/textbook memories will be consolidated.

A cross-disciplinary challenge for educational, cognitive, and sleep sciences is to devise methods that increase sleep-dependent memory consolidation of educational concepts in students. Declarative memories, such as those learned in the classroom, are theorized to initially be encoded into a temporary store in the hippocampus (Born and Wilhelm 2012), and during slow-wave sleep (SWS), the memories are reactivated and redistributed to more permanent stores in the neocortex (for an alternative view, see Yonelinas et al. 2019). Critically, memory consolidation is theorized to be a selective process (Cairney et al. 2014; Saletin et al. 2011). Only the daily experiences that are perceived as most important, emotionally-salient, or future-relevant are those that are spontaneously reactivated and consolidated (Bennion et al. 2015). Therein lies the problem for students. Though educators might perceive that classroom learning is highly important, classroom/textbook memories compete against everyday social, entertainment, and extracurricular memories.

A potential solution is to bias which daily experiences an individual consolidates during SWS via targeted memory reactivation (TMR; Rudoy et al. 2009). In a seminal experiment, Rasch et al. (2007) had participants learn spatial card locations while smelling a rose odor. Subsequently, researchers presented the same odor (or control) during SWS, rapid eye movement (REM) sleep, or wakefulness prior to testing memory again the next day. Only odor re-exposure during SWS significantly improved memory consolidation. This TMR finding has now been replicated in more than 30 published studies (for review, see Schouten et al. 2017), most commonly using laboratory-based learning materials (e.g., cat picture location) paired with discrete sound auditory stimuli (e.g., cat’s meow). Such experiments have been useful in identifying that TMR can trigger changes in SWS theta power, spindles, or slow oscillations en route to enhancing retention (Cellini and Capuozzo, 2018; Schreiner and Rasch 2014).

Despite TMR’s robustness in the laboratory, there are three issues that limit its translational value to addressing educational challenges. The first issue is that student success in higher education is often determined by conceptual learning and integration. Such learning processes are not well-captured by the research tradition of having participants learn word lists, images, or spatial locations. This distinction between rote learning and conceptual learning is quite critical, as underscored by recent findings in the “testing effect” literature (Karpicke and Roediger 2008). The testing effect literature has shown that retrieving a memory while awake reliably improves rote learning (item memory), but that simple memory retrieval while awake does not normally benefit conceptual learning or integration (Miyatsu et al. 2018; Pan and Rickard 2018). Thus, if TMR is simply and only reactivating item memories, then it should not benefit conceptual learning.

A second issue for translating TMR from the laboratory to educational settings is that most students study in dormitory or library settings. In such settings, it is doubtful that most students would attempt to pair their studying with odors or to attempt to match exact study content to exact discrete sounds. A more translatable approach would be to capitalize on existing habits of listening to music while studying, a “sensory-study pairing” habit shared by more than half of college students (Danhauer et al. 2009). Listening to lyrical music might impede studying, but empirical work shows that listening to instrumental classical music does not harm encoding (Jäncke et al. 2014) or affect nighttime sleep (Harmat et al. 2008). Thus, if instrumental music can be demonstrated as an effective TMR tool, then instrumental music would be an optimal candidate for broadly implementing TMR into educational settings.

A third issue for understanding TMR’s translational value concerns whether TMR effects are acute or sustained. Whereas most students are primarily concerned with strategies that immediately enhance performance (i.e. acute effects; Miyatsu et al. 2018), most educators are interested in techniques that prevent the steep forgetting rates that occur across a semester or academic year (i.e. sustained effects; Conway et al. 1991; Ebbinghaus 1885). TMR studies generally only test memory the next morning (a few studies used a 1-week delay; e.g., Hu et al. 2015). Thus, there is a need for data on whether TMR buffers against typical long-term-forgetting curves, or conversely, if TMR should only be applied the night before select tests.

The current work was a double-blind, placebo-controlled study in which college students took a virtual lecture on microeconomics while listening to classical music. Later that night, when participants entered stable SWS, they were re-exposed to the classical music (or a white noise control). The next morning, as well as approximately 9-months later (akin to the length of one academic year), participants took a microeconomics test that included concepts they were explicitly trained to solve as well as problems that required novel conceptual integration (hereafter, knowledge transfer).

