Youth Depression Symptoms During COVID-19
A Longitudinal Twin Study on Resilience Factors
Abstract
Abstract: The COVID-19 pandemic has induced new and exacerbated existing stressors. From a resilience framework perspective, we investigated which potentially protective individual and family factors are negatively associated with youth depression symptoms (DS) during COVID-19 and to what extent these associations are attributable to genetic and environmental factors. We considered 3,025 monozygotic and dizygotic twins in their adolescence and early adulthood from a representative German twin family sample. Multiple regression models yielded significant effects of prepandemic DS, life satisfaction, openness to experience, and internalizing behavior. We found a substantially smaller explanatory power of the considered predictors for pandemic compared to prepandemic DS. Twin analyses showed major time-specific environmental effects. Genetic variance was fully explained by prepandemic DS, life satisfaction, openness to experience, and internalizing behavior. Consecutive increases in explanatory power across pandemic waves point toward plasticity. The findings are discussed regarding the specificity of the pandemic and the importance of individual social settings in adaptation to pandemic adversity.
Since the onset of the coronavirus disease 2019 (COVID-19) pandemic, young people worldwide have experienced an increase in depression symptoms (DS; Robinson et al., 2022). Alongside physical health and financial concerns, social distancing policies following the pandemic onset have burdened youth mental health (Jin et al., 2021; Prati & Mancini, 2021). Youth’s adaptation was found to considerably vary between individuals, corroborating that certain personality and social factors affect how young people deal with pandemic stressors and uncertainties (Prime et al., 2020; Shanahan et al., 2020). Adolescence and young adulthood are pivotal phases in development (Bleidorn et al., 2022). There is thus a pressing need for knowledge on how such risk and resilience factors are associated with the stability and change of youth DS during the COVID-19 pandemic and how nature and nurture jointly contribute to it.
The present study focused on adversity from and adaptation to pandemic strains from a resilience perspective. A common denominator of resilience theories is that they pose a set of intraindividual, interpersonal, and socioecological resources as helpful to adapt to adversity. They are deemed to protect from adversity when present and increase the risk of adverse effects when missing (Zimmerman, 2013). Genetically informative designs can further enrich the understanding of genetic and environmental effects underlying the level and change of youth DS and the degree to which individual and social characteristics mediate or moderate these effects.
This study addressed three research questions. Which individual and family factors are negatively associated with DS in adolescents and young adults during the COVID-19 pandemic? How do the effects of these factors compare: Are some more, some less capable to offer protection against the emergence of DS? And, to what extent are these associations attributable to genetic and environmental factors?
The Concept of Resilience
Resilience research examines psychological health and its stability as trait configurations and as outcomes despite adversities (Harms et al., 2018). Adversity has been defined as severe (e.g., bereavement, injury, childhood maltreatment) or ordinary (e.g., modern life stress) experiences that are commonly considered unpleasant and/or challenging. Adaptation has been measured as absence of psychopathology, such as DS, and as positive adaptation, e.g., stable or pronounced subjective well-being (Cosco et al., 2017). Adversity/adaptation dyads are continuous in their level of manifestation and should be treated as such (Masten, 2013; Yates et al., 2015).
Most discussions on resilience are based on an amalgamation of developmental theories, which can be understood as a resilience framework. Theories on competence and adversity, resources and risks, and protective factors and vulnerabilities have been described as comprehensive for understanding resilience (Yates et al., 2015), as have been theories on developmental attachment, agency and mastery, learning, and self-regulation (Masten, 2013). For example, self-determination theory and the theory of self-efficacy conceptualize how humans strive for subjective agency as a means to achieve psychological health (Bandura, 1997; Ryan & Deci, 2000). To incorporate the numerous dimensions necessary to understand resilience, theory is currently moving toward a socioecological multisystem perspective (Masten et al., 2021; Vella & Pai, 2019). Such a framework approach is the analytical base for our study.
Under the umbrella term of resilience theory, resources and assets are differentiated (Zimmerman, 2013). Resources are external factors that potentially help to overcome adversity (e.g., socioeconomic status, regional support structures). Assets are personality traits such as self-efficacy or self-esteem. Protective models of resilience theory postulate that assets and resources moderate adversity risks (Zolkoski & Bullock, 2012). Thus, when coping with the psychological challenges from a pandemic, both individual and familial resource capacities should distinguish (non)resilient individuals.
