Emotion dynamics prospectively predict depressive symptoms in adolescents: findings from intensive longitudinal data

Abstract

Background

The incidence of depression among adolescents has risen significantly over the past decade. Emotional dynamics, including variability, instability, and inertia of positive affect (PA) and negative affect (NA), are potential risk factors for depressive psychopathology. However, limited longitudinal evidence exists on how these emotion dynamics relate to depression, particularly in collectivistic cultural contexts, where emotional expression and regulation are shaped by social and familial expectations. This study aimed to investigate the longitudinal associations between emotion dynamics—variability, instability, and inertia of PA and NA—and subsequent depressive symptoms among Chinese adolescents.

Methods

Data were collected from middle school students in Taizhou, China, between November 2021 and April 2022. Participants completed baseline surveys and experience sampling assessments, reporting their emotional states ten times daily over five consecutive weekdays. Depressive symptoms were assessed using the Children’s Depression Inventory (CDI), while emotional dynamics (variability, instability, and inertia) were derived from the experience sampling data. Logistic regression models were employed to examine whether emotion dynamics predicted depressive symptoms at 1-month and 3-month follow-ups.

Results

A total of 448 participants completed all study procedures and were included in the analysis. Emotional variability and instability in PA and NA were longitudinally associated with depressive symptoms at both 1-month and 3-month follow-ups. After controlling for mean affect levels, PA variability and instability, but not NA, were uniquely linked to depressive symptoms. Emotional inertia showed no significant association with subsequent depressive symptoms. Emotional variability and instability in PA and NA also predicted the development of new symptoms in adolescents without baseline depression (n = 372).

Conclusions

Emotional variability and instability of PA and NA were longitudinally associated with changes in depressive symptoms and the development of new symptoms among adolescents. These emotion dynamics provide insights into real-world emotional processing and offer important targets for adolescent depression interventions.

Background

Adolescent depression rates have surged over the past decade, imposing significant personal, societal, and economic cost [1]. A recent systematic review and meta-analysis found that the global point prevalence of elevated self-reported depressive symptoms among adolescents from 2001 to 2020 was 34% (95% CI: 0.30–0.38) [2]. Similarly alarming, a nationwide study reported a pooled prevalence of 26.17% (95% CI: 25.00–27.41%) for depressive symptoms among children and adolescents in China [3]. Despite these concerning statistics, early detection and intervention for adolescent depression remain insufficient, underscoring the urgent need for improved strategies to address this growing public health issue.

Emotions play a crucial role in our daily lives, which strongly influence psychological health and well-being across the life span. Emotions are not static but dynamic and time-dependent in response to internal and external events [4]. The way emotions unfold over time reflects how individuals cope with environmental changes and regulate their emotions [4, 5]. Traditional research on habitual or single transient emotional experiences offers limited evidence of the actual function of emotions and how they determine people’s psychological well-being and mental health. As a result of these limitations, there is growing interest in the relation between emotion dynamics or patterns of emotional change over time and psychological well-being in recent decades.

Emotion dynamics refer to the trajectories, patterns, and regularities that characterize the changes and fluctuations in emotions across multiple timescales (moment-to-moment, day-to-day, week-to-week, etc.) [6, 7]. Researchers have proposed various measures of emotion dynamics based on intensive longitudinal data collected through techniques such as the experience sampling methodology (ESM) [8, 9]. Among emotion dynamic measures, emotional variability, instability, and inertia are most often used and widely linked to psychological well-being [7]. Emotional variability refers to how much an individual’s emotions diverge from their average or core emotion. Variability is typically calculated by the standard deviation or within-person variance (WPV) of an individual’s emotion ratings over time [10]. Emotional instability reflects how emotions change from one moment to the next and is often calculated by the mean square of successive differences (MSSD) [11, 12]. Unlike variability, metrics of instability capture the temporal changes by quantifying moment-to-moment fluctuations rather than changes over an entire period [8]. Emotional inertia captures the degree to which emotions carry over from one moment to the next. How much emotions are self-predictive and resistant to change is typically measured as the autocorrelation (ACORR) of emotions over time [13, 14].

