Psycho-social factors associated with mental resilience in the Corona lockdown

The SARS-CoV-2 pandemic is not only a threat to physical health but is also having severe impacts on mental health. Although increases in stress-related symptomatology and other adverse psycho-social outcomes, as well as their most important risk factors have been described, hardly anything is known about potential protective factors. Resilience refers to the maintenance of mental health despite adversity. To gain mechanistic insights about the relationship between described psycho-social resilience factors and resilience specifically in the current crisis, we assessed resilience factors, exposure to Corona crisis-specific and general stressors, as well as internalizing symptoms in a cross-sectional online survey conducted in 24 languages during the most intense phase of the lockdown in Europe (22 March to 19 April) in a convenience sample of N = 15,970 adults. Resilience, as an outcome, was conceptualized as good mental health despite stressor exposure and measured as the inverse residual between actual and predicted symptom total score. Preregistered hypotheses (osf.io/r6btn) were tested with multiple regression models and mediation analyses. Results confirmed our primary hypothesis that positive appraisal style (PAS) is positively associated with resilience (p < 0.0001). The resilience factor PAS also partly mediated the positive association between perceived social support and resilience, and its association with resilience was in turn partly mediated by the ability to easily recover from stress (both p < 0.0001). In comparison with other resilience factors, good stress response recovery and positive appraisal specifically of the consequences of the Corona crisis were the strongest factors. Preregistered exploratory subgroup analyses (osf.io/thka9) showed that all tested resilience factors generalize across major socio-demographic categories. This research identifies modifiable protective factors that can be targeted by public mental health efforts in this and in future pandemics.

• Table S2: Questionnaire elements: psychological constructs and used instruments.
• Table S4: Past or present diagnosed mental health conditions. • Table S5: A priori hypotheses.

Survey questionnaire
"DynaCORE-C -the DynaMORE cross-sectional survey study on psychological resilience to the mental health consequences of the Corona crisis" was conducted by the EU project DynaMORE (Dynamic MOdelling of REsilience), see www.dynamore-project.eu. Study participation was possible at bit.ly/DynaCORE-C or www.dynacore.info. At the time when data were pulled for the second interim analysis reported in this paper (April 20 th 2020), available languages were Arabic, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hebrew, Hungarian, Italian, Lithuanian, Norwegian, Polish, Portuguese, Serbian, Slovak, Spanish, Swedish, and Ukrainian. The 127-item survey questionnaire contains a combination of self-generated questions and questions from existing instruments (see Table S2). In some instances, instruments were shortened or instructions or questions were adapted to better fit the study context. The quality of translations from the original English version into other languages was assured by using back-translation by independent translators. Where available, local published validated versions of the instruments included in the questionnaire were used. The questionnaires are privately stored in osf.io/5xq9p/ and are made available upon request. Before the final form of the questionnaire was reached, near-identical prefinal versions were already employed in which answer options to the household income question were less fine-grained (the lowest range "0 to 24.999 €" was overly broad: first 355 complete valid data sets, of whom 346 from European residents) and in which previous or current psychiatric disorder was not enquired (first n=406 complete valid data sets, or n=512 complete European data sets that were partially invalid due to the prefinal income range options).