Transactional theory poses stress, a typical concomitant of adversity, as a distinctly subjective phenomenon (Biggs et al., 2017). Stress management is negotiated between stressor interpretation, resource equipment, and individual coping styles. Similarly, in the framework of salutogenesis, stressors are conceptualized to promote tension, whereas different resistance factors are theorized to ease the burdens of such strains (Eriksson, 2017) and can even be capable of transforming tension positively. Consequently, the capacity (and usage) of individual assets and resources for coping with stressors is key. Stress reactions are then transactional in that they are characterized by plasticity: Individuals can learn and improve upon their capacity to deal with upcoming strains. Resilience theory calls this a challenge model (Zolkoski & Bullock, 2012), akin to stress inoculation theory (Meichenbaum & Deffenbacher, 1988). Therein, moderate stressors (not too mild as not to incite a learning reaction, not too severe as to cause collapse) are postulated to act as inoculants that may help to deal with future stressors. If stress inoculation properties exist in pandemic settings, pandemic effects on DS should decrease over time, as individuals had chances to learn and act out their sets of protective factors.
COVID-19 and Youth Resilience
Many studies reported a decline in youth mental health at the onset of the COVID-19 pandemic (Robinson et al., 2022; Samji et al., 2021). Except for DS, the levels of anxiety and other mental health symptoms diminished later in 2020, mostly returning to prepandemic levels (Robinson et al., 2022). As Robinson et al. discussed, this pattern may reflect an initially severe response to an unexpected distressing event, followed by a lagged adaptation. Previous studies found several individual characteristics to predict youth pandemic DS, such as Big Five personality traits (Mourelatos, 2021), or previous psychopathological symptoms (Samji et al., 2021; Shanahan et al., 2020). In line with resilience theory, family characteristics, such as socioeconomic status (Daly et al., 2020) or parental support and closeness (Vacaru et al., 2022), were also found to play a crucial role for pandemic DS.
The contributions of genetic and environmental factors to variance in youth DS or related constructs of psychological health during COVID-19 have rarely been investigated (e.g., Carroll et al., 2021; Rimfeld et al., 2021). Theories on genotype–environment interplay depict genetic differences to account for divergent individual environment choices and differential environmental reactions to certain genotypes that manifest through behavior (Kandler & Zapko-Willmes, 2017). For example, various exposures, such as social support and socioeconomic adversity, have been found to interact with genetic risks for depression (Kendall et al., 2021). Van de Weijer et al. (2022) found large effects of environmental factors not shared within twin pairs to account for variance in quality of life. They argued that this finding is in line with the bioecological model (Bronfenbrenner & Ceci, 1994), which suggests that genetic effects are amplified in stable environments and, following gene–environment interaction, should thus be impeded in higher-risk environments in favor of unique environmental effects (South et al., 2017).
The Present Study
We examined the protective effects of individual and family characteristics on the level and change of youth DS during the COVID-19 pandemic and how nature and nurture jointly contribute to it. Based on a number of theoretically and empirically underpinned resilience factors, we hypothesized that the level and change of youth DS after the onset of the pandemic significantly varies between individuals (Hypothesis 1; H1). We proposed that individual (H2) and family (H3) characteristics affect the level and change in pandemic DS. Regarding individual characteristics, we assumed a positive association of DS with neuroticism (H2a), externalizing and internalizing problem behavior (H2b), and emotion-oriented coping (H2c), and a negative association of DS with task-oriented coping (H2d), self-efficacy and self-esteem (H2e), life satisfaction (H2f), and optimism (H2g). Following stress inoculation theory, we expected a U-shaped association of DS with negative life experiences (H2h). Regarding family characteristics, we hypothesized a negative association of DS with parental emotional support (H3a) and family socioeconomic status (H3b), and a positive association of DS with a chaotic home environment (H3c). We further hypothesized that unfolding genetic factors and accumulating environmental factors affect the change in youth DS during the COVID-19 pandemic (H4) and that the effect of individual characteristics is primarily attributable to genetic factors (H5). See Table E1 for an overview of the hypotheses’ backgrounds (all table and figure names with letters refer to the Electronic Supplementary Material, ESM 1).
We tested H1 based on latent growth curve models and H2a–h and H3a–c using multiple regression analyses with the stable trait component of DS as the outcome measure, based on the augmented backward elimination algorithm (Dunkler et al., 2014), quasi-cross-validating findings across twins. To test H4 and H5, we ran multivariate twin models. All hypotheses of the study were pre-registered (https://osf.io/wurx2). Deviations from the pre-registration are noted on p. 1 in ESM 1.
Method
Sample
We used data from the ongoing German twin family panel study TwinLife (Hahn et al., 2016). Subjects participated in face-to-face, telephone, and web interviews in up to three surveys before the onset of the COVID-19 pandemic (2014–2016, 2016–2018, 2018–2020) and up to three surveys during the pandemic. We chose participations from narrower time frames within the pandemic surveys to capture specific pandemic phases, namely, the first pandemic wave, assessed in retrospect (study participations June 2020–November 2020), the (partial) lockdown during the second wave and subsequent third wave (November 2020–April 2021), and the onset of the fourth wave (September 2021–November 2021). As a cutoff date for prepandemic information, we considered data from the third prepandemic survey up to 10 March 2020 (the day of the official WHO classification of SARS-CoV-2 as a pandemic).
