School engagement and mental health in adolescence: longitudinal and bidirectional relationships

Abstract

Background:

School engagement and mental health are frequently linked, but within-person associations over time are largely unstudied. Emerging statistical techniques can better gauge how longitudinal changes in school engagement or mental health influence individual-level outcomes.

Methods:

Two cohorts of students (recruited in 2017 or 2018) in 5 Los Angeles high schools completed baseline surveys at the high school transition and 3 annual follow-up surveys through 11^th grade. Random-intercept cross-lagged panel models explored the strength and directionality of associations between school engagement (Student Engagement Instrument) and mental health (Mental Health Inventory).

Results:

Among 431 participants, we observed between-person and within-person correlations between school engagement and mental health. Autoregressive effects of school engagement and mental health on future levels of these variables were identified, with more consistent effects in late high school. A single cross-lagged effect from mental health to school engagement across the high school transition was identified.

Conclusion:

Decreased school engagement may signal corresponding cross-sectional changes in mental health, presenting opportunities for monitoring and intervention. Changes in school engagement or mental health may influence later within-person changes in these constructs, but associations are likely weaker and less consistent than previously assumed based on techniques that conflate between-person and within-person effects.

Education influences health trajectories from childhood through adulthood and has a unique relationship with health in adolescence.^1,2 The adolescent years are accompanied by the emergence of some chronic physical or mental health conditions, the onset of health-affecting behaviors that may persist into adulthood, and a heightened risk of intentional or unintentional injury.³ At the population level, school engagement generally declines during this life stage, sometimes resulting in significantly worsened educational outcomes including failure to graduate high school.⁴ Studies taking a person-oriented approach to study individual-level change in school engagement during adolescence have identified more variable school engagement trajectories with some studies suggesting links between low or decreasing engagement trajectories and both individual (e.g., male youth, Black and Latino youth) and family factors (e.g., lower parental education, lower socioeconomic status).^5–8 School disengagement and resulting low academic achievement in adolescence have been commonly linked to worsened mental health and increased health-affecting behaviors that may be detrimental to adolescent health including substance use, risky sexual behaviors, low physical activity, and unhealthy diet.¹ Reinforcing cycles of school disengagement, mental health concerns, and negative health-affecting behaviors during adolescence can amplify risks for poor health and educational outcomes making the high school years a critical time to intervene to maintain school connection and enhance adolescent health and wellness.⁹ Both school-level universal prevention interventions and targeted individual-level treatment interventions hold promise for addressing adolescents’ interrelated health and educational needs to reduce co-occurring health and educational inequities, support youth to thrive, and promote healthy adulthood.^9,10

School engagement is the extent to which students ascribe value to the process and goals of education (cognitive engagement), feel connection to their school and school community (affective engagement), and participate in the academic and extracurricular activities of schooling (behavioral engagement).^4,11 Mental health and wellbeing includes, but is not limited to, the absence of emotional distress and presence of positive affect, and is critical for positive youth development.¹² Adolescent school engagement has been linked with symptoms of anxiety or depression and general mental health in numerous prior cross-sectional studies.¹³ Some longitudinal studies have suggested reduced risk of future mental health concerns among highly school engaged youth but other studies have failed to identify significant associations.^13,14 A bidirectional relationship between school engagement and mental health is theoretically plausible. Being physically and psychologically well can increase students’ ability to regularly attend and meaningfully engage in school.¹⁵ Conversely, students who are highly engaged in school may have increased opportunities to form social connections with supportive peers and adult mentors or achieve academic success which can enhance mental health.^9,16,17 Both school disengagement and mental health concerns may emerge or increase in adolescence and are of particular concern in the wake of the COVID-19 pandemic with national data indicating elevated rates of both among U.S. high school students.¹⁸ Improved understanding of the strength and directionality of the relationship between school engagement and mental health at the individual level may shed light on the varied trajectories of these interrelated factors during the developmental phase of adolescence and across the COVID-19 pandemic. Further, school engagement and mental health are individually promotive of positive health and educational outcomes and if bidirectional interactions exist between these factors at the individual level, they may be uniquely powerful targets for intervention to establish reinforcing cycles and amplify desired outcomes.⁹

Some prior studies have investigated bidirectional relationships between school engagement and adolescent mental health using cross-lagged panel models (CLPMs), a form of structural equation modeling suited to examine longitudinal associations from a between-person perspective.^19–22 These results have largely suggested bidirectional relationships between various measures of school engagement and components of mental health or wellbeing, with varied strength and directionality of associations across age and gender.^19–21 One study suggested that school connectedness may buffer future mental health concerns only for males while mental health concerns may impair future school connectedness only for females.¹⁹ Another study revealed associations between depressive symptoms and future school connectedness for males and females but again suggested a stronger association for females.²¹ Associations between school connectedness and future depressive symptoms were generally stronger for males but were observed only among females across the transition from primary to secondary school.²¹

