Worries about the COVID-19 pandemic and the dynamic regulation of emotions in the general population: A network analysis study


                  Background
                  The impact of the COVID-19 pandemic on mental health has been widely reported. Yet, little remains known about the psychological mechanisms associated with changes in mental well-being during the currently ongoing pandemic.
               
                  Methods
                  Here, we use a network analysis to unravel complex relationships between COVID-19 related stressors and emotional states during the initial phase of the COVID-19 (April 2020). Adults living in the Netherlands and Belgium (N = 1145, age 16 and older) (repeatedly) completed an online survey (approximate survey completion rate = 66.2%) about COVID-19 (over a 5-day maximum sampling period).
               
                  Results
                  Partial correlations and contemporaneous networks illustrated that worries about the impact of the COVID-19 pandemic were primarily associated with distress and mood ratings, which were subsequently associated with other indicators of well-being. Temporal network analysis revealed that COVID-19 worries were selectively associated with the reciprocal interplay between high distress and low positive mood (https://osf.io/vtdkr/).
               
                  Limitations
                  Short-term temporal intervals were evaluated. A small percentage of participants completed the survey repeatedly (35.63% of the total sample), yielding to a relatively small sample size for repeated measures online research. The sample was self-selected.
               
                  Conclusion
                  These results may point to potential mechanisms by which initial worries about the COVID-19 pandemic may have impacted psychological well-being.
               


Introduction
The coronavirus disease 2019 (COVID-19) pandemic, and its associated socioeconomic consequences, can be considered a stressor of unprecedented, global scale.Although still ongoing, experts have expressed concerns about the potential adverse effects of pandemicrelated stressors on well-being and mental health (Leach et al., 2021;Cénat et al., 2021).These worries stem from various aspects related to the COVID-19 pandemic.
For instance, the implementation of nationwide lockdowns and curfews, aimed at mitigating the spread of the virus, have profoundly disrupted social activities, thereby aggravating feelings of loneliness (Tull et al., 2020;O'Sullivan et al., 2021).Recent evidence suggests that social isolation and feelings of loneliness during the COVID-19 pandemic were associated with depressed mood, heightened anxiety, and poorer sleep quality (Hwang et al., 2020;Santini and Koyanagi, 2021;Meda et al., 2021;Grey et al., 2020): psychological changes that are commonly seen in affective and stress-related disorders.In addition, the unpredictable course of the pandemic has caused unbridled uncertainty (e.g., regarding the impact of the pandemic on a personal, professional, and societal level) (Koffman et al., 2020).Even before the occurrence of COVID-19, it was known that poor coping with uncertainty magnifies worries, anxiety, and avoidance behavior (Norr et al., 2013;Hunt et al., 2019).Such excessive uncertainty and worrying can put a considerable strain on mental health and well-being (Nitschke et al., 2021;Varga et al., 2021).
Past epidemics that share important similarities to the current COVID-19 pandemic in terms of mitigation measures and uncertainty, like Ebola, SARS, and H1N1 Influenza, have been associated with a spike in psychological distress, low mood, and emotional exhaustion (Brooks et al., 2020).Interestingly, studies in UK adults in the initial stage of the COVID-19 outbreak also found elevated levels of anxiety, traumatic stress, depression (Shevlin et al., 2020), and even suicidal ideation (O'Connor et al., 2021).Similarly, studies conducted in Italy, Spain, Germany, and China observed, among others, increased levels of distress and heightened affective symptoms (Losada-Baltar et al., 2021;Wang et al., 2020;Mazza et al., 2020;Rauschenberg et al., 2021).Put together, these results highlight how major health crises can be accompanied by increased distress and altered emotional states.
Given its recency, much remains unknown about the psychological impact of COVID-19 pandemic, for example whether such effects are more specific to emotional states, or changes in mental well-being more generally.Network models (Hoffart et al., 2021;S. Epskamp et al., 2018a) can provide important insights into complex relationships among COVID-19 related stressors (e.g., social isolation, worries about the virus, worries about loved ones) and more general indicators of mental well-being (e.g., low mood, distress, loneliness).A great advantage of network analysis is that it allows investigation of direct and indirect interactions among all variables of interest.Importantly, this approach can help identify key variables or clusters of variables that have a strong influence on other variables within a given network.As such, the network approach facilitates understanding of the complex interactions that underlie (changes in) psychological variables, which may provide clues on important variables/clusters that could serve as the target of, for example, interventions (S.Epskamp et al., 2018a).Some initial evidence from COVID-19 network studies in the general population suggests high interconnectivity between COVID-19 related stressors and symptoms of anxiety and depression (Zavlis et al., 2021;Hoffart et al., 2021), pointing to a potentially important role for altered emotional states during the COVID-19 pandemic.
Fewer studies, however, have assessed longitudinal networks.These studies are important because they reveal how changes in emotional states unfold over time.Among the few studies conducted, Fried et al. (2020) observed mixed emotional changes (e.g., increased depression ratings, lower anxiety ratings) over the course of a two-week measurement window early in the COVID-19 pandemic.On the other hand, Martin-Brufay et al. (2020) suggested that the impact of pandemic-related stressors on emotional states was dependent on adaptation strategies using also a two-week measurement window (i.e., negative expectations in the beginning of quarantine lead to better adaption, while positive expectations in the beginning of quarantine lead to poorer adaptation over time).Lastly, Zavlis et al. (2021) observed connectivity between pandemic-related anxiety, and symptoms of generalized anxiety disorder and depression over a one-month interval.These studies once more point to associations between pandemic-related stressors and changes in emotional states, yet much remains inconclusive about their temporal interactions (i.e., whether they predict each other over time).
In an attempt to contribute to and expand on these recent insights, we investigated the psychological state of adults in the Netherlands and Belgium in the earliest phase of COVID-19 pandemic, when social isolation and uncertainty were high.Utilizing cross-sectional and (a maximum of five) repeated measures of psychological variables associated more generally with mental well-being (e.g., low mood, distress, energy, loneliness) and pandemic-specific variables (i.e., worries about the virus, social distancing), in combination with network analyses, we sought to establish the temporal dynamics of emotional states during the initial phase of the COVID-19 pandemic.

