Supporting healthy development in adolescence: Technology-supported cooperative learning can reduce stress and increase sleep quality

Abstract

Peer relationships are a significant source of stress for adolescents that can negatively impact sleep quality. Cooperative learning can reduce adolescent stress by enhancing positive social interactions in school, which may improve adolescent sleep quality. This study evaluated (a) the effects of technology-assisted cooperative learning (i.e., PeerLearning.net) on adolescents’ sleep quality, sleep duration, and sleep onset latency; (b) whether these effects were mediated by reductions in stress; and (c) whether effects were moderated by race and ethnicity, sex, grade level, and dosage. We conducted a cluster randomized trial with 12 middle and high schools in the Pacific Northwest (n = 6 intervention, n = 6 control) and collected two waves of data from a sample of 813 students (50.2% female, 70.7% White, US Grades 6–9, ages 12–16 years). Results indicated significantly reduced stress (R² = .80) and increased perceived sleep quality (R² = .47) among adolescents after implementing technology-assisted cooperative learning, including a significant effect for dosage, but no effects on sleep duration or sleep onset latency. Effects on perceived sleep quality were mediated by effects on stress. No moderation by sex, grade, or race/ethnicity was found. Our findings (and those from previous research) suggested that technology-assisted cooperative learning is a promising universal school-based prevention program that can impact a wide range of student (and teacher) outcomes.

Introduction

It is recommended that adolescents ages 13–18 years should sleep for 8–10 hr per night (Paruthi et al., 2016), but in a recent national survey, 72.7% of high school students reported getting insufficient sleep on school nights (Wheaton et al., 2018). Sleep problems often emerge in the adolescent years due to later circadian timing and reductions in homeostatic sleep pressure (Dahl & Lewin, 2002). From a developmental perspective, poor or insufficient sleep in adolescence is associated with a variety of negative outcomes, including increased emotional problems (e.g., anxiety, depression), impaired cognitive functioning (e.g., attention, regulation), and increased behavioral problems (e.g., self-harm, aggression; Beebe, 2011; Kansagra, 2020; Wang et al., 2016). Insufficient sleep, sleepiness, and poor sleep quality can also negatively impact adolescents’ school performance (Dewald et al., 2010). Given the importance of sleep to adolescent development and the chronic lack of sleep that has been reported among adolescent populations, the American Academy of Pediatrics (2014) formally recognized sleep problems in adolescence as an international public health concern.

Stress and Sleep

Among the factors that can impact sleep quality in adolescence, stress has emerged as one of the most important (Åkerstedt et al., 2012). Stress causes increased psychological and physiological activation that is incompatible with the deactivation required for sleep (Åkerstedt, 2006). In fact, stress was the strongest predictor of poor sleep quality in a large study of college students, with stress acting as a predisposing, precipitating, and perpetuating factor of poor sleep quality (Lund et al., 2010). The relationship between stress and sleep may be reciprocal, with 36% of adolescents reporting sleepiness due to stress and 18% of adolescents reporting being more stressed due to lack of sleep (Bethune, 2014). Thus, reducing stress may be an avenue to improving adolescent sleep.

Peer Relations as a Source of Stress

Peer relations are a notable source of stress for adolescents. As youth begin to separate from families, peers become increasingly important as a source of support and affiliation (Steinberg & Morris, 2001). During this developmental period, youth often form close, supportive relationships outside the family unit as a preparation for pair bonding and, eventually, parenthood (Allen & Land, 1999). Indeed, the achievement of positive peer relations is seen as a critical developmental milestone in adolescence (Bukowski et al., 2011; Collins & Steinberg, 2006), and as a result, youth during this period are highly attuned to social reward and vigilant for signs of peer indifference or rejection (Fareri et al., 2008; Masten et al., 2009; Spear, 2000). At the same time, adolescents are less able to self-regulate due to developmental deficits in cognition and, as a result, have a more difficult time managing their emotional response to stressful situations (Casey et al., 2000; Yurgelun-Todd, 2007). Thus, difficulties with peers are a significant source of stress for adolescents that can create risk for a range of psychosomatic symptoms (Murberg & Bru, 2004), including increased anxiety and depression and low self-esteem (Hankin et al., 2015; Moksnes et al., 2010; Platt et al., 2013; Rudolph, 2002). Stress can also interfere with academic engagement (Raufelder et al., 2014) and, as noted above, can cause problems with sleep (Lund et al., 2010). From a prevention perspective, if the risk for negative peer interactions could be reduced, then risk for these negative outcomes (including poor sleep) could also be reduced.