An important limitation of traditional CLPMs—and the associated evidence base on the bidirectional relationship between school engagement and mental health—is the inability to provide estimates of effects at a within-person level.^23,24 Specifically, these CLPMs have tested whether students with greater (or lower) school engagement relative to their peers will also experience greater (or lower) mental health relative to their peers in the future, and vice versa.^23,24 Newer statistical techniques including the random-intercept cross-lagged panel model (RI-CLPM) can better disentangle between-person effects (i.e., the stable “trait-like” component reflecting each student’s baseline tendency toward school engagement and mental health) and within-person effects (i.e., the time-varying “state-like” fluctuations in each student’s school engagement and mental health over time) to produce less biased and more interpretable estimates.^23–25 In this study, we will use RI-CLPMs to newly investigate how intraindividual changes in school engagement impact changes in mental health relative to each student’s own unique baseline, and vice versa.^23,24

Increased understanding of the directionality of associations between school engagement and mental health at this within-person level may have important implications for designing and prioritizing targeted individual-level treatment interventions to concurrently promote adolescents’ connection to school and mental health and wellness. In addition, the use of RI-CLPMs to decompose trait-like and state-effects at the individual level can be beneficial when studying constructs expected to follow variable trajectories across participants, such as may be expected with school engagement and mental health as students traverse different ages, developmental stages, school settings, and the onset of the COVID-19 pandemic during the course of this study.^23,24,26 In this study, we used a 4-wave random-intercept cross-lagged panel model (RI-CLPM) including 431 adolescents surveyed annually from the transition to high school to the end of 11^th grade to investigate the longitudinal relationship between school engagement and mental health. We hypothesized that:

Hypothesis 1 (H1): At the between-person level, students with greater school engagement as compared to their peers would display greater mental health as compared to their peers (i.e., the extracted between-person trait-like components of school engagement and mental health would be positively correlated).

Hypothesis 2 (H2): At the within-person level, greater than expected individual levels of school engagement would be associated with greater than expected individual levels of mental health at the same time point (i.e., the cross-sectional within-person state-like components of school engagement and mental health would be positively correlated).

Hypothesis 3 (H3): At the within-person level, greater than expected individual levels of school engagement would be associated with greater than expected individual levels of school engagement in the future and greater than expected individual levels of mental health would be associated with greater than expected individual levels of mental health in the future (i.e., the longitudinal within-person auto-regressive effects would be positive).

Hypothesis 4 (H4): At the within-person level, greater than expected individual levels of school engagement would be associated with greater than expected individual levels of mental health in the future, and vice versa (i.e., the longitudinal within-person cross-lagged effects would be positive).

METHODS

Study Design and Procedure

We conducted secondary analysis of survey data from a randomized trial of Advancement Via Individual Determination (AVID; NCT03059433), a college readiness program operating in elementary, middle, and high schools nationwide.²⁷ Study participants were students across 5 public high schools within a large urban school district in Southern California. In the study school district, the high school AVID program focuses on interested students with middle school grade point average (GPA) of 2.0 to 3.5 from backgrounds underrepresented in higher education, such as those identifying as Black or Latino. Study participants were recruited in two cohorts in 2017 and 2018. Eligible students were those entering 9^th grade at a study high school who either: 1) applied for AVID and were selected to participate via their school’s AVID random selection admission process, 2) applied for AVID and were not selected to participate, or 3) were ineligible for AVID based on a middle school GPA above 3.5. Parent consent and student assent were required for participation. The larger trial investigated the effects of AVID on students’ social networks, substance use, and health behaviors with further details of study procedure and results presented elsewhere.²⁸ The study was approved by the overseeing institutional review board and participating school district.

Data Collection

Participants completed electronic surveys at 4 time points each approximately 1 year apart. Time points of survey collection were: T1) at the high school transition at the end of 8^th grade or beginning of 9^th grade (April-October 2017 or 2018), T2) at the end of 9^th grade (May-June 2018 or 2019), T3) at the end of 10^th grade (May-June 2019 or 2020), and T4) at the end of 11^th grade (May-June 2020 or 2021). All surveys were conducted at school prior to 2020. The 2020 and 2021 surveys were conducted remotely due to the COVID-19 pandemic. Schools closed to in-person learning in March 2020 and instruction continued remotely for the vast majority of students through June 2021.

Measures

School engagement.