One remaining note is that the current study considered gender differences in TMR, as motivated by two literatures. First, a recent meta-analysis on TMR found divergent trends of TMR effect across males and females (Hu et al. 2020). Second, research at the intersection of music arts and brain sciences (Cheever et al. 2018) has indicated that females process music more efficiently than males, with better association and recognition of familiar music (Koelsch et al. 2003; Feizpour et al. 2018; Fancourt et al. 2016; Miles et al. 2016), raising the possibility that musical cues could be especially strong retrieval cues for females. Based on these literatures, we supplemented our primarily analyses with gender-stratified analyses to investigate whether classical-music TMR affected males and females similarly.

2. Materials and Methods

2.1. Participants

Fifty college students between the ages of 18 and 33 (M_age = 21.16, SD_age = 2.77, 70% female) were recruited via campus flyers to complete a two-night study protocol consisting of educational tasks and polysomnography procedures. Participants were not eligible if they had taken a college-level Economics course, had a history of psychiatric, neurological, or sleep disorders, or were younger than 18 years old. The Baylor University Institutional Review Board approved this study and all participants provided written informed consent.

2.2. Material and Measurements

2.2.1. Educational materials.

The educational learning task resembled course material from an undergraduate microeconomics class, with the virtual lecture being developed based on topics from a leading undergraduate textbook (Baumol and Blinder 2007). During the learning phase, participants navigated through a self-paced, computer-based interactive lecture for up to 30 minutes. The lecture included a section on supply and a section on demand in a counterbalanced order. Then, participants navigated through the lecture and learned to identify and calculate changes in supply or demand in a given scenario, followed by 6 practice questions (Gao et al. 2019).

During the initial testing phase, participants completed a microeconomics test with 18 trained questions (9 supply and 9 demand) and 9 knowledge-transfer questions. The 18 supply/demand questions were similar to lecture example questions where participants needed to shift only the supply curve or only the demand curve to calculate the correct answer. The 9 knowledge-transfer questions required integration of supply and demand concepts, which was not explicitly taught in the lecture. Participants needed to shift both the supply curve and the demand curve and identify that the correct answer was the intersection of the two curves. All questions on the test were open-ended questions scored using a rubric.

To evaluate metacognition (i.e., self-awareness of learning), participants reported from 0% to 100% their predictions for learning (start of lecture sections) and postdictions for learning (conclusion of supply and demand sections). They repeated these prediction and postdiction estimates before and after taking the next-day test and the follow up test. Table S1 summarizes the meta-cognitive assessment outcomes, which were consistent in pattern with the objective test outcomes.

2.2.2. Questionnaires.

Participants kept a sleep diary for a week prior to the first session (Carney et al. 2012). During the study, participants completed questionnaires and intelligence scales. Questionnaires and tests that were relevant to sleep and cognitive functioning included the Pittsburgh Sleep Quality Index (Buysse et al. 1989), Epworth Sleepiness Scale (Johns 1991), Morningness-Eveningness Questionnaire (Horne and Östberg 1976), a vocabulary test, and the Raven’s Advanced Progressive Matrices test (Raven 1938). Participant responses are summarized in Table 1.

Table 1.

Demographic Characteristics and Learning Outcomes

	TMR (n = 19)	Control (n = 22)	Comparison between TMR and control groups (p-value)
Age	20.89 (2.18)	21.05 (2.08)	.82
Gender (female)	68.42%	77.27%	.52
Race/Ethnicity (White)	63.16%	63.64%	.98
Participants with music experience	73.68%	81.82%	.75
Fluid intelligence (RAPM) (%)	65.12 (14.36)	59.85 (12.36)	.22
Lecture Duration (min)	16.18 (3.99)	17.80 (5.79)	.31
Vocabulary test (%)	72.89 (8.22)	70.45 (9.69)	.40
MEQ score	47.05 (10.53)	49.32 (6.43)	.42
PSQI	5.22 (2.44)	6.61 (2.69)	.10
ESS	9.53 (4.64)	10.14 (3.71)	.64
Sleep diary total sleep time (min)	426.20 (44.44)	443.71 (56.68)	.30
Habitual bedtime (PSQI)	11:42 pm	11:31 pm	.56
Difference between habitual bedtime and experimental night bedtime (min)	87.05 (56.23)	70.82 (55.70)	.37
Completed the 9-month follow-up session	63.16%	68.18%	.74

Note. Data presented as percentage or mean (standard deviation).