Extreme patterns of emotional change may indicate maladaptive emotional responses and regulation, and play a key role in the development of psychopathology [15]. A high-quality meta-analysis revealed that emotion dynamics were indeed moderately but consistently linked to several indicators of psychological well-being and psychopathology [7]. These associations were found to be particularly strong for depression in both clinical and nonclinical populations characterized by more variable, unstable, and inert positive and negative emotions. These patterns were more obvious for negative emotions in terms of clinical levels of depression. The research on emotion dynamics can provide valuable insights into basic emotional processing in depressive psychopathology and may also inform strategies for early identification and interventions of depression in adolescents.

Adolescence is a critical period for studying emotion dynamics and their associations with psychopathology due to the increase in the intensity and frequency of emotions during adolescence and higher emotional response to daily events than adults (e.g., having an exam, talking to friends) [16, 17]. Surprisingly, a recent meta-analysis revealed that few studies have examined emotion dynamics in childhood and adolescence, and the available evidence is scarce and fragmented apart from some literature on emotional intensity and variability [18].

Despite advances in modelling and computing emotion dynamics, several important questions remain regarding the associations between emotion dynamics and mental health in adolescents. First, most studies in youth explored the relationships between emotion dynamics and depression in a cross-sectional way, and the few prospective studies yielded inconsistent results. Neumann et al. [16] found that higher variability in positive affect (PA) and negative affect (NA) predicted the development of depression during adolescence, while Maciejewski et al. [19]reported that emotional instability prospectively predicted depressive symptoms between the ages of 14 and 16. Van Roekel et al. [20] studied inertia of PA in adolescents, but observed no concurrent association with depressive symptoms. Kuppens et al. [14] found that higher inertia in both negative and positive emotional behaviours during interactions with their parents predicted the onset of clinical depression 2.5 years later in early adolescents. However, only behavioural indicators of inertia in interpersonal contexts were studied in their research rather than the endogenous nature of emotional inertia. Overall, there is a dearth in longitudinal studies of emotion dynamics during adolescence, particularly in emotional inertia [18]. Second, dynamic measures of emotions can be significantly confounded by mean levels of the same variable [21]. Because of this fact, it remains unclear what emotion dynamic patterns could have robust relationships with depressive symptoms over-and-above mean levels. Last, existing studies among adolescents were all conducted in North American or European countries, however the experience, expression, and regulation of emotions have cultural differences [18]. For example, people in more individualistic cultures are more likely to just feel positive and focus on standing out during good events, whereas people from more collectivistic cultures tend to feel mixed emotions (both negative and positive) during good events based on their belief that negative emotions help people fit in socially [22]. Still, it is not clear whether there are cultural differences in the associations between emotion dynamics and their psychological consequences.

Thus, the goals of the present study were to investigate the associations between different emotion dynamical patterns and depressive symptoms 1-month and 3-months later among Chinese adolescents. The central research question of this study is: How do emotion dynamics, including variability, instability, and inertia in positive affect (PA) and negative affect (NA), predict depressive symptoms and their onset in Chinese adolescents over 1-month and 3-month follow-ups? Specifically, we were interested in (1) how the emotion dynamics of variability, instability, and inertia of PA and NA predicted depressive symptoms at 1-month and 3-month follow-ups; (2) which emotion dynamics could predict the onset of depressive symptoms during follow-ups in participants without depressive psychopathology at baseline; and (3) which dynamic patterns were uniquely linked to changes in depressive symptoms or development of new symptoms after correction for overlap with average affect levels. Inspired by the research of Houben and Kuppens [21], our study focused on a non-clinical adolescent population with a broad range of different mental health levels, representing varying intensities of depressive symptoms, rather than individuals with clinical diagnoses. This decision was made due to the importance of understanding subclinical psychopathology. By investigating adolescents most of whom have not yet reached a clinical diagnosis, we can identify processes that might predict an increase in the severity of depressive symptoms and the development of new symptoms. This predictive insight should be beneficial to future prevention efforts for depression in adolescent populations.