School engagement can be categorized as observable (i.e., behavioral engagement) and internal subtypes (i.e., cognitive engagement, affective engagement).¹¹ Observable subtypes can be assessed with outward indicators including school attendance and classroom participation, while internal subtypes are best measured through students’ self-report of their own experiences and perspectives.¹¹ Internal subtypes of disengagement may be less readily apparent and are therefore critical to intentionally assess as they have been associated with poor academic outcomes and may have unique relationships with internalizing mental health symptoms.^11,29,30 In this study, we assessed internal subtypes of school engagement at all time points via 29 items of the Student Engagement Instrument.^11,31,32 Participants indicated their agreement to items assessing 2 domains of affective engagement (i.e., teacher-student relationships, peer support for learning) and 2 domains of cognitive engagement (i.e., control and relevance of schoolwork, future aspirations and goals). All item responses were on a 4-point scale ranging from “strongly disagree” to “strongly agree”. Subscale means of affective and cognitive engagement were highly correlated (r(429) = 0.80, p < 0.0001) at baseline and Cronbach’s alpha indicated excellent internal consistency reliability for the full scale within our sample at all time points (α_T1 = 0.96; α_T2 = 0.95; α_T3 = 0.95; α_T4 = 0.95). We therefore calculated a total mean school engagement score by averaging all items which ranged from 1–4 with higher values representing greater school engagement. This overall score on the Student Engagement Instrument captures both internal subtypes of school engagement and has been previously used in school settings to guide assessment and intervention efforts.³³

Mental health.

Mental health was assessed at all time points with the 5-item version of the Mental Health Inventory (MHI-5).^34,35 The MHI-5 is considered a measure of general mental health with scores representing a continuum of psychological distress to psychological wellbeing.^34,35 Participants indicated how frequently they experienced each of 5 different aspects of mental health in the past month (e.g., How much of the time have you felt calm and peaceful?”). All item responses were on a 6-point scale ranging from “none of the time” to “all of the time”. The MHI-5 provides a scaled total score ranging from 0–100 with higher values representing greater mental health. Cronbach’s alpha indicated good internal consistency reliability within our sample at all time points (α_T1 = 0.81; α_T2 = 0.82; α_T3 = 0.82; α_T4 = 0.85).

Demographic characteristics.

Participants self-reported their demographic characteristics at baseline. Participants were asked to identify their gender with response options including male or female. This single question did not allow for expression of a full range of gender identities and is likely more closely aligned with participants’ sex assigned at birth as it will be considered here.³⁶ Participants were asked to describe their race and ethnicity by selecting one or more of the following identities: White non-Hispanic, Hispanic, Black or African American, Asian or Pacific Islander, or American Indian or Native American. Participants additionally reported caregiver characteristics including if at least one of their primary caregivers was born in the U.S. (yes/no), graduated high school or received a GED (yes/no), and currently had any part- or full-time employment (yes/no). School, cohort (cohort 1 recruited in 2017, cohort 2 recruited in 2018), and study arm (AVID group, control group, high performing group) were recorded from study records.

Statistical Analysis

School engagement, mental health, and demographic characteristics were summarized for the overall sample and compared across male and female participants using t-tests for continuous variables or chi-square tests for categorical variables. These analyses were completed using Stata v17 (StataCorp LLC, College Station, TX).

All remaining analyses were conducted in Mplus version 8.7 (Muthén & Muthén, Los Angeles, CA). We examined associations between school engagement and mental health using RI-CLPMs.²⁴ Traditional CLPMs evaluate longitudinal between-person relationships between two or more variables including autoregressive effects and cross-lagged effects.²⁴ Autoregressive effects characterize the influence of a variable on future occurrences of that same variable or “carry-over” effects (e.g., the influence of school engagement at T1 on school engagement at T2). Cross-lagged effects characterize the influence of a variable on future occurrences of a different variable or “spill-over” effects (e.g., the influence of school engagement at T1 on mental health at T2). The RI-CLPMs used in this study extend this approach by disentangling stable trait-like between-person effects and time-varying state-like within-person effects.²⁴

Following the modeling procedure described by Hamaker and colleagues²⁴ and subsequent extensions,²³ we first established a standard RI-CLPM in which all parameters were allowed to freely vary over time as a base model to examine the relationship between school engagement and mental health. The base model included a between-person component represented by two latent random intercepts loading on school engagement and mental health at all time points, with their covariance representing the association between the extracted between-person trait-like components of school engagement and mental health (H1). By decomposing these between-person effects, the within-person component of the RI-CLPM can be understood to evaluate how individual-level deviations in school engagement or mental health from one’s expected level of school engagement or mental health influence future values of these variables relative to an individual’s own unique baseline. The within-person component included covariances between school engagement and mental health at each time point (H2), autoregressive effects among school engagement and mental health over time (H3), and cross-lagged effects between school engagement and mental health over time (H4).