Cooperative Learning as a Universal School-Based Prevention Strategy

Stress related to peer relationships may be reduced through Cooperative Learning (CL; Johnson et al., 2013), which is a collaborative group-based instructional technique. CL reduces adolescent stress by promoting positive social interactions in small group learning situations (Van Ryzin & Roseth, 2021). However, to be effective, CL lessons must be structured and delivered according to four key design principles. First, CL lessons must establish positive interdependence among students in learning situations where students can gain personally from helping others (Johnson et al., 2013). For example, teachers may require a single finished product from a group (i.e., goal interdependence) or offer a reward to the group if everyone achieves above a certain threshold (i.e., reward interdependence). Alternatively, the lesson may specify that each member of the group fulfill a different role (i.e., role interdependence) or complete a unique task (i.e., task interdependence) for a lesson to be completed successfully. Second, CL lessons must ensure individual accountability such that students have an incentive to contribute to the success of the group. This may take the form of an end-of-unit test to be taken individually (with the potential for group rewards via reward interdependence) or an oral quiz of a random group member by the teacher as they supervise the group work during the lesson. Third, CL lessons must include explicit development of group social skills (e.g., checking for understanding among group members, summarizing group discussions), which includes scaffolding targeted skills and monitoring and observation by the teacher with positive reinforcement for examples of such behavior. Fourth, CL must include guided post-lesson reflection and processing of group performance after the lesson in which the group discusses what they did well, sets targets for improvement, and provides one another with positive reinforcement for behavior that contributed to group success. These design principles are summarized in Table 1.

Table 1.

Cooperative Learning Design Principles and Strategies

Principle	Strategies
Positive Interdependence	Goal interdependence: Teachers require a single finished product from a group. Reward interdependence: A reinforcement for the group if everyone achieves above a certain threshold on an assessment. Role interdependence: Each member of the group must fulfill a different role for the lesson to be completed successfully. Task interdependence: Each member of the group must complete a unique task for the lesson to be completed successfully.
Individual Accountability	Individual contributions to group success graded separately. An end-of-unit test to be taken individually. An oral quiz of a random group member by the teacher as he/she supervises the group work during the lesson.
Group Social Skill Development	Scaffolding of the targeted skill before the lesson, including sentence fragments indicating what the skill “sounds like”. Monitoring and observation of the target skill during the lesson. Positive reinforcement for examples of such behavior both during the lesson and directly afterward.
Post-lesson Reflection and Group Processing	The group discusses what they did well, what they could do better next time, and sets targets for improvement. Group members provide one another with positive reinforcement for behavior that contributed to group success.

When these four key design principles are established in CL lessons, students are more likely to promote the success of one another through instrumental and emotional support and sharing of information and resources (Johnson et al., 2013). These positive social interactions during learning situations can enhance peer relations (Hedges’s g = .42–.48; Roseth et al., 2008) and peer support (R² = .33; Van Ryzin et al., 2020) and reduce negative peer interactions (e.g., victimization; R² = .36; Van Ryzin & Roseth, 2018). Because of these more positive peer interactions, CL can reduce adolescent stress (R² = .24) and, in turn, mitigate typical consequences of stress, both within the school setting (i.e., enhancements to academic engagement; R² = .35) and outside the school setting (i.e., reductions in mental health problems; R² = .29; Van Ryzin & Roseth, 2021).

Support for Implementing Cooperative Learning

Notably, implementing the key design principles does not always occur during CL lessons, and when some principles are left out, the impact of CL is greatly reduced (McMaster & Fuchs, 2002; Roseth et al., 2008). Unfortunately, evidence indicates that teachers rarely include all key design principles when delivering CL lessons due to implementation burdens such as limited time to plan, difficulties managing lesson delivery, and lack of understanding of the key design principles (Abrami et al., 2004; Buchs et al., 2017; Gilles & Boyle, 2010). Thus, this study leveraged a Web-based technology platform (i.e., PeerLearning.net) to support teachers in implementing CL with fidelity to all the key design principles, ensuring that CL would have the maximum possible impact.

PeerLearning.net was developed to promote equitable access to evidence-based practices in instruction for students in under-resourced school environments while also supporting greater implementation fidelity and scalability and reducing implementation costs. It contains several lesson templates, such as jigsaw, group projects, and peer tutoring, that promote fidelity to the key CL design principles outlined above. The lesson design process is implemented as a series of Wizard-like screens that ask users to (a) specify lesson details, including lesson title, subject, grade, and topic/domain; (b) upload the lesson content and materials; (c) create an outline for student note-taking through PeerLearning.net; (d) add a follow-up activity to the lesson if desired; (e) insert information for students, including a description of the learning goal for the students, a plan for assessment, and any group-level bonuses that are offered to enhance positive interdependence; and (f) specify the length of each phase of the lesson to ensure that it can finish on time.

During lesson delivery, PeerLearning.net enables teachers to manage membership in the learning groups through a drag-and-drop user interface and assists teachers in distributing the uploaded instructional materials, directing student activities, supporting observations of student behavior, and delivering post-lesson group reflection activities. PeerLearning.net removes the logistical burden for teachers by ensuring that each student (a) is working with the correct collaborators, (b) possesses the correct learning materials, and (c) is aware of the goals and tasks for the current phase of the lesson and the amount of time remaining before transitioning to the next phase. For each lesson template (e.g., jigsaw, group project), PeerLearning.net supports the teacher in following a specific outline of lesson phases that represent evidence-based best practice (see Johnson et al., 2013). By ensuring that students have all the information they need to proceed through the lesson, teachers are freed to observe and reinforce the targeted social skill by providing positive reinforcement for examples of such behavior (e.g., written or verbal recognition). Teachers also have additional time to support students and groups that need extra help with lesson content or group processes.

