Introduction

It is well established that the presence of both acute and chronic stress pose risk for the development of affective symptoms and related disorders (Grant et al., 2004; Shields & Slavich, 2017). Yet how daily reports of emotion during those moments may or may not inform adjustment remains still less-well understood. Some evidence suggests that emotional responses drive key features of mental health trajectories, from adaptive and protective coping such as support seeking or physical exercise (Aurora et al., 2022; Nylocks et al., 2018) to maladaptive risk behaviors, such as substance use or binge eating (Coifman & Aurora, 2022; Johnson et al., 2013). Moreover, affective disorders broadly defined (e.g., depression, anxiety, stress, substance and eating disorders: Kring, 2008; Barlow et al., 2014) are characterized by high and sustained levels of negative affect and low but variable levels of positive affect (Barlow et al., 2014; Clark & Watson, 1991; Khazanov & Ruscio, 2016). Indeed, the capacity to generate and maintain positive emotional experiences is considered an etiological risk factor for psychiatric illness broadly, predicting risk over decades (Kendall et al., 2015). However, when considering the scientific meaning of emotion reported in-vivo, self-reports of emotion reflect not only the relative intensity -but- the manner in which emotional experiences are conceptualized (Barrett et al., 2001). For example, the Conceptual Act model of emotion suggests that emotions are constructed based on how an individual conceptualizes core affective phenomena, including arousal (e.g., based on interoceptive cues), valence, and contextual factors (Barrett, 2006). Hence, how reports of emotion are associated with long-term psychological outcomes may also be particularly relevant to clinical models of emotion and emotion-related risk. Indeed, the subjective experience (or conceptualization) of emotion are the primary target of dominant psychotherapeutic interventions for affective disorders, including Cognitive Behavioral Therapy (Beck, 1995); Dialectical Behavioral Therapy (Linehan, 1993); and Acceptance and Commitment Therapy (Hayes et al., 1999).

Methodological developments over the past two decades have facilitated increased research examining the dynamic nature of emotion in daily life. Daily emotion reports can be interpreted in terms of both static features, such as person-mean level, as well as dynamic indices, with increasing scientific emphasis placed on testing the relevance of dynamic indicators to psychological functioning (Trull et al., 2015). However, both static and dynamic indices of emotion tend to overlap (Dejonckheere et al., 2019a, 2019b), and research is needed to determine their unique contributions to health—as well as how they interact to predict longitudinal outcomes. Importantly, dynamic indices of affective conceptualization can capture affective phenomenon, such as emotion flexibility, that have been broadly attributed to psychological health, in ways not typically evident for static indices (Coifman & Summers, 2019). This is likely because dynamic constructs capture the individuals’ conceptualization of emotion with time in mind, indexing many moments of emotion report, across variable contexts, within one individual to derive an estimate.

Emotional inertia is a dynamic index that captures the tendency for emotions to be reported as persistent or rigid over time and is measured by indexing the within-person, serial, autocorrelation of reported negative—or—positive affect (Koval et al., 2016; Kuppens et al., 2009). For example, high inertia in reports of negative affect from moment-to-moment likely suggest persistent inflexibility in the appraisal of emotional events and experiences across contexts, with greater negative emotional inertia predicting lower wellbeing and risk for psychopathology (Koval et al., 2016; Kuppens et al., 2009). Emotional inertia may reflect a disconnection between the conceptualization of emotional experiences and the typical variability in contextual demands that is present in daily life (Coifman & Summers, 2019) as the individual’s emotional state is resistant to change even when contexts shift (Kuppens et al., 2009). Indeed, research has shown that context-insensitive emotional responses (e.g., low sadness in response to sad stimuli, low positive affect in response to positive stimuli) are a characteristic of psychiatric disorders, like depression (Rottenberg et al., 2005), as well as maladjustment in times of stress (Coifman et al., 2012). Although the negative effects of negative emotional inertia are well-established, research on the role of positive emotional inertia in wellbeing and psychological functioning has been mixed. Prima facie, persistent positive emotions might be a sign of psychological health, rather than inflexibility, as positive emotions tend to be highly adaptive (Fredrickson, 2001; Gruber et al., 2013), even when expressed in negative contexts (Coifman & Bonanno, 2010; Harvey et al., 2016; Papa & Bonanno, 2008). Indeed, some research suggests that greater positive emotional inertia predicts fewer depression symptoms in the general population and greater response to treatment in patients with depression (Höhn et al., 2013). However, other research on positive emotional inertia has found an association with negative indicators of well-being, including increased depressive symptoms (e.g., Koval et al., 2016). The adaptive (or maladaptive) role of positive emotional inertia may depend on individual-level factors, such as trait level or dispositional tendencies to appraise experiences. Importantly, positive emotional inertia indices can reflect either persistent high or persistent low levels of positive emotion over time. Thus, within a person, positive emotional inertia is most likely only adaptive when mean levels of positive emotion are also high. However, limited research has tested interactions between dynamic emotion indices (e.g., emotional inertia) and more static indices (e.g., mean intensity of positive or negative affect).

Emotion polarity is an alternative dynamic index which captures the momentary within-person association of negative to positive emotional experiences (Rafaeli et al., 2007). Highly consistent with clinical models of dichotomous thinking (Beck, 1995), polarized reports of emotional experience reflect a rigid appraisal of an experience as either all bad or all good. A broad literature has examined how this phenomenon is typically more prevalent when individuals are under greater stress, varying both within and between individuals (Dowd et al., 2010; Zautra et al., 2001). Importantly, highly polarized emotional reports during stress can predict maladaptive outcomes including increased depression eighteen months following bereavement (Coifman et al., 2007) and are characteristic of specific high-risk patient groups, predicting increased reliance on maladaptive behaviors (i.e., substance use, binge eating, self-injury, and risky sexual behavior in Borderline Personality, Coifman et al., 2012). Generally, individuals demonstrate less polarity in their emotional conceptualization during periods of low perceived stress or when psychological resources are abundant (Reich et al., 2001; Zautra et al., 2001). There are also key individual differences and evidence that some individuals can maintain low polarity even during heightened stress (Coifman et al., 2007). However, it may also be that polarized responses reflect active coping during acute periods of stress. For example, elevated reports of both negative and positive affect are associated with greater cardiovascular recovery following an acute stressor in lab (Dowd et al., 2010) and daily sampling evidence suggests that polarized reports of negative and positive affect may facilitate coping, oscillating widely when needed, gradually abating over days and weeks (e.g., in bereavement: Bisconti et al., 2004). Indeed, polarized affective responses experienced in daily life may be a key component of psychological health and wellbeing. However, their influence on long-term mental health outcomes, particularly when other indices of affect reporting are considered, remains unclear.