We next established various nested models constraining assumptions regarding model parameters including autoregressive effects and cross-lagged effects of school engagement and mental health over time. This technique allowed us to test various assumptions about these effects and present well-fitting and theoretically plausible models of the relationship between school engagement and mental health during adolescence. Further details of the modeling procedure are presented in Supplement 1.

All models used full information maximum likelihood to account for missing data and maximum likelihood estimation with robust standard errors to account for non-normality of data.³⁷ Fit statistics were calculated for all models and assessed for goodness of fit based on Akaike Information Criteria (AIC), Bayesian Information Criteria (BIC), comparative fit index (CFI) ≥ 0.95, Tucker-Lewis index (TLI) ≥ 0.95, root mean square error of approximation (RMSEA) ≤ 0.06 and confidence interval from 0.00 to 0.8, and standardized root mean square residual (SRMR) ≤ 0.08.³⁸ Satorra-Bentler scaled chi-square values are reported and nested models were compared to the base model using chi-square difference testing, with the absence of a significant test indicating that a constrained model is tenable.³⁹

RESULTS

431 participants completed the initial survey (T1). Participant retention was 418 at T2 (97%), 377 at T3 (87%), and 324 at T4 (75%). 261 participants identified as female (61%) and 170 as male (39%; Table 1). 357 participants identified as Hispanic/Latino (83%). 35% had at least one primary caregiver who was born in the U.S. and 54% had at least one primary caregiver who graduated high school. Mean school engagement was 3.3 (standard deviation (SD) 0.5) at T1 and remained relatively stable throughout high school (Table 2). Male participants reported higher overall school engagement (mean 3.2, SD 0.5) compared to female participants (mean 3.1, SD 0.5; p = .03) at T2. Mean mental health was 68.0 (SD 20.6) at T1 and generally declined throughout high school. Male participants reported higher mental health as compared to female participants at all time points.

Table 1.

Participant Demographic Characteristics

Characteristic	Participants N (%) (N=431)
Female	261 (61%)
Race/Ethnicitya
Hispanic/Latino	357 (83%)
Asian or Pacific Islander	53 (12%)
White	19 (4%)
Black	18 (4%)
American Indian or Native American	13 (3%)
≥ 1 caregiver who was born in the U.S.	166 (39%)
≥ 1 caregiver who graduated high school or received GED	231 (54%)
≥ 1 caregiver who is employed	415 (96%)
Cohort
Cohort 1	198 (46%)
Cohort 2	233 (54%)
Study arm
AVID Group	124 (23%)
Control Group	146 (34%)
High Performing Group	161 (37%)

^aParticipants could select more than one response.

Table 2.

School Engagement and Mental Health Overall and by Sex Assigned at Birth

Variable	Time	N	Overall	Males	Females	P value
School engagementa	T1	430b	3.3 (0.5)	3.2 (0.6)	3.3 (0.5)	0.55
School engagement	T2	418	3.1 (0.5)	3.2 (0.5)	3.1 (0.5)	0.03
School engagement	T3	377	3.2 (0.5)	3.2 (0.5)	3.2 (0.5)	0.65
School engagement	T4	324	3.2 (0.5)	3.2 (0.5)	3.1 (0.5)	0.25
Mental healthc	T1	430b	68.0 (20.6)	73.3 (17.8)	64.5 (21.5)	<0.0001
Mental health	T2	418	64.4 (21.0)	71.2 (18.9)	60.1 (21.1)	<0.0001
Mental health	T3	377	65.7 (21.2)	72.9 (17.7)	61.3 (21.8)	<0.0001
Mental health	T4	324	61.0 (21.1)	68.5 (18.7)	56.8 (21.2)	<0.0001

T1 = end of 8th grade or beginning of 9th grade; 2017–2108

T2 = end of 9^th grade; 2018–2019

T3 = end of 10^th grade; 2019–2020

T4 = end of 11^th grade; 2020–2021

^aSchool engagement has a possible range of 1–4.

^bOne participant did not complete school engagement or mental health measures at T1.

^cMental health has a possible range of 0–100.