Data are publicly available (Diewald et al., 2022). We used pre-release internal data to include the third pandemic survey, plus survey dates and regions (to add COVID-19 incidence rates). We considered adolescent and young adult twins from three cohorts (born 1990–1993, 1997–1998, and 2003–2004) who participated in at least one pandemic survey. A total of 3,025 individuals remained, with 1,384 complete same-sex twin pairs. See Table 1 for sample characteristics. Participants’ pandemic DS were similar across waves (self-reported on a four-point scale from 1 = not at all to 4 = almost every day: M = 1.69–1.84, SD = 0.61–0.69; see Table E4 and Figure E1 in ESM 1). See Table E5 in ESM 1 for the rank-order stability of DS.
Measures
DS were measured based on an adaptation of the Beck Depression Inventory-Fast Screen (BDI-FS) before the pandemic and via adapted items of the Patient Health Questionnaire (PHQ-2) during the pandemic (Beck et al., 2000; Löwe et al., 2005). See Table E2 for details on the considered measures for potential predictors and Table E6 for internal consistency estimates for each survey. We also included the participants’ lifetime experience of positive and negative life events. This was operationalized as two indices of positive (from +1 = rather positive to +3 = very positive) and negative (from –1 = rather negative to –3 = very negative) life event evaluations, derived from a list of 30 life events developed for TwinLife and collected during the prepandemic waves.
We used age, sex, dichotomous indicators of a SARS-CoV-2 infection, and quarantine in the household (overall, 9.56% reported an infection and 24.1% reported a quarantine measure during any of the pandemic surveys – see Table E3 for details) as covariates in our regression analyses. Seven-day COVID-19 incidence rates and the survey month were used in survey-specific control analyses to adjust for pandemic status and seasonality effects (see Table E26).
Preparatory Analyses
Construct Validity
To confirm factorial validity, we ran factor analyses for all scales (see ESM 1, parts B and C). For adapted measures, we conducted exploratory factor analyses (EFAs) for each survey. For validated scales, we ran confirmatory factor analyses (CFAs) across surveys, specifying a higher-order latent factor and using a full information maximum likelihood algorithm to handle missingness. Goodness of fit was estimated based on χ2, the root mean square error of approximation (RMSEA), and the comparative fit index (CFI); we considered the model fit sufficient if RMSEA < .06 and CFI > .95 (Hu & Bentler, 1999).
All CFAs showed a satisfactory model fit, except for coping styles, which we subsequently dropped from further analyses (CFI = .929 and .946; see Table E20). EFAs showed the expected factor loading pattern with mostly one-factor solutions, except for internalizing and externalizing problem behavior and coping styles.
Since prepandemic and pandemic DS were measured with different instruments (BDI-FS and PHQ-2, respectively; see Table E2 and E3 for details), we tested whether the DS instruments were convergent by comparing a one-factor model (with a single higher-order latent factor) with a bifactor model (see Table E16). The one-factor model showed a comparably worse fit (Δχ2 = 124, p < .001), suggesting that, despite considerable correlation (r = .63), we cannot treat the measurements as fully convergently valid. Nonetheless, both DS measures are valid instruments (Beck et al., 2000; Richardson et al., 2010) and yield highly correlated measurements. See Table E35 for phenotypic twin correlations of DS.
We further tested for measurement invariance (MI) across waves and cohorts (see ESM 1, part C). In addition to the χ2 difference test, we considered other changes in goodness-of-fit indices as indicators of MI due to the sample size sensitivity of the χ2 difference test, namely, ΔRMSEA ≤ −.015 (Chen, 2007) and ΔCFI ≤ −.01 (Cheung & Rensvold, 2002). Our analyses required scalar MI and invariance of factor covariances to ensure that the factor score estimation is invariant across cohorts and time. When invariance was not found, we tested for partial invariance using the backward method to identify noninvariant parameters (Jung & Yoon, 2016). Our analyses confirmed at least scalar MI across time for most measures, with partial scalar MI for self-esteem (see Table E17). We also found either scalar or partial scalar MI and invariance of factor covariances across cohorts for all measures, with partial invariance of factor covariances for family socioeconomic status (see Table E18).
Factor Scores
Following the CFAs, we extracted latent factor scores based on the partial least squares regression method (Thomson, 1934; Thurstone, 1935). This procedure yielded factor scores for the stable trait component of the repeatedly measured constructs (optimism, home environment, and parental emotional support were measured once).
Main Analyses
To test H1, we ran latent growth curve (LGC) models based on the three pandemic DS measurements. We set the first pandemic measurement as the reference point and compared a level-only model with a model including the level and linear slope (linear change model). The χ2 difference test was used for model selection.
For regression and genetically informative analyses, we considered nonresponse survey weights. They accounted for selective response patterns based on the first prepandemic survey (Krell et al., 2022).