Present Study

In this study, we evaluated whether CL, as delivered with PeerLearning.net, can improve sleep in adolescents via a reduction in stress. As presented in Figure 1, existing curricula and learning materials are modified to suit a small-group format and delivered via PeerLearning.net, which supports teachers in delivering lessons with fidelity to evidence-based best practices in small-group instruction. The positive social interactions during these small-group lessons promote positive peer relations in the classroom and reduce student stress (Van Ryzin & Roseth, 2021), which can have beneficial effects outside the educational context (e.g., mental health; Zagni & Van Ryzin, 2024). In this study, we evaluated sleep quality as our key outcome.

Figure 1.

Hypothesized Model

As far as we are aware there is no research examining the relationship between CL and sleep; however, as noted above, stress is a key predictor of sleep quality (Åkerstedt et al., 2012; Lund et al., 2010). Therefore, we hypothesized that (a) CL/PeerLearning.net would reduce stress in adolescents, which in turn would (b) improve adolescent sleep. In measuring sleep, we used the PSQI, which is a subjective measure of sleep, because previous research has found inconsistent results when examining links between stress and sleep with alternative measures such as sleep diary and actigraphy (Slavish et al., 2021). This study did not include the impact of CL/PeerLearning.net on peer relations because those results have been reported elsewhere, finding that CL/PeerLearning.net can promote significantly more positive peer relations (Zagni & Van Ryzin, 2024).

We also evaluated the impact of demographics. Given that some effects of CL have been found to be stronger for students of color (Van Ryzin et al., 2020), we evaluated student race and ethnicity as a moderator of effects. Considering this is the first study examining CL and sleep, we also conducted an exploratory analysis to evaluate whether results were moderated by student sex and grade level, even though there is no evidence for moderation by sex or grade level in previous research on CL (Roseth et al., 2008). Finally, we evaluated whether there were effects of dosage (i.e., whether the number of lessons taught with PeerLearning.net in each school impacted the degree of change in student outcomes).

Method

Sample

All aspects of this study were approved by the Institutional Review Board (IRB) at the Oregon Research Institute and the study was performed in accordance with the ethical standards from the 1964 Declaration of Helsinki and its later amendments. The sample was derived from a 1-year cluster randomized trial of PeerLearning.net in 12 middle and high schools in the Pacific Northwest (registered on ClinicalTrials.gov; NCT04478240). Schools were matched based upon school level (i.e., middle vs. high), size, and demographics (i.e., free or reduced lunch percentage and race and ethnicity) and one school from each matched pair was randomized to the intervention condition using a random number generator, with the paired school being assigned to the control condition.

Our sample included 813 students who participated in the project during the 2021–2022 school year. Students were in middle or high school when they were initially enrolled in the project; we intended to enroll eighth graders (approximately ages 14–15 years) in middle schools and ninth graders (approximately ages 15–16 years) in high schools, but two middle schools requested that we enroll sixth graders instead (approximately ages 12–13 years). Any student who participated in at least one wave of data collection was included in the analytic sample, which was 50.2% female (n = 408) and 70.7% white (n = 575). Other racial/ethnic groups included Hispanic/Latino (n = 146, 18.0%) and multi-racial (n = 61, 7.5%); our sample included less than 2% Asian (n = 12), African-American (n = 9), American Indian/Alaska Native (n = 8), and Native Hawaiian/Pacific Islander (n = 2). Student demographics by school and the number of lessons taught in each school with PeerLearning.net (this total includes only the teachers in the relevant grade) are provided in Table 2. We note that Table 2 provides school-level percentages for White, Latinx, and multi-racial students only (given the extremely small numbers of students who were Asian, African-American, American Indian/Alaska Native, and Native Hawaiian/Pacific Islander). We were able to successfully recruit between 79%–97% of eligible students in each school to participate (the non-participating students either opted out themselves or were opted out by their parents). Student free and reduced-price lunch status and special education status were not made available by the schools, so we added school-level figures (obtained from state records) to Table 2.

Table 2.

Intervention Condition, Sample Size (number of students), Grade Level, Sex, Race/Ethnicity, Special Education Status, Free/Reduced Price Lunch Percentage, and Lessons Taught by School

School	Intervention	Grade	n	% Female	% White	% Latinx	% Multi	% SPEDa	% FRPLa	Lessons
1	Yes	MS (6)	70	52.9	62.9	25.7	7.1	22	> 95	31
2	Yes	MS (8)	25	44.0	76.0	20.0	4.0	13	> 95	45
3	Yes	MS (8)	83	47.0	71.1	14.5	12.0	13	> 95	86
4	Yes	HS (9)	53	47.2	73.6	15.1	9.4	21	82	108
5	Yes	HS (9)	67	50.7	68.7	20.9	6.0	16	> 95	75
6	Yes	HS (9)	91	51.6	72.5	18.7	6.6	14	> 95	71
7	No	MS (6)	78	50.0	71.8	15.4	7.7	15	> 95	0
8	No	MS (8)	53	37.7	62.3	24.5	9.4	13	> 95	0
9	No	MS (8)	87	54.0	70.1	18.4	6.9	15	> 95	0
10	No	HS (9)	33	48.5	66.7	21.2	9.1	7	> 95	0
11	No	HS (9)	52	59.6	71.2	17.3	< 2.0	15	> 95	0
12	No	HS (9)	121	51.2	76.9	12.4	7.4	14	> 95	0

Note. MS (6) = middle school, 6^th grade; MS (8) = middle school, 8^th grade; HS (9) = high school, 9^th grade; SPED = special education; FRPL = free/reduced price lunch; Multi = Multi-racial; Lessons = the number of CL lessons delivered with PeerLearning.net.