Study outline
To investigate (temporal) associations between COVID-19 related items and psychological indicators of mental well-being during the initial phase of the COVID-19 pandemic, we conducted an online survey among individuals living in the Netherlands and Belgium.The study was active between 2020/03/31 and 2020/04/30, a timeframe during which both countries implemented stringent measures to contain the spread of COVID-19 (see Fig. 1 for an overview of daily infections and examples of regulations implemented by the Dutch government to curtail COVID-19 infections).
Participants were recruited via social media (e.g., Facebook, Twitter), university media (website, mailing list), and local news and media (newspaper, television).Dedicated efforts were made to ensure that older adults were also represented in the final sample, including advertisements and availability of tablets for questionnaire completion, and support from staff, in local elderly communities and nursing homes.The study was approved by the Faculty of Psychology and Neuroscience ethical review committee of Maastricht University (protocol number: 221 62 03 2020).

Survey procedure
The survey was hosted on Qualtrics (Qualtrics, Provo, UT) and was accessible to anyone with a digital device with an Internet connection.Participants accessed the survey via an anonymous link and were invited to complete a three-part survey consisting of 1) demographics and COVID-19 status, 2) ratings of psychological indicators of mental wellbeing (low mood, distress, loneliness, energy, motivation), and 3) ratings of worries about/preoccupation with COVID-19 and adherence to widely disseminated infection-mitigation guidelines that necessitate social distancing.
The order of block two and three, as well the item order within each block, was randomized between sessions.We used a structured diary approach (Bolger et al., 2003;Shiffman et al., 2008) to investigate within-person changes in ratings over time, as well as associations among items ratings.Prior to and following completion of the survey, all participants were reminded that the survey could be completed once per day for a maximum of five days.We opted for a short sampling period due to the high uncertainty at time of conducting the study (i.e., April 2020).At this point it was unclear for how long certain policies and restrictions were going to be implemented.Next, to boost compliance, participants were invited to set an alarm on their phone or computer for the next day, as a reminder to complete the next survey 12-24h later.This approach ensured that future prompt times were acceptable to participants, and has been associated with a questionnaire completion rate similar to that of random prompts (Burke et al., 2017).To complete the next report upon being prompted by the alarm, participants accessed the link, filled in a unique participant-generated code (generated during the first session), and completed block two and three again.Demographics were not collected during these follow-up measurements.
The average survey completion time in the entire sample with valid reports (see section, "3.1 Sample characteristics") was 9.40 min (SD=4.40)for the first session and 5.80 min (SD=3.04)for follow-up prompts.Surveys were generally completed within 1-1.5 day intervals (M = 1.41,SD=1.04).