Multiple Regression Analyses
To test H2 and H3, we employed a variable selection procedure through multiple regression analyses based on the augmented backward elimination (ABE) algorithm (Dunkler et al., 2014). ABE is helpful for variable selection in that it offers a reproducible, data-based, and less biased option compared to selection only on theoretical grounds. At the same time, it tends to include more relevant variables than backward elimination, which does not consider a change-in-estimate criterion. Specifically, we first ran the full model (i.e., including all independent variables; see Table E28) with the stable trait component of DS across the pandemic context as the dependent variable and subsequently dropped variables from the model by combining the selection by significance (α = .10) and change-in-estimate criterion (τ = .05). The alpha threshold is set more strictly than recommended to achieve a more parsimonious model. α = .20 or the inclusion of all considered variables did not yield substantial increases in explanatory power (see Table E28).
The results were quasi-cross-validated across twins as follows: We first performed individual ABEs for Twin 1 and Twin 2, which resulted in different variables that remained in each model. The found final model for each twin was applied to the twin and respective co-twin subsample. From this, the model with the highest adjusted R2 across both twins was chosen. Predictors that were significant across both twins were then taken forward into genetically informative analyses.
Genetically Informative Analyses
To test H4 and H5, we conducted twin analyses that allow decomposing interindividual variance into genetic and environmental variance components. Twin analyses are based on the differences in the covariance between and within twin pairs reared together. Under the assumption that monozygotic (MZ) and dizygotic (DZ) twins experience equal environments, additive genetic (A), nonadditive genetic (D), shared environmental (C), and unique (i.e., nonshared) environmental (E; including measurement error) components can be distinguished. Specifically, differences between MZ and DZ pair covariance can be attributed to genetic effects, since MZ twins are genetically identical, whereas DZ twins share, on average, 50% of their segregating genes and 25% of nonadditive genetic effects regarding allelic dominance. A similar pair covariance reflects effects of environmental factors shared by twin siblings reared together. Differences within MZ pairs can be attributed to environmental effects not shared by twin siblings.
We aimed to compare three twin models (a Cholesky decomposition model, a common factor model, and an LGC model) based on Akaike weights (Wagenmakers & Farrell, 2004) to find the model that best describes the genetic and environmental effects on the change of DS. A Cholesky decomposition model allows an atheoretical estimation of genetic and environmental sources of (co)variance of the DS measurements and thus presents a baseline model against which the other models can be compared. A common factor model assumes that the DS measurements share a common latent factor for which genetic and environmental effects are estimated; contributions of genetic and environmental effects on measurement-point-specific variance are also estimated. A genetically informative LGC model enables one to estimate genetic and environmental effects on a specified reference level and the intraindividual change trajectory to or from this level (the slope). See ESM 1, part E, for a graphical representation of the models. The models assume the absence of assortative mating and gene–environment interplay. Subsequently, we fit multivariate twin models to identify the pathways through which the found predictors influence the variance of DS during the COVID-19 pandemic, considering prepandemic DS as a covariate. In the twin analyses, all considered variables were adjusted for significant effects of twins’ sex, age (linear and quadratic), and sex × age interaction using a standard regression procedure (McGue & Bouchard, 1984).
Results
The phenotypic LGC model analyses showed that the level-only model was not significantly worse than the linear change model (Δχ2 = 1.262, p = .738; see Table E30). Thus, we found significant interindividual variance in the (initial) level of pandemic DS (intercept = 1.811, σ2 = 0.186, p < .001) but did not find linear changes over the pandemic. Due to this result, partially confirming H1, we did not compute a genetically informed LGC model, as it would essentially equal a common factor model (see Figures E3 and E4). Thus, we could not test H4.
Multiple Regression Analyses
The regression models that were considered by the ABE algorithm showed surprisingly low explanatory power toward the stable trait component of pandemic DS (adjusted R2 = .16; see Tables E21 and E22 for separate twin analyses and Table E23 for the final model). In the final model, only prepandemic DS, life satisfaction, openness to experience, and internalizing behavior remained as significant predictors across both twins, confirming H2f and partially confirming H2b. The highest positive association, by far, lies with prepandemic DS, while many typical predictors of DS fell out of our consideration criteria. A control analysis without prepandemic DS as a predictor did not yield substantially different results (see Table E24). H2a, H2c–e, H2g, and H3a–c were thus rejected.
To check our results, we conducted regression analyses using the same procedure with prepandemic DS as criterion. We found substantial explanatory power (adjusted R2 = .57; see Table E25). This was a stark contrast to the low prediction of pandemic DS (see Figure 1). A second control analysis compared the prediction of survey-specific pandemic DS states, additionally controlling for seasonality (via survey month) and regional 7-day COVID-19 incidence rates. This showed similar results to the main analyses (adjusted R2 = .12–.17; see Table E26). There were minor seasonality effects (that are accounted for in our longitudinal design) and no effects from the 7-day incidence rate. A third control analysis compared the results with the state scores of the third prepandemic survey. This yielded similar results to the main analysis (adjusted R2 = .16; see Table E27). Another control analysis excluded participants who ever experienced an infection in the household. These results were again comparable, ruling out effects of the infection status (adjusted R2 = .15; see Table E29).
