Emotional Responses to Stressors in Everyday Life Predict Long-Term Trajectories of Depressive Symptoms

Abstract

Background

Individuals’ emotional responses to stressors in everyday life are associated with long-term physical and mental health. Among many possible risk factors, the stressor-related emotional responses may play an important role in future development of depressive symptoms.

Purpose

The current study examined how individuals’ positive and negative emotional responses to everyday stressors predicted their subsequent changes in depressive symptoms over 18 months.

Methods

Using an ecological momentary assessment approach, participants (n = 176) reported stressor exposure, positive affect (PA), and negative affect (NA) five times a day for 1 week (n = 5,483 observations) and provided longitudinal reports of depressive symptoms over the subsequent 18 months. A multivariate multilevel latent growth curve model was used to directly link the fluctuations in emotions in response to momentary stressors in everyday life with the long-term trajectory of depressive symptoms.

Results

Adults who demonstrated a greater difference in stressor-related PA (i.e., relatively lower PA on stressor vs. nonstressor moments) reported larger increases in depressive symptoms over 18 months. Those with greater NA responses to everyday stressors (i.e., relatively higher NA on stressor vs. nonstressor moments), however, did not exhibit differential long-term changes in depressive symptoms.

Conclusions

Adults showed a pattern consistent with both PA and NA responses to stressors in everyday life, but only the stressor-related changes in PA (but not in NA) predicted the growth of depressive symptoms over time. These findings highlight the important—but often overlooked—role of positive emotional responses to everyday stressors in long-term mental health.

Adults who demonstrated greater stressor-related changes in everyday positive affect reported larger increases in depressive symptoms over the next 18 months.

Introduction

Depression is a highly prevalent emotion dysregulation disorder that impairs physical function, social skills, and quality of life and is also associated with elevated morbidity and mortality [1]. Subclinical levels of depressive symptoms also have significant negative impact on people’s daily living and well-being [2] and are associated with elevated risk of major depressive disorders [2, 3]. A wide range of risk factors for the development of depressive symptoms and disorders have been identified, including the biological (e.g., chronic diseases, functional limitations), psychological (e.g., personality traits, negative self-image, psychopathology), and social factors (e.g., smaller social networks, lack of social support, stressful life events, see [4] for a review). There is a large literature supporting the association between “stress” (often broadly used to refer to stressors or stressor-related responses) and the exacerbation of depressive symptoms and the risk for major depressive disorders (see [5, 6] for reviews). However, it is still unclear why some people develop depressive symptoms or disorders following stressful events and others do not.

Recent research indicates that individual differences in the way people emotionally respond to stressors in daily life, often referred to as affective reactivity or stress sensitivity in the literature, is one process that plays an important role in the development of depressive symptoms and the risk for major depressive disorder over time [7–9]. For example, previous daily diary studies found that higher levels of negative affect (NA) that individuals experienced in response to a daily stressor (vs. no stressor) predict greater depressive symptoms or disorder 2 months [8, 9], 1 year [10], and 10 years later [7]. Yet, little research exists examining how positive emotional responses to everyday stressors (i.e., lower levels of positive affect [PA] when a stressor occurs vs. not) impact subsequent changes in depressive symptoms. This is particularly surprising given that several theories suggest that positive emotional status is a valuable psychological resource during times of stress, could “undo” the harmful aftereffects of NA [11], and also serves as a protective factor for depression [12, 13]. To address this gap in the literature, the present study used ecological momentary assessments (EMAs) and a novel analytical approach to simultaneously examine how individuals’ positive and negative emotional responses to naturally occurring stressors in everyday life predict their long-term trajectories of depressive symptoms. This knowledge is essential for a better understanding of the emotional mechanisms through which everyday stressors exert their effects on later development of depressive symptoms.

Emotional Responses to Stressors in Everyday Life and Long-Term Health Outcomes

The emotional responses to everyday stressors capture individuals’ emotion regulation patterns in the context of aversive or stressful events in daily life. At least under some assumptions (e.g., an external environment—such as stressor intensity and type—that is fairly similar between individuals), it has been argued that larger emotional responses to a stressor in daily life (i.e., relatively higher NA or lower PA in response to a stressor) may indicate ineffective emotional regulation, which could result in negative psychological and physiological health outcomes in the long term [14]. Several daily diary studies have used repeated measurements and within-person associations between naturally occurring daily stressors and affect at the day level to capture individuals’ typical pattern of stressor-related emotional responses across a wide range of situations and contexts in daily life. Increasing evidence using this approach has linked the heightened stressor-related (positive or negative) emotional responses in daily life with a variety of health outcomes, including elevated inflammation [15], diminished sleep efficiency [16], increased risk of chronic health conditions [17], and higher risk of mortality in 10 years [18]. Importantly, these findings of the risk associated with greater emotional responses to stressors hold even when accounting for differences in the frequency of stressors exposure.