Understanding the unique role of daily emotion in long-term psychological health and adjustment is critical to further developing targets for treatment of many common and burdensome emotion-linked psychiatric conditions. Although many psychotherapies target moment-level conceptualization of emotion, including many “third-wave” cognitive and behavioral therapies (e.g., dialectical behavioral therapy; Dimidjian et al., 2016; Linehan, 1993), these techniques and their emphasis vary widely. Prior research has shown that emotion dynamics can predict the course of psychopathology. For example, daily emotion dynamics have been shown to predict the course of depressive symptoms (Wichers et al., 2010) and influence treatment responsiveness to antidepressants and psychotherapy (Husen et al., 2016; Wichers et al., 2012). However, how dynamic emotion indices uniquely influence long-term psychological health may also simply be attributable to dispositional tendencies that are commonly present in clinical samples. Increasingly, models of psychopathological risk and emotion-related disease states attribute a large proportion of the variance to common underlying general factors (Conway et al., 2022; Kotov et al., 2017). Indeed, relatively little research has considered how in-vivo emotional processes, including moment-level conceptualizations of emotional experience, can reflect and/or can be influenced by more stable dispositional tendencies that are known to drive risk and relapse for common affective disorders (Barlow et al., 2014). Considerable research has converged around the relative importance of trait negative affectivity (Shackman et al., 2016) in driving both momentary reports of emotion and in long-term processes of psychological adjustment. Indeed, evidence across disciplines has demonstrated the role of general affective tendencies in shaping patterns of emotion-related processing and health, even decades later (e.g., Behavioral Genetics: Gilman et al., 2015; Caspi et al., 2010; Neuro-imaging: Hostinar et al., 2017; Personality: Wrzus et al., 2021). In sum, dynamic and static reports of daily emotion may have important relevance to models of psychological health but this must also be considered in relation to dispositional factors. Unpacking these interacting relationships will further the understanding of emotion-related risk, and is essential for tailoring treatments to specific patients, and to develop more effective tools to assess and identify individuals at high risk.

Current Investigation

In this investigation, two distinct cohorts of US adults completed 10 days of daily diaries at key points in the COVID-19 pandemic and then were re-contacted one year later to assess psychological adjustment. These cohorts were recruited as part of a larger experimental investigation on mathematical understanding of COVID-19 risk information in US adults (reported here: Thompson et al., 2021). The primary aim of the current investigation was to determine how daily emotion reports indexed at an initial assessment could predict longitudinal psychological health and wellbeing, one year later. We focused on dynamic emotion indices that represent emotion flexibility, including flexibility in emotional experience across time (e.g., emotional inertia) and flexibility in emotional appraisal (e.g., experiencing affective states as all bad or all good, versus experiencing positive and negative affect simultaneously; emotion polarity), consistent with current perspectives on emotion flexibility in psychological health (e.g., Coifman & Summers, 2019) and current psychotherapeutic approaches (e.g., Linehan, 1993). We considered three outcomes that are highly relevant to both physical and mental health: common symptoms of psychiatric disorders, reports of loneliness, and psychological well-being.

For added context, as part of the parent project, initial assessments of daily emotion were timed to coincide with peak moments of disease-related stress during the COVID-19 pandemic whereas follow-up assessments one year later were timed to correspond with periods that were relatively calmer. Cohort 1 completed daily diaries in March and April of 2020 corresponding to the initial lock-down periods in the US when uncertainty about the pandemic was quite high (CDC 2021). Cohort 2 completed daily diaries in January and February of 2021 corresponding to the first winter peak of disease and importantly, just prior to wide-spread vaccine availability (CDC, 2023). Moreover, we explicitly compared the role of specific dynamic indices (emotion polarity, emotional inertia) of daily emotion with relatively more static indicators (person-mean intensity and within-person standard deviation) to determine their relative influence on outcomes (Dejonckheere et al., 2019a, 2019b). Although the standard deviation of affect might be considered “dynamic”, it is considered the simplest index (after the mean) and is therefore an important control variable for determining the unique contributions of more complex, dynamic indices (Dejonckheere et al., 2019a, 2019b). Hence, for simplicity, we will label the mean and standard deviation as “static” indices, and polarity and inertia as “dynamic” indices, to distinguish conventional affect constructs from more complex ones in a manner that is aligned with prior research.

In addition, we tested all emotion indices (dynamic and static) while considering trait anxiety as an index of dispositional negative affectivity. Dispositional negative affectivity describes a general tendency to experience and express negative affect with greater frequency and intensity, and is a robust predictor of symptoms of common psychiatric disorders, including depression and anxiety (Barlow et al., 2014; Shackman et al., 2016). Importantly, we were unable to include baseline measures of our outcomes (symptoms, loneliness, and wellbeing) as covariates in our models. When the baseline data collections were conducted, we had not yet executed plans for longitudinal follow-up assessments and did not anticipate needing baseline measurements for these outcomes since they were not germane to the parent project. However, because trait anxiety is a strong predictor of all three outcomes (Abdellaoui et al., 2019; Aldinger et al., 2014; Barlow et al., 2014; Griffith et al., 2010; Shackman et al., 2016), by controlling for trait anxiety in our analyses, we were able to compensate to some degree for this lack of baseline measurement. Indeed, there is strong meta-analytic evidence which suggests that dispositional negative affectivity, indexed here by trait anxiety, can explain a large share (30–50%) of individual variance in negative affect, positive affect, and subjective wellbeing (Steel et al., 2008).

Transparency and Openness

The current investigation used data from two larger investigations that were pre-registered (see Thompson et al., 2021; Fitzsimmons et al., 2023). Analyses for both cohorts were also pre-registered. Pre-registrations (cohort 1: https://osf.io/6k73y; cohort 2: https://osf.io/8svn), materials, data and code (https://osf.io/4hyk6/) are all available at Open Science Framework. All study procedures were approved by the Kent State University Institutional Review Board.

Methods

Participants

We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study. Participants were recruited via Qualtrics panels and provided online written informed consent. Initial recruitment was stratified by age, gender, and education to match the U.S. distribution. Participants were recruited in two large-scale cohorts in March/April of 2020 and again in January of 2021. Following the online assessment, participants were invited to continue with a 10-day daily diary, but not all participants opted to continue. For Cohort 1, of the n = 1297 who were initially recruited, 709 participants completed at least some of the daily diary, 523 completed the 1-year follow-up assessment, and 420 completed both. In this investigation we include n = 378Footnote 1 individuals who completed four or more days of the diary and the follow-up assessment. For cohort 2, of the initial n = 1820 recruited, 810 participants completed at least some of the diary, 662 completed the 1-year follow-up assessment, and 507 completed both. We included, n = 460Footnote 2 participants who completed four or more diaries and the follow-up.