Evaluating model fit and comparing nested models

The base RI-CLPM of school engagement and mental health had good model fit overall with RMSEA = 0.042, CFI = 0.985, TLI = 0.955, and SRMR = 0.04 (Supplement 2). This model implies that the within-person autoregressive and cross-lagged effects between school engagement and mental health varied over time as youth progressed through high school and across the COVID-19 pandemic. Two nested RI-CLPMs, one with constrained cross-lagged effects alone and one with constrained autoregressive and cross-lagged effects, were tenable based on the chi-square difference test (Δ χ2(4) = 3.86, p = 0.43 and Δ χ2(8) = 15.03, p = 0.06 respectively). On review of overall model fit statistics, the RI-CLPM with constrained cross-lagged effects alone had better fit and was retained as a plausible model. This model implies that the within-person cross-lagged effects between school engagement and mental health were constant over time. Guidelines for constraining model parameters and model selection in analyses using RI-CLPMs suggest that decisions should be based on theoretical relationships between variables over time as well as goodness of fit to the data.²³ We therefore present the results of both the base RI-CLPM and the nested RI-CLPM with cross-lagged effects based on perceived theoretical plausibility given the fluctuating nature of school engagement and mental health during the high school years and overall goodness of fit to the data respectively. Further details comparing base and nested models are presented in Supplement 1.

RI-CLPMs of school engagement and mental health

Unstandardized parameter estimates of the base RI-CLPM and nested RI-CLPM with constrained cross-lagged effects are presented in Table 3. Standardized parameter estimates and visual depictions of the base RI-CLPM and nested RI-CLPM with constrained cross-lagged effects are presented in Figure 1 and Figure 2, respectively.

Table 3.

Unstandardized parameter estimates for RI-CLPMs

	RI-CLPM	RI-CLPM constrained CL
Parameter	Estimate (SE)a	Estimate (SE)a
Covariances
RI E – RI M	1.829 (.43)	2.111 (.45)
E1 – M1	0.884 (.45)	0.815 (.42)
E2 – M2	1.47 (.49)	1.023 (.56)
E3 – M3	1.253 (.48)	1.134 (.46)
E4 – M4	0.428 (.42)	0.568 (.39)
AR effects
E1→E2	−0.042 (.07)	0.013 (.08)
E2→E3	0.068 (.09)	0.082 (.08)
E3→E4	0.253 (.10)	0.209 (.10)
M1→M2	0.221 (.11)	0.160 (.13)
M2→M3	0.042 (.11)	0.008 (.12)
M3→M4	0.310 (.09)	0.310 (.09)
CL effects
E1→M2	1.517 (2.77)	1.395 (2.23)
E2→M3	4.432 (3.35)	1.395 (2.23)
E3→M4	0.237 (3.75)	1.395 (2.23)
M1→E2	0.005 (.002)	0.001 (0.001)
M2→E3	0.004 (.002)	0.001 (0.001)
M3→E4	0.000 (.002)	0.001 (0.001)

RI-CLPM: Random intercept cross-lagged panel model

AR: Autoregressive

CL: Cross-lagged

SE: Standard error

E: School engagement

M: Mental health

RI E: Random intercept of school engagement

RI M: Random intercept of mental health

^aBolded estimates are significant at p < .05.

Figure 1.

Random intercept cross-lagged panel model of school engagement (SE) and mental health (M) across 4 time points. Time 1 (T1) is the end of 8^th grade or beginning of 9^th grade; 2017–2018. Time 2 (T2) is the end of 9^th grade; 2018–2019. Time 3 (T3) is the end of 10^th grade; 2019–2020. Time 4 (T4) is the end of 11^th grade; 2020–2021. Standardized parameter estimates are shown. Bolded arrows represent statistical significance at p < .05. Dotted errors represent non-statistically significant relationships.

Figure 2.

Random intercept cross-lagged panel model of school engagement (SE) and mental health (M) with constrained cross-lagged effects across 4 time points. Time 1 (T1) is the end of 8^th grade or beginning of 9^th grade; 2017–2018. Time 2 (T2) is the end of 9^th grade; 2018–2019. Time 3 (T3) is the end of 10^th grade; 2019–2020. Time 4 (T4) is the end of 11^th grade; 2020–2021. Standardized parameter estimates are shown. Bolded arrows represent statistical significance at p < .05. Dotted errors represent non-statistically significant relationships.