We hypothesized that variability and instability of PA and NA would be associated with depressive symptoms 1-month and 3-months later. Moreover, we predicted that variability and instability of PA and NA would predict development of new depressive symptoms. Based on the weaker correlations between mean PA and emotion dynamics of PA compared to those of NA [10], we predicted that variability and instability of PA, but not NA, would be uniquely associated with depressive symptoms after correction for overlap with average affect levels. However, we were unable to provide a clear hypothesis regarding emotional inertia due to a scarcity of evidence.

Methods

Participants and procedure

We conducted a longitudinal investigation among junior and senior high school students in Taizhou, Zhejiang Province, China. Six schools in Taizhou were invited to participate, of which one junior high school and one senior high school agreed to join the study. Within the participating schools, two classes from each grade (7 to 9) in the junior high school and three classes from each grade (10 to 11) in the senior high school were selected using cluster random sampling. Participants eligible for the study should be students in the selected classes with the ability to independently read, comprehend, and complete the questionnaires. Additionally, both the adolescents and their parents were required to provide online informed consent for participation. The study was approved by the Ethics Committee of Taizhou Central Hospital (2022 L-01-17), and all methods were carried out in accordance with relevant guidelines and regulations.

Data collection took place between November 2021 and April 2022. A total of 533 students who were assessed for eligibility participated in the baseline survey, followed by a period of assessment using experience sampling method (ESM). While ESM studies typically require a minimum sample size of 30 participants with at least 10 assessments per person [23], we aimed to enrol as many participants as possible to strengthen the robustness and generalizability of our findings. During the study, participants completed a 6-item ESM questionnaire assessing their current emotional state 10 times daily over five consecutive weekdays (Monday to Friday), starting at 7:00 am and recurring every 1.5 h. To minimize predictable response patterns, missed entries, or premature completion of questionnaires, a teacher or class president in each class reminded students and ensured accurate completion of the questionnaires. A researcher also remained on-site at each school during the ESM assessment period to oversee the process and ensure data quality. Of the 533 students, one was excluded for not providing informed consent, and 72 were excluded for either refusing to continue participation or completing fewer than 10 ESM questionnaires (out of 50 prompts), which raised concerns about data validity. This resulted in 460 students completed the baseline assessment and provided sufficient ESM data for analysis. Among them, 458 completed the 1-month follow-up, and 448 completed the 3-month follow-up. The final sample consisted of 448 participants who successfully completed all study procedures (see Fig. 1). All baseline and follow-up surveys were conducted online via the Wenjuanxing platform (https://www.wjx.cn), which effectively minimized missing values.

Fig. 1

Flowchart showing sample size included in the study

Measures

Depressive symptoms

Depressive symptoms were assessed using the Chinese version of the 27-item Children’s Depression Inventory (CDI) [24]which was specifically developed to evaluate depression in childhood and adolescence. Each CDI item evaluates one symptom by offering three choices, graded from 0 to 2 in the direction of increasing psychopathology. The total score of CDI ranges from 0 to 54, with higher scores indicating more severe depressive symptoms. Those who had scores of 19 or above were considered to have depressive symptoms [25]. The Cronbach’s α of the measurement showed good internal consistency (beasline:0.87; 1-month follow-up: 0.89; 3-month follow-up:0.90).

ESM items

The selection of ESM items was guided by established frameworks and validated measures in affective science, including emotion differentiation tasks [26], the circumplex model of affect [27], and the Positive and Negative Affect Schedule [28], To capture key affective dimensions of depression and to ensure age-appropriateness by tailoring to adolescents’ developmental characteristics, the final selection included four negative emotions (feeling down, meaninglessness, anxiety, loneliness) and two positive emotions (happiness, contentment).

During the ESM phase, participants rated their current experience of these six emotions multiple times per day using a 7-point Likert scale (0 = “completely disagree” to 7 = “completely agree”). Example items included “I am feeling down.” This design enabled real-time, ecologically valid data collection, minimizing recall bias and capturing the dynamic nature of emotional experiences.

Sociodemographic characteristics

Sociodemographic characteristics of participants were collected from the baseline questionnaire, including age (years), sex (female, male), grade (junior high school, senior high school), and residence (rural, urban).