General note on resilience
We here base our efforts to gain insight into protective factors in the Corona crisis on a resilience framework (Kalisch et al., 2017). The science of resilience is based on the welldocumented observation that some individuals maintain mental health despite exposure to severe psychological or physical adversity (Bonanno and Mancini, 2011;Kalisch et al., 2015).
Resilience research aims to understand why some people do not (or only temporarily) develop stress-related mental dysfunction, despite being subject to the same kind of challenges that cause long-term dysfunction in others. This approach is naturally linked to the question of how to prevent stress-related mental health conditions, rather than attempting to treat them at a later stage when significant individual suffering and societal and economic costs have already occurred (Sapienza and Masten, 2011). Hence, resilience research is health-and preventionfocused and, thereby, an alternative approach to classical disease-focused research in the battle against stress-related disorders (Kalisch et al., 2015). Prevention is one of the main under-researched and under-exploited strategies in current efforts to promote mental health (Jorm et al., 2017) and has been highlighted as one key area of research and public health measures in the European roadmap for mental health (http://www.roamermh.org/files/DocRoamer_Roadmap2015_FINALISSIMA_050615.pdf).
There is growing consensus that psychological (individual) resilience is best defined as the maintenance or quick recovery of mental health during and after times of adversity, which may consist in a potentially traumatizing event, challenging life circumstances, a critical life transition phase, or physical illness (Kalisch et al., 2017). Hence, resilience is conceptualized as an outcome (maintained mental health). Resilient outcomes are partly determined by pre-existing factors or predispositions, such as (trait-like as well as more modifiable and time-variant stylelike) characteristics of the individual, individual resources, or features of the individual's environment. These "resilience factors" are incompletely understood, and it is unclear if factors that are protective against one type of adversity are also protective against other types of adversity. It is also unclear if resilience factors that are protective in one population (culture, nation, ethnicity, gender, social class, profession etc.) are also protective in other populations. Finally, different resilience factors may protect differentially against different types of mental health problems (anxiety, depression, compulsive symptoms, …) and such types of problems may occur at different rates in different populations during the Corona crisis (Kalisch et al., 2015(Kalisch et al., , 2017. This makes it highly relevant to rapidly collect reliable information on effective resilience factors in a maximum possible number of subjects and populations world-wide in the current global crisis.

Hypothesized resilience factors (independent variables)
Despite uncertainties about effective resilience factors in the Corona pandemic, existing knowledge about resilience factors from a large variety of different types of individual-level and society-level adversities and crises can be used to formulate testable hypotheses and thereby reduce the search space.