Quality control and final sample
We implemented a number of quality control criteria for survey (attempts), which we represent visually in a Sankey chart (Fig. 2).
First, survey attempts that were largely incomplete (i.e., ≤50% questions completed) were not considered for analysis (n = 862).In most cases, these incomplete reports indicated a lack of engagement: for example, in 90.26% of these 862 survey attempts, participants did not move past the introduction screen.Remaining survey attempts (i.e., >50% finished) were excluded if the participant did not reach the final prompt reminder screen (n = 21), leaving a set of 2460 (73.60%) completed surveys.
Completed surveys with a completion time that exceeded 30 min were also excluded (n = 203), with 30 min being approximately three times the average completion time.Moreover, a small proportion of completed surveys from participants that only completed the survey once without providing sociodemographic details were excluded from the analysis (n = 18).This could have occurred if participants inadvertently indicated during their first survey that they had previously completed the survey.Participants that completed the survey multiple times without providing demographic details, however, were not removed from the dataset, and used in some analyses.Finally, we removed surveys for which at least one of the rating variables of interest, listed in the next section, was missing (n = 12).
After removing the ineligible surveys, a final sample of 2227 completed surveys remained (66.62% of opened surveys; 89.76% of surveys that participants engaged with), obtained from 1145 unique participants (1089 with completed demographics), of which 408 participants completed the survey more than once.

Survey items and outcome measures
Survey items focused on psychological indicators more generally associated with mental well-being as well as items that were specific to the COVID-19 pandemic.An overview of all self-rated items is available in Table 1.All items were rated on a 0-100 slider scale with an anchor at both ends describing the intensity of the rating (for subjective ratings 0=completely absent/ not at all, 100=very much so; for COVID-19 related items 0=completely disagree, 100=completely agree).The only exception to this format was the social isolation item, which was rated on a 0-24-hour scale and recoded post hoc to a 0-100 scale for consistency with the other items for the analyses (i.e., hours × 4.167).

Mental well-being items
All internal consistency and factor analyses reported below and under Section 2.4.3Factor Analysis were conducted on the eligible 1145 first reports.We used previously-validated items from ecological momentary assessment (EMA) studies (Myin-Germeys et al., 2009) to assess emotional states which we categorized into measures more closely associated with negative/positive mood (three items, Cronbach's α=0.61) and distress (five items, α=0.81).Participants additionally provided subjective momentary ratings of motivation (two items, α=0.79), energy (two items, α=0.50), and loneliness (one item).For each of these five mental well-being domains, item ratings were averaged (see Table 1 for individual items).

COVID-19 items
Participants rated ten COVID-19 related statements about their perceived risk of infection, worries about and preoccupation with (the potential impact) of the virus, and fear of (contracting) the virus (COVID-19 worries; α=0.70).An additional set of four statements was used to assess the degree to which participants adhered to infectionmitigation procedures that involved social distancing, which were disseminated by each country's respective government (COVID-19 guideline adherence; α=0.69).For each of these two COVID-19 domains, item ratings were averaged (see Table 1 for individual items).

Factor analysis
To confirm the existence of a more general mental well-being and COVID-19 domain we conducted an exploratory factor analysis (varimax rotation, here and below) using all rating items listed in Table 1.All 13 mental well-being items loaded more strongly onto factor 1 (0.22 proportion of variance explained) than factor 2, while 12 out of the 14 COVID-19 items loaded more strongly onto factor 2 (0.11 proportion of variance explained) than factor 1 (see Supplemental Table 1 for factor loadings).A similar exploratory factor analysis, but this time using all COVID-19 items, revealed that the 4 social distancing guideline adherence items loaded more strongly onto factor 1 (0.17 proportion of variance explained) and 7 out of the 10 worry/preoccupation items loaded more strongly onto factor 2 (0.14 proportion of variance explained).A final exploratory factor analysis using all 13 mental wellbeing items revealed the existence of two more general psychological constructs, one of which seemingly associated with negative (sad, annoyed, stressed, anxious, lack of undertaking activities/motivation, lower energy, loneliness; 0.27 proportion of variance explained) and the other with positive (cheerful, carefree, relaxed, calm, well-rested; 0.20 proportion of variance explained) psychological states.All in all, these results provide some evidence for the item groupings we discussed in Section 2.4.1, 2.4.2 and Table 1, and for all (network) analyses referenced below we consistently used these item groupings.