Emotional Responses to Stressors in Everyday Life and Depressive Symptoms

Traditional diathesis-stress models of depression emphasize the role played by emotional regulation of NA in the context of stress [9, 19]. Individuals with predispositional vulnerabilities (e.g., genetic and biological vulnerabilities, personality, negative self-schemas; see [5] for a review) are more likely to struggle with regulating their NA when facing a stressor; and the difficulties with NA regulation in daily life could, over time, cause wear and tear on individuals’ psychological well-being and contribute to the future development of depression. In other words, larger NA responses to stressors in daily life are hypothesized as one important mechanism through which vulnerability factors eventuate in depressive symptoms [20]. Guided by this theoretical perspective, some existing studies have focused on the role of NA responses to everyday stressors in predicting later mental health outcomes, including depression. For example, the magnitude of NA in response to momentary stressors in daily life was positively related to depressive symptoms and increased risk for major depression 1 year later [10]. In another study, greater NA responses to daily stressors, but not the exposure to daily stressors, predicted increased likelihood of affective disorders, including depressive disorder, 10 years later [7]. Neither of these studies, however, considered the role of stressor-related PA responses in predicting future depressive symptoms.

More recent theories have highlighted positive emotions as a valuable psychological resource during times of stress. For example, the Broaden-and-Build theory posits that positive emotions can bolster the person’s adaptation during times of stress by buffering or protecting against the harmful effects of NA and broadening the scopes of the person’s attention, cognition, and action [11]. More importantly, the momentary “broadening” effects of positive emotions could build enduring personal resources (e.g., physical, intellectual, and social resources) over time and carry long-term adaptive benefits. Similarly, the dynamic model of affect also emphasizes the important function of PA in preserving well-being when stressors occur by minimizing NA [18, 21]. Therefore, it is critical to examine the unique variance in both PA and NA in response to stressors simultaneously in order to capture the associations between them. According to these theories, the capacity to maintain PA in the face of stressors in everyday life (i.e., smaller decreases in PA in response to a stressor) should have a salutary effect on long-term depressive symptomatology, independent of the influence of NA responses to stressors.

Despite the theoretical proposition, very few empirical studies have examined both PA and NA responses to naturally occurring stressors in daily life as predictors of future depressive symptoms. One exception is a daily diary study using a college-student sample and a later replication study to examine whether PA or NA responses to daily interpersonal stressors predicted depressive symptoms 2 months later [8, 9]. Both studies found that students who showed relatively higher NA in response to daily interpersonal stressors reported higher scores of depressive symptoms 2 months later than their counterparts. Only O’Neill and colleagues’ study [8], but not the later replication [9], found a significant predictive effect of stressor-related PA responses on later depressive symptoms. Because PA or NA responses to stressors were not examined in the same model, neither of these studies was able to take into account the covariation of PA and NA, especially during stress periods [22], nor to compare the relative influence of PA and NA responses on the development of depressive symptoms. In sum, the scarce and inconsistent findings in the literature highlight the need to clarify the roles of PA and NA responses to everyday stressors in predicting future changes in depressive symptoms.

The Present Study

The primary goal of this study is to simultaneously examine PA and NA responses to stressors in everyday life and how individual differences in stressor-related PA and NA responses predict the trajectories of depressive symptoms in the following 18 months. The present study advances past research in three key ways. First, this study used EMAs to measure stressors and affect multiple times a day for 1 week, which enabled us to capture individuals’ experiences and feelings in near real-time with less retrospective biases than daily diary approaches. Second, this study used a novel joint-model statistical approach linking momentary fluctuations of emotional responses with long-term trajectory of depressive symptoms in one model [23]. This joint-model approach is more powerful than the traditional two-stage approach in which a first model was estimated and output the emotional responses index, then these outputted estimates were used as predictors in other models. Third, this study provided a particularly strict test of the unique roles of emotional responses to everyday stressors in predicting future depressive symptoms by controlling for not only baseline depressive symptoms and frequency of stressors exposure in everyday life, but also individual differences in trait negative emotionality (i.e., neuroticism), major life events in the past year, and global levels of perceived stress, each of which have been linked with depressive symptomatology [8, 24]. Thus, this study sheds light on how emotional processes in everyday life contribute to long-term mental health beyond the influences of trait-level and contextual vulnerabilities.

Based on previous theories and evidence, this study expected to find both lower levels of PA and higher levels of NA in response to stressors (vs. nonstressor occasions) at the momentary level; and it was also expected that individuals’ typical stressor-related emotional responses (i.e., relatively lower PA and relatively higher NA when stressors occurred) would predict their long-term growth of depressive symptoms. Specifically, individuals who experienced heightened emotional responses to momentary stressors in everyday life would experience greater increases in depressive symptoms over the next 18 months than individuals who experienced less pronounced emotional responses to momentary stressors.