Genetically Informative Analyses
Based on Akaike weights, a Cholesky ADE model was found to fit the data best (wAIC = > .99; see Tables E32 and E33) and better than a common factor model reduced of nonsignificant paths (wAIC = >.99; see Table E34). Thus, we chose a six-factor Cholesky ADE model to estimate the variance contributions of the predictors that were significant across both twins (prepandemic DS, life satisfaction, openness to experience, and internalizing behavior). See Table E35 for parameter estimates and Figure 2A for the proportions of genetic and environmental variance of pandemic DS.
In line with H5, genetic variance in the three pandemic DS measurements was attributable to genetic variance in prepandemic DS (37%–43%), followed by life satisfaction (17%–39%), openness to experience (5%–33%), and internalizing behavior (7%–15%). The amount of genetic variance increased over time through the course of the pandemic (26%, 38%, 36%). However, the methodological difference between the first (retrospective) measurement and following (concurrent) measurements has to be noted here (see the limitations section). Unique environmental variance was substantial (61%–74%) and mostly unexplained. Former pandemic measurements accounted for 8–10% of latter pandemic measurements’ unique environmental variance.
As a control analysis, we ran a Cholesky model for prepandemic DS and its found predictors. We found genetic variance in internalizing problem behavior and conscientiousness to account for most genetic variance in prepandemic DS and about one fifth of unique environmental effects on prepandemic DS to be attributable to individual characteristics (see Figure 2B).
Discussion
Of the wide range of considered resilience factors, we only found prepandemic DS, life satisfaction, openness to experience, and internalizing behavior to predict the stable trait component of youth pandemic DS across both twins. This profoundly differed from the predictability of prepandemic DS in both number of predictors and amount of explained variance in youth DS. Twin analyses revealed a considerable amount of unique environmental variance in youth pandemic DS and that the amount of genetic variance explained by the predictors increased with the progress of the pandemic.
Youth Pandemic DS Differed From Prepandemic DS
In contrast to prepandemic DS and despite the consideration of theoretically and empirically established predictors, we found mostly small to null effects of individual and family characteristics on youth pandemic DS. Prepandemic DS were by far the strongest predictor for pandemic DS, along with internalizing behavior before the pandemic, of which DS are a defining characteristic (e.g., Zahn-Waxler et al., 2000). Linking our findings back to salutogenesis theory, we can only confirm prior life satisfaction as a resistance factor. This would be considered an asset in resilience theory. Openness was positively associated with pandemic (and prepandemic) DS. Past literature has been inconclusive regarding the role of openness for mental health (e.g., Khoo & Simms, 2018; Lyon et al., 2021). This may be due to different effects of openness facets. For example, openness to fantasy and openness to feelings may affect how deeply individuals engage with stressful situations, both cognitively and emotionally. The familial resources that we considered did not show significant protective effects. Youth DS thus appear to be a more specific phenomenon when occurring in a pandemic context. This adds to the growing body of evidence that found high unexplained heterogeneity and variability in pandemic DS (Robinson et al., 2022).
Notably, we did find the amount of explained variance to slightly increase with every subsequent pandemic survey (see Table E26). In line with transactional theory and stress inoculation theory, this may reflect that young people exceedingly acclimated themselves to the pandemic, with commonly important individual and social characteristics reclaiming their protective function, suggesting plasticity in the developmental phases we observed. The exceedingly higher amount of explained interindividual variance in pandemic DS approximated the results for prepandemic DS, suggesting that the later stage pandemic setting approached the baseline. In other words, individual differences in DS were decreasingly attributable to pandemic(-related) factors. This plasticity can also be seen in the higher genetic contribution to variance in DS during the third and fourth compared to the first pandemic wave. The increasing amount of genetic variance in pandemic DS measurements accounted for by prepandemic DS, life satisfaction, openness to experience, and internalizing behavior may suggest an unfolding of inherent individual genetic differences.
Large Time-Specific Unique Environmental Effects
Our results showed a large proportion of mostly unexplained unique environmental effects, in line with previous findings on youth psychopathology (Carroll et al., 2021; van de Weijer et al., 2022). Apart from measurement error, this could reflect time-specific individual environmental factors which we did not account for. This could entail, for example, friendship characteristics and its subjective quality (Juvonen et al, 2022; Vacaru et al., 2022), institutional trust (Ochnik et al., 2022), physical activity (Lippke et al., 2021), partnerships, joblessness, or workplace settings. For example, research on chronic stress has found friendship and family relationship problems to function as unique environmental influences on individual differences in youth depression (Eley & Stevenson, 2000). Household characteristics other than those we considered might matter as well, especially for twins who have already moved out of their parental home. All of these factors are unique environmental characteristics with a conceivably large influence on individual adaptation to pandemic uncertainties. Changes in these factors could explain the small proportion of unique environmental variance in pandemic DS attributable to (i.e., shared with) former pandemic DS measurements.