^aState records

At least 95% of the participating students in each school completed the surveys at Wave 1 to create our initial sample of 813, but we missed 93 of these students at Wave 2 (11.4%). The missing students were not significantly different from the students who participated in Wave 1 in terms of sex (r = .01, p = .88), race/ethnicity (r = .01, p = .85), or intervention condition (r = .06, p = .11).

Teacher Training

Training in PeerLearning.net for teachers in intervention schools began in the fall and consisted of two independent 2-hr sessions roughly 2 months apart, periodic check-ins via videoconference, and support via email. The training was experiential as well as informational, with teachers actively involved in their own learning through CL lessons delivered with PeerLearning.net in which teachers participated as students. Teachers were given multiple lessons covering foundational theory, key design dimensions, and numerous examples, and each lesson was implemented using a form of CL. In this way, teachers developed not only an understanding of CL, but also experienced the lessons from the student perspective, developing an appreciation for the social nature of CL how it is active and student-centered as well as highly structured, and how this contributes to positive social and academic outcomes (Roseth et al., 2008).

After the training, teachers were asked to adapt existing lessons to be delivered through PeerLearning.net. For example, a teacher may have an existing lecture with a whole-class discussion on a certain topic that can be reconfigured into a jigsaw lesson where students teach and learn from one another in small groups. In such a scenario, the lesson material would be divided into two to four independent subtopics and each student in a jigsaw group would be made responsible for learning one subtopic in the company of other students who are assigned the same material. Once the subtopic is learned, students then teach what they know to other students in a jigsaw group. As each student in a jigsaw group teaches their topic, students are exposed to all the lesson material.

We asked teachers in intervention schools to use PeerLearning.net to deliver a lesson at least twice per month. At the conclusion of the school year, data from PeerLearning.net indicated that 29 of 52 (56%) of eligible teachers in intervention schools delivered more than one lesson. On average, these participating teachers each delivered about 2–3 lessons per month with PeerLearning.net.

Measures

Student data collection was conducted in September/October and March/April (two waves in total about 6 months apart) using online surveys (i.e., Qualtrics; https://www.qualtrics.com/).

Stress

We used three items from the Perceived Stress Scale (PSS; Cohen et al., 1983), a common measure of stress among adolescents that included the following three items: “In the last month, how often have you been upset because of something that happened unexpectedly?”, “In the last month, how often have you felt nervous and stressed?”, and “In the last month, how often have you felt that things were going your way?” (the latter being reverse scored). Students selected among five Likert-style options from 0 (Never) to 4 (Very often) and scale items were averaged to arrive at the scale score. The PSS items had adequate internal reliability (McDonald’s omega = .67 at Wave 1 and .72 at Wave 2). The validity of the PSS has been established in previous research (Lee, 2012).

Sleep quality

We used several items from the Pittsburgh Sleep Quality Index (PSQI; Buysse et al., 1989), including sleep onset latency (i.e., “During the past month, how long in minutes has it usually taken you to fall asleep?”), sleep duration (i.e., “During the past month, how many hours of actual sleep have you usually gotten each night?”), and perceived sleep quality (i.e., “During the past month, how would you rate your sleep quality overall?”). For sleep onset latency, students were given options in 5-min increments ranging from 0 (0 minutes) to 19 (90 minutes); there was also an option for More than 90 minutes (20). For sleep duration, students were given options in 30-min increments from 4 hours (1) to 10 hours (13); there was also an option for More than 10 hours (14). For perceived sleep quality, students selected among four Likert-style options from Very bad (1) to Very good (4). The PSQI is frequently used to assess sleep quality in adolescents (e.g., Blake et al., 2016; John et al., 2016; Tan et al., 2012; Zhang et al., 2012) and it has been validated in children, adolescents, and young adults (De La Vega et al., 2015; Larche et al., 2021; Raniti et al., 2018).

Demographics

Sex and race/ethnicity were collected from school records and coded Male (0) or Female (1) and White (0) or Student of Color (1). The latter coding was necessary given the small size of most of our racial/ethnic groups that did not present sufficient statistical power to conduct a detailed examination of group differences.

Use of Cooperative Learning

We asked teachers in control schools to indicate the frequency with which they used CL principles using a standard assessment (presented in Veenman et al., 2002). This survey asks teachers to indicate, when using CL, whether they felt competent, whether it required extra preparation time, etc. Because these questions would only be meaningful if CL was actually being used, we also provided an option for teachers to indicate whether they used CL; if they indicated that they did not, then they were not asked to respond to questions about their experiences. Budget constraints prevented us from making any in-class observations of teacher behavior.