Method

Participants

This study utilized data from a longitudinal measurement burst study of everyday experiences and health, approved by the Institutional Review Board of Syracuse University (see [25] for details). Potential participants for the larger study (N = 214) were recruited from community-dwelling adults in upstate New York area via a diverse array of advertisements from 2010 to 2012. Eligibility requirements for the study included age (20–80 years old), fluency in English, physical ability to operate a palmtop computer, and absence of major cognitive impairment. The final sample for this study included 176 participants (49% men) who were eligible, completed the EMA component of the study, and provided information on depressive symptoms at baseline. In the final sample, participants ranged in age from 20 to 79 years old (M = 49.47, SD = 16.98), and 57% of them self-identified as Caucasian, 32% as African American, 3.5% as Hispanic, and 7.5% as others. About 74% of the participants had a high school degree or less and 26% reported a bachelor or higher degree.

Procedure

Following the initial phone screening, eligible participants completed the consent process, a demographics questionnaire, and training in the use of the palmtop computer in the lab session. Participants were then invited to participate in a 2-day prescreening session to practice and habituate to the EMA protocol. During the prescreening session, participants completed a morning assessment, an evening assessment and up to five randomly beeped momentary assessments each day using a palmtop computer. The five beeped assessments were scheduled for pseudo-random times spaced approximately 2–3 hr apart throughout the day. Participants who correctly completed at least one morning assessment, one evening assessment, and 6 out of 10 beeped assessments (i.e., ≥60% momentary compliance) during the 2 days were invited to complete the full EMA session. To be considered correctly completed, beeped assessments had to be started within 30 min of the scheduled beep and could last no longer than 30 min from start to finish. Participants returned the palmtop computer within 4–5 days of the prescreening session. Of the 188 individuals who participated in the prescreening session, 177 (94%) were eligible to participate in the full EMA session. These participants did not differ significantly from those who participated in prescreening session but not the full EMA session regarding demographic characteristics and baseline depressive symptoms. As in the prescreening session, participants completed a morning assessment, an evening assessment, and up to five randomly prompted assessments each day using a palmtop computer for seven consecutive days. After the EMA session, participants returned to the lab for a series of cognitive tests, and were debriefed and compensated for their participation ($50 for the completion of two lab sessions and $50 for completion of the EMA session). Participants were interviewed again at 9-month and 18-month follow-ups, during which their depressive symptoms were again assessed.

This study used data from the beeped assessments from the baseline EMA session and the depressive symptoms assessments from baseline (N = 176), 9-month (N = 135), and 18-month (N = 114) interviews. Out of a potential 6,160 beeped momentary assessments (176 individuals × 7 days × 5 beeped assessments per day), a total of 5,483 assessments were completed (89%). At the person level, each participant completed on average 31 (SD = 6.31, Range = 4–35) momentary assessments over the 7-day EMA session. More than 94% of the participants (N = 165) completed at least 60% of potential momentary assessments (n = 21) over the 7-day EMA session. Sensitive analyses suggested that excluding participants who completed less than 60% of the assessments did not change any results. Missing data analyses at the person level revealed that the number of completed momentary assessments was not significantly associated with participants’ age, gender, race, education level and income. The long-term retention rate over 18 months was only significantly associated with age (r = −.24, p = .001), but not with other demographics or any key study variables (e.g., baseline momentary PA, NA, and stressor aggregated at the person level).

Measures

Momentary stressor

At each momentary assessment, participants were asked to indicate whether anything stressful had occurred since the last assessment (0 = No, 1 = Yes).

Momentary affect

At each momentary assessment, participants were asked to report their feelings “right now” on eight affect items (i.e., happy, enthusiastic, content, excited, tense, upset, sad, disappointed) [26] using a scale from 1 (Not at all) to 7 (Extremely). Responses on tense, upset, sad, and disappointed were averaged as a measure of current NA (within-person reliability is 0.85, between-person reliability is 0.99) [27], and responses on happy, enthusiastic, content, and excited were averaged as a measure of current PA (within-person reliability is 0.82, between-person reliability is 0.99). Consistent with other EMA or daily diary studies (e.g., [27]), intraclass correlations (ICC) indicated that about half of the variability in PA was due to differences between persons (51%) and the other portion was due to fluctuations in PA within individuals across occasions and days (49%); for NA, 41% of the variability was due to between-person differences and 59% of the variability was due to within-person differences.

Depressive symptoms

The Center for Epidemiologic Studies Depression Scale (CES-D) [28] was used to assess depressive symptoms. Participants rated 20 statements of ways they might have felt or behaved in the past week on a 4-point scale (1 = Rarely or none of the time to 4 = Most of the time). Example items include “I felt depressed” and “I was bothered by things that don’t usually bother me.” Responses on 20-items were averaged to create a mean score of depressive symptoms (between-person reliability Cronbach’s αs are 0.91, 0.90, and 0.92 for baseline, 9-month, and 18-month follow-up measures, respectively).