Strengths, Limitations, and Future Outlook
The main strength of our study lies in the use of genetically informative, longitudinal data of three cohorts across five measurement points, with trait assessments from up to a 6-year time frame as predictors for DS across 18 months of the COVID-19 pandemic. The DS instruments differed between prepandemic and pandemic surveys. This diagnostic discrepancy might have contributed to the differences in explanatory power. Specifically, due to the double- and triple-barreled item wordings (i.e., little interest or pleasure in your activities vs. I find it difficult to enjoy anything and dejection, melancholy or hopelessness vs. I am pessimistic about my future), the PHQ-2 measurements might have suffered from a higher measurement error or they might have tapped into additional affective aspects not covered by and/or covering the item content of the BDI-FS. The discrepancy also restricted our modeling options, not allowing for more theoretically informative model tests (e.g., a dual change score model; Gillespie et al., 2015). Nonetheless, both DS measures are valid instruments (Beck et al., 2000; Richardson et al., 2010) and yield highly correlated measurements.
The first pandemic wave was assessed in retrospect. This could have led to a decrease in recall accuracy, which could have affected our results, for example, in the form of an underestimation of the amount of change in DS between the first and second pandemic surveys (Hipp et al. 2020). However, the specifically anchored time frame in the instructions (during the initial period of the Corona pandemic, i.e., from March 2020 until the first relaxations) may have mitigated the recall bias (Hipp et al., 2020).
Although we considered many theoretically and empirically established DS predictors, other sociocultural aspects might play a role as well, such as friendship networks, relationship status, family closeness, and the subjective quality of these resources. Although several of these are included in the TwinLife data, we did not consider them in this pre-registered, hypothesis-driven study, as we included individual and family characteristics that were (a) theoretically underpinned, (b) supported by previous research (on the pandemic situation), and (c) measured with validated or adapted instruments (except for the life event list). Similarly, we did not consider self-reports on experienced pandemic stress, as they were measured with newly developed items currently awaiting validation. This might partially account for the found low explanatory power of our models, since subjective pandemic burden could be part of the unshared environmental variance. However, prior research suggests that most young people feel burdened by the pandemic (Ravens-Sieberer et al., 2022). Future research should investigate these aspects.
Adolescents and young adults, on average, still see greater and more frequent changes in their personality characteristics, especially compared to middle and later adulthood (Bleidorn et al., 2022). This could lead to prepandemic traits showing lower explanatory power toward pandemic DS due to more recent trait-level changes that we could not account for. The ongoing TwinLife study will allow researchers to account for longer time periods with future waves, spanning into or across the COVID-19 pandemic.
Finally, the assumption of an absence of assortative mating and gene–environment interplay, necessary for the used twin analyses, is very strict and rarely holds true. This might have led to an underestimation of genetic effects and could not provide insight into how, for example, the genetically influenced differences in life satisfaction lead to an active (re-)shaping of certain environmental settings during the pandemic (e.g., a “make the best of it” mentality resulting in doing fun home activities with one's partner). Future research could employ models that inform about genotype–environment correlation (e.g., Dolan et al., 2014) and interaction (e.g., Purcell, 2002).
Our findings call for further research into the matter of youth depression symptoms during the COVID-19 pandemic and what might be particular about them. This appears all the more pressing given the possibility of future pandemics or similar global disruptions that youths might have to face. Adolescents and young adults are, after all, living through pivotal developmental stages of their lives. Future studies could focus on the suggested characteristics and employ models with more measurement points. Researchers could also shed light on the found small mean-level change of youth DS across the pandemic waves. In the future, the data from the ongoing TwinLife panel will allow analyses of pandemic DS with PHQ-2 measurements over up to five time points, depending on how long the pandemic will continue. Given the found differences between prepandemic and pandemic DS in this study, it appears valuable for researchers to identify pandemic-specific DS predictors, which might not be as important for DS outside of pandemic contexts.
Electronic Supplementary Material
The electronic supplementary material is available with the online version of the article at https://doi.org/10.1027/2151-2604/a000521
The authors wish to thank all participants for their valuable contributions to this study and the TwinLife research team for the data collection and management.
References
(1997). Self-efficacy: The exercise of control. W. H. Freeman.
(2000). Manual for the BDI-Fast Screen for medical patients. The Psychological Corporation.