Statistical analysis

Mean, standard deviation (SD), frequency, and percentage were calculated to describe the sample characteristics. All statistical analyses were performed using R version 4.3.1. To address multiple comparisons and reduce the risk of Type 1 errors, the Holm-Bonferroni correction was applied where appropriate. Statistical significance was determined based on adjusted p-values, with dynamically corrected thresholds to maintain a family-wise error rate (FWER) of 0.05.

Calculation of composite scores for positive and negative affect

To compute emotion dynamic measures, it is common practice to create composite scores for positive affect (PA) and negative affect (NA) by averaging same-valenced emotion items at each measurement occasion [8, 18]. In the study, composite scores for PA and NA were calculated by averaging the corresponding emotion items at each measurement occasion. Specifically, happiness and contentment were averaged to form the PA score, while feeling down, meaningless, anxiety, and loneliness were averaged to form the NA score. Internal consistency reliability for PA and NA was assessed using multilevel confirmatory factor analysis with the lavaan package and the compRelSEM function from the semTools package in R. This method calculates point estimates for multilevel composite reliability as defined by Lai [29]. The NA index demonstrated satisfactory within-person reliability (ω^w = 0.75) and good between-person reliability (ω^b = 0.89). The PA index showed good within-person reliability (ω^w = 0.85) and excellent between-person reliability (ω^b = 0.95).

Computation of emotion dynamics

Emotion dynamics of variability (within-person variance), instability (mean square of successive differences), and inertia (autocorrelation) were computed as follows:

Emotional variability refers to the range or amplitude of a person’s emotional states across time. Within-person variance (WPV) often serves as a measure of emotional variability. WPV was calculated as the standard deviation of each participant’s PA and NA scores [10, 11]. Higher WPV values indicate more emotional variability. People with higher levels of emotional variability experience more extreme emotions and show larger emotional deviations from their average emotional level.

Emotional instability describes the magnitude of emotional changes from one moment to the next. Mean square of successive differences (MSSD) typically serves as a measure of emotional instability. It quantifies emotional changes between time_i and time_{i + 1} instead of over an entire period of time [8]. MSSD was calculated separately for PA and NA following the equation proposed by Sperry [10]. Higher MSSD values indicate more emotional instability. In other words, people with higher levels of instability experience larger emotional shifts from one moment to the next.

Emotional inertia describes how well the intensity of an emotional state is predicted from the emotion at a previous moment. Autocorrelation (ACORR) is used to assess emotional inertia, which captures the degree to which an emotion persists or carries over from one moment to the next [8]. In the study, ACORR was computed with a time lag of one using the acf function estimation of the stats package in R [30]. High values of ACORR represent resistance to change or inertia (high temporal dependency), resulting in emotions that are more self-predictive and self-perpetuating across time [10].

The study calculated emotion dynamics from moment-to-moment rather than day-to-day. However, measurements of moment-to-moment dynamics were aggregated across days. Therefore, data from the first ESM measurement of each day was not included in the computation of MSSD and ACORR to ensure that indices did not include differences between the last measurement of the prior day and the first measurement of the following day. Each participant was assigned one value of WPV, MSSD, and ACORR for both PA and NA, respectively. These measures were used in several sets of binary logistic regression models to examine whether emotion dynamics prospectively predict depressive symptoms.

Emotion dynamics and depressive symptoms

In a first set of analyses, we explored the association between emotion dynamics and depressive symptoms in the entire sample. In a first step, emotional variability, instability, and inertia of PA and NA were individually entered the models as predictors of subsequent depressive symptoms. Separate models were estimated for PA and NA. In a second step, baseline depressive symptoms were added to the models to control for baseline mood psychopathology. In a third step, we examined the associations between dynamic measures and depressive symptoms with correction for mean affect, by adding both emotional dynamics and average affect levels (mean PA or mean NA) as simultaneous predictors of subsequent depressive symptoms. These three steps were repetitively conducted for 1-month and 3-month follow-ups.