Described resilience factors
We hypothesized that the described resilience factors perceived social support, optimism, and perceived general self-efficacy (Bonanno et al., 2015) will be positively associated with outcome-based resilience (in short, 'resilience' in the further). We also hypothesized that a perceived increase in social support during the crisis will be positively associated with resilience. We further hypothesized that perceived good stress recovery, as assessed by a questionnaire that asks subjects about how quickly and easily they recover from stress responses (Smith et al., 2008), will be positively associated with resilience. See Table S5 Kampa et al. (2018Kampa et al. ( , 2020 and Kalisch et al. (2020).
As briefly outlined in the main text, PASTOR claims that the common final pathway to maintained mental health in the face of adversity (i.e., resilience) lies in the positive appraisal of potential stressors (threats to one's goals and needs). 'Positive appraisal' sets the values that an individual attributes to a stressor on the key threat appraisal dimensions threat magnitude or cost, threat probability, and coping potential to levels that realistically reflect the threat or even slightly underestimate it. That is, positive appraisal avoids catastrophizing (magnitude/cost dimension), pessimism (probability dimension) and unnecessarily low selfefficacy or control perceptions (coping dimension). At the same time, positive appraisal avoids unrealistically positive (delusional) threat perceptions that might lead to trivialization or blind optimism. Such mildly positive appraisal permits the organism to fine-tune stress responses to optimal levels, producing stress reactions when necessary but avoiding unnecessary stress, inefficient deployment of resources and concomitant deleterious allostatic load effects. PASTOR assumes that an individual usually appraises similar situations in a similar fashion, and therefore can be characterized by his or her typical appraisal tendencies ('appraisal style'). A negative appraisal style (NAS) consists in a propensity to overestimate the aversive consequences and the probability of challenging situations and to underestimate one's coping potential. An NAS results in consistent over-reactions to threats and increases the likelihood of an individual developing stress-related mental problems during and after adversity. A positive appraisal style (PAS), by contrast, is defined as the absence of such negative biases, but also by the absence of delusional positive appraisal tendencies. A PAS typically results in realistic threat estimations or mild underestimations of threat and is claimed to be protective for mental health. Hence, PAS includes the constructs of optimism and general self-efficacy, but is broader (Kalisch et al., 2015).
Importantly, a PAS is believed to be determined by the efficacy and efficiency of the neural and cognitive processes that underlie positive stressor appraisal (that produce appraisals when subjects are challenged). This opens two different roads towards assessing appraisal style with self-report instruments: one can assess the typical appraisal contents that an individual produces in response to stressors, and one can assess the typical thinking processes, or mental operations, that individuals perform when they are challenged. An example for the former would be that a subject reports to see a positive aspect in a threatening situation or to find that there are worse things in life. An example of the latter would be that a subject reports to usually try to think of positive aspects of a situation or to try to tell oneself that there are worse things in life. Everyday language often does not clearly differentiate between appraisal contents and processes, and also pre-formulated questions in existing questionnaire instruments are often a mix of more content-and more process-focused propositions.
In MARP and LORA, we employ two established instruments developed to capture various behavioral and cognitive coping and emotion regulation strategies and that include, among others, questions assessing various appraisal contents and processes, often in a mixed fashion: the brief COPE (Carver, 1997) and the CERQ-short (Garnefski and Kraaij, 2006). The brief COPE has 14, the CERQ-short has nine two-item subscales. In addition, we amended the CERQ-short with two own-formulated questions on a distanced (detached) stressor appraisal.
In the baseline (T0) data from both samples, we conducted a factor analysis on the subscalelevel including both questionnaires and (in MARP) the additional distancing subscale. This reliably identified only three factors spanning the questionnaires. Several subscales loaded predominantly on one of the three factors, such that they could be qualified as factorcharacteristic subscales. Of these, the subscales distancing (dis; own-formulated), positive reappraisal (pra; CERQ), acceptance (acc; CERQ, less COPE), putting into perspective (per; CERQ), positive reframing (posref; COPE), andto a lesser extent -refocus on planning (rfp; CERQ), positive refocusing (prf; CERQ) and humor (hum; COPE) strongly positively loaded on the first factor. Another factor was dominated by less cognitive, more behavioral coping subscales, namely instrumental support seeking (iss), emotional support seeking (ess), venting of emotions (vent), and acting out (act), all from the COPE. Both factors were positively predictive of resilience (outcome-based, for details see below), though the first more so. Detailed results from this factor analysis and the MARP and LORA studies will be published elsewhere.
We interpret the first, cognitive factor as reflecting positive appraisal contents and processes and therefore consider it an appropriate self-report measure of PAS. For the current project, we dropped the posref and acc subscales from the COPE, given that they conceptually overlap highly with pra and acc from the CERQ but loaded less on the first factor. This left the seven subscales dis, pra, acc, per, rfp, prf, and hum, which we henceforth refer to as PASS, for 'positive appraisal style scale'. We confirmed that a sum score from these subscales also positively predict resilience in MARP and LORA. We hypothesize that PASS will be positively associated with resilient outcome in the current project.
Because of a (lesser) positive association of the behavioral coping factor with resilience in MARP and LORA, we also include the eight questions for iss, ess, vent and act, which together we here term BCS, for 'behavioral coping scale'. Behavioral coping may also be helpful in the Corona crisis, and we therefore also hypothesize that BCS will be positively associated with resilience.
Another, yet unpublished result from MARP is that personality instruments including the PANAS (Watson et al., 1988) and the BFI-10 ( Rammstedt and John, 2007) could be summarized factoranalytically in one common factor that reflects a positive vs. a negative trait affect (Jeronimus et al., 2016;Schenk et al., 2018). The factor was positively loaded by positive affect items from the PANAS and highly negatively by the neuroticism subscale of the BFI-10 and also predicted resilience. For brevity, we here selected two neuroticism items and hypothesize that their sum score will be negatively associated with resilience. Because in MARP and LORA, PASS and neuroticism are negatively correlated, they may reflect overlapping constructs that could be summarized as a general positive emotional style.
Finally, we hypothesize that the positive appraisal specifically of the current pandemic will be positively associated with resilience. To test this, two custom-made questions were generated.
('I expect that I will learn something positive from of the Corona pandemic for my own life.' and 'In the long run, I think that society will change for the better because of the Corona pandemic.') The survey questionnaire also contained room for open answers, for hypothesis generation about resilience factors. Results from this analysis will be reported elsewhere.