Covariates

Participants’ gender, age, and psychological characteristics that have been linked with depressive symptoms in past research were included as covariates: (i) Neuroticism. Past research demonstrated that neuroticism is a well-established predictor of daily NA and PA, as well as depressive symptoms [8]. Thus, the trait level of neuroticism measured by the subscale of the Big Five Inventory [29] at baseline was controlled as a covariate. Participants were asked to rate to what degree they agree with each of eight statements on a 5-point scale (1 = Disagree strongly to 5 = Agree strongly). Sample item includes “I see myself as someone who worries a lot.” Responses to the eight items were averaged to form a composite score (Cronbach’s α = 0.83). (ii) Global Perceived Stress. Overall perceptions of current life stress were assessed using the Perceived Stress Scale (PSS) [30] at baseline. Participants were asked to indicate their perceptions of stress in the past month on 14 items on a 5-point scale (1 = Never to 5 = Very often). Sample items include “In the last month, how often have you felt difficulties were piling up so high that you could not overcome them?” The composite score for global perceived stress was calculated by taking the mean across the 14 items (Cronbach’s α = 0.80). (iii) Major Stressful Life Events. Participants also reported the presence of major stressful life events occurred during the past 1 year on a checklist based on the Life Events Survey (LES) [31] at baseline. The checklist includes 46 stressful life events such as divorce, being fired from job, major change in living conditions, major personal illness or injury, and death of spouse, family, or close friend. A count variable was created to represent participants’ exposure to major stressful life events in the past year. The major stressful life events and global perceived stress represent the individual differences in the overall levels of objective exposure to stressors and subjective appraisal of stressors, as well as general levels of coping and responses to these stressors in life, respectively. As both measures have been linked with depressive symptoms in the past research (see [5] for a review), controlling for these two measures generates a particularly strict test of PA and NA changes in response to everyday stressors and their predictive effects on depressive symptoms over time.

Data Analysis

Consistent with previous stress research [17, 18], emotional response to stressors in daily life was operationalized as a within-person slope between a recent stressor (occurred in the past 2 or 3 hr) and the current level of PA or NA. More precisely, this within-person slope represents the difference in affect (i.e., higher NA or lower PA) when stressor was reported relative to when no stressor was reported. Two stressor–affect slopes estimated by a multivariate multilevel model (MLM) directly predicted the growth factors (i.e., intercept and slope) of a latent growth curve model (LGCM) which estimated the trajectories of depressive symptoms over the next 18 months. As illustrated in Fig. 1, these two parts of analysis (MLM and LGCM) were estimated simultaneously in a single, joint model in Mplus Version 8 [32].

Fig. 1.

Schematic representation of the joint model used to test the primary research questions. The circles and oval denote latent factors; the squares represent manifest variables. The double-headed curved arrow represents covariance, and the circular arrows signify estimated variances. The numbers next to the arrows index the coding used to capture the intercept and linear slope of the growth curve model.

This joint model took into account the nested structure of the data by using a multilevel extension of a first-order vector autoregressive model (VAR(1)), in which momentary assessment occasions (level-1) were nested within individuals (level-2) and the random lagged parameters, random innovation variances, and random innovation covariance of PA and NA were included in the within-person part of the model [33]. The missing assessments and uneven time intervals between assessments (including the overnight time lag) were also taken into account in this model by making the observations (approximately) equally spaced through adding missing values between subsequent observations that were further apart in time (see [34] for details). Following Hamaker et al.’s notation [33], the model first decomposed PA and NA into within-person and between-person components, as shown in equations 1 and 2:

$${\displaystyle \begin{array}{l}{\displaystyle P{A}_{it}={\mathrm{\mu }}_{PA,i}+P{A}_{it}^{(w)}}\end{array}}$$ (1)

$${\displaystyle \begin{array}{l}{\displaystyle N{A}_{it}={\mathrm{\mu }}_{NA,i}+N{A}_{it}^{(w)}}\end{array}}$$ (2)

where the within-person components, $P{A}_{it}^{(w)}$ and $N{A}_{it}^{(w)}$ represent the temporal deviations of individual i’s affect at occasion t from individual i’s average score of PA (${\mu }_{PA,i}$) or NA (${\mu }_{NA,i}$) across all occasions. The within-person components were further decomposed as in following equation 3 and 4:

$${\displaystyle \begin{array}{l}{\displaystyle P{A}_{it}^{(w)}={\beta }_{PA,i}Stres{s}_{it}^{(w)}+{\varnothing }_{PA,i}P{A}_{it-1}^{(w)}+{\epsilon }_{PA,it}}\end{array}}$$ (3)

$${\displaystyle \begin{array}{l}{\displaystyle N{A}_{it}^{(w)}={\beta }_{NA,i}Stres{s}_{it}^{(w)}+{\varnothing }_{NA,i}P{A}_{it-1}^{(w)}+{\epsilon }_{NA,it}}\end{array}}$$ (4)