Statistical analyses
We first describe sociodemographic characteristics of the final sample using descriptive statistics.Associations (Spearman's ρ for age bracket, χ 2 for categorical predictors) between sample characteristics and survey-related details, such as the number of repeated measurements, and date of first report, were also assessed.
Next, we obtained insights into general, time-related trends in the item ratings.We, therefore, used linear mixed-effects models with surveys nested within participants to investigate associations between average item ratings from the 7 domains (i.e., mood, distress, motivation, energy, loneliness, COVID-19 related worries, COVID-19 guideline adherence) and a) day of first survey (a proxy of more general betweensubjects changes in ratings during the measurement window, i.e., April 2020) and b) day number relative to day of first survey (a proxy of more general within-subjects changes in ratings during the study participation window).These analyses were Bonferroni-corrected for the number of dependent variables tested (α=0.05/7).
Next, we carried out two types of network analyses (Borsboom and Cramer, 2013).The first analysis focused on data from all eligible first surveys (n surveys =1145), allowing us to examine cross-sectional associations among all domains of interest.The second analysis focused on longitudinal associations.To limit potential effects of very short or long temporal delays between subsequent surveys, we restricted this longitudinal analysis to surveys with temporal delays of 12 h to 4 days, leaving a sample of 395 (out of 408) participants with multiple eligible timepoints (n surveys =1038).
For the cross-sectional analyses, we computed a) the productmoment correlations and b) the partial correlations between the 7 item domains.Significant correlations at a Bonferroni-corrected α=0.05/21 were visualized in a network graph.For the longitudinal analyses, we fitted a multilevel lag-1 vector-autoregressive model (Bringmann et al., 2013) that provides information on the a) contemporaneous associations at a given time point (7 × 6/2 = 21 correlations) and b) lagged associations between each variable and the values of all variables from the previous report (7 × 7 = 49 coefficients).We used a fully multivariate model, in which all variables simultaneously acted as outcomes and all lagged variables were used as predictors.The temporal (lagged) associations were of particular interest given the possible causal insights that might be derived from these analyses (S.Epskamp et al., 2018b).
Two adjustments were made when fitting this model.First, to  (r)  Consider: many COVID-related deaths, many COVID-19 infections, and hospitals with max-out capacities.These are things that won't happen in the country I live in (r) I am not worried about the repercussions of the COVID-19 pandemic on work, income, or future perspective (r) Others have a greater chance of contracting COVID-19 than I do (r)  COVID-19 is not worse/more dangerous than the flu (r)  I am scared of contracting COVID-19 I am scared that my colleagues, friends and/or family will contract COVID-19 COVID-19 guideline adherence ("in the past 24 h, I:") have followed COVID-19 hygiene guidelines to the best of my ability (1.5 distance, sneezing in elbow, no handshakes, washing hands) have deliberately not taken the initiative to meet with other people; to minimize the risk of COVID-19 spread a have declined people's invitations (to physically meet) to the best of my ability; to minimize the risk of COVID-19 spread a have not left my home for [XX] consecutive hours b (r) = reverse-scored.a Participants were instructed to exclude individuals that they lived with from their answers.
b Rated on a 0-24-hour scale.

S.D. Voulgaropoulou et al.
account for differences in the lag between adjacent reports, we included the time lag as a predictor in the model and allowed it to interact with the coefficients that represent the temporal associations.The lagged associations we report represent those for a 24-hour time lag.Second, since fitting the model with random effects for each of the lagged coefficients led to convergence problems, we removed these random effects and instead used cluster-robust inference methods (Pustejovsky and Tipton, 2018) to test temporal associations in a model that still included random item intercepts at the subject level.Network graphs for significant contemporaneous (α=0.05/21correlation pairs) and temporal (α=0.05/49coefficients) associations were used to visualize these results.
All models were fitted twice; once without including additional covariates and once when controlling for age, gender, education, country, and confirmed daily COVID-19 cases (day prior to measurement).All reported analyses were carried out using R (Team, 2013) using package nlme (Pinheiro and Bates, 2000) for fitting the multilevel vector autoregressive models, package clubSandwich for the cluster-robust inferences (Pustejovsky, 2020), and package qgraph (Epskamp et al., 2012) for the visualizations of the networks.

Sample characteristics
In the final sample of 1145 participants, 35.66% of participants completed two or more surveys, with most of these participants completing five surveys.Demographic variables, available for 1089 participants, are reported in Table 2.In general, participants were more likely to be women (than men) and living in the Netherlands (compared to Belgium).The various education levels and age groups were evenly distributed across the sample, with only some underrepresentation of younger (<20) and older (>70) participants, although participants older than 60 years of age still made up 22.50% of the total sample.
Only a small subgroup of participants self-reported having a formal positive test result for COVID-19 (6.89%), with an additional 1.93% being suspected of having COVID-19 by a physician (see Table 2).Using a rating item that asked about the intensity of influenza-like symptoms, we confirmed that COVID-19 positive and suspected COVID-19 positive participants on average experienced greater flu-like symptoms than COVID-19 negative participants (В raw =33.60, 95% CI=[30.07-37.13], t 1082 =18.69, p<0.001).