(2017). Lazarus and Folkman’s psychological stress and coping theory. In C. L. CooperJ. C. Quick (Eds.), The handbook of stress and health: A guide to research and practice (pp. 351–364). Wiley. 10.1002/9781118993811.ch21
(2022). Personality stability and change: A meta-analysis of longitudinal studies. Psychological Bulletin, 148(7-8), 588–619. 10.1037/bul0000365
(1994). Nature-nurture reconceptualized in developmental perspective: A bioecological model. Psychological Review, 101(4), 568–586. 10.1037/0033-295x.101.4.568
(2021). Understanding the effects of the COVID-19 pandemic on youth psychopathology: Genotype-environment interplay. Biological Psychiatry Global Open Science, 1(4), 345–353. 10.1016/j.bpsgos.2021.07.004
(2007). Sensitivity of goodness of fit indexes to lack of measurement invariance. Structural Equation Modeling, 14(3), 464–504. 10.1080/10705510701301834
(2002). Evaluating goodness-of fit indexes for testing measurement invariance. Structural Equation Modeling, 9(2), 233–255. 10.1207/s15328007sem0902_5
(2017). Operationalising resilience in longitudinal studies: A systematic review of methodological approaches. Journal of Epidemiology and Community Health, 71(1), 98–104. 10.1136/jech-2015-206980
(2020). Longitudinal changes in mental health and the COVID-19 pandemic: Evidence from the UK Household Longitudinal Study. Psychological Medicine, 52(13), 2549-2558. 10.1017/s0033291720004432
(2023). Supplementary materials to “Youth depression symptoms during COVID-19: A longitudinal twin study on resilience factors”. https://osf.io/gu4yk
(2022). TwinLife. GESIS Data Archive, Cologne. (ZA6701 Data File Version 6.0.0).
(2014). GE covariance through phenotype to environment transmission: An assessment in longitudinal twin data and application to childhood anxiety. Behavior Genetics, 44(3), 240–253. 10.1007/s10519-014-9659-5
(2014). Augmented backward elimination: A pragmatic and purposeful way to develop statistical models. PLoS ONE, 9(11),
Article e113677 . 10.1371/journal.pone.0113677(2000). Specific life events and chronic experiences differentially associated with depression and anxiety in young twins. Journal of Abnormal Child Psychology, 28(4), 383–394. 10.1023/a:1005173127117
(2017). The sense of coherence in the salutogenic model of health. In M. B. MittelmarkS. SagyM. ErikssonG. F. BauerJ. M. PelikanB. LindströmG. A. Espnes (Eds.), The handbook of salutogenesis (pp. 91–96). Springer Nature. 10.1007/978-3-319-04600-6_11
(2015). Testing models for the contributions of genes and environment to developmental change in adolescent depression. Behavior Genetics, 45(4), 382–393. 10.1007/s10519-015-9715-9
(2016). What drives the development of social inequality over the life course? The German TwinLife Study. Twin Research and Human Genetics, 19(6), 659–672. 10.1017/thg.2016.76
(2018). Resilience and well-being. In E. DienerS. OishiL. Tay (Eds.), Handbook of well-being. DEF Publishers. https://nobascholar.com/chapters/4
(2020). Problems and pitfalls of retrospective survey questions in COVID-19 studies. Survey Research Methods, 14(2), 109–114. 10.18148/srm/2020.v14i2.7741
(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6(1), 1–55. 10.1080/10705519909540118
(2021). Mass quarantine and mental health during COVID-19: A meta-analysis. Journal of Affective Disorders, 295, 1335–1346. 10.1016/j.jad.2021.08.067
(2016). Comparisons of three empirical methods for partial factorial invariance: Forward, backward, and factor-ratio tests. Structural Equation Modeling: A Multidisciplinary Journal, 23(4), 567–584. 10.1080/10705511.2015.1138092
(2022). Young adult adaptability to the social challenges of the COVID-19 pandemic: The protective role of friendships. Journal of Youth and Adolescence, 51(3), 585–597. 10.1007/s10964-022-01573-w
(2017). Theoretical perspectives on the interplay of nature and nurture in personality development. In J. Specht (Ed.), Personality development Across the lifespan (pp. 101–115). Academic Press. 10.1016/b978-0-12-804674-6.00008-9
(2021). The genetic basis of major depression. Psychological Medicine, 51(13), 2217–2230. 10.1017/s0033291721000441
(2018). Links between depression and openness and its facets, Personality and Mental Health, 12(3), 203–215. 10.1002/pmh.1417
(2022). TwinLife Data Documentation: ShortGuide v1.2.0. University of Bielefeld; University of Bremen; Saarland University. https://access.gesis.org/dbk/70534
(2021). Physical activity, loneliness, and meaning of friendship in young individuals – a mixed-methods investigation prior to and during the COVID-19 pandemic with three cross-sectional studies. Frontiers in Psychology, 12,
Article 617267 . 10.3389/fpsyg.2021.617267(2005). Detecting and monitoring depression with a two-item questionnaire (PHQ-2). Journal of Psychosomatic Research, 58(2), 163–171. 10.1016/j.jpsychores.2004.09.006