where ${\beta }_{PA,i}$ and ${\beta }_{NA,i}$ represent the within-person association between momentary stressors occurred 2–3 hr before occasion t and individual i’s PA and NA at occasion t, respectively. In other words, ${\beta }_{PA,i}$ and ${\beta }_{NA,i}$ capture the PA responses and NA responses to recent stressors for individual i, respectively. The ${\varnothing }_{PA,i}$ and ${\varnothing }_{NA,i}$ are the autoregressive parameters for PA and NA, indicating the “carryover” effects of PA and NA at the preceding occasion t − 1. Parameters ${\epsilon }_{PA,it}$ and ${\epsilon }_{NA,it}$ represent the residuals or innovations of PA and NA, including variations in PA and NA that were not explained by stressors at occasion t and the carryover effects of PA and NA from occasion t − 1. Individuals may differ in their exposure and response to the variability of unobserved external influences, such as social interactions, physiological status, weather, which are captured by ${\epsilon }_{PA,it}$and ${\epsilon }_{NA,it}$. Hence, the variances and covariance for ${\epsilon }_{PA,it}$ and ${\epsilon }_{NA,it}$ were allowed to vary across persons. The random effects are indicated by a subject index i and all random effects (i.e., two random intercepts, two random slopes of emotional responses, two random slopes of autoregressive parameters, two random innovation variances, and one random covariance of PA and NA) were allowed to correlate to each other in the model. Following previous research [35], the dichotomous variable of stressor was used as level-1 predictor while the person-mean stressor exposure was included in the model to separate individual differences in the frequency of stressor exposure from the within-person effect of experiencing a stressor on current affect. This approach keeps the meaningful unit of the dichotomous variable of stressor (e.g., 0 = no, stressor, 1 = yes, stressor) and thus provided more interpretable results compared with a person-mean-centered approach. Beep order within a day and study day were included as level-1 covariates to control for potential changes in PA and NA across beeps and study days [36].

At the between-person level of the model, the level-2 outcome—depressive symptoms, was regressed on the parameters of PA responses and NA responses to stressors as follow:

$${\displaystyle \begin{array}{ll}{\displaystyle DepressiveSymptoms{}_{ij}=}& {\gamma }_{0i}+{\gamma }_{1i}Tim{e}_{ij}+{\gamma }_2{\beta }_{PA,i}+{\gamma }_3{\beta }_{NA,i}\\ +{\gamma }_4({\beta }_{PA,i}\times Tim{e}_{ij})+{\gamma }_5({\beta }_{NA,i}\times Tim{e}_{ij})+{\upsilon }_{it}\end{array}}$$ (5)

where Depressive Symptoms_ij represents individual i’s depressive symptoms measured at measurement wave j; Time_ij indicates the measurement wave when depressive symptoms were reported for person i and was coded as j = 0, 1, and 2 to represent baseline, 9-month, and 18-month measurement wave, respectively. Hence, ${\gamma }_{0i}$ and ${\gamma }_{1i}$ are the random parameters that define the conditional growth trajectory (i.e., intercept and linear slope) of depressive symptoms; ${\gamma }_2$ and ${\gamma }_3$ represent the predictive effects of PA responses to everyday stressors (i.e., ${\beta }_{PA,i}$) and NA responses to everyday stressors (i.e., ${\beta }_{NA,i}$) on the baseline depressive symptoms, respectively; and ${\gamma }_4$ and ${\gamma }_5$ represent the predictive effects of PA responses to everyday stressors (i.e., ${\beta }_{PA,i}$) and NA responses to everyday stressors (i.e., ${\beta }_{NA,i}$) on the linear growth slope of depressive symptoms, respectively. The covariates capturing individual differences in gender, age, neuroticism, global perceived stress, and major stressful life events were included as level-2 predictors for both the intercept and growth slope of depressive symptoms in order to test the unique predictive effects of PA and NA responses to everyday stressors on depressive symptoms beyond the influences of these covariates. The effects of interest did not differ when these covariates were omitted from the model.

The model was implemented in a Bayesian framework such that parameter estimates were obtained from posterior distributions (based on noninformative priors following a normal distribution N (0, 10¹⁰)) and inferential conclusions were drawn based on the credible intervals (CIs) of these posterior distributions. In Bayesian analyses, missing data at all levels are assumed as missing at random and are treated in the same way as random effects and model parameters, which implies that at each iteration of the Markov chain Monte Carlo (MCMC) algorithm, missing data are sampled from their conditional posterior. This conditional distribution takes the autocorrelation structure of the individual’s data into account and thus is appropriate for the intensive longitudinal data in the present study [33].

Results

Descriptive Statistics

Participants in the current sample (N = 176) completed 5,483 momentary assessments over 7 days at baseline and reported a stressor on 15% of the momentary assessments (42% of the days in study). As shown in Table 1, participants reported moderately high levels of momentary PA (M = 4.35, Range: [1, 7]), low levels of momentary NA (M = 1.98, Range: [1, 7]), and low levels of depressive symptoms (Ms ≤ 1.67, Range: [1, 4]), on average. The correlations at the person level revealed that averaged frequency of momentary stressors (i.e., proportion of beep surveys with stressors reported) was related to higher average levels of NA, but unrelated to the average levels of PA. In addition, depressive symptoms measured at three time points were each related to lower levels of everyday PA and higher levels of everyday NA averaged across all reported occasions.

Table 1.