In the base RI-CLPM, the random intercepts were significantly positively correlated (Figure 1; standardized β = 0.454, standard error (SE) .08; p < 0.001) suggesting that on average individuals with greater school engagement as compared to their peers also have greater mental health as compared to their peers (H1). Within-person correlations between school engagement and mental health were significant at T1, T2, and T3 (T1: β = 0.130, SE .07; p = 0.05; T2: β = 0.266, SE .08; p = 0.003; T3: β = 0.219, SE .08; p = 0.004) suggesting that greater than usual levels of school engagement are associated with greater than usual levels of mental health at the same time point (H2). Autoregressive effects of school engagement were significant only at T4 (;β = 0.24, SE .10; p = 0.02) suggesting that individuals with greater school engagement at T3 relative to their expected score will likely experience greater school engagement at T4 (H3). Autoregressive effects of mental health were significant at T2 and T4 (T2: β = 0.213, SE .10; p = 0.04; T4: β = 0.285, SE .08; p < 0.001) suggesting that individuals with greater mental health at T1 and T3 relative to their expected mean will likely experience greater mental health at T2 and T4 respectively (H3). There were no significant cross-lagged effects of school engagement on mental health (β range: 0.005 – 0.122) suggesting that transient elevations in school engagement relative to an individual’s expected score do not influence subsequent changes in mental health (H4). There was a single significant cross-lagged effect of mental health on school engagement at T2 (β = 0.170, SE .08; p = 0.04) suggesting that individuals with transient elevations in mental health at T1 relative to their expected score will likely experience relatively greater school engagement at T2 (H4).

In the nested RI-CLPM with constrained cross-lagged effects, the random intercepts remained significantly correlated (Figure 2; β = 0.494, SE .08; p < 0.001; H1). Within-person correlations between school engagement and mental health remained significant at T2 and T3 (T2: β = 0.171, SE .08; p = 0.02; T3: β = 0.200, SE .08; p = 0.007; H2). Autoregressive effects of school engagement and mental health remained significant only at T4 (T4 school engagement: β = 0.206, SE .10; p = 0.03; T4 mental health: β = 0.285, SE .08; p < 0.001; H3). No significant cross-lagged effects were identified (β range: 0.028 –0.089; H4).

DISCUSSION

In this study, we used random intercept cross-lagged panel modeling to examine longitudinal relationships between school engagement and mental health from the high school transition to the end of 11^th grade among a sample of predominantly Hispanic- or Latino-identified youth. Building on existing research using traditional CLPMs,^19–21 we identified two well-fitting and theoretically plausible RI-CLPMs to contribute new understanding of the relationship between school engagement and mental health in adolescence by differentiating effects at the between-person (i.e., stable trait-like differences between students) and within-person (i.e., time-varying state-like deviations within individuals) levels. Results of these models partially supported our hypotheses. Between school engagement and mental health, between-person time-invariant correlations (i.e., individuals with greater school engagement on average also have greater mental health on average as compared to their peers) and within-person correlations across time points (i.e., individual-level changes in school engagement are positively associated with changes in mental health at the same time point) were frequently observed largely supporting hypotheses 1 and 2. Within-person autoregressive (i.e., individual-level changes in school engagement or mental health are positively associated with subsequent changes in the same variable) or cross-lagged effects (i.e., individual-level changes in school engagement or mental health are positively associated with subsequent changes in the other variable) were less often significant largely rejecting hypotheses 3 and 4 with some exceptions.

Despite frequently observed declines in school engagement throughout high school at the population level, overall mean school engagement in our sample remained relatively constant across all time points and observed values reflected those found in prior studies using the Student Engagement Instrument among high school students.^31,40 Overall mean mental health in our sample generally declined throughout the study period from 68 at T1 to 61 at T4. Notable differences were observed by sex assigned at birth, with females experiencing lower mental health at all time points ranging from 64.5 at T1 to 56.8 at T4. Although the MHI-5 was used to represent a continuum of mental health from psychological distress to psychological wellbeing in this study, various prior studies have sought to establish cutoffs for the MHI-5 to aid in the diagnosis of psychiatric conditions among adults with typically suggested cut-offs ranging from 52 to 76.^41,42 If these findings hold for adolescents, our results suggest that there may be a substantial amount of clinically significant mental health concerns within this sample. This finding echoes recent national data collected in the wake of the COVID-19 pandemic in the fall of 2021 shortly after our final time point in this study.¹⁸ This data revealed that 42% of U.S. high school students felt persistently sad or hopeless and 22% seriously considered suicide in the past year.¹⁸ Female youth and youth who identified as lesbian, gay, bisexual, questioning, or another non-heterosexual identity (LGBQ+) experienced higher rates of psychological distress with 24% and 37% having made a suicide plan in the past year respectively.¹⁸ There is a clear and urgent need to improve youth mental health which reinforces the importance of research that seeks to identify untapped pathways for intervention such as that presented here.