Mediation effects
Based on Kalisch et al. (2015) (PASTOR), we hypothesize that the expected positive effect of social support on resilience is positively mediated by its effect on PAS. PASTOR further claims that positive appraisal permits the organism to fine-tune stress responses to optimal levels, thus avoiding unnecessary stress. This includes having stress responses that are not higher in magnitude and especially not longer in duration than necessary. It is through this pathway that positive appraisal eventually results in maintained mental health despite stressor exposure (i.e., resilience). This response pattern of usually only having quickly recovering stress responses is assessed by the Brief Resilience Scale (BRS; Smith et al., 2008). It can therefore be hypothesized that the effect of PAS on resilience is positively mediated by its effect on selfperceived good stress recovery.

Resilience outcome measures (dependent variables)
Our outcome-based definition of resilience as maintenance or quick recovery of mental health during and after adversity (Kalisch et al., 2017; see 2.1) implies that adversity is necessarily part of the equation. Only registering mental health outcomes without taking into account the adversity a subject was or is exposed to may be informative about mental health, but is not informative about resilience, which in its essence is mental health despite adversity (Mancini and Bonanno, 2009;Kalisch et al., 2017). Ideally, this is achieved by measuring mental health problems (P) before (T0) and after (T1) some time of stressor exposure (E) and normalizing changes in mental health between T0 and T1 by the stressor exposure that has occurred in between (Kalisch et al., 2015(Kalisch et al., , 2017. In such a scenario, of two individuals having experienced comparable stressor exposure E, the individual with fewer increases in mental health problems P would be the more resilient individual. Another scenario would be two individuals showing similar increase in P with different E. Here, the individual with higher E would be the more resilient one. A detailed description of how we measure mental health and stressors in the longitudinal studies MARP and LORA and use those data to calculate resilience outcomes is available in Kalisch et al. (2020).

Mental health measure
For practical reasons, the current study uses a cross-sectional design. We can thus not perform repeated mental health assessments. We therefore make use of a specific feature of our mental health instrument, the short 12-item version of the General Health Questionnaire (GHQ-12; Goldberg et al., 1997), which is designed to ask subjects about how they felt in the past two weeks relative to how they normally feel. Thus, the GHQ assesses a perceived change in mental health, which we will use as an approximation of a measured difference in mental health between two time points. See Table S3 for scoring of the GHQ, to derive the mental health problem score P.