Descriptive Statistics on Study Variables

	M	SD	Correlations
			1	2	3	4	5	6	7	8	9	10
1. Gendera	0.512	0.501
2. Age	49.471	16.975	−.014
3. Neuroticism	2.618	0.806	.011	−.187*
4. Global perceived stress	2.693	0.466	−.038	−.330***	.622***
5. Stressful life events	5.119	4.240	−.099	−.386***	.278***	.479***
6. Momentary stressorb	0.153	0.161	.060	.106	.098	.184*	.281***
7. Positive affect	4.352	0.918	.035	.155*	−.344***	−.322***	−.111	−.142
8. Negative affect	1.975	0.843	−.122	−.203**	.423***	.423***	.310***	.315***	−.455***
9. Depressive symptoms (T1)	1.610	0.525	−.142	−.188*	.730***	.631***	.345***	.121	−.399***	.461***
10. Depressive symptoms (T2)	1.669	0.518	−.142	−.291***	.505***	.444***	.327***	−.021	−.260**	.304***	.674***
11. Depressive symptoms (T3)	1.659	0.550	−.026	−.324***	.532***	.530***	.281**	.081	−.185*	.363***	.720***	.721***

N = 176. Descriptive statistics were calculated at the person level. T1 = baseline; T2 = 9-month follow up; T3 = 18-month follow up.

^aGender: 0 = male; 1 = female.

^bProportion of beeps with stressor reported.

*p < .05. **p < .01. ***p < .001.

PA and NA Responses to Everyday Stressors

The PA and NA responses to recent stressors that occurred in the past 2–3 hr were examined simultaneously in the MLM part of the joint model. As shown in the top panel of Table 2, the posterior mean of PA responses to stressors was −0.584, 95% CI [−0.693, −0.474], indicating on average a 0.584 point difference (lower) in PA (on a 7-point scale) on occasions on which a recent stressor was reported compared with occasions on which no stressor was reported. The posterior mean of NA responses to recent stressors was 1.078, 95% CI [0.927, 1.234], indicating on average a 1.078 point difference (higher) in NA (on a 7-point scale) on occasions on which a recent stressor was reported compared with occasions on which no stressor was reported. Both PA and NA responses to recent stressors (i.e., the estimates of within-person slopes) had significant variances (0.291, 95% CI [0.201, 0.399] for PA responses; 0.753, 95% CI [0.569, 0.933] for NA responses), reflecting individual differences in stressor-related responses in momentary PA and NA. Stressor-related changes in PA and NA were moderately correlated (r = −.375, 95% CI [−0.491, −0.255]). In addition, the magnitude of stressor-related changes in NA was significantly larger than the magnitude of stressor-related changes in PA (difference score = 0.495, 95% CI [0.383, 0.609]). These results suggest that recent stressors had significant effects on both current PA and NA, but the stressor effect was (on average) larger on NA responses than on PA responses.

Table 2.

Results for the Fixed Effects from the Joint Model Predicting the Growth of Depressive Symptoms from Emotional Responses to Daily Stressors

Short-term outcomes	Momentary positive affect (PA)			Momentary negative affect (NA)
	Est.	Posterior SD	95% CI	Est.	Posterior SD	95% CI
Daily stressors (within-person)	−0.584*	0.056	[−0.693, −0.474]	1.078*	0.078	[0.927, 1.234]
Daily stressors (between-person)	−0.502	0.455	[−1.369, 0.395]	0.806*	0.388	[0.036, 1.575]
Beep	−0.021*	0.005	[−0.030, −0.012]	−0.002	0.002	[−0.007, 0.003]
Day	−0.009*	0.003	[−0.016, −0.003]	0.000	0.002	[−0.003, 0.004]
AR(1) parameter	0.260*	0.019	[0.222, 0.298]	0.209*	0.021	[0.168, 0.249]
Long-term outcomes	Intercept of the growth of depressive symptoms			Slope of the growth of depressive symptoms
PA responses to daily stressors	0.257*	0.132	[0.011, 0.532]	−0.213*	0.094	[−0.409, −0.039]
NA responses to daily stressors	0.123	0.080	[−0.024, 0.291]	−0.056	0.059	[−0.181, 0.051]
Gendera	−0.111*	0.052	[−0.213, −0.010]	0.026	0.037	[−0.047, 0.099]
Age	0.001	0.002	[−0.002, 0.004]	−0.002	0.001	[−0.004, 0.001]
Global perceived stress	0.220*	0.076	[0.072, 0.372]	−0.003	0.054	[−0.110, 0.104]
Stressful life events	0.013	0.007	[−0.001, 0.027]	0.001	0.005	[−0.009, 0.011]
Neuroticism	0.338*	0.040	[0.257, 0.417]	−0.061*	0.031	[−0.122, −0.001]
Mean Stress	0.003	0.176	[−0.334, 0.356]	0.045	0.129	[−0.212, 0.293]

N = 176 subjects, n = 5,483 observations. Est. = unstandardized coefficients of posterior means. Estimate method = Bayes with noninformative priors. CI credible interval.

^aGender: 0 = male, 1 = female.

*Significant at α = 0.05 level.