Cohort 1 had a mean age of 51.97 (M = 51.97, SD = 15.98). 80.4% of participants identified as White, 7.1% as Black or African American, 5.3% as Mixed (reported 2 or more racial identities), 4.2% as Asian, 2.1% as Hispanic or Latino, 0.5% as American Indian or Alaskan Native, and 0.3% as Native Hawaiian or Pacific Islander. In addition, 25.9% of the sample had a high school diploma or less, 34.2% had up to an Associate’s degree, 25.9% had a Bachelor’s degree, and 14% had a graduate degree. The sample was also majority female (58.7%), with 40.7% identifying as male, and 0.6% identifying as non-binary or other. Cohort 2 had similar demographics, with a mean age of 52.87 (M = 52.87, SD = 15.57), 81.1% of participants identifying as White, 6.9% as Black or African-American, 5.1% as Asian, 2.9% as Hispanic or Latino, 1.8% as 1.6, 1.3% as Mixed, 0.7% as Other, and 0.2% as American Indian or Alaskan Native. In addition, 22.8% had a high school diploma or less, 31.3% had up to an Associate’s degree, 27% had a Bachelor’s degree, and 18.3% had a graduate degree. The sample was also predominantly female (54.7%), with 44.9% identifying as male, and 0.4% as non-binary other.

Procedure

Details regarding the parent study design and procedures are reported in Thompson and colleagues (2021; Coifman et al., 2021; Fitzsimmons et al., 2023) and all materials were provided at pre-registration (Parent project: https://osf.io/4hyk6/). Participants completed an online assessment in which they reported demographics, completed questionnaire measures (e.g., trait anxiety), and a brief math intervention. Following the initial assessment, participants were invited to complete 10-days of daily diaries. This portion of the study was optional, and participants had the opportunity to opt in or out. Hence, only a proportion of the original sample continued onto the diary portion. These diaries were administered via Qualtrics and were sent via email at the same time each day (mid-late afternoon). Approximately 1 year later, participants were invited to complete a follow-up assessment through Qualtrics. The follow-up assessment included measures of psychological functioning (i.e., psychological symptoms, psychological well-being, loneliness). For the current investigation, we include only demographics, report of trait anxiety, reports of affect in the daily diary and follow-up assessments of symptoms, wellbeing, and loneliness.

Materials

Trait Anxiety

During the initial assessment, participants from both cohorts completed the 20-item State-Trait Anxiety Inventory (Spielberger et al., 1971). Internal consistency was excellent for both cohorts (Cohort 1: α = 0.903; Cohort 2: α = 0.912). Mean levels of trait anxiety for cohort 1 (M = 40.11, SD = 11.87) and cohort 2 (M = 40.18, SD = 12.03) reflect some elevations in anxiety, consistent with prior community samples during stressful circumstances (Barnes et al., 2002). Elevations may have been the result of peak periods of uncertainty due to the COVID-19 pandemic.

Daily Diary

At each diary, participants first reported current emotional experience, then COVID-19 specific cognitions and behaviors not included in this investigation (see Thompson et al., 2021; Coifman et al., 2021). Cohort 1 had a diary completion rate of 89.66% (M = 8.97, SD = 1.55), with a total of 3389 diary reports across n = 378 participants. Diary compliance for cohort 2 was 88.50% (M = 8.85, SD = 1.61) with a total of 4070 diary reports across n = 460 participants.

Daily Emotion Reports

Participants reported the extent to which they currently felt both positive and negative emotion using a 5-point Likert-type scale (1 “Not at all” to 5 “Extremely”). For cohort 1, we used the following emotion terms to index negative affect: anger, disgust, distress, fear, guilt, sadness. We used the following emotion terms to index positive affect: relief, enjoyment, amusement, happiness, affection, joy. For cohort 2, we used slightly different emotion words to index negative (anger, disgust, fear, guilt, sadness) and positive affect (relief, happiness, calm, excitement, thankful). The emotion terms used for both cohorts were consistent with dominant circumplex models of affect (Rafaeli et al., 2007). Reliability for each cohort was assessed at the within-person (Rc) and between-person (Rkf) level, consistent with recommendations (Cranford et al., 2006). For cohort 1, between-person reliability was excellent (negative affect: Rkf = 0.99; positive affect: Rkf = 0.99), and within-person reliability was good (negative affect: Rc = 0.66; positive affect: Rc = 0.67). For cohort 2, between-person reliability was excellent (negative affect: Rkf = 0.99; positive affect: Rkf = 0.99) and within-person reliability was good (negative affect: Rc = 0.55; positive affect: Rc = 0.59).

From daily diary reports of emotion, we derived two static indices of emotional experience across both cohorts. This included person-level mean and person-level standard deviations by valence (per recommendations Dejonckheere et al., 2019a, 2019b).

We also derived two dynamic indices of emotional experience that considered time and sequence and reflect the process of appraising daily experiences. Emotion polarity (Coifman et al., 2012; Rafaeli et al., 2007) is an estimate of the within-person momentary association of negative to positive emotional experience. To compute, we create person-centered positive affect scores that are entered as a fixed effect predicting moment-level report of negative affect, both at diary signal t (Proc Mixed; SAS Institute Inc, 2013). An autoregressive error structure is used, and the random effects for the slope coefficient are retained as an individual estimate of polarity. Lower scores reflect a highly polarized conceptualization of emotional experience. Consistent with prior research (Rafaeli et al., 2007), mean emotion polarity for both cohorts was near zero (Cohort 1: M = 0.01, SD = 0.19, range −0.76 to 0.75; Cohort 2: M = 0.01, SD = 0.17, range −0.59 to 0.58).

Emotional inertia (Kuppens et al., 2009; Koval et al., 2016) is an estimate of the within-person serial autocorrelation of affective experience by valence (positive or negative). Both negative and positive emotional inertia were derived via the same mixed-multilevel modelling framework. We entered the person-centered indicator of state affect (either person-centered negative or positive) at diary signal t predicting the moment-level report of the same-valenced affect at diary signal t + 1. Again, we retain the random effects for the slope coefficient as individual-level estimates. Higher scores of emotional inertia (NE inertia or PE inertia) suggests a stronger association from moment to moment in affect. Consistent with prior research (Kuppens et al., 2009), both NE inertia (Cohort 1: M = −0.00, SD = 0.02; Cohort 2: M = −0.00, SD = 0.07) and PE inertia (Cohort 1: M = −0.00, SD = 0.01; Cohort 2: M = 0.00, SD = 0.03) had means approximating zero. A brief description of each emotion construct and its respective equation can be found in Table 1.