RAPM = Raven’s Advanced Progressive Matrices

MEQ = Morningness-Eveningness Questionnaire

PSQI = Pittsburgh Sleep Quality Index

ESS = Epworth Sleepiness Scale

2.2.3. Polysomnography.

We used the Grass Comet XL Plus system to objectively record sleep in a sound-attenuated sleep laboratory. For the experimental night, we recorded sleep from F3, F4, FZ, C3, C4, CZ, P3, P4, PZ, T3, T4, T5, T6, O1, O2, and OZ sites (200 Hz), left and right EOG, mentalis EMG, forehead cerebral oximetry, and fingertip pulse oximetry. A certified polysomnography technician scored the sleep stages in 30 second epochs, and according to AASM guidelines (Iber et al. 2007).

2.2.4. Spectral power analysis.

Spectral analysis characterizes the power of EEG waves and has been used to investigate the mechanisms of TMR (Schouten et al. 2017). We used BrainVision Analyzer 2.0 software to conduct spectral analyses. First, trained research personnel visually inspected each epoch and excluded epochs containing artifacts. Second, we filtered EEG data with high- and low-pass cutoffs of 0.3 Hz and 35 Hz. We then modified the sampling rate to 128 Hz. Next, we segmented each epoch into four-second segments with 50% overlap. We applied a symmetric Hanning window and performed Fast Fourier Transformation with 0.25 Hz resolution. Last, we computed spectral power values for each sleep stage for the relevant frequency ranges: 0.5–1 Hz for slow oscillation; 1–4 Hz for delta; 4–8 Hz for theta; 8–12 Hz for alpha; 12–16 Hz for sigma; and 16–32 Hz for beta. We analyzed spectral power in the slow oscillation, delta, and theta bands averaged across all frontal sites during SWS, averaged across all sites during SWS, and averaged across all sites during all sleep epochs. For visual comparison across groups, we normalized power across the whole night within each frequency band to highlight the relative spatial power distribution across the head.

2.2.5. Spindle detection.

Using Matlab 9.0, we implemented a previously-validated, wavelet-based algorithm shown to have high agreement with expert visual identification (Wamsley et al. 2012; Warby et al. 2014). First, we modified the sampling rate to 100 Hz. Second, we performed a time-frequency transformation on the EEG data using Morlet wavelet. Third, we classified spindles as EEG events that occurred in the 10–16 Hz frequency range, that exceeded 4.5 times the mean amplitude of all artifact-free epochs, and that lasted between 300 and 3000 ms (Scullin et al. 2019).

2.3. Procedure

Participants completed an overnight polysomnography-adaptation visit prior to returning for the experimental night. During the experimental night, research personnel set the laboratory environment brightness to 45 lux to simulate evening studying with a desk lamp. The room temperature was set to 68 degrees Fahrenheit (20 degrees Celsius), with blankets provided as needed. Participants arrived at approximately 8:45 pm. Both TMR and control condition participants completed the economics virtual lecture while listening to background classical music played from the computer. The music was played to all participants in loop at 40 decibels, which is the volume equivalent of soft background noise in a library setting. Decibel level was confirmed each evening prior to participant arrival using an Extech 407735 Digital Sound Meter.

The classical music consisted of three pieces: Moonlight Sonata 1^st movement (duration: 6:04; Beethoven 1802), Spring movement 1: Allegro (duration: 3:13; Vivaldi 1725), and Nocturne in E-Flat Major, Op.9, No.2 (duration: 4:28; Chopin 2005). These pieces were selected for their distinctive melodies, because previous TMR work found that sensory stimuli must be distinctive to become associated with learning materials (cf. ocean wave sensory stimuli; Donohue and Spencer 2011; music and economics learning materials are publicly available at https://osf.io/ems7z/). Using Audacity software (version 2.1.3; https://audacityteam.org/), we digitally smoothed the volume and applied fade-in and fade-out to each song to remove sudden volume fluctuations (Sanchez and Bootzin 1985). After completing the lecture, trained personnel applied polysomnography (PSG) electrodes while participants completed questionnaires. Lights out was at approximately 10:30 pm.