Stressor exposure measures
For stressor exposure, we employ two instruments as part of the survey questionnaire. The first instrument is a list of eleven categories of daily hassles (DHs) and life events (LEs), which we condensed specifically for this study from the much more extensive DH (58 items) and LE (27 items) lists employed in MARP and LORA (general stressor exposure instrument, for calculation of stressor exposure score EG). We were mainly guided by the frequency of DHs reported at study inclusion by the LORA participants, which is the larger and more age-representative sample than MARP (N=1200 at inclusion, age range at inclusion 18-50 years). The most frequently reported DHs in LORA were household chores, commuting, bad weather, interruption in activities, performance pressure, negative events in the media, time pressure, and waiting time or delay. The second instrument employed in the current study to assess stressors is a list of 29 stressors that we collated to reflect situations of adversity related to the Corona pandemic (Corona pandemic-specific stressor exposure, for calculation of stressor exposure score ES). In both instruments, subjects are requested to indicate whether they experience a given stressor currently or have experienced it during the last two weeks. If they report to have experienced the stressor, they are further asked to report how burdensome it was to them (severity rating). (In addition, for both EG and ES, respondents have the possibility to add and rate one stressor that was not mentioned in our predefined lists.) In this way, we obtain an overview of recent stressor exposure E roughly in the time period across which subjects also report changes in P, using the GHQ. We can then meaningfully relate both variables to capture stressor-related changes in mental health.
To derive E from participants' responses, we here use two alternative scoring approaches. In the first approach, we simply count the reported listed stressors (stressor count method); in the second approach, we weigh each reported stressor (including 'other') by its severity rating and form a weighted sum (stressor severity method). The advantage of the stressor count method is that it restricts E to reflect the mere occurrence of stressors, not being confounded by individual differences in the way stressors are perceived (appraised). We use this scoring method in MARP and LORA, where we have long stressor lists (85 items) and can thus assume that the mere frequency of stressors is a good reflection of objective stressor exposure E. Factual differences in the severity of a given stressor can be ignored, given the sheer number of reportable stressors (Chmitorz et al., 2020). This assumption may hold only partly for the current study, which uses a shorter list of only 40 stressors. The advantage of the stressor severity method is that factually (objectively) strong individual stressors are appropriately taken into account. For instance, own serious health problems due to a COVID-19 infection or a job loss that threatens material survival may be objectively highly burdensome and will be taken into account as such with the severity method. The downside is that subjective over-reactions (i.e., overly negative appraisal) may inflate the E measure. This is a relevant potential confound, because relating a change in P to an inflated E measure may lead to an overestimation of a subject's resilience.
Another question regarding E scoring is whether general and Corona pandemic-specific stressors should be combined, to derive a combined stressor exposure score EC and, consequentially, to determine subjects' resilience to all stressors; or whether EG and ES should be treated separately. In our preregistered analysis plan, we leave this question as well as the question of the optimal scoring method open and plan to answer them based on data, using below decision rule (see 2.3.3).
To summarize, our data allow us to calculate six different stressor exposure scores: • a general stressor exposure score based on the stressor count method (EG,SCM), • a general stressor exposure score based on the stressor severity method (EG,SSM), • a Corona pandemic-specific stressor exposure score based on the stressor count method (ES,SCM), • a Corona pandemic-specific stressor exposure score based on the stressor severity method (ES,SSM), • a combined stressor exposure score based on the stressor count method (EC,SCM), • a combined stressor exposure score based on the stressor severity method (EC,SSM).

Relationships between mental health and stressor exposure measures
In MARP and LORA, we observe predominantly monotone E-P relationships in the range of R=0.3 to 0.4. This applies to both scoring methods and to combined as well as DH-and LEspecific E scores. The strongest E-P relationship is consistently observed in both samples for combined stressor exposure EC, calculated based on the theoretically preferred stressor count method (see 2.3.2), which is why we use this score as a basis for calculating resilience outcomes in MARP and LORA . We expected similar E-P relationships in the current study. As in MARP and LORA, we therefore explored which E score explains most variance in P and based our main resilience measure on this score. This applied to EC,SSM (EC,SSM: R 2 =0.21; EG,SSM: R 2 =0.17; ES,SSM: R 2 =0.18; comp. EC,SCM: R 2 =0.07; EG,SCM: R 2 =0.08; ES,SCM: R 2 =0.04). We had defined in the preregistered analysis plan (osf.io/r6btn): "If the chosen E score happens to combine general and Corona pandemic-specific stressors, we will subsequently compute separate analyses for both stressor categories, to also learn about potential differential effectiveness of resilience factors for each category. We will use the same scoring method (either count or severity) for these secondary scores as for the chosen primary score." This strategy is realized in the main text (EC,SSM, followed by EG,SSM and ES,SSM). We further defined: "If the E-P relationship turns out not to be predominantly linear, the decision will be made based on the predominant form of relationship (e.g., quadratic)." We observed that a model with linear and quadratic effects had a much better model fit (RESC: F=556.68; RESG: F=637.82; RESS: 353.78; all p<0.001) than a model with a linear effect only (comp. van Harmelen et al., 2017) and therefore used the former to calculate individual RES scores.