Table 1 Overview of emotion constructs computed from the diary affect ratings
Full size table

1-Year Follow-Up Assessments

Psychological Symptoms

Approximately 1-year following each diary data collection, participants completed the combined scales of the Patient Health Questionnaire–8 (PHQ-8; Kroenke et al., 2009) and the Generalized Anxiety Disorder–7 (GAD-7; Spitzer et al., 2006). Previous research using a combined score from the PHQ-9 and GAD-7 has demonstrated strong reliability and construct validity (Chilcot et al., 2018; Kroenke et al., 2016). Descriptive statistics suggested that participants were moderately distressed at each time point relative to other community samples: Cohort 1: M = 8.13, SD = 9.82; Cohort 2: M = 6.96, SD = 9.04 (Patel et al., 2022; Rettie & Daniels, 2021). Internal consistency of this measure was excellent (Cohort 1: α = 0.93; Cohort 2; α = 0.90).

Loneliness

Participants completed the short version of Social and Emotional Loneliness Scale for Adults (SELSA-S; DiTommaso et al., 2004), a 15-item measure of loneliness. Items were rated on a 7-point Likert-type scale. The SESLA-S has demonstrated strong discriminant and convergent validity, as well as high test–retest reliability (DiTommaso et al., 2004). Internal consistency of this measure was good (Cohort 1: α = 0.89; Cohort 2 α = 0.89). Mean reports of loneliness (Cohort 1: M = 45.86, SD = 19.24; Cohort 2: M = 44.25, SD = 19.06) were higher than other samples indexed in research prior to the pandemic (see Ernst et al., 2022 for a recent meta-analysis).

Psychological Wellbeing

Participants completed the Riverside Life Satisfaction Scale (Margolis, et al., 2019) as an index of psychological wellbeing. Participants rated their agreement with 6 items on a 7-point Likert-type scale. This measure has demonstrated high internal consistency and strong test–retest reliability (Margolis et al., 2019). Internal consistency of this measure was very good (Cohort 1: α = 0.90; Cohort 2: α = 0.89). Mean reports of wellbeing (Cohort 1: M = 25.65, SD = 8.94; Cohort 2: M = 27.22, SD = 8.68) were consistent with prior research (Datu et al., 2022).

Data Analytic Plan

Analyses for both cohorts were pre-registered at (Cohort 1: https://osf.io/6k73y; Cohort 2: https://osf.io/8svn5). As preliminary analyses we conducted independent samples t-tests to check for within-cohort differences in any key variables based on whether participants received the initial online math intervention (versus the control condition: Thompson et al., 2021). Zero-order correlations between all key variables for cohort 1 are presented in the supplemental material (Tables S1 and S2).

Ordinary least squared (OLS) regression models were applied to test each outcome (psychological symptoms, loneliness, and psychological wellbeing). Models tested emotion polarity, NE inertia, and PE inertia as predictors of all three outcome variables separately. Therefore, we ran 9 models in total for each cohort. Each predictor was tested against static indicators including mean affect intensity and affect variability to determine if dynamic indices predicted unique variance in outcomes above and beyond static measures of affect. We further tested whether dynamic and stable indices of emotion predicted outcomes above and beyond trait anxiety—the most temporally stable indicator included in the model and a factor known to have robust associations with psychopathology (e.g., Barlow et al., 2014), loneliness (e.g., Ikizer et al., 2022) and wellbeing (e.g., Emma et al. 2017). For all models, we controlled for diary compliance, age, gender, and education. Each model was run incrementally, adding variables at each step. Covariates were entered in step 1, a dynamic emotion index was entered in step 2, person-level mean NA and PA were entered in step 3, person-level standard deviations for NA and PA were entered in step 4, and trait anxiety was entered in step 5. Lastly, we explored interactions between emotion indices and affect intensity of similar valence (e.g., negative emotion inertia with mean negative affect). We added the interaction term to the final step of each model (step 6). All predictors were mean-centered, and we confirmed that the data met the assumptions of regression. Because we conducted several separate OLS regression models, we used the Benjamini–Hochberg False Discovery Rate (FDR) Correction procedure to correct for Type 1 error (Benjamini & Hochberg, 1995). Effects that survive this correction are described below and indicated in bold in tables. See Supplementary Tables S3–S8 for a full summary of the Benjamini–Hochberg Method.

Results

Preliminary Analyses

We first conducted independent samples t-tests to check for group differences in key variables based on the initial math intervention condition (Fitzsimmons et al., in press; Thompson et al., 2021). The only group difference was for PE inertia in cohort 1, t(376) = −2.19, p = 0.029, d = 0.23, suggesting that those in the math condition (M = 0.000, SD = 0.008) had slightly greater PE inertia than the control group (M = −0.001, SD = 0.007). Hence, we ran all models with and without condition as a covariate and the results were identical. For the sake of parsimony, we only include models without condition here.

Next, zero-order correlations showed associations between all three indices of emotion dynamics in both cohorts. Emotion polarity was negatively associated with NE inertia, but its association with PE inertia differed by cohort. In cohort 1, emotion polarity was positively associated with PE inertia (r = 0.15, p = 0.003), whereas in cohort 2 it was negatively associated with PE inertia (r = −0.19, p < 0.001). More polarized emotional reports (lower estimates) and higher NE inertia were generally associated with higher symptoms, loneliness, and lower wellbeing as prior research would suggest, but there were some non-significant associations (e.g., polarity and well-being in cohort 1). The effects of PE inertia differed across the two cohorts—consistent with mixed results in prior research (Höhn et al., 2013; Koval et al., 2016). In cohort 1, PE inertia was negatively associated with mean NA, trait anxiety, psychological symptoms, and loneliness, but positively associated with mean PA and wellbeing. In contrast, in cohort 2, PE inertia was positively associated with mean NA, trait anxiety, psychological symptoms, and loneliness, but negatively associated with mean PA and wellbeing. For both cohorts, trait anxiety had expected medium to large correlations with symptoms (rs = 0.57-0.62), loneliness (rs = 0.45–0.46), and wellbeing (r = −0.61 both cohorts). See Tables S1 and S2 for all zero-order correlations for each cohort.

Primary Analyses

Emotion Polarity as a Predictor of Psychological Symptoms, loneliness, and Wellbeing (Tables 2 and 3)

Table 2 Regression results of emotion polarity predicting psychological symptoms, loneliness, and psychological wellbeing (Cohort 1)
Full size table
Table 3 Regression results of emotion polarity predicting psychological symptoms, loneliness, and psychological wellbeing (Cohort 2)
Full size table
Psychological Symptoms

For cohort 1, emotion polarity had no unique association with symptoms in any step of the model. Instead, mean NA (β = 0.24, p < 0.001, sr2 = 0.03) had a small positive association to symptoms that remained even when considering trait anxiety (β = 0.52, p < 0.001, sr2 = 0.14). There was also a small inverse effect of education (β = −0.11, p = 0.007, sr2 = 0.01) on symptoms. For cohort 2, emotion polarity initially had a significant inverse effect on symptoms (more polarized daily affect was associated with higher symptoms) in step 2 of the model (β = −0.27, p < 0.001, sr2 = 0.07), but the effect became non-significant after adding mean NA (β = 0.42, p < 0.001, sr2 = 0.12) and mean PA (β = −0.18, p < 0.001, sr2 = 0.03) in step 3. After adding trait anxiety in step 5 of the model, only age (β = −0.17, p < 0.001, sr2 = 0.02), mean NA (β = 0.28, p < 0.001, sr2 = 0.04), and trait anxiety (β = 0.36, p < 0.001, sr2 = 0.06) had unique effects on symptoms.