Analysis Plan

Because students were nested within schools in our sample, we fit a 2-level hierarchical linear model (HLM) using R (lme4 package; Bates et al., 2014) to evaluate our study hypotheses. This package uses Maximum Likelihood (ML) analysis to produced unbiased estimates in the presence of missing data (Baraldi & Enders, 2010). In all tests, our threshold for significance was α = .05. For each outcome, we initially calculated the Intraclass Correlation Coefficient (ICC), which indicates the proportion of variance that is present at Level 2 (i.e., between schools, instead of within schools). The larger the ICC, the more variance is present between schools.

Tests of Pre-Existing Group Differences

We initially fit models with intervention condition predicting student outcomes from Wave 1 to ascertain whether we had any pre-existing differences between our intervention and control groups. We found that students in intervention and control schools did not differ in terms of Wave 1 levels of stress (B = .10, SE = .10, p = .33), sleep latency (B = .27, SE = .41, p = .52), amount of sleep (B = .17, SE = .29, p = .57), or perceived sleep quality (B = −.03, SE = .09, p = .72), indicating that our sample did not have pre-existing group differences.

Tests of Main Effects on Stress and Sleep

Next, we evaluated the ability of intervention condition to predict student outcomes at Wave 2, controlling for Wave 1 values, as follows:

$$\begin{array}{ll}\left(\text{level}1-\text{student}\right)\hfill & {\text{Y}}_{\text{ij}}={\pi }_{0\text{j}}+{\pi }_{1\text{j}}{\left(\text{Wave}1\text{value}\right)}_{\text{ij}}+{\text{e}}_{\text{ij}}\hfill \\ \left(\text{level}2-\text{school}\right)\hfill & {\pi }_{0\text{j}}={\beta }_{00}+{\beta }_{01}{\left(\text{Intervention}\right)}_{\text{j}}+{\text{r}}_{0\text{j}}\hfill \\ \left(\text{level}2\right)\hfill & {\pi }_{1\text{j}}={\beta }_{10}\hfill \end{array}$$

In this model, Y_ij is the outcome value for student i in school j. The mean school-level effect of PeerLearning.net on student outcomes is represented by β₀₁. Given the small size of our sample (12 schools), we were not able to fit random effects for Wave 1 values in these models, so the impact of Wave 1 measures on Wave 2 measures (i.e., continuity) was fixed across our sample. As is standard for HLM, separate models were run for each outcome (i.e., perceived stress and each of the three sleep outcomes [sleep onset latency, sleep duration, perceived sleep quality]). For significant effects, we calculated an effect size in terms of the percentage of variance explained at the appropriate level of the model. We also evaluated the effects of dosage by substituting the number of lessons taught with PeerLearning.net in each school (see Table 2) in place of the intervention condition.

Tests of Moderation by Demographics

To evaluate moderation by student demographics (e.g., sex, race/ethnicity), we inserted a cross-level interaction effect into the model as follows:

$$\begin{array}{ll}\left(\text{level}1-\text{student}\right)\hfill & {\text{Y}}_{\text{ij}}={\pi }_{0\text{j}}+{\pi }_{1\text{j}}{\left(\text{Wave}1\text{value}\right)}_{\text{ij}}+{\pi }_{2\text{j}}{\left(\text{Demographics}\right)}_{\text{ij}}+{\text{e}}_{\text{ij}}\hfill \\ \left(\text{level}2-\text{school}\right)\hfill & {\pi }_{0\text{j}}={\beta }_{00}+{\beta }_{01}{\left(\text{Intervention}\right)}_{\text{j}}+{\text{r}}_{0\text{j}}\hfill \\ \left(\text{level}2\right)\hfill & {\pi }_{1\text{j}}={\beta }_{10}\hfill \\ \left(\text{level}2\right)\hfill & {\pi }_{2\text{j}}={\beta }_{20}+{\beta }_{21}{\left(\text{Intervention}\right)}_{\text{j}}+{\text{r}}_{2\text{j}}\hfill \end{array}$$

In this model, the interaction is comprised of the intervention condition at Level 2 and (via π_2j) student demographics (e.g., sex, race/ethnicity) at Level 1. The interaction effect represented by β₂₁ was evaluated for significance; if significant, it would indicate that intervention effects differed by student demographics (i.e., moderation).

Tests of Mediation of Effects on Sleep Outcomes

Finally, we evaluated whether change in stress from Wave 1 to Wave 2 would mediate intervention effects on sleep outcomes at Wave 2. Tests of mediation used the standard four-step process (Judd et al., 2001) in which we evaluated (a) main effects of the intervention on outcomes, (b) effects of the intervention on the mediator, (c) effects on outcomes with the mediator in the model, and (d) the indirect effect of the intervention on the outcome via the mediator using on-line tools (i.e., Preacher’s Monte Carlo method for assessing mediation in multi-level models, available as an on-line calculator). The first two steps were addressed above; for the third step, we inserted a variable represented change in stress (the mediator) into our model and evaluated the effect on the outcome. Change in stress was calculated by regressing stress at Wave 2 on stress at Wave 1 and retaining the residuals. The revised model was:

$$\begin{array}{ll}\left(\text{level}1-\text{student}\right)\hfill & {\text{Y}}_{\text{ij}}={\pi }_{0\text{j}}+{\pi }_{1\text{j}}{\left(\text{Wave}1\text{value}\right)}_{\text{ij}}+{\pi }_{2\text{j}}{\left(\text{Change in Stress}\right)}_{\text{ij}}+{\text{e}}_{\text{ij}}\hfill \\ \left(\text{level}2-\text{school}\right)\hfill & {\pi }_{0\text{j}}={\beta }_{00}+{\beta }_{01}{\left(\text{Intervention}\right)}_{\text{j}}+{\text{r}}_{0\text{j}}\hfill \\ \left(\text{level}2\right)\hfill & {\pi }_{1\text{j}}={\beta }_{10}\hfill \\ \left(\text{level}2\right)\hfill & {\pi }_{2\text{j}}={\beta }_{20}\hfill \end{array}$$