Stability of item ratings during the study and participation window
Rating distributions of individual and averaged items from the first survey are presented in Supplemental Fig. 1.
Mixed-effects model analyses revealed that COVID-19 guideline adherence ratings were associated with day of first survey (B raw =− 0.27, 95% CI=[− 0.39 -− 0.16], t 1081 =− 4.71, p bonf <0.001), corresponding to a difference of 7.83 percentage points (on the 0-100 scale) between participants that completed their first rating on April 1st versus April 30th.No mental well-being domains nor COVID-19 related worries were associated with day of first survey following a Bonferroni correction.
These results provide some evidence for systematic between-and within-person trends during the study and participation window.We next turn to the network analyses to investigate (temporal) associations among COVID-19 and mental well-being item ratings.
Fig. 3C and 3D provide a visual overview of the contemporaneous and time-lagged associations, respectively, as found in the longitudinal network model using all eligible timepoints from the repeated measures data; n surveys =1038.The network of contemporaneous associations was highly consistent with the results from the cross-sectional partial correlation analyses (Fig. 3B versus Fig. 3C), with greater COVID-19 related worries being associated with increased distress (r = 0.10) and lower positive mood (r= − 0.14) ratings.Results from these two analyses, moreover, provide converging evidence that associations among mental well-being items were particularly pronounced in the moment.Importantly, however, most of the associations among mental wellbeing items were no longer significant when examining the temporal relationships between these items (Fig. 3D).Time-lagged associations were primarily observed between COVID-19 related worries and mental well-being items.Specifically, greater worries related to the COVID-19 pandemic at timepoint t were associated with greater distress (B raw =0.17, 95% CI=[0.10 -0.25], t 125 =4.48, p bonf <0.001) and lower positive mood (B raw = − 0.17 Although temporal associations were mostly small-to-modest, these results suggest that COVID-19 related worries may strengthen the reciprocal (negative) interplay between positive mood and distress.When repeating the analyses while controlling for several demographic covariates (n surveys =898, Supplemental Fig. 2), we observed a highly similar network of time-lagged associations, emphasizing the selective temporal dynamics involving positive mood, distress, and COVID-19 related worries.