(2021). Associations between Facets and Aspects of Big Five Personality and Affective Disorders: A Systematic Review and Best Evidence Synthesis. Journal of Affective Disorders, 288, 175–188. 10.1016/j.jad.2021.03.061
(2021). Resilience in development and psychopathology: Multisystem perspectives. Annual Review of Clinical Psychology, 17, 521–549. 10.1146/annurev-clinpsy-081219-120307
(2013). Risk and resilience in development. In P. D. Zelazo (Ed.), The Oxford handbook of developmental psychology, Vol. 2: Self and other (pp. 579–608). Oxford Academic. 10.1093/oxfordhb/9780199958474.013.0023
(1984). Adjustment of twin data for the effects of age and sex. Behavior Genetics, 14(4), 325–343. 10.1007/bf01080045
(1988). Stress inoculation training. The Counseling Psychologist, 16(1), 69–90. 10.1177/0011000088161005
(2021). How personality affects reaction. A mental health behavioral insight review during the Pandemic. Current Psychology. 10.1007/s12144-021-02425-9
(2022). Longitudinal predictors of Coronavirus-related PTSD among young adults from Poland, Germany, Slovenia, and Israel. International Journal of Environmental Research and Public Health, 19(12),
Article 7207 . 10.3390/ijerph19127207(2021). The psychological impact of COVID-19 pandemic lockdowns: A review and meta-analysis of longitudinal studies and natural experiments. Psychological Medicine, 51(2), 201–211. 10.1017/S0033291721000015
(2020). Risk and resilience in family well-being during the COVID-19 pandemic. American Psychologist, 75(5), 631–643. 10.1037/amp0000660
(2002). Variance components models for gene–environment interaction in twin analysis. Twin Research and Human Genetics, 5(6), 554–571. 10.1375/136905202762342026
(2022). Impact of the COVID-19 pandemic on quality of life and mental health in children and adolescents in Germany. European Child & Adolescent Psychiatry, 31, 879–889. 10.1007/s00787-021-01726-5
(2010). Evaluation of the PHQ-2 as a brief screen for detecting major depression among adolescents. Pediatrics, 125(5), e1097–e1103. 10.1542/peds.2009-2712
(2021). Genetic correlates of psychological responses to the COVID-19 crisis in young adult twins in Great Britain. Behavior Genetics, 51(2), 110–124. 10.1007/s10519-021-10050-2
(2022). A systematic review and meta-analysis of longitudinal cohort studies comparing mental health before versus during the COVID-19 pandemic in 2020. Journal of Affective Disorders, 296, 567–576. 10.1016/j.jad.2021.09.098
(2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist, 55(1), 68–78. 10.1037/0003-066x.55.1.68
(2021). Review: Mental health impacts of the COVID-19 pandemic on children and youth—A systematic review. Child and Adolescent Mental Health, 27(2), 173–189. 10.1111/camh.12501
(2020). Emotional distress in young adults during the COVID-19 pandemic: Evidence of risk and resilience from a longitudinal cohort study. Psychological Medicine, 52(5), 824–833. 10.1017/s003329172000241x
(2017). Biometric modeling of gene-environment interplay: The intersection of theory and method and applications for social inequality. Journal of Personality, 85(1), 22–37. 10.1111/jopy.12231
(1934). The meaning of ‘i’ in the estimate of ‘g’. British Journal of Psychology, 25(1), 92–99. 10.1111/j.2044-8295.1934.tb00728.x
(1935). The vectors of mind: Multiple-factor analysis for the isolation of primary traits. University of Chicago Press. 10.1037/10018-000
(2022). Internalizing symptoms and family functioning predict adolescent depressive symptoms during COVID-19: A longitudinal study in a community sample. PLoS ONE, 17(3),
Article e0264962 . 10.1371/journal.pone.0264962(2022). Genetic and environmental influences on quality of life: The COVID-19 pandemic as a natural experiment. Genes, Brain and Behavior, 21(8),
Article e12796 . 10.1111/gbb.12796(2019). A theoretical review of psychological resilience: Defining resilience and resilience research over the decades. Archives of Medicine and Health Sciences, 7(2), 233–239. 10.4103/amhs.amhs_119_19
(2004). AIC model selection using Akaike weights. Psychonomic Bulletin & Review, 11(1), 192–196. 10.3758/bf03206482
(2015). Resilience theory and the practice of positive psychology from individuals to societies. In S. Joseph (Ed.), Positive psychology in practice (pp. 773–788). Wiley. 10.1002/9781118996874.ch44
(2000). Internalizing problems of childhood and adolescence: Prospects, pitfalls, and progress in understanding the development of anxiety and depression. Development and Psychopathology, 12(3), 443–466. 10.1017/S0954579400003102
(2013). Resiliency theory: A strengths-based approach to research and practice for adolescent health. Health Education & Behavior, 40(4), 381–383. 10.1177/1090198113493782
(2012). Resilience in children and youth: A review. Children and Youth Services Review, 34(12), 2295–2303. 10.1016/j.childyouth.2012.08.009