To minimize expectation effects and other potential sources of bias, the participants and experimenters were masked to the experimental conditions. Condition assignments were determined via blocked randomization methodology (block sizes of 2 and 4). Once participants reached their first bout of stable slow wave sleep, the overnight technician opened a sealed envelope to determine whether to play the classical music or white noise (matched in decibel level). Classical music or white noise was played through speakers at the bedside (Bose Companion 2 Series III). The overnight technician was not involved in administering any lecture or test materials.

Each classical music piece was played once (total duration = 13.75 minutes); and white noise was played to the control group for the same duration. If a participant showed an EEG arousal, the technician paused the music/white noise until after the participant returned to stable SWS. Music/white noise started to play 33.81 (SD = 29.66) minutes after sleep onset in the TMR condition and 34.24 (SD = 31.26) minutes after sleep onset in the control condition, which did not differ between conditions, t(39) = .05, p = .96. Music/white noise presentation time also did not differ across males (M = 32.72, SD = 34.08) and females (M = 34.53, SD = 29.19), t(39) = 0.17, p = .87.

The next day, lights on was at approximately 7:30 am, with the exact timing determined by spontaneous awakenings or by waiting until the participant reached stage 1 sleep. Participants reported whether they heard any sounds during the night and then started the microeconomics test at approximately 7:45 am. They were given up to 45 minutes to complete the test. Approximately nine months after the initial session (M_TMR = 263.58, SD_TMR = 84.88 days; M_Control = 292.93, SD_Control = 116.37 days; t(25) = 0.73, p = .47), 27 participants completed a second microeconomics test that included different test items, but was similar in difficulty and structure (i.e., consisting of 9 supply and 9 demand questions that participants were trained to solve and 9 knowledge-transfer questions that required application of learned concepts; all questions were open-ended). Fourteen participants did not complete the follow-up test because they had enrolled in economics courses (n = 2), were not available to participate (n = 4), or did not respond to requests for follow-up (n = 8). The probability of completing the follow-up session did not differ across TMR and control conditions (12 TMR participants, χ²(1) = 0.11, p = .74) or across gender groups (20 female participants, χ²(1) = 0.03, p = .86).

All microeconomics questions were scored independently by two raters and blinded to the experimental conditions. Discrepancies were resolved by discussion among raters. Overall performance was calculated as the percentage of correctly answered questions.

2.4. Statistical Analysis

Participants were excluded from statistical analyses if there was a protocol deviation (n=5 had the music/control noise played prematurely during light sleep; n=1 had the speakers turned off during the initial lecture). Two other participants did not complete the study and one wrote responses that were ungradable by independent, blinded experimenters. Exclusions did not differ across conditions (χ²(1) = 1.22, p=.27).

We conducted statistical analyses using SPSS (version 25) and JASP (version 0.10.2.0) software. We used Pearson’s correlations to investigate associations between test performance and only a small group of sleep variables that were previously implicated in TMR studies (frontal theta, delta, spindle, and slow oscillation activity; see Cellini and Capuozzo, 2018, review). We conducted t-tests and ANOVAs to compare test performance between the TMR and control conditions, first collapsed across genders and then stratified by males and females (following the approach taken by Gao et al. 2019). EEG data were analyzed and presented in aggregate form (averaged across relevant channels), and displayed as individual channels for visualization of spatial patterns. We conducted additional Bayesian analyses to illustrate the effects on EEG data. Though all TMR effects were hypothesized to occur in one direction (TMR > Control), we report the statistics as two-tailed tests with results of p ≤ .05 considered to be statistically significant. Bayesian analyses results with Bayes factors greater than 3 were considered as moderate evidence that the data favor the alternative hypothesis (Lee and Wagenmakers, 2014; Dienes, 2014).