Loneliness

For cohort 1, emotion polarity had no effect on loneliness in any step of the model (β = 0.01, p = 0.816). After adding mean NA and PA to the model in step 3, mean PA predicted lower levels of loneliness (β = −0.28, p < 0.001, sr2 = 0.06). However, this effect became non-significant after adding trait anxiety in step 5, with trait anxiety becoming the only significant predictor of loneliness (β = 0.48, p < 0.001, sr2 = 0.12). For cohort 2, emotion polarity had no association with loneliness (β = 0.09, p = 0.095). However, in this model, mean PA was retained as a significant predictor and in the final step (when also considering trait anxiety: β = 0.42, p < 0.001, sr2 = 0.08) higher mean PA predicted lower reports of loneliness (β = −0.18, p = 0.001, sr2 = 0.02).

Psychological Wellbeing

Results from cohort 1 indicated that emotion polarity was only a significant predictor of wellbeing in step 4 when controlling for affect intensity and variability (β = −0.17, p = 003, sr2 = 0.002). After adding mean NA and PA to the model in step 3, mean NA predicted lower wellbeing (β = −0.19, p < 0.001, sr2 = 0.03) and mean PA predicted higher wellbeing (β = 0.32, p < 0.001, sr2 = 0.008). However, once trait anxiety was added in step 5, trait anxiety became the only significant predictor of wellbeing (β = −0.58, p < 0.001, sr2 = 0.18). Results from cohort 2 indicated that in step 2 of the model, emotion polarity predicted greater reports of wellbeing (β = 0.25, p < 0.001, sr2 = 0.06). However, this effect became non-significant after adding mean NA and PA to model in step 3, with mean NA predicting lower wellbeing (β = −0.17, p = 0.001, sr2 = 0.02) and mean PA predicting greater wellbeing (β = 0.39, p < 0.001, sr2 = 0.13). After adding trait anxiety in step 5, mean PA continued to predict greater wellbeing (β = 0.17, p < 0.001, sr2 = 0.02), trait anxiety predicted lower wellbeing (β = −0.54, p < 0.001, sr2 = 0.13), and the effect of mean NA became nonsignificant (β = 0.01, p = 0.898). There was some evidence of an emotion polarity by mean NA interaction (β = −0.13, p = 0.019, sr2 = 0.01), but this effect did not survive the Benjamini–Hochberg correction.

In summary, there was some evidence in cohort 2 that emotion polarity was protective against symptoms and promoted wellbeing, with the inverse relationship in cohort 1. However, these effects were non-significant when we accounted for static indices of affect intensity and variability. In general, mean PA was protective and mean NA conferred risk to long-term psychological health, and the specific effects were sometimes retained even if trait anxiety was considered. Additionally, the person-level standard deviation for NA and PA exerted no influence in any model. As expected, trait anxiety reported at the initial assessment was the strongest predictor of long-term outcomes for both cohorts.

Negative Emotional Inertia as a Predictor of Psychological Symptoms, Loneliness, and Wellbeing (Tables 4 and 5)

Table 4 Regression results of NE inertia predicting psychological symptoms, loneliness, and psychological wellbeing (Cohort 1)
Full size table
Table 5 Regression results of NE inertia predicting psychological symptoms, loneliness, and psychological wellbeing (Cohort 2)
Full size table
Psychological Symptoms

Results from cohort 1 indicated that NE inertia predicted greater reports of symptoms in step 2 of the model (β = 0.30, p < 0.001, sr2 = 0.09). However, once the mean affect intensity variables were added, the effect of NE inertia become non-significant. In the final model, mean NA (β = 0.23, p < 0.001, sr2 = 0.02) and trait anxiety (β = 0.53, p < 0.001, sr2 = 0.15) predicted greater reports of symptoms, and education was protective (β = −0.11, p = 0.006, sr2 = 0.01). The results were similar for cohort 2. In step 2 of the model, NE inertia predicted greater reports of symptoms (β = 0.30, p < 0.001, sr2 = 0.09), but the effect became non-significant after adding the mean affect intensity variables in step 3, in which mean NA predicted greater symptoms (β = 0.46, p < 0.001, sr2 = 0.10) and mean PA predicted fewer symptoms (β = −0.18, p < 0.001, sr2 = 0.03). In step 5, mean NA continued to predict greater symptoms (β = 0.29, p < 0.001, sr2 = 0.03), along with trait anxiety (β = 0.34, p < 0.001, sr2 = 0.06), but the effect of mean PA became non-significant (β = −0.03, p = 0.450). For all steps in the model, older age was negatively associated with symptoms (Step 5: β = −0.17, p < 0.001, sr2 = 0.02). Lastly, there was some evidence for a NE inertia by mean NA interaction (β = 0.13, p = 0.035, sr2 = 0.01), but the effect did not survive the Benjamini–Hochberg correction.

Loneliness

Results from cohort 1 showed no effect of NE inertia on loneliness (β = 0.06, p = 0.230). Instead, the only significant predictor of loneliness was trait anxiety (β = 0.49, p < 0.001, sr2 = 0.13). Results from cohort 2 were similar, but in the final model, mean PA predicted lower reports of loneliness (β = −0.17, p = 0.001, sr2 = 0.02) and trait anxiety predicted greater loneliness (β = 0.40, p < 0.001, sr2 = 0.08).

Psychological Wellbeing

Results from cohort 1 indicated that NE inertia initially predicted lower wellbeing in step 2 of the model (β = −0.16, p = 0.002, sr2 = 0.03), but this effect became non-significant after adding mean affect intensity in step 3 of the model, in which mean PA predicted greater wellbeing (β = 0.29, p < 0.001, sr2 = 0.07). However, after adding trait anxiety to the model in step 5, trait anxiety was the only unique predictor of wellbeing (β = −0.60, p < 0.001, sr2 = 0.19). from cohort 2 were similar. Higher NE inertia initially predicted lower wellbeing in step 2 of the model (β = −0.23, p < 0.001, sr2 = 0.05), but it had no unique effect after adding mean NA (β = −0.19, p = 0.001, sr2 = 0.02) and mean PA (β = 0.40, p < 0.001, sr2 = 0.14). When trait anxiety was added in step 5 of the model, the protective effect of mean PA remained (β = 0.17, p < 0.001, sr2 = 0.02), trait anxiety predicted lower wellbeing (β = −0.53, p < 0.001, sr2 = 0.13), but there was no unique effect of mean NA (β = 0.01, p = 0.826).