The effect of change in stress on the outcome, represented by β₂₀, was evaluated for significance; if significant, it would indicate that the mediator had a significant impact on the outcome, satisfying the third requirement for mediation. If the intervention effect (β₀₁) was no longer significant with the mediator in the model, this would suggest full mediation; if the intervention effect was still significant, this would suggest partial mediation.

If mediation was present, we then evaluated the significance of the indirect effect (i.e., the combined effect of the intervention on the mediator, and the mediator on the outcome) with an online tool that uses Monte Carlo methods to calculate a 95% confidence interval for the indirect effect (Selig & Preacher, 2008). If the indirect effect was significant (i.e., if it had a 95% confidence interval that did not contain zero), then this would satisfy the fourth and final requirement for mediation.

Results

Descriptive Statistics

We provide descriptive data and correlations for student data in Table 3. In general, female (vs. male) students reported more stress and more trouble with sleep, including longer sleep onset latency, lower amounts of sleep, and lower perceived sleep quality. There were no differences by race or ethnicity in stress or sleep. When evaluating teacher responses, we found that teachers in control schools indicated no use of CL.

Table 3.

Correlations and Descriptive Data

	1	2	3	4	5	6	7	8	9	10
1. Stress (Wave 1)	—
2. Stress (Wave 2)	.27***	—
3. Sleep onset latency (Wave 1)	.20***	.19**	—
4. Sleep onset latency (Wave 2)	.10**	.39***	.45***	—
5. Sleep duration (Wave 1)	−.37***	−.20***	−.26***	−.11**	—
6. Sleep duration (Wave 2)	−.18***	−.39***	−.21***	−.38***	.35***	—
7. Perceived sleep quality (Wave 1)	−.47***	−.19***	−.37***	−.14***	.47***	.22***	—
8. Perceived sleep quality (Wave 2)	−.14***	−.63***	−.22***	−.43***	.21***	.47***	.22***	—
9. Sex	.15***	.24***	.11**	.04	−.12***	−.06	−.13***	−.09*	—
10. Race/ethnicity	−.08	.05	.03	−.08	.02	.01	.03	.01	.12***	—
N (overall)	805	713	806	714	808	707	810	716	813	813
M (overall)	2.40	2.16	7.09	6.58	7.21	7.49	2.70	2.74	.50	.29
SD (overall)	.86	1.00	5.04	5.02	3.04	2.92	.78	.94	-	-
N (control group)	396	344	395	342	397	340	397	342	399	399
M (control group)	2.35	2.41	6.94	6.83	7.11	7.37	2.72	2.51	.51	.28
SD (control group)	.85	.86	4.92	4.63	3.14	2.82	.81	.95	-	-
N (intervention group)	409	369	411	372	411	367	413	374	414	414
M (intervention group)	2.44	1.91	7.23	6.35	7.30	7.59	2.69	2.94	.49	.30
SD (intervention group)	.87	1.06	5.15	5.34	2.93	3.01	.75	.88	-	-

Note. The variables are numbered on the left and these numbers are repeated across the top of the table. Sex is coded as Female (1) or Male (0). Race/Ethnicity is coded as Student of Color (1) or White (0). Standard deviations were not calculated for dichotomous variables.

Tests of Main Effects and Moderation

When evaluating study hypotheses, we found that students in schools where teachers were given access to PeerLearning.net reported (a) significantly less stress (effect size [ES] = .80) and (b) more positive perceived sleep quality (ES = .47); however, there were no differences by intervention condition in sleep onset latency or duration (see Table 4). As noted above, ES was calculated as the amount of variance explained at Level 2 (i.e., between schools) by the intervention condition. For both significant outcomes, the proportion of variance between schools (i.e., ICC) was moderate (ICC = .08 and .10, respectively), indicating that most variance in these outcomes was at Level 1 (i.e., within schools). For our non-significant outcomes, ICCs were very low (ICC = .01 and .02), indicating that there were almost no differences between schools and thus very little variance to be explained by the intervention condition.

Table 4.

Intervention Effects at Follow-Up

Effect	Estimate (SE)	p	Effect size
Model #1: Predicting Stress (Wave 2; ICC = .08)
Level 1: Stress (Wave 1)	.33 (.04)	p < .001	.08
Level 2: Intervention condition	−.49 (.10)	p = .002	.80
Model #2: Predicting Sleep onset latency (Wave 2; ICC = .01)
Level 1: Sleep onset latency (Wave 1)	.45 (.03)	p < .001	.20
Level 2: Intervention condition	−.60 (.37)	p = .152	-
Model #3: Predicting Sleep duration (Wave 2; ICC = .02)
Level 1: Sleep duration (Wave 1)	.33 (.03)	p < .001	.12
Level 2: Intervention condition	.07 (.28)	p = .806	-
Model #4: Predicting Perceived sleep quality (Wave 2; ICC = .10)
Level 1: Perceived sleep quality (Wave 1)	.26 (.04)	p < .001	.07
Level 2: Intervention condition	.38 (.13)	p = .017	.47

Note. Estimates are unstandardized betas. Effect sizes were calculated as the percentage of variance explained at the appropriate level of the model. No effect sizes were calculated for non-significant effects.