In a second set of analyses, we explored whether emotional dynamics could predict the onset of depressive symptoms at follow-ups among adolescents without baseline depressive symptoms. Similarly, emotional variability, instability, and inertia of PA and NA were individually entered the models as predictors of the development of depressive symptoms at 1-month and 3-month follow-ups. Subsequently, mean PA and mean NA were added in the respective models to correct for mean affect levels.

In a final set of analyses, we explored which emotional dynamics were more accurate and effective in predicting the development of depressive symptoms. Receiver Operating Characteristic (ROC) curves for the prediction models were plotted, and the area under the curve (AUC) was estimated using the multipleROC package in R.

Results

A total of 448 participants completed all procedures of our study (Fig. 1) and were included in the current analysis. Table 1 presents the characteristics of the participants. The mean age of respondents was 15.1 years (SD 1.4), with 203 (45.3%) being male. The average CDI score in our sample was 12.4 (SD 7.0) at baseline, 12.6 (SD 7.8) at 1-month follow-up, and 10.6 (SD 7.5) at 3-month follow-up. At baseline, 17.0% of participants scored at or above the cut-off of 19 for having depressive symptoms. The percentage of participants with depressive symptoms were 21.2% at 1-month follow-up and 16.7% at 3-month follow-up. Table S1 [see Additional file 1] shows the correlations of all the PA and NA emotion dynamics, as well as mean PA and NA.

Table 1 Characteristics of participants in the analysis sample

Results of the associations between emotion dynamics and depressive symptoms for the entire sample (n = 448) are presented in Table 2. Emotional variability of PA was positively associated with depressive symptoms in Step 1. This result was found at 1-month follow-up and replicated at 3-month follow-up. Correcting for baseline depressive symptoms did not alter the results in Step 2 (1-month follow-up: OR 2.02, 95% CI [1.10, 3.73]; 3-month follow-up: OR 2.24, 95% CI [1.18, 4.29]). After correction for mean PA in Step 3, PA variability still predicted depressive symptoms at 3-month follow-up (OR 1.88, 95% CI [1.04, 3.41]). Considering variability of NA, in both waves, higher variability of NA was associated with presence of depressive symptoms in Step 1. These results were robust after controlling for baseline depressive symptoms (1-month follow-up: OR 2.41, 95% CI [1.26, 4.63]; 3-month follow-up: OR 2.84, 95% CI [1.43, 5.69]). However, the unique predictive value of NA variability became weaker and did not reach statistical significance anymore after correction for mean NA levels in Step 3 in both waves. Emotional instability of PA and NA was positively associated with depressive symptoms at both follow-ups in Step 1. When correcting for baseline depressive symptoms in Step 2, PA and NA instability predicted depressive symptoms at 1-month follow-up (PA: OR 1.27, 95% CI [1.08, 1.50]; NA: OR 1.32, 95% CI [1.02, 1.68]) but not at 3-month follow-up (PA: OR 1.17, 95% CI [0.98, 1.39]; NA: OR 1.22, 95% CI [0.93, 1.57]). Only PA instability was a significant predictor of depressive symptoms even after correction for mean PA in Step 3. However, this result was only found at 1-month follow-up (OR 1.22, 95% CI [1.05, 1.42]). Emotional inertia of PA and NA did not show significant associations with depressive symptoms at either follow-up. The results remained consistent when correcting for baseline depressive symptoms or mean emotion levels.

Table 2 Association of positive affect (PA) and negative affect (NA) dynamics with depressive symptoms at 1- and 3-month follow-ups in the entire sample (n = 448)

Results of the associations between emotion dynamics and development of new depressive symptoms for adolescents without baseline depressive symptoms (n = 372) are reported in Table 3. Both PA and NA variability were found to predict the onset of depressive symptoms at 1-month and 3-month follow-ups. When correcting for mean emotion levels, only PA variability predicted the development of depressive symptoms at 3-month follow-up (OR 3.59, 95% CI [1.58, 8.46]). PA and NA instability were found to be positively associated with new depressive symptoms at both follow-ups. Instability of PA predicted depressive symptoms at 1-month and 3-month follow-ups even after correction for mean PA levels (OR 1.25, 95% CI [1.02, 1.52]; OR 1.28, 95% CI [1.03, 1.58]). Neither PA nor NA inertia showed significant associations with depressive symptoms at any follow-ups, and the results remained identical when adjusting for mean emotion levels.