In summary, in both cohorts, there was initial evidence that NE inertia predicted greater symptoms and lower wellbeing. However, like emotion polarity, NE inertia did not predict unique variance in outcomes when accounting for static indices of affect intensity. Instead, as in the prior set of models, mean NA was generally maladaptive and mean PA was generally protective and these effects survived the inclusion of trait anxiety.

Positive Emotional Inertia as a Predictor of Psychological Symptoms, Loneliness, and Wellbeing (Tables 6 and 7)

Table 6 Regression results of PE inertia predicting psychological symptoms, loneliness, and psychological wellbeing (Cohort 1)
Full size table
Table 7 Regression results of PE inertia predicting psychological symptoms, loneliness, and psychological wellbeing (Cohort 2)
Full size table
Psychological Symptoms

Results from cohort 1 indicated that PE inertia initially predicted fewer symptoms in step 2 of the model (β = 0.23, p < 0.001, sr2 = 0.05). However, this effect became non-significant after accounting for mean affect intensity in step 3, in which mean NA predicted greater reports of symptoms (β = 0.40, p < 0.001, sr2 = 0.14). After adding trait anxiety to the model in step 5, the effect of mean NA remained (β = 0.22, p < 0.001, sr2 = 0.03). In this step, trait anxiety predicted greater reports of symptoms (β = 0.52, p < 0.001, sr2 = 0.14) and education was protective against symptoms (β = −0.11, p = 0.006, sr2 = 0.01). In contrast, results from cohort 2 indicated that PE inertia initially predicted greater reports of symptoms (β = 0.25, p < 0.001, sr2 = 0.06), but this effect became non-significant when accounting for mean affect intensity in step 3, in which mean NA predicted greater symptoms (β = 0.37, p < 0.001, sr2 = 0.08) and mean PA predicted fewer symptoms (β = −0.19, p = 0.004, sr2 = 0.01). In step 5 of the model, the unique effect of mean NA remained (β = 0.26, p < 0.001, sr2 = 0.04), trait anxiety predicted greater symptoms (β = 0.35, p < 0.001, sr2 = 0.06), and older age was protective against symptoms (β = −0.17, p < 0.001, sr2 = 0.02). However, in step 6 of the model, there was a significant PE inertia by mean PA interaction (β = −0.15, p < 0.001, sr2 = 0.02). Tests of simple slopes indicated that PE inertia predicted lower reports of symptoms at higher levels of mean PA (1 SD above the mean; B = 45.35, p = 0.035, 95% CI [−87.46, −3.24]), but had no effect on symptoms at low levels of mean PA (B = 28.87, p = 0.181, 95% CI [−13.45, 71.18]). See Figure S1 in the supplemental material for the plotted interaction.

Loneliness

Results from cohort 1 indicated that higher PE inertia predicted lower levels of loneliness in step 2 of the model (β = −0.30, p < 0.001, sr2 = 0.09). Moreover, this unique effect of PE inertia remained significant even when controlling for mean affect intensity, person-level standard deviations of NA and PA, and trait anxiety in step 5 of the model. In this final step, the unique predictors of loneliness were PE inertia (β = −0.20, p = 0.002, sr2 = 0.02) and trait anxiety (β = 0.48, p < 0.001, sr2 = 0.12). There was no evidence of a PE inertia by mean PA interaction. By contrast, results from cohort 2 indicated that PE inertia predicted greater loneliness initially in step 2 of the model (β = 0.30, p < 0.001, sr2 = 0.09), but the effect became non-significant after adding affect intensity in step 3 of the model (β = 0.00, p = 0.986), in which higher mean PA was significant predictor of lower reports of loneliness (β = −0.34, p < 0.001, sr2 = 0.04). Mean PA remained a unique inverse predictor of loneliness in step 5 (β = −0.19, p = 0.010, sr2 = 0.01), along with trait anxiety as a predictor of greater loneliness (β = 0.40, p < 0.001, sr2 = 0.08). Lastly, there was some evidence of a PE inertia by mean PA interaction (β = −0.10, p = 0.028, sr2 = 0.01), but the effect did not survive the Benjamini–Hochberg correction.

Psychological Wellbeing

Results from cohort 1 indicated that higher PE inertia predicted greater reports of wellbeing in step 2 of the model (β = 0.36, p < 0.001, sr2 = 0.13), and this effect remained significant when controlling for all other variables in step 5. In step 3 of the model, when affect intensity was added, mean NA was a unique inverse predictor of lower levels of wellbeing (β = −0.14, p = 0.006, sr2 = 0.02), but the effect became non-significant when controlling for person-level standard deviations in NA and PA and trait anxiety in step 5. In the final model, only PE inertia (β = 0.19, p = 0.002, sr2 = 0.02) and trait anxiety (β = −0.58, p < 0.001, sr2 = 0.18) had unique effects on wellbeing. Moreover, there was no evidence of an interaction by mean levels. In contrast, results from cohort 2 indicated that PE inertia initially predicted lower reports of psychological wellbeing in step 2 of the model (β = −0.39, p < 0.001, sr2 = 0.14), but the effect became non-significant when controlling for affect intensity in step three of the model, in which mean NA predicted lower wellbeing (β = −0.20, p < 0.001, sr2 = 0.03) and mean PA predicted greater wellbeing (β = 0.36, p < 0.001, sr2 = 0.05). These effects remained significant in step 4 when adding person-level standard deviations for NA and PA to the model. However, in step 5 of the model, trait anxiety became the only unique predictor of wellbeing (β = −0.53, p < 0.001, sr2 = 0.13).

In summary, PE inertia was the only emotion dynamic index with unique effects on outcomes when considering static indices, such as mean affect intensity, and dispositional risk, trait anxiety. In cohort 1, it uniquely predicted lower loneliness and higher wellbeing, suggesting persistent PA was generally protective. Moreover, in cohort 2, there was a significant PE inertia by mean PA interaction which suggested that PE inertia was protective against symptoms when mean levels of PA were higher. Like all previous models, indices of affect variability had modest influence on outcomes. Trait anxiety consistently had the strongest effects, predicting poorer outcomes in all models.