These results were not moderated by sex, grade level, or race-ethnicity. When substituting the number of lessons taught for the intervention condition in our multi-level models, we found effects for dosage such that each lesson taught resulted in a significant reduction in stress and increase in perceived sleep quality (see Table 5). Mirroring our main effects analysis, we found no effects for dosage on sleep onset latency or duration.

Table 5.

Dosage Effects at Follow-Up

Model	B (SE)	Sig
Stress	−.0079 (.0009)	p < .001
Sleep onset latency	−.0083 (.0046)	p = .118
Sleep duration	.0018 (.0035)	p = .619
Perceived sleep quality	.0056 (.0015)	p < .001

Test of Mediation

We evaluated mediation of the intervention effect on perceived sleep quality by change in stress (we did not evaluate mediation for sleep onset latency or duration, as the intervention effects were not significant). We found that the effect of change in stress, when inserted into the model predicting sleep quality, was significant (B = −.56, SE = .03, p < .001). The negative regression coefficient indicated that reductions in stress predicted improvements in sleep quality. The effect of the intervention condition was no longer significant in this model (B = .12, SE = .10, p = .23), suggesting full mediation. The indirect effect of CL on sleep quality via reductions in stress (-.49 * −.56 = .27) had a 95% confidence interval of .22 to .31; because this interval does not include zero, the indirect effect can be considered statistically significant.

Discussion

Sleep is crucial to adolescent development, and a key disruptor of sleep in adolescence is stress (Åkerstedt et al., 2012; Lund et al., 2010). In this study, we found some evidence to support our hypothesis presented in Figure 1 that CL lessons, as delivered by PeerLearning.net, can reduce stress and, in turn, improve sleep. Specifically, students in schools where teachers used PeerLearning.net to deliver CL lessons with their own curricula and learning materials reported (a) significantly less stress and (b) an increase in their perceived sleep quality compared to students in the control schools where teachers reported no use of CL. Based upon previous research, we argue that CL confers these benefits through significant improvements in peer relations, which has been found to reduce stress (i.e., improvements in peer relations mediated the relationship between CL and reduced stress; Van Ryzin & Roseth, 2021). There were no group differences in these results by race or ethnicity, sex, or grade level. We found effects of dosage on these outcomes, indicating that every lesson taught with PeerLearning.net had salutary effects on student stress and perceived sleep quality.

In contrast, CL/PeerLearning.net had no impact on sleep duration and sleep onset latency, which was contrary to our hypotheses. Individual differences in sleep characteristics, such as insomnia symptoms, may have attenuated intervention effects. For example, sleep onset latency increases after experiencing cognitive stress before bed in individuals without insomnia, but decreases in those with insomnia (De Koninck, & Koulack, 1975). Cognitive stress before bed may act as a distractor from typical stressors, such as worries regarding sleep, in individuals with insomnia (Haynes et al., 1981). In addition, there are neurophysiological differences between people with normal sleep and those suffering from insomnia such that those with insomnia may have a distorted judgement of when sleep is initiated and the duration of sleep (Espie, 2007). Notably, an estimated 23.8% of adolescents have insomnia (Donskoy & Loghmanee, 2018). Insomnia and other individual differences in the effects of CL/PeerLearning.net on sleep duration and sleep onset latency may have obscured mean effects of the intervention and should be examined in future research.

In general, adolescent stress and negative peer relations have been associated with multiple sleep characteristics, such as sleep duration, quality, and disturbances (De Lise et al., 2023), although findings are mixed. Previous longitudinal research that used the same measures for stress and sleep (i.e., PSS and PSQI) found similar results whereby stress was predictive of sleep quality rather than quantity among young adults (Wallace et al., 2017). Similarly, research has also found that only sleep quality, not quantity, was associated with cortisol responses to stress in a sample of 10–17-year-olds (Capaldi et al., 2005). In addition, several studies have found that sleep quality, relative to sleep quantity, appears to be more strongly associated with other negative outcomes (e.g., depression, anxiety, health complaints; Vazsonyi et al., 2021), which might provide a potential explanation regarding the differential associations between adolescent stress and sleep quality versus sleep latency and duration that were found in this study.