Table 3 Association of positive affect (PA) and negative affect (NA) dynamics with new depressive symptoms at 1- and 3-month follow-ups among adolescents without baseline depressive symptoms (n = 372)

Based on data from adolescents without baseline depressive symptoms, the ROC curves and AUC for the prediction models of PA and NA variability and instability are shown in Fig.S1 [see Additional file 1]. Emotional inertia was not included in the prediction models since it was not associated with new depressive symptoms as shown in Table 3. Our findings indicated that variability of PA and NA exhibited moderate levels of discriminative accuracy in predicting new depressive symptoms at both follow-ups (AUC 0.63–0.69) [31]. The discriminative accuracy of variability was slightly higher than instability of PA and NA (AUC 0.61–0.64), particularly at 3-month follow-up.

Post-hoc analyses

To more accurately analyze the continuous depression scores, we conducted post-hoc analyses using linear regression models to examine the relationship between emotion dynamics (variability, instability, and inertia of positive and negative affect) and depressive symptoms, as measured by the total score of the Children’s Depression Inventory (CDI), at both the 1-month and 3-month follow-ups in the entire sample (n = 448). The results from these linear regression analyses, presented in Table S2 and S3 [Additional file 1], were consistent with those from the previous logistic regression analyses, confirming the robustness of our findings. Variability in negative affect (NA) was significantly associated with depressive symptoms at both the 1-month (β = 6.00, 95% CI [4.34, 7.67]) and 3-month follow-ups (β = 5.39, 95% CI [3.79, 6.98]). Similarly, variability in positive affect (PA) was significantly associated with depressive symptoms at both the 1-month (β = 3.18, 95% CI [1.58, 4.77]) and 3-month follow-ups (β = 2.73, 95% CI [1.19, 4.26]). These consistent results further support the validity of both approaches in capturing the relationship between emotional dynamics and depressive symptoms.

Discussion

Our study, among a sample of Chinese adolescents, prospectively investigated the associations between different emotion dynamical patterns and depressive symptoms. The main findings revealed that emotional variability and instability of PA and NA significantly predicted adolescent depressive symptoms at both 1-month and 3-month follow-ups, whereas emotional inertia did not show a notable association. Among adolescents without baseline depressive symptoms, both PA and NA variability and instability predicted the onset of depressive symptoms, and the discriminative accuracy was slightly higher for variability. Furthermore, PA variability and instability, but not NA, remained robust predictors for the presence and onset of depressive symptoms after controlling for mean emotion levels.

Our findings highlight the pivotal role of emotional variability in predicting depressive symptoms among adolescents. Consistent with our hypotheses, both positive and negative emotional variability exhibited significant prospective associations with depressive symptoms. These associations persisted after controlling for baseline depressive symptoms, suggesting that adolescents with a wider overall range of emotional experiences (emotional variability) were more likely to have depressive symptoms in the future. The current findings confirmed those of prior research that variability of emotions contributed to changes in depression during adolescence [16, 19]. Some emotional variability in response to environmental stressors is normal, while too much variability may indicate heightened emotional reactivity and ineffective attempts at emotional regulation [16]. Excessive emotional variability within a relatively brief time interval can have detrimental effects on psychological well-being because they involve extreme lows and highs [32, 33].

Emotional instability also plays a significant role in predicting depressive symptoms. Our study found that the instability of both positive and negative emotions was prospectively associated with depressive symptoms. These findings might be explained by abnormalities in the function and structure of the posterior parietal cortex, a key attentional control region involved in emotion regulation. A recent neuroimaging study showed that increased emotional instability was associated with decreased neural activation for the left inferior parietal cortex during an emotion regulation task [34]. Emotional instability also correlated with cortical hypogyria of the same region [34]. This suggests that increased emotional instability may reflect a reduced capacity to manage and regulate emotions, thereby increasing vulnerability to depressive symptoms.