3. Results

3.1. Baseline Measures

Table 1 presents the demographic characteristics of the sample. The TMR and control conditions were similar in age, gender, race/ethnicity, music-related experience (i.e., instrument playing and choir singing experience), fluid intelligence, chronotype, sleep quality in the past month, daytime sleepiness, and total sleep time from the sleep diary (all ps > .05). Table 2 shows that PSG sleep architecture variables and report of hearing sounds during the night did not differ significantly between TMR and control groups (all ps > .10). The TMR and control groups took equally long to complete the microeconomics lecture (t(39) = 1.03, p = .31; Table 1), and they showed equal levels of initial learning on practice questions during the lecture, t(39) = 0.19, p = .85 (Figure 1a; Creery et al. 2015). Male and female students spent similar amounts of time on the lecture (t(39) = 0.70, p = .49) and showed equal levels of initial learning (t(39) = 0.29, p = .78).

Table 2.

Polysomnography Measures

	TMR (n = 19)	Control (n = 22)	Comparison between TMR and control groups (p-value)
Stage 1 sleep (min)	25.34 (16.61)	23.29 (14.62)	.68
Stage 2 sleep (min)	289.11 (37.42)	282.96 (40.99)	.62
Slow wave sleep (min)	88.53 (27.14)	92.09 (28.50)	.69
REM sleep (min)	106.32 (31.39)	111.36 (32.42)	.62
Total sleep time (min)	509.28 (33.97)	509.87 (37.79)	.96
Sleep latency (min)	11.50 (10.35)	13.86 (12.25)	.51
Sleep efficiency (%)	92.14 (5.09)	93.02 (5.35)	.59
WASO (min)	32.66 (26.74)	24.57 (25.75)	.33
Number of wake epochs when auditory stimuli were presented	1.16 (2.67)	0.14 (0.35)	.12
Number of arousals when auditory stimuli were presented	2.47 (4.51)	1.09 (1.74)	.19
Reported hearing sounds at night	36.84%	36.36%	.98

Note. Data presented as percentage or mean (standard deviation).

WASO = Wake After Sleep Onset

Figure 1.

(a) Performance on lecture practice questions was not different between TMR and control groups; (b) Proportion of participants who passed the test with a total score ≥70% was significantly higher in the TMR group; TMR benefited performance on (c) the next-day microeconomics test, but performance on (d) the follow-up microeconomics test dropped to floor levels. Error bars: standard error. * p ≤ .05.

3.2. Next-day Test

At most colleges in the United States, a grade of 70 is considered the cutoff for passing a test, and thus we used this criterion to define a passing score, representative of successful learning. As shown in Figure 1b, TMR increased the likelihood of “passing” the microeconomics test (57.89% of participants) relative to the control condition (22.73% of participants; Odds Ratio = 4.68, 95%CI: 1.21, 18.04, χ²(1)= 5.30, p = .02). Specifically, this was driven by the TMR group showing higher performance than the control group on the challenging, integrative knowledge-transfer questions, by 18.58 percentage points (t(39) = 2.03, p = .050, d = 0.63; Figure 1c). Figure 2a illustrates how the benefit of TMR on knowledge transfer questions seemed to be more pronounced in female participants (Figure 2a; t(28) = 3.05, p = .005, d = 1.13) than in male participants (t(9) = 0.40, p = .70, d = 0.24), though the direct comparison across genders was only marginally significant, F(1,37) = 3.46, p = .07, partial η² = 0.09. Performance on basic/trained questions showed similar patterns but was less sensitive to condition (t(39) = 1.56, p = .13, d = 0.50; Figure 2b). By the 9-month follow-up visit, test performance dropped to floor levels in both the TMR and control conditions (Figure 1d: basic/trained questions: t(25) = 0.27, p = .79, d = 0.10; knowledge-transfer questions: t(25) = 1.34, p = .20, d = 0.56), with no evidence for a gender by condition interaction, F(1,23) = 0.18, p = .68, partial η² = 0.01.

Figure 2.