Calculation of resilience outcome measures based on mental health and stressor exposure measures
Based on the expected observation of a robust linear E-P relationship and on above decision on the chosen E score (EC,SSM), the distance of an individual's P score to the E-P regression line can be considered informative about the reactivity of his/her mental health to his/her stressor exposure. The regression line is viewed as the normative reactivity of mental health to stressor exposure (in short: 'stressor reactivity', SR) in the whole sample. A subject's residual onto the regression line expresses to what extent the subject deviates from that normal E-P relationship. Individuals with positive residual values show "too many" mental health problems, given their level of stressor exposure; individuals with negative values show "too few" mental health problems, given their stressor exposure (ignoring random variability for the moment). In other words: a positive residual reflects over-reactivity of mental health to stressor exposure (high stressor reactivity, high SR); a negative residual reflects under-reactivity (low SR) .
The advantage of using residualization is that the method inherently corrects for individual differences in the level of stressor exposure. Two subjects 1 and 2 that have developed comparable mental health problems P over the last two weeks may still diverge in their SR score, for instance, if subject 1 had higher exposure E than subject 2. Subject 1 would then obtain a lower SR score and could be classified as more resilient than subject 2, in accordance with our definition of resilience as an outcome (Kalisch et al., 2015(Kalisch et al., , 2017, that is, as mental health maintenance despite stressor exposure. Two subjects 3 and 4 with comparable stressor exposure E would also obtain different SR scores, for instance, when 3 has developed fewer problems P than subject 4. Subject 3 would then obtain a lower SR score and could be regarded more resilient. On this basis, we defined the inverse of SR as our outcome-based resilience measure RES . Residualization approaches have been introduced into the resilience literature by Amstadter et al. (2014) and van Harmelen et al. (2017).
A potential criticism of the SR score might be that they are also calculated for individuals with very low stressor exposure E, and that resilience in the absence of significant adversity is not a meaningful concept (Mancini and Bonanno, 2009). We therefore adopted as a general rule for the analyses of SR data that a) primary analyses based on the entire sample be complemented by secondary analyses of those two thirds of the subjects with the highest overall stressor exposure E, and that b) the results of those secondary analyses should go in the same direction as those of the primary analyses, in order for the primary analysis results to be considered valid . All results of these secondary analyses went into the same direction as those of the primary analyses (not shown).

Variable/index Used questionnaire element(s) Mental health problems (P)
Mental health (sum score, item scoring 0 to 3) Perceived social support (PSS) Perceived social support (sum score, item scoring 1 to 5) Perceived change in social support during the Corona crisis (CSS) Perceived change in social support during the Corona crisis (item scoring 1 to 5)

Optimism (OPT)
Optimism (item scoring 1 to 7) Perceived general self-efficacy (GSE) Perceived general self-efficacy (sum score, item scoring 1-5) Perceived good stress recovery (REC) Perceived good stress recovery (mean score, item scoring 1 to 5) Neuroticism (NEU) Neuroticism (sum score, item scoring -2 to 2) Positive appraisal style (PAS) Positive appraisal style (composite score, taking the average of the z-normalized scores of the COPE items (item scoring 1 to 4), the CERQ items (item scoring 1 to 5), and the self-generated items (item scoring 1 to 5)).

Incomplete data
After data cleaning (see main text), there were n=2905 valid but incomplete data sets. Of these, 2849 did not contain complete answers to the questions on stressor exposure, which were placed at the end of the questionnaire. As this precludes the calculation of the outcome variable resilience (mental health normalized to stressor exposure, see 2.3.4), the incomplete data sets were as a whole not included in the analyses. Qualitative comparison with the participants providing complete data (Table S1) indicated that the participants providing incomplete data were more likely male and younger, were more likely not to report a mental health diagnosis, to live in a household with 3-4 people and with underage people, to have fewer years of education and a lower average household income and to undergo education, while being less likely to work in education or research. They were also more likely to be Polish residents or nationals and less likely to be German residents or nationals. We suspect that younger participants may have been less patient with filling in the 127-item questionnaire.