At the between-person level, we identified moderate positive correlations between school engagement and mental health, indicating that adolescents with greater (or lower) school engagement on average also have greater (or lower) mental health on average as compared to their peers. This finding aligns with substantial prior literature using variable-centered approaches to link school engagement and measures of mental health and wellbeing.^13,14,43 Through the use of the RI-CLPM approach, our finding specifically suggests a relationship between the time-stable trait-like components of school engagement and mental health. This may be related to the existence of more constant individual (e.g., self-efficacy⁴⁴), interpersonal (e.g., peer and teacher support^45,46), family (e.g., parental involvement⁴⁴), school (e.g., school climate¹³), or community (e.g., violence exposure^47,48) factors that dually influence school engagement and mental health contributing to an adolescent’s shared baseline tendency towards both.

At the within-person level, we found positive cross-sectional correlations between school engagement and mental health, indicating that an adolescent experiencing greater (or lower) than expected school engagement is likely to experience greater (or lower) than expected mental health at the same time point. Notably, significant correlations were identified at all time points except T4, the time point following the onset of the COVID-19 pandemic in both cohorts. The identified associations between the time-varying state-like components of school engagement and mental health may be related to similarly time-varying shared predictive factors. The transition to virtual schooling during the COVID-19 pandemic may have removed shared contextual drivers of school engagement and mental health (e.g., peer and teacher connections⁴⁹) thereby attenuating the association observed at other time points. Alternatively, this shift may have increased the importance of protective factors related to the home environment that differentially impacted school engagement and mental health during the COVID-19 pandemic and disrupted this previously observed within-person association. A prior study among high school students supports this idea, revealing that positive family environment had a similar association with wellbeing but an amplified association with school engagement after the onset of the COVID-19 pandemic as compared to pre-pandemic.⁵⁰

The observed between- and within-person correlations between school engagement and mental health have relevance for educators and clinicians. Educators can recognize that education and health are closely interrelated and specifically consider that school disengagement and its potential manifestations including disruptive school behaviors or school absenteeism may be a signal of concurrent diminished wellbeing or inadequately addressed mental health concerns as opposed to willful disobedience.¹⁵ Signs of school disengagement can be considered calls to increase support to youth who may be struggling and facilitate referral to school- or community-based health care providers to support any physical or mental health needs.¹⁵ Clinicians can consider school engagement as a “vital sign” of youth wellbeing and regularly ask about school attendance and performance.¹⁵ Clinicians treating youth for mental health conditions can recognize their potential educational impacts and provide anticipatory guidance or any needed documentation to ensure that students receive appropriate academic supports.¹⁵

Our RI-CLPM results regarding autoregressive and cross-lagged associations between school engagement and mental health diverge from prior studies using traditional CLPMs which have more consistently revealed longitudinal and bidirectional interactions.^19–22 This discrepancy is not entirely surprising as these related but distinct methodologies support theoretically different hypothesis testing.^23,24 Our findings indicate that changes in an adolescent’s school engagement or mental health relative to their predicted individual-level score may influence later within-person changes in these constructs but these associations are likely weaker and less consistent than previously assumed based on traditional CLPMs. Traditional CLPMs do not separate between- and within-person effects and attempts to use these methods to evaluate intraindividual relationships may lead to overestimation and confounding of effects.^23,24 Studies directly comparing the results of traditional CLPMs and RI-CLPMs for related constructs such as depression,⁵¹ academic achievement,⁵² or both,⁵³ have found similarly reduced or absent autoregressive and cross-lagged effects in RI-CLPMs.

In our, study, we identified variable autoregressive effects of school engagement and mental health during the high school years. In only the base RI-CLPM, an autoregressive effect of mental health was seen across the high school transition, suggesting that adolescents with greater (or lower) than expected mental health at the high school transition (i.e., end of 8^th grade or beginning or 9^th grade) experience greater (or lower) than expected mental health at the end of 9^th grade. This implies that the high school transition may be a time when trajectories of mental health are more amenable to change, an important finding considering the increasing prevalence of depression throughout middle and late adolescence.⁵⁴ Prior studies have also identified shifting depression trajectories at the high school transition mediated by changes in social support and school-based social connections^55,56 and instituted prevention programs to improve mental health outcomes during this time.⁵⁷ Autoregressive effects of both school engagement and mental health were most consistently observed in later high school years (i.e., end of 10^th to end of 11^th grade). Autoregressive effects may emerge or strengthen in later years of high school as youth establish reinforcing patterns of school engagement or mental health that may be nurtured with support, or allowed to dwindle in its absence, as youth navigate the challenges of emerging adulthood. Notably, this period of stronger and more consistent autoregressive effects occurred following the onset of the COVID-19 pandemic in both cohorts. It is also possible that the observed effects are a result of this pandemic context. The transition to virtual schooling and associated severing of social connections, disruption of routine, and introduction of new stressors may have increased instability in both school engagement and mental health during this time.^58,59 Future research can seek to establish if these effects are sustained or increased through the final year of high school, across the transition to postsecondary education or careers, and following the COVID-19 pandemic.