Performance on the next-day test in male and female students, separated by (a) knowledge-transfer questions and (b) trained questions. Error bars: standard error. * p ≤ .05.

3.3. Spectral Analysis Results

Table 3 shows that TMR affected sleep microarchitecture. Performance on lecture practice questions did not predict frontal theta activity across SWS (r(39) = −.19, p = .25). However, TMR specifically increased frontal theta activity averaged across SWS epochs throughout the night, t(39) = 2.67, p = .01, d = 0.87 (the next largest effect was for spindle density, t(39) = 1.89, p = .07, d = 0.55). To restrict the potential inflation in family-wise error rates for conducting multiple null hypothesis significance tests, we complemented our analyses with Bayesian analyses. Note that because corrections for family-wise type 1 error rates are adjustments of p-values, and the Bayesian analytical approach does not depend on this limitation of p-values, corrections for family-wise type I error rates are unnecessary in Bayesian analyses (Bender and Lange 2001). Table 3 illustrates that Bayesian analyses yielded similar findings. Gender stratified analyses indicated similar trends in both gender groups for theta activity during SWS at frontal sites (Figure 3a; Females: t(28) = 1.78, p = .09, d = 0.69; Males: t(9) = 1.99, p = .08, d = 1.14; Gender by Condition interaction: F(1,37) = 2.07, p = .16, partial η² = 0.05). Next, we tested whether frontal theta activity during SWS predicted test scores, with the scatterplots shown in Figure 4. Although frontal theta activity was not associated with levels of learning during the lecture in females (r(28) = −.01, p = .95), Figure 3b shows that greater frontal theta activity significantly predicted higher performance on knowledge transfer questions on the next-day test (r(28) = .42, 95% CI: .07, .68, p =.02). We observed a similar effect size at the 9-month follow-up test, albeit with marginal significance given the sample size attrition (r(18) = .42, 95% CI: −.02, .73, p = .06, Figure 4). There were too few male participants to conduct definitive correlational tests in this subgroup, and therefore we only report the correlation values for archival purposes (lecture performance: r(9) = −.34, p = .30; next-day test: r(9) = −.01, 95%CI: −.61, .59, p = .98; follow-up test: r(5) = .11, 95% CI: −.70, .80, p = .82; Fisher’s r-to-z-transformation test across the gender groups showed Z = 1.14, p = .25 for next-day test; Z = 0.61, p = .54 for follow-up test; Fisher, 1915).

Table 3.

Spectral Power and Spindle Density in TMR and Control Conditions

	TMR (n = 19)	Control (n = 22)	Condition Effect (p-value)	Condition Effect (Bayes Factor10)
Slow oscillation activity (0.5–1 Hz; μV2)
SWS at frontal sites	34.58 (12.87)	33.19 (15.16)	.76	0.32
SWS at all sites	20.39 (6.40)	20.38 (7.61)	>.99	0.31
All sleep epochs at all sites	7.40 (1.99)	8.29 (5.26)	.47	0.37
Delta activity (1–4 Hz; μ,V2)
SWS at frontal sites	87.38 (38.24)	75.99 (37.33)	.34	0.44
SWS at all sites	41.55 (14.32)	38.86 (16.12)	.58	0.35
All sleep epochs at all sites	14.37 (4.50)	13.65 (5.42)	.65	0.33
Theta activity (4–8 Hz; μ,V2)
SWS at frontal sites	6.51 (2.58)	4.72 (1.48)	.01*	5.64†
SWS at all sites	3.97 (1.44)	3.28 (0.91)	.08	1.19
All sleep epochs at all sites	2.29 (0.82)	1.95 (0.64)	.14	0.74
Spindle density during N2 at frontal sites	2.90 (0.32)	2.57 (0.75)	.07	1.07
Spindle density during SWS at frontal sites	1.54 (0.74)	1.35 (0.76)	.43	0.39

Note. Data presented as mean (standard deviation).

^*p ≤ .05

^†Bayes factor > 3 indicates that the data favor the alternative hypothesis over the null hypothesis.

Figure 3.

(a) The TMR group showed increased frontal midline theta power during SWS for both males and females. Values shown are normalized to z-scores, to highlight relative spatial power distributions across groups. (b) Knowledge transfer test performance was positively correlated with theta power during SWS in females, but not males. Significant correlations are marked with dark dots.