Emotional Responses to Everyday Stressors Predicting Future Depressive Symptoms

The LGCM part of the same joint model used the means of the stressor-related emotional responses to predict the growth trajectories of depressive symptoms while adjusting for the individual differences in age, gender, neuroticism, major stressful life events, global level of perceived stress as well as the person-mean of everyday stressors exposure during the baseline EMA session. As shown in the bottom panel of Table 2, PA responses to everyday stressors were positively associated with the baseline levels of depressive symptoms, indicating that individuals with higher levels of depressive symptoms at baseline reported smaller stressor-related PA responses compared to individuals with lower levels of depressive symptoms. The magnitude of PA responses to everyday stressors at baseline negatively predicted the growth of depressive symptoms over the subsequent 18 months (−0.213, 95% CI [−0.409, −0.039]), with those individuals exhibiting relatively lower PA in response to stressors (compared to PA when no stressors were reported) appearing to be at risk. Specifically, a 1-unit difference in stressor-related response in PA (i.e., lower) was associated with 0.213 increases in the linear slope of depressive symptoms, that is, 0.426 points increases (on a 4-point scale) in depressive symptoms over 18 months totally. It is worth noting that the unconditional LGCM of depressive symptoms (i.e., the model without any covariates) revealed that on average the linear growth rate of depressive symptoms was not significantly different from zero (mean of slope = 0.022, 95% CI [−0.016, 0.064]), but there were significant individual differences in the growth rates of depressive symptoms (variance of slope = 0.009, 95% CI [0.002, 0.034]). Therefore, the results from the joint model in Table 2 suggested that, although the average trajectory of depressive symptoms was not different from zero, individuals who had heightened PA responses to everyday stressors at baseline experienced steeper increases in depressive symptoms over time relative to individuals with less PA responses to everyday stressors (as shown in Fig. 2). Contrary to predictions, however, greater stressor-related NA responses at baseline did not significantly predict either the intercept or the growth rate of depressive symptoms over time when examined in the same model with PA responses.

Fig. 2.

Schematic representation of the model-implied growth trajectories of depressive symptoms over 18-month for individuals who have low levels of stressor-related positive affect (PA) response (1 SD below mean); mean levels of stressor-related PA response; and high levels of stressor-related PA response (1 SD above mean).

The R² of the joint model was 0.591 for the intercept factor and 0.427 for the slope factor of depressive symptoms, indicating that 59% of the between-person variability in the initial levels of depressive symptoms and 43% of the between-person variability in the growth rates of depressive symptoms were explained by the model. Further analyses which compared the R² values from models with versus without emotional responses to everyday stressors as predictors for depressive symptoms revealed that emotional responses to everyday stressors added to the model’s explanatory power: 4.5% for the variabilities in the initial levels and 24% for the growth rates of depressive symptoms, respectively, beyond other covariates.

Discussion

How people emotionally respond to stressors in daily life has important implications for their long-term health and well-being. Previous research has demonstrated the associations between the negative emotional responses to daily stressors and depressive symptoms (e.g., [7–10]). However, little is known about how stressor-related responses in positive emotions influence the development of depressive symptoms. The current study thus examined how stressor-related changes in both PA and NA in daily life predicted long-term growth trajectories of depressive symptoms. Using EMAs and a novel analytical approach, the present study directly linked everyday stressors and emotional processes to the development of depressive symptoms over 18 months. The results revealed that recent stressors appeared to exert an influence on mood, relating to significantly lower PA and higher NA. Importantly, individuals who showed larger average PA responses to everyday stressors (i.e., presumptively greater stressor-related decreases in PA) experienced steeper growth in depressive symptoms over the next 18 months compared with those who exhibited less pronounced PA responses. Moreover, the predictive effect of stressor-related PA responses on changes in depressive symptoms was independent of the frequency of everyday stressors exposure and individual differences (including baseline depressive symptoms, age, gender, neuroticism, major stressful life events, and global level of perceived stress). Unexpectedly, however, the magnitude of individuals’ typical NA responses to everyday stressors did not significantly predict the growth of depressive symptoms over time when examined in the same model with PA responses and other covariates. This study highlights the important contributions of positive emotional responses in the context of everyday stress to long-term mental health.

Why do individuals’ everyday stressor-related decreases in PA, but not increases in NA, significantly predict the growth in depressive symptoms over time? One possible explanation is that PA mitigates NA during periods of stress, according to the dynamic model of affect [18, 21]. Thus, individuals who can maintain the PA in the face of everyday stressors are more likely to counteract NA as well as its detrimental influences on health and well-being. Alternatively, individuals who are able to maintain their PA in the face of everyday stressors are more likely to increment coping resources over time due to the “broaden-and-build” effects of positive emotions [11], which may help them better manage stress and stress-related NA and reduce the risk of future depression. Finally, although being able to maintain PA in the face of stress may always be beneficial, the normative levels of distress (e.g., increased NA) following a stressor, especially a recent one, may not necessarily be maladaptive. More recent theory and research suggests that the prolonged negative emotional responses to a stressor (i.e., slow recovery) [37] or failure to shift negative emotions according to changing contextual demands [38], rather than larger acute NA responses, is a clear sign of emotion dysregulation and carry negative long-term health consequences.