Discussion

In this investigation, we examined emotional experiences reported in daily life as predictors of psychological health one year later, in two representative community samples of U.S. adults. Using a multi-cohort longitudinal design, we investigated how dynamic conceptualizations of emotion, operationalized as emotion polarity, NE and PE inertia, and static indices of affect intensity and variability, extracted from daily reports of affect over 10 days, were associated with psychological symptoms, loneliness, and psychological wellbeing 1 year later. Results from both cohorts indicated that dynamic indices of affect were associated with long-term psychological health, but that they only rarely accounted for unique variance in outcomes when controlling for static indices of affect intensity. Rather, mean negative daily affect predicted higher reports of symptoms and mean positive daily affect tended to predict lower reports of loneliness and higher reports of wellbeing. The one dynamic emotion indicator that was an exception was that greater PE inertia predicted lower loneliness and higher wellbeing, above and beyond static indices (as well as trait anxiety) in cohort 1. However, in cohort 2, the protective effect of PE inertia was moderated by mean intensity of positive emotion, exerting benefits only for those with high intensity PA across days. Our results also demonstrated that although trait anxiety was the strongest and most consistent predictor of maladaptive longitudinal outcomes (as expected), daily reports of affect, particularly static, but in some cases dynamic (PE inertia), did still have unique, albeit small, associations with outcomes. Because daily affective experience, and in particular the ways in which individuals understand their affective experience, is a common focus in contemporary psychotherapies, these findings have clear relevance as they relate to targets for intervention and models of risk.

Although some effects were robust and consistent across samples (e.g., effects of trait anxiety, mean NA, and mean PA), it is possible that some of the subtler discrepancies (e.g., PE inertia) are due to contextual differences related to the COVID-19 pandemic. Indeed, we tested these questions at two distinct timepoints: the initial months in which lockdowns were in place across the country (March and April 2020) and during the first winter peak in cases (January and February 2021). Therefore, results could be due to differences in the experience of stress, or in the amount of social interaction participants were engaging in during these unique pandemic contexts.

Recent methodological and theoretical innovations have led affective scientists to emphasize the importance of emotion flexibility, with a focus on the role of individual conceptualizations of affective experience (Coifman & Summers, 2019). With this framework in mind, we investigated if reports of more flexible affective experiences (e.g., lower NE inertia and less polarized reports of negative and positive emotion) would be adaptive for long-term mental health outcomes when considered against static indices of affect—and—stable and robust predictors of outcomes.

Results from the current investigation suggested that more flexible conceptualizations of affective experiences—in the form of lower emotion polarity—were generally protective. However, when static indicators were also accounted for, those impacts fell away. This finding is surprising, given that past research has shown meaningful relationships between emotion polarity and psychological health even when controlling for mean affect intensity and variability (Dejonckheere et al., 2018, 2019a, 2019b). However, unique effects in past research were small, and more recent research did not find evidence for longitudinal associations between polarity and wellbeing (also during the pandemic; Dawel et al., 2023). Moreover, there are important methodological differences between the present investigation and past research worth noting. Specifically, past research has conceptualized emotion polarity as the outcome, rather than the predictor (e.g., Dejonckheere et al., 2018, 2019a, 2019b), which might explain some of these discrepancies. On the other hand, the COVID-19 pandemic created an unprecedented source of stress, and past research has suggested that less polarized affective experiences are hard to retain under stressful conditions (Coifman et al., 2012; Zautra, 2003) likely due to depleted psychological resources (Davis et al., 2004). Indeed, in now two investigations administered during the pandemic (that we are aware of), the relationship between polarity and wellbeing-related measures has not been robust when controlling for other factors (Dawel et al., 2023). To attempt to reconcile disparate findings in the literature, future research should consider moderators and mediators of the relationship between emotion polarity and wellbeing-related factors, including stressful contexts (e.g., Coifman et al., 2012) and emotion regulation capacity or flexibility (e.g., Dejonckheere et al., 2018), as well as the directionality of these relationships within a longitudinal framework.

Consistent with previous research, greater NE inertia predicted greater reports of symptoms, loneliness, and lower wellbeing (Koval et al., 2016; Kuppens et al., 2009) one year later. However, like the results of emotion polarity, these effects did not survive when controlling for static levels of NA and PA. Indeed, much of the past work on emotional inertia has not controlled for both mean levels of NA and PA, and thus has not accounted for the shared variance between these sometimes, interdependent constructs (Rafaeli et al., 2007). A recent investigation into the relative importance of dynamic indices of affect suggested that emotional inertia may have limited added value for the prediction of outcomes such as depression symptoms, borderline symptoms, and life satisfaction (Dejonckheere et al., 2019a, b). However, we would argue that these findings still have clinical importance. NE inertia reflects persistent reports of negative emotion across time, reflecting rigidity in the conceptualization of emotion. This rigid pattern of emotion is consistent with dominant models of emotion-related risk for affective disorders (i.e., Joormann & Gotlib, 2010). However, our results are also consistent with recent research (e.g., Dejonckheere et al., 2019a, b) that suggest NE inertia provides only limited incremental value in the prediction of longitudinal outcomes.

Unlike the NE inertia findings that were broadly consistent across cohorts, PE inertia did have interesting diverse effects by cohort. In cohort 1, greater PE inertia predicted lower loneliness and higher wellbeing, and these effects remained significant after controlling for (1) affect intensity, (2) affect variability, and even (3) trait anxiety, suggesting that the effects were robust. Hence, reporting persistent levels of positive affect during the baseline measurement was linked to better psychological health one year later. Interestingly, the effects were more complicated in cohort 2. Indeed, the inconsistency between cohorts 1 and 2 is consistent with the broader PE inertia literature (e.g., Höhn et al., 2013; Koval et al., 2016). These mixed findings may be the result of how emotional inertia is computed, where greater scores on the PE inertia index reflect only the persistence of reports from one moment to the next. Hence, greater PE inertia scores can reflect both persistent, but low, or persistent, but high reports of PA, making the interpretation of findings difficult. Thus, by testing PE inertia’s interaction with mean intensity levels of PA, the results gained added nuance. Specifically, in cohort 2, PE inertia was protective against symptoms only at higher mean intensity PA. These findings have clear clinical implications, suggesting that persistent elevated PE in daily experiences can be adaptive and is associated with psychological health and wellbeing. Although the test of moderation did not reach significance in cohort 1, the same relationships appeared to be present. Future research on PE inertia should continue to consider its interaction with PA intensity to better parse its effects on indices of adjustment and wellbeing, which can depend on overall PA intensity.