Figure 4.

(a) Theta power at frontal sites during slow wave sleep predicted performance on next-day knowledge-transfer questions in females, but not in males; (b) Theta power trended toward predicting knowledge-transfer question performance on the follow-up test in females, but not in males. Circles and solid lines represent females; triangles and dash lines represent males.

4. Discussion

This double-blinded TMR study showed that naturalistic sensory stimuli (classical music) can be leveraged to promote integration of college-level educational concepts. TMR increased theta activity during SWS, and greater theta activity across conditions was associated with better subsequent test performance. Interestingly, some results suggested that classical music TMR might particularly benefit females, a finding that converges with the literature on music neuroscience and that has implications for addressing achievement gaps in economics or other mathematics-based fields.

4.1. TMR Mechanisms

Most previous TMR work has focused on the mechanisms, or EEG signature, of memory reactivation during SWS. For example, studies of odor and discrete-sound TMR cues indicated that TMR increased spindle density (Creery et al. 2015; Oyarzún et al. 2017), delta/slow oscillation activity (Creery et al. 2015), or frontal theta activity (Farthouat et al. 2016; Oyarzún et al. 2017), and that such increases predicted subsequent performance. In the current study, using more continuous auditory stimuli, classical music TMR had no discernible impact on delta/slow oscillations, and there was only a marginal trend for an increase in spindle activity. However, classical music TMR was associated with significantly more frontal theta activity than the white-noise control group, and frontal theta activity predicted later test performance.

Increased theta activity during SWS can indicate successful processing of auditory cues (Cox et al. 2014; Farthouat et al. 2016). By some views, theta activity is also theorized to contribute to memory formation through hippocampal-neocortical long-term potentiation and synaptic plasticity mechanisms (Axmacher et al. 2006; Cantero et al. 2003; Nyhus and Curran 2010). Consistent with this latter view, the re-occurrence of theta rhythms during SWS was predictive of retention of specific memories in at least two studies (vocabulary/translation words and paired associative learning in Schreiner and Rasch 2014 and Farthouat et al. 2016, respectively). The current study indicates that similar theta-dependent mechanisms can be triggered across bouts of SWS by classical music, with the outcome being the integration of learned educational concepts rather than solely the consolidation of specific discrete memories.

An alternative view is that classical music during SWS benefits test performance not because of frontal theta activity or other memory reactivation mechanisms (i.e. TMR), but instead because there is some intrinsic cognitive benefit to listening to classical music. This notion, which has been popularized as the “Mozart effect,” is rooted in a few studies that found that listening to Mozart (or other energetic classical music) improved performance on the Stanford-Binet intelligence test (Rauscher et al. 1993). However, all reported benefits to performance have been temporary (15 minutes), meaning that this literature would not predict that listening to classical music while sleeping would help test performance 9 hours later. In addition, since the original media-popularized studies in the 1990s, most researchers have concluded that the Mozart effect is an artifact of arousal and mood (Thompson et al. 2001), with one meta-analysis finding that evidence for the Mozart effect was largely restricted to one laboratory (Pietschnig et al. 2010). In other words, Mozart does not make memories; but, pairing music with studying and then listening to that music again during sleep can be memory-promoting via TMR mechanisms.

4.2. Gender Differences in Music Processing and Academic Achievement

The convergent fields of music and neuroscience are developing rapidly (Cheever et al. 2018), and there are several studies demonstrating that females process music differently than males (Koelsch et al. 2003; Gaab et al. 2003). There are at least two mechanisms by which gender differences in music processing would explain why classical music TMR effects seemed very strong in females. First, in multitasking scenarios, females show that they can maintain cognitive task performance levels while listening to music whereas males cannot (Fancourt et al. 2016; Feizpour et al. 2018), implying that females may be able to study more effectively than males while listening to music, a pre-requisite for TMR to be successful (Creery et al. 2015). Second, females outperform males at later recognizing familiar music (Miles et al. 2016), implying that when familiar classical music was played during SWS, females’ ability to efficiently recognize the music would facilitate reactivation of the associated educational content. Of course, future studies with considerable statistical power to detect gender interactions are required to confirm (or falsify) this hypothesis that gender differences in musical processing allow females to specifically benefit from classical-music TMR.