Importantly, our findings demonstrated that both variability and instability of PA and NA predicted the onset of new depressive symptoms in participants without baseline depressive symptoms. They all showed moderate levels of predictive accuracy, among which variability presented particularly higher AUC at 3-month follow-up. Our results are contrary to Sperry et al.‘s research [10], which found that altered emotional variability and instability were not associated with new diagnoses of major depressive disorder or major depressive episodes three years later. However, there are discrepancies between design of the two studies. Our sample consisted of high school students rather than university students, our follow-up period was shorter, and we focused on depressive symptoms rather than clinical depression diagnoses. Although more evidence is needed from future research, our findings in adolescents are promising as real-world momentary assessments of emotions were related to the development of depressive symptoms three months later, which may facilitate early identification of adolescents at risk for depression. Emotion dynamics, which can be easily monitored using electronic devices, may serve as a feasible target for prophylactic interventions.

Notably, our findings revealed that the variability and instability of PA were uniquely linked to changes in depressive symptoms and the development of new symptoms, even after accounting for average PA levels. Recently, there has been a debate on whether assessing multiple emotion dynamics is necessary or whether specific emotion dynamical patterns have unique predictive validity over mean levels. Our findings suggested that fluctuations in PA was a crucial component to describe the emotional functions of adolescents characterized by more depressive symptoms. These adolescents may experience PA that deviated strongly from their baseline mean PA over time, which resulted in the unique predictive validity for depression. In contrast, the predictive power of NA variability and instability weakened when controlling for mean NA levels, indicating that PA fluctuations might have a more unique relationship with depressive symptoms above average affect levels. A meta-analysis in children and adolescents reported that differences in emotion dynamics between typically developing youth and those with mental health problems were consistently more evident for valence of negative emotions than positive emotions [18]. The difference in cultural context may be relevant here, as existing research predominantly comes from North American and European settings. In collectivistic cultures like China, where emotional restraint and the belief that negative emotions help people fit in socially are emphasized, adolescents may suppress fluctuations in PA due to cultural norms. This suppression may increase emotional variability and instability and contribute to the development of depressive symptoms. These cultural factors may help explain why PA dynamics, rather than NA dynamics, were more robust predictors of depression in our study, in contrast to findings from studies in Western contexts where NA plays a more prominent role. Our study adds valuable insights into the unique role of emotion dynamics of PA in predicting depressive symptoms among Chinese adolescents, emphasizing the need for targeted emotional management and culturally relevant intervention strategies.

In addition, our study found limited evidence for the role of emotional inertia in predicting depressive symptoms. At any follow-ups, neither the inertia of negative nor positive emotions significantly predicted depressive symptoms. Similarly, van Roekel et al. [20] reported that depressive symptoms were associated with higher variability and instability in early and late adolescence, but not with inertia. These findings make theoretical sense, as greater variability and instability are often considered as the inverse of resistance to emotional change (i.e., autocorrelation of emotions) [18]. Kuppens et al. [14] found that increased inertia in both negative and positive emotional behaviours predicted clinical depression in adolescents 2.5 years later, however emotional behaviours rather than emotional experiences were used in their study. Overall, estimates of emotional inertia in adolescents are scarce. Furthermore, the different time intervals during the ESM assessment period make comparisons between studies difficult, as autocorrelations based on longer intervals are likely to be weaker than those with shorter intervals (e.g., hours versus seconds) [35]. Further replication is needed, particularly with clinical samples as emotional inertia might be stronger in persons with a diagnosis of major depressive disorder.

Strengths and limitations

The strengths of this study include its prospective longitudinal design and detailed measurement of multiple emotion dynamics. Our study added evidence for the predictive effects of emotional variability, instability, and inertia on adolescent depressive symptoms. By employing the ESM to record dynamic emotional states, the study provided high temporal resolution data, ensuring the accuracy of momentary emotional assessment. By controlling for baseline depressive symptoms and average emotional levels, the study minimized confounding effect. Focusing on Chinese adolescents, it filled a gap in existing research which was primarily conducted in North American and European countries.