Our findings also suggest a possible but inconsistently observed cross-lagged effect of mental health on school engagement in early high school (i.e., high school transition to end of 9^th grade), suggesting that individuals with transient elevations (or declines) in mental health relative to their expected score experience relatively greater (or lower) subsequent school engagement. The transition to high school can be an important inflection point in adolescents’ health and education trajectories.⁶⁰ There is growing understanding of the potential for school-based interventions to concurrently enhance adolescent health and education and the transition to high school may be a critical time to leverage these.^28,61,62 If our findings are replicated, this will support the utility of interventions that target mental health as a pathway to improved school engagement and educational success. Interventions should seek to expand beyond promoting only the absence of diagnosable mental health conditions towards a goal of comprehensively fostering adolescent wellbeing and supporting youth to thrive.

Our study has several limitations. The study sample includes students within one large public school district in Los Angeles, California and most participants identified as Hispanic or Latino. This may limit the generalizability of findings but allows us to investigate relationships between school engagement and mental health in a population that may experience co-occurring health and educational inequities. Additionally, the recruitment strategies for the larger trial selected a sample of students who were either academically high performing in middle school or who sought entry to a college readiness program in their high school. Students in these groups may not be representative of the larger student population including with respect to school engagement or mental health profiles. Guidelines for RI-CLPMs recommend special attention to maintaining approximately equal time intervals when constraining lagged parameters.²³ In this study, timing of T1 data collection (end of 8^th grade or beginning of 9^th grade) resulted in a shorter T1 to T2 time interval relative to all other time intervals for some participants. Although the relative difference is small compared to the one-year time intervals at all other time points, we present results of both the fully unconstrained base RI-CLPM and the nested RI-CLPM with constrained cross-lagged effects to further increase the reliability of our results given this limitation. Data collection for this study was partially following the onset of the COVID-19 pandemic (T4 for cohort 1, T3 and T4 for cohort 2). High school students’ education and mental health have been deeply impacted during the COVID-19 pandemic which speaks to the relevancy of this work but also may somewhat limit the generalizability of these findings during other times.^18,58,63 Our analysis may also lack sufficient statistical power to detect small autoregressive or cross-lagged effects as prior evidence suggests that sample sizes of 1000 or more may be required in RI-CLPMs.⁶⁴ Readers should interpret our negative findings with caution and future research may seek to replicate this technique with larger sample sizes to determine if small but meaningful associations exist between school engagement and mental health during this unique developmental period. Additionally, although RI-CLPMs offer some unique benefits not accomplished with traditional CLPMs, there is still much debate about the relative appropriateness of these two methods and results should be considered within these confines.^65,66 Overall, the use of RI-CLPMs is a strength of this study that allows a nuanced exploration of longitudinal relationships between school engagement and mental health throughout the high school years in a manner not previously accomplished.

In conclusion, we assessed longitudinal relationships between school engagement and mental health from the transition to high school through 11^th grade using random intercept cross-lagged panel modeling. We confirmed between-person correlations and newly identified within-person correlations between school engagement and mental health, highlighting the importance of recognizing these as interrelated factors in the lives of adolescents. We identified within-person autoregressive effects of both school engagement and mental health on future levels of these variables, with suggestion of a more significant effect in the later high school years and during the COVID-19 pandemic. We identified a significant autoregressive effect of mental health and cross-lagged effect from mental health to school engagement across the high school transition that was not consistently reproduced in all models. These findings suggest the importance of the high school transition as a time during which health and education trajectories may be more amenable to change and, if confirmed, would offer potential pathways for intervention. Overall, these results have implications for educational and clinical practice aimed to collectively promote educational success, good health, and wellbeing in adolescence and beyond.

Impact:

• This study assesses bidirectional and longitudinal relationships between school engagement and mental health among rising high school students.

• Both school engagement and mental health impacted future levels of these variables with stronger effects observed in the later high school years.

• Relationships between school engagement and mental health were less frequently observed than in prior studies, with an effect of mental health on future school engagement seen only across the high school transition.

• Interventions that promote both school engagement and mental health may be particularly beneficial, and the high school transition may be a promising time to leverage these.

Data availability statement:

Per the University of California, Los Angeles Institutional Review Board guidelines, the authors are unable to provide data from this study because it contains potentially identifying information, in addition to restrictions that research participants consented to. Further, the Committee for External Research Review has denied permission to share data as it contains potentially identifying information. Data sharing requests can be made to the UCLA South General Institutional Review Board at gcirb@research.ucla.edu.