The current finding that individuals’ stressor-related PA responses (but not NA responses) predicted the growth in depressive symptoms is also consistent with other empirical studies that simultaneously examined the influences of both PA and NA responses to daily stressors on health outcomes. For example, Mroczek et al. reported that greater decreases in PA in response to daily stressors, but not the stressor-related increases in NA, predicted increased mortality risk over 10 years [18]. In a national sample of midlife adults, Sin et al. found that adults who fail to maintain PA when faced with minor stressors in everyday life appear to have elevated levels of inflammation [15]. The NA responses to daily stressors, however, were not significantly related to the inflammation levels when examined in the same model. The current finding, along with these studies, highlights the unique and more robust role of PA responses (relative to NA responses) to everyday stressors in influencing health outcomes.

The prospective effects of PA responses to everyday stressors on subsequent development of depressive symptoms were found despite the statistical control of individual differences in the initial levels of depressive symptoms and other covariates. In fact, the concurrent associations between the initial levels of depressive symptoms (i.e., the intercept factor) and emotional responses to everyday stressors were also examined in the same model. The results (see Table 2) suggest that individuals with higher (vs. lower) levels of depressive symptoms at baseline demonstrated less positive emotional responses to everyday stressors. This result is consistent with the Emotion Context Insensitivity (ECI) theory and previous studies suggesting that depressed individuals would exhibit reduced reactivity to emotion cues because depression as a defensive motivational state could inhibit ongoing emotional responses [39].

In sum, the current study extends the literature on stress and health by showing that the emotional processes in the context of everyday stress, beyond mere stressor exposure and a host of important individual difference characteristics, have important implications for long-term mental health. The finding that relatively lower stressor-related PA significantly predicted subsequent increases in depressive symptoms over 18 months not only supports previous theories emphasizing the importance of positive emotions in promoting health and well-being, but also suggests that depression theories may benefit from considering the role of PA responses to everyday stressors as a vulnerability factor for depressive symptoms. From a clinical perspective, the current findings suggest that future interventions for depressive symptoms might consider targeting individuals’ loss of pleasure/interest [13] in the context of everyday stress.

Limitations and Future Directions

Despite the important theoretical and clinical implications and methodological strengths, the current study also has several limitations. First, the current study used a binary variable to assess whether anything stressful had occurred since the last assessment (i.e., the occurrence of a stressor) and did not capture other features, such as the nature, severity, and persistence, of the stressors. It is possible that certain type(s) of stressors (e.g., interpersonal conflicts), or stressors that are more severe or more frequent/persistent may cause larger or prolonged emotional responses than other stressors. Also, individuals may respond to different types of stressors in different ways. Thus, further work is needed to investigate how the emotional response to stressors varies across different types or features of stressors and across individuals. Second, although the intense sampling procedure used in the current study had a higher temporal resolution than daily diary studies, the assessments were on average between 2 and 3 hr apart, which precludes a clear demarcation between the magnitude versus duration of the emotional responses to the stressors (i.e., reactivity may be somewhat confounded with lack of recovery). Recent stress theory emphasizes the distinction among different components of stress response: reactivity (i.e., the magnitude of the initial responses to a stressor), recovery (i.e., the duration or persistence of responses after initial reactivity), and pile-up (i.e., repeated experiencing and/or responding to and recovering from stressors) [40]. It has also been proposed that prolonged, rather than larger acute responses to stressors, may carry more negative health consequences (e.g., [37]). Thus, future studies should use more frequent assessments and also measure the timing of the occurrence of stressors, which could allow a more fine-grained analysis to differentiate the influences of distinct components of stress-related emotional responses on long-term health and well-being. Third, the current study focused on the role of emotional responses to everyday life as a predictor of future depressive symptoms, but did not examine the possibility that heightened emotional responses to stressors in daily life may also serve as a mechanism through which vulnerability factors eventuate in depressive symptoms. For example, there is evidence that emotional responses to everyday stressors represent, at least in part, the behavioral expression of genetic risk of depression [41]. Thus, future research would benefit from further examination of how emotional responses to everyday stressors might mediate the effects of other vulnerability factors on the development of depressive symptoms or disorders. Finally, the current study used a convenience sample of community-dwelling adults, which may limit the generalizability of the findings. For example, the adults, especially the older adults, who chose to participate in the current study and completed the longitudinal follow-up assessments may have reflected the best functional group of their cohort (e.g., lower levels of or less increases in depressive symptoms; best health status—survivor effect). Thus, it is critical to extend this work in more representative samples of the population to determine the validity of the current findings.

Conclusion

The current study examined how emotional experiences in the context of everyday stressors that take place over “micro” periods (i.e., 2–3 hr) contribute to long-term mental health. Results suggest that how people typically emotionally respond to stressors in everyday life matters more for their future mental health outcome than the frequency of exposure to such stressors. Particularly, individuals who experienced relatively lower PA in response to everyday stressors reported steeper growth in their depressive symptoms over time compared with their counterparts, highlighting the importance of maintaining PA in the face of everyday challenges to future health and well-being.

Compliance with Ethical Standards

Conflict of Interest The authors declare that they have no conflict of interest.