Some limitations should be considered when interpreting the results. Firstly, the study’s observational design precludes definitive conclusions about causality. While associations between emotion dynamics and depressive symptoms were observed, causation cannot be established due to the potential influence of unmeasured confounding variables. Future research using experimental or longitudinal designs could better assess causal relationships between emotion dynamics and psychopathology. Secondly, the sample was drawn from only two schools in Taizhou City, which may limit the generalizability of the findings to other adolescent populations across China. To enhance the external validity, future research should include a more diverse geographical and cultural range. Thirdly, the study had a relatively short follow-up period of 1 month and 3 months, which may not adequately capture the long-term effects of emotion dynamics on depressive symptoms. Future studies could extend the follow-up duration to more thoroughly explore whether patterns of emotion dynamics can prospectively predict changes in psychopathology over longer periods. Additionally, while the Ecological Momentary Assessment (ESM) provided data with high temporal resolution, its complexity might reduce participant compliance, affecting the integrity of the data. The limited duration of the ESM (five days) is another constraint. The selected week might not reflect a typical week for each adolescent, and atypical events or circumstances could introduce bias. Finally, while the ESM allowed for rich, real-time data collection, its reliance on self-reports could be influenced by biases such as recall bias or social desirability, which could further limit the conclusions that can be drawn about the relationship between emotions and depressive symptoms. Furthermore, Time Series Analysis could offer valuable insights into the temporal patterns and interactions between emotion dynamics and depressive symptoms. This method would be especially useful for understanding the longitudinal fluctuations in depressive symptoms and how they are influenced by emotional dynamics. Future research could consider utilizing advanced analytical frameworks, such as Dynamic Structural Equation Modeling (DSEM), to more effectively examine the complex temporal relationships between emotion dynamics and depressive symptoms. Such frameworks would allow for a more robust modeling of the reciprocal influences between these variables over time.

Implications

Our study indicates that emotional variability and instability may be measurable risk factors that are identifiable prior to changes or onset of depressive symptoms. Phone-based apps that aim to prevent depression could specifically assess moment-to-moment fluctuations in PA and NA apart from mood symptoms. Ideally, both active (e.g., ecological momentary assessments) and passive data collection (e.g., duration and frequency of social media use) should be used to improve the accurate measurement of emotions. From the perspective of adolescent development and public health, this study underscores the critical importance of emotional regulation during adolescence. Adolescence may be a time of vulnerability to emotional dysregulation, and high emotional variability and instability can predispose adolescents to depressive symptoms [18, 36, 37]. Therefore, providing effective emotional management technique and psychological support is beneficial. Families and schools are of great importance to the improvement in adolescent emotion regulation and management to help them cope effectively with daily challenges [38,39,40,41]. Phone-based interventions for depression are another potentially favourable approach and should make use of evidence-based principles to target emotion dysregulation. Interventions can include teaching mindfulness and cognitive-behavioural skills such as attentional focus, emotional acceptance, and cognitive restructuring, which have been proven to aid in emotion regulation [42]. For example, the World Health Organization’s ‘Doing What Matters in Times of Stress’ [43] is a self-help stress management guide based on Acceptance and Commitment Therapy, which uses illustrations and pre-recorded audio exercises to teach five different stress-relief and emotion regulation strategies, such as grounding and unhooking. Adapted as a mobile-supported, 5-week program with distance support, this guide has been proved effective in reducing psychological distress [44,45,46]. More useful strategies and tools should be designed to mitigate high emotional variability and instability, and promote better emotional health for adolescents.

Conclusions

The study indicated that emotional variability and instability of PA and NA were longitudinally associated with changes in depressive symptoms and the development of new symptoms among adolescents. When overlap with mean affect levels was taken into account, variability and instability of PA were uniquely and specifically linked to depressive symptoms. Assessing emotion dynamics is a promising approach for understanding real-world emotional processing and can be implemented using phone-based apps. While these findings suggest potential targets for adolescent depression interventions, further research is needed to explore causal mechanisms and assess the clinical utility of these approaches.