Mediation

We found that change in stress fully mediated the impact of CL/PeerLearning.net on change in perceived sleep quality. Specifically, there were no longer any direct effects of the intervention, suggesting that the mechanism of stress reduction fully accounted for the effects of CL/PeerLearning.net on perceived sleep quality. The associations between CL and perceived stress, as well as that between perceived stress and subjective sleep quality, are consistent with previous research (Åkerstedt, 2006; Van Ryzin & Roseth, 2021; Zhao et al., 2021). This finding provides support for our hypothesis (see Figure 1) that the primary mechanism by which CL impacts sleep quality is by reducing adolescent stress. Given the reciprocal relationship found in previous research between stress and sleep (Bethune, 2014), future research should evaluate the potential for effects of CL/PeerLearning.net on these outcomes to strengthen over time in a positive feedback loop. In addition, it is likely that reductions in stress can have further benefits beyond sleep, such as improved behavior, academic performance, and mental health (Hankin et al., 2015; Moksnes et al., 2010; Raufelder et al., 2014; Van Ryzin & Roseth, 2021), and future research should evaluate the long-term effects of CL/PeerLearning.net on these outcomes.

Teacher Uptake and Assessing Lesson Success

The school differences in dosage come down to individual teacher choices regarding the frequency with which to use CL/PeerLearning.net; some teachers believed strongly in student-centered active learning, and others did not, and such teachers are often difficult to convince otherwise. Even among teachers that were willing, there was variability in the degree to which they could commit to spending time to build their own lessons. In the future, we hope to have lessons available for teachers to use, thereby reducing this barrier. It would also be advantageous in future research to collect data on teachers in advance of the study to enable an assessment of the factors that contribute to the uptake of CL/PeerLearning.net.

We note that PeerLearning.net captures certain data during the lesson that could support conclusions regarding the success of the lesson and whether the teacher delivered it as intended (which could, in some cases, be considered a measure of fidelity). For example, as part of each lesson, the teacher sets a class goal for the targeted social skill (in terms of the number of times they would like to observe the skill in the class during the lesson); the teacher also records observations of the skill in PeerLearning.net during the lesson. These data can indicate the degree to which the students are using the skill during the lesson and could potentially be used as a measure of the success of the lesson in terms of student social skill development (see Table 1). PeerLearning.net also captures the projected and elapsed time for the lesson, and if the elapsed time is much smaller than the projected timeframe, it could be because the teacher skipped or cut short an important phase of the lesson (which could be considered an indication of fidelity of lesson delivery). Finally, student feedback on the lesson is captured during the reflection and group processing phase (see Table 1), indicating the degree to which the students felt that they were engaged during the lesson, learned a lot, collaborated well, and enjoyed the lesson. These data could be useful in gauging the degree to which the lesson was successful in achieving its learning goals, and low scores could prompt a reconsideration of the lesson design.

Limitations

A few study limitations should be considered. First, this study was based upon a relatively homogeneous sample of rural students that was about three-quarters white, which limits the external validity (generalizability) of the results. Second, all student measures were self-report and included a circumscribed number of items, which limits internal validity. This approach was taken to maximize the feasibility of data collection in schools by minimizing student and school measurement burden, but future research should consider additional data sources, such as teachers and parents, as well as validated behavioral measures of sleep and stress. Although sleep outcomes were measured by a single item, this is consistent with previous use of the PSQI to assess individual sleep constructs (e.g., Blake et al., 2016; John et al., 2016). Third, the reliability of our measure of stress was somewhat low, but we note that this scale performed as theoretically expected, thus reducing concerns about low reliability (i.e., the signal was sufficient to overcome the noise). Fourth, the small number of schools in our sample (i.e., 12) limited the complexity of the models that we were able to fit to the data; as a result, some of our null effects (e.g., lack of moderation by race and ethnicity) may be due to limitations in statistical power. Fifth, the mediation results should be evaluated with caution given that we only had two waves of data collection. Thus, the direction of effects could be the opposite of that proposed, or (more likely) it could be bi-directional. Future research should include larger, more diverse samples assessed over longer periods of time to enable mediation analysis with temporal precedence between the mediator and outcome. Sixth, we were unable to calculate dosage by student; to pinpoint the students who were present in the classroom when each lesson was delivered would require access to detailed attendance data, which the schools were unwilling to provide. Future research should consider alternative mechanisms for evaluating dosage by student. Finally, we were only able to successfully engage approximately half of the teachers in actively using PeerLearning.net. Future research should explore whether techniques such as Motivational Interviewing (MI) can incentivize more teachers to participate.

Conclusion

In summary, the present study provides new evidence that CL, delivered with the aid of PeerLearning.net, is a promising approach to instruction that reduces perceived stress and improves subjective sleep quality among middle school and high school students, in addition to its established benefits for academic achievement, behavior, and mental health (Chirimwami & Van Ryzin, 2024; Low & Van Ryzin, 2023; Roseth et al., 2008; Zagni & Van Ryzin, 2024). Thus, CL/PeerLearning.net could serve as a comprehensive behavioral and mental health prevention strategy that also supports academic goals. Interestingly, the dosage effects suggest that it may be possible to generate an even larger impact on stress and sleep by increasing the number of lessons taught. Given that CL can be used at any grade level and in any subject, it seems quite possible to increase usage of CL/PeerLearning.net.

As far as we are aware, this is the first study investigating the association between CL and sleep. Sleep is highly intertwined with our physical and mental wellness (Worley, 2018), so examining potential interventions that improve sleep can have great significance for adolescents. These results, and future research with larger, more diverse samples and longitudinal measurement, can help to further clarify the long-term effects of CL/PeerLearning.net on stress, sleep, and adolescent well-being.