Advancing knowledge of TMR and music processing might inform pathways to bridging achievement gaps in STEM learning. Gender differences have long been observed in mathematics-based disciplines (Fryer and Levitt 2010). There are numerous factors that contribute to the underrepresentation of women in some STEM fields (e.g., disposition, occupational interests); however, many of the factors known to influence STEM achievement gaps are themselves influenced by sleep and sleep restriction: cognitive abilities/relative strengths, field-specific ability beliefs, and gender-related stereotypes (Wang and Degol 2017). Restricting sleep amplifies stress/emotional reactivity and activates stereotypes (Ghumman and Barnes 2013; Goldstein-Piekarski et al. 2018), which in a cyclical fashion can contribute to more chronic sleep disturbances, thereby impairing attention and memory consolidation. If females are more susceptible to these detrimental effects than males (Gao et al. 2019; Hajali et al. 2012; Rångtell et al. 2019), then addressing sleep disparities may bridge STEM achievement gaps (Heissel et al. 2017). Classical music TMR may be one sleep-based approach to reducing sleep disparities and achievement gaps.

4.3. Limitations and conclusions

Limitations of the current study include attrition, sample size inequality, and low performance at the nine-month follow-up. Although a single TMR session may help students to cram for the next-day test, for effects to persist throughout the academic year, TMR sessions may need to be repeated and progressively spaced apart (Kornell 2009). Furthermore, while the current study used an ecological research paradigm, the research procedures were carried out in a controlled-laboratory setting. Future research that translates TMR procedures to home settings will be an important step towards remediating deficits in memory consolidation caused by insufficient sleep in college students (Goldi and Rasch 2019).

It may also be fruitful for future studies to include multiple control groups, for example, a group that listens to classical music during sleep but not during the lecture. While more a question of mechanistic import than of educational value, adding such a condition would disentangle TMR-based explanations from brainwave-entraining explanations. According to the brainwave-entraining view, listening to classical music during sleep entrains brainwave oscillations in a manner that leads to memory enhancement (Obleser and Kayser 2019). However, for this explanation to be viable, several methodological conditions must be met. First, when music entrains brainwaves, the elicited state evoked potentials will synchronize to the tempo of the music (Daly et al. 2014); but, in the current study, the classical music tempos never matched theta frequency, meaning that they should not entrain theta waves (110 to 147 beats per minute, corresponding to 1.83 to 2.45 Hz). Second, for auditory stimulation during sleep to entrain specific brainwave frequencies, the stimulation must consistently occur in-phase with ongoing oscillations (Ngo et al. 2013). However, the Chopin, Beethoven, and Vivaldi pieces used here lacked consistency in interstimulus intervals, again disfavoring an entrainment explanation. The practical recommendation, therefore, is that students only listen to classical music while sleeping if they first paired that music with studying.

The current work showed that TMR is achievable with continuous auditory cues (classical music) and integrative educational concepts. Classical music TMR can improve performance on knowledge-transfer questions on the next-day test. Some findings indicated that females were particularly benefited by classical music TMR, but definitive gender-based conclusions will require additional future testing. Future work at the intersection of music, memory theory, and neuroscience is needed to enhance educational outcomes and bridge achievement gaps in STEM learning.

Highlights:

• Classical music TMR improves next-day performance on STEM learning content.

• TMR is particularly beneficial for test items that measure knowledge transfer/integration.

• TMR techniques can potentially be used to bridge gender achievement gaps in STEM.

Abbreviations:

TMR

targeted memory reactivation

STEM

science, technology, engineering, and mathematics

SWS

slow wave sleep

REM

rapid eye movement

PSG

polysomnography

RAPM

Raven’s Advanced Progressive Matrices

MEQ

Morningness-Eveningness Questionnaire

PSQI

Pittsburgh Sleep Quality Index

ESS

Epworth Sleepiness Scale

WASO

wake after sleep onset