Finally, mean levels of affect, reflecting broader estimates of mood during our 10-day assessment, remained consistent and robust predictors of longitudinal health, even when controlling for dispositional risk, operationalized as trait anxiety. Negative affectivity is well-established for its stable relationship with psychopathology and other indices of wellbeing across the lifespan (Griffith et al., 2010; Shackman et al., 2016). Thus, by controlling for trait anxiety, we were able to isolate the relative role of state emotion indicators reported in the diary. As expected, trait anxiety was indeed the most consistent predictor of all three outcomes. However, mean levels of NA and PA also exerted small but meaningful effects on the outcomes. Specifically, mean NA was most predictive of psychological symptoms, whereas mean PA was most predictive of loneliness and wellbeing. Given that the shared variance between dispositional negative affectivity and affect reports is considerable (here rs at approximately 0.50), this suggests that daily affect remains a useful target for interventions targeting adjustment processes over time and psychological health. Several clinical implications follow from the results of the current investigation. First, modern psychotherapy approaches (e.g., CBT) emphasize more diverse and flexible conceptualizations of emotional experience, consistent with our findings. Although dynamic indices (polarity and NE or PE inertia) were relatively less meaningful than affect intensity in our fully inclusive models of long-term psychological health, our findings suggest that daily conceptualizations of affective experience are still important and can predict important variance in outcomes one year later. However, our results suggest that interventions targeted at decreasing negative affect and increasing positive affect, at daily intervals, may be most effective. Indeed, research has shown that daily negative affect is associated with increases in maladaptive cognition and coping in patients with psychopathology (e.g., depression), which in turn maintains disease states, whereas daily positive affect is associated with more adaptive coping (Dunkley et al., 2017). In addition, the reduction of daily negative affect is a mechanism of symptom reduction for several treatments (e.g., CBT: McIntyre et al., 2019; Mindfulness-based stress reduction: Snippe et al., 2017) and the increased generation of positive emotion, a cornerstone of positive psychological intervention (e.g., Seligman et al., 2005).

Findings from the current study relating to PE inertia and mean PA have additional clinical implications. First, it has long been established that positive emotions promote physical health and wellbeing through several channels (Fredrickson, 2004). For instance, experiences of positive emotion serve to down-regulate (or “undo”) the negative effects of bouts of negative emotion (e.g., cardiovascular recovery; Behnke et al., 2023; Fredrickson et al., 2000). In addition, positive emotions foster social connectedness, which in turn, promotes increased positive emotion, access to social resources, and wellbeing (Kok & Fredrickson, 2010; Ong & Allaire, 2005). Moreover, daily positive emotions promote healthy behaviors (e.g., exercise, relaxation and hobbies) in both healthy and impaired (e.g., depression and anxiety) individuals, likely through increased motivation to approach goals (e.g., exercise, support-seeking; Aurora et al., 2022; Nylocks et al., 2018). Results from the current investigation build on these findings by showing that both mean PA and elevated PE inertia over 10 days of reports were protective against loneliness and were associated with greater psychological wellbeing one year later during a period of reduced access to social interaction (e.g., social distancing and lock-down orders).

Importantly, the findings relating to PE Inertia are consistent with broader evidence that it is not only the intensity of positive emotional experiences that matter, but the frequency of positive emotion that promote psychological health and wellbeing (Diener et al., 2009; Fredrickson & Joiner, 2018). For example, individuals who report a high frequency of positive moments are buffered against risk for psychopathology during times of high perceived stress (Bränström, 2013). Research targeting an increase in moments of positive emotion have shown promise for stress reduction (e.g., in medical and emergency personnel during COVID-19; Coifman et al., 2021) and findings are consistent with theories underlying behavioral treatments for disorders like depression (e.g., behavioral activation; Dimidjian et al., 2014). Importantly, there is also increasing evidence suggesting that positive emotion is linked to circadian regulation (Murray et al., 2002; 2009) and that the salutary motivational influences of positive emotion are limited to within a day, rather than across days (Aurora et al., 2022). Taken together, results from the current investigation highlight the importance of both frequent and persistent daily positive affect as a target for treatment to promote greater wellbeing and psychological health.

Study strengths include the pre-registered, multi-cohort longitudinal design using two large, nationally representative U.S. samples stratified by key demographic variables. Another strength is the use of a 10-day daily diary which allowed us to measure both static and dynamic indices of affective experience in daily life. By doing this, we were able to directly compare mean levels of NA and PA (as well as variability in affect) with dynamic indices that reflect clinically relevant conceptualizations of emotional experience as predictors of long-term psychological health, a necessary step in research focusing on dynamic indices (Dawel et al., 2023; Dejonckheere et al., 2019a, b). Furthermore, by considering affect within the context of dispositional negative affectivity, we were able to evaluate the unique effects of state-level affect indicators, including how one conceptualizes emotional experience, a common target in psychotherapy. In addition, because we considered multiple indices of emotion flexibility—an uncommon approach—we were able to examine the covariation between indices, as well as how they differ in their predictions of psychological outcomes. Finally, we investigated both positive (psychological wellbeing) and negative (psychological symptoms and loneliness) indicators of psychological health.

Limitations include single diaries per day, which may have limited our estimates of the dynamic indices of flexibility (polarity and inertia). In addition, our investigation focused on broad dimensions of psychological functioning and the unique effects of dynamic indices were relatively small when predicting these distal outcomes. Further explication of dynamic emotion indices could be oriented to more proximal outcomes (e.g., behaviors in daily life) as has been demonstrated in some prior work (Coifman et al., 2012). Another key limitation relates to our lack of baseline measurement of the longitudinal outcomes (symptoms, loneliness, and wellbeing). To attempt to compensate for this limitation, we included a robust and stable indicator of trait negative affectivity in trait anxiety (Shackman et al., 2016). By controlling for trait anxiety in our models, we likely accounted for some of the stable variance in the outcomes. This nevertheless is an important limitation and these findings warrant replication in future more conventional longitudinal design frameworks. Finally, only a portion of the samples completed the diary or follow-up portions of the study, and there was evidence that those who opted out of them were more distressed, least educated, and more racially diverse participants. This may be due in part to the unique circumstances of this investigation during key moments of the COVID-19 pandemic. Relatedly, because we do not have data from before the COVID-19 pandemic that could serve as a control group, it is unclear what impact the context had on the results. Therefore, both the pattern of attrition and the broader context of the investigation (e.g., COVID-19 pandemic) are important features that constrain the ability to generalize from these results and suggest additional replication is needed.

Conclusion

This investigation aimed to test how daily reports of emotion reported in two distinct periods of the COVID-19 pandemic were predictive of psychological symptoms, loneliness, and wellbeing one year later. By measuring emotion reports in daily life, we were able to test both conventional static measures of negative and positive affect and dynamic measures (polarity and inertia) reflecting specific types of cognitive conceptualizations of emotional experience as predictors of long-term psychological health one year later. Moreover, we tested the relative importance of daily affect when also considering a robust indicator of dispositional risk for emotion-related dysfunction. Results suggest that both stable and dynamic indices were associated with outcomes one year later, predicting small but meaningful variance. In particular, PE inertia was the sole dynamic index predicting unique variance in outcomes after accounting for all other factors. Results were interpreted within the context of common psychotherapeutic targets and suggest continued focus on daily affect experiences, including focus on shifting intensity levels (reducing negative and increasing positive daily emotions). However, our results also shed light on how to consider the dynamic ways in which individuals conceptualize experience, targeting less polarized and less persistent reports of negative affect, and more elevated and persistent reports of positive affect. Finally, these results highlight the importance of daily experiences of emotion during periods of stress in long-term adjustment when considering a range of outcomes, including psychological symptoms, loneliness, and well-being.