Discussion
Here we investigated the psychological state of adults in the Netherlands and Belgium during the initial phase of the COVID-19 pandemic, a time of drastic changes in daily life routines due to uncertainty surrounding the COVID-19 pandemic and preventive measures taken to curtail the spread of the virus.Using network analyses, we found evidence for selective dynamic temporal interplay between worries about the COVID-19 pandemic and negative emotional states characterized by higher distress and lower positive mood.
Our cross-sectional results involving Pearson product-moment correlations revealed associations between COVID-19 related worriese.g., about infection risk, future repercussions, and impact on loved oneswith all mental well-being items.However, cross-sectional associations after controlling for correlations among rating items and contemporaneous associations (using longitudinal data) revealed a more nuanced pattern of results.In these analyses, COVID-19 related worries were consistently associated with higher ratings of distress and lower positive mood ratings.In turn, positive mood and distress ratings were associated with other indicators of mental well-being, such as loneliness, motivation, and energy.
Previous work has reported associations between COVID-19 stressors and a range of mental health proxies, including loneliness/social behavior, anxiety, and energy (O' Sullivan et al., 2021;Fried et al., 2020;Ryu et al., 2021).Interestingly, in studies employing network methodologyeven when using heterogenous samples in terms of participant characteristics and/or (subclinical) psychopathologylow positive mood and distress exhibit high centrality within depression-anxiety symptom networks, followed by other symptoms such as anhedonia, low energy, worthlessness, and nervousness (Beard et al., 2016;Bai et al., 2021).These observations are consistent with the notion that stress reactivity and low positive mood are paramount to the regulation of mental well-being (Flores-Kanter et al., 2021;Olff et al., 2021).For example, affective states are strong predictors of social behavior, daily-life activities and routines, and subsequent stress coping (Quoidbach et al., 2019;Flores-Kanter et al., 2021).Collectively, our cross-sectional and contemporaneous findings point to the presence of a relationship between COVID-19 stressors and heightened negative emotional states, which may exert secondary influences on other components of well-being, such as energy, motivation, and loneliness.
Importantly, associations among indicators of mental well-being were primarily observed in the same measurement window.Our timelagged network analysiswhich illustrates how variables predict each other in subsequent measurement windows (S.Epskamp et al., 2018a)revealed temporal associations within a selective cluster of items including COVID-19 related worries, distress, and positive mood.These ratings were not only (positively) autocorrelated, indicating a degree of similarity for ratings of a given item across time, but they also fueled each other over time.Specifically, COVID-19 related worries at timepoint t (e.g., day 1) were linked to lower positive mood and increased distress at t + 1 (e.g., day 2).These results could indicate that increased COVID-19 related worries may impact the dynamic regulation of emotional states over longer temporal windows.In addition, low positive mood and increased distress reciprocally interacted, resulting in a vicious cycle (i.e., high distress ⇄ low positive mood; Fig. 3D).Previous studies have postulated the existence of a bidirectional relationship between stress and (negative) affect, which can express itself in a downward spiral characterized by low positive mood and high distress (Langens and Stucke, 2005;Martinowich and Lu, 2008;Wolk et al., 2016;Wichers et al., 2009).Such interactions between low positive mood and distress can also be observed in the flow of daily life (Bos et al., 2018).Moreover, enhanced stress reactivity is associated with more severe depression and anxiety levels (van Winkel et al., 2015).Thus, our results suggest that worries about the current COVID-19 pandemic have the potential to selectively accentuate the negative interplay between low positive mood and distress.It would, therefore, be interesting for future studies to evaluate whether low positive mood and distress could keep reinforcing each other, even when initial worries about the pandemic start to fade away.
Despite the impact of COVID-19 worries on positive mood and distress, ratings of some mental well-being items (e.g., positive mood, distress, and energy) slightly improved throughout the measurement window.These findings are in agreement with previous studies reporting that after an initial increase in negative emotional states, the intensity of self-ratings may subside over time (Fried et al., 2020;Bendau et al., 2021).These relative improvements in emotional states could suggest the presence of an initial elevation bias in negative psychological states, which is often observed in self-report studies (Shrout et al., 2018).Alternatively, improvements could be indicative of successful adaptation or resilience (Veer et al., 2021).
Interestingly, adherence to hygiene and social distancing measures slightly decreased during the measurement window.This is in agreement with a gradual decline in adherence to protective measures reported in previous studies (Petherick et al., 2021;Scandurra et al., 2021), and may have been associated with a drop in cases and/or and good news reports (e.g., planned reopening of public institutions announced later in the measurement window).Given the moderate decrease in distress over time, this finding corroborates the observed positive association between distress and COVID-19 guideline adherence in the contemporaneous network.That is, lower distress may lead to reduced guideline adherenceeither directly or indirectly via the COVID-related worries node.
All in all, our data collected in the initial stages of the COVID-19 pandemic suggest that increased pandemic-related worries are associated with heightened negative emotional states.Temporal associations among COVID-19 related worries, distress, and positive mood may constitute a mechanism by which the ongoing pandemic could impact mental well-being, although studies with longer measurement intervals and knowledge of underlying resilience determinants would be necessary to support such conclusions.If confirmed, this mechanism may provide one explanation for the increased prevalence of affective/stressrelated disorders reported during the COVID-19 pandemic (Salari et al., 2020;Qi et al., 2021).

Strengths and limitations
An advantage of the network approach used in this study is that it highlights the complex associations that collectively influence mental well-being.Our results provide initial insights into the psychological mechanisms that may be impacted by the currently on-going COVID-19 pandemic, and, as such, provides clues on potential risk or resilience mechanisms during major health crises.
However, whether and under which circumstances these emotional changes will meaningfully contribute to psychopathology remains elusive and speculative.One caveat of this study is its short duration, which resulted in short intervals between subsequent measurements in conjunction with a relatively small sample size for online research.Although in this study we observed short-term temporal stability of the network, it would have been helpful to evaluate associations among mental well-being and COVID-19 specific items over longer intervals.Longer-term measurement windows with larger sample size could reveal whether COVID-19 worries, distress, and positive mood dynamics, uncovered here, persist over the course of the pandemic.Furthermore, caution is warranted regarding the generalizability of findings due to potential attrition bias, as observed when comparing cross-sectional and longitudinal data (35.63% of the total sample completed the survey more than once).Past research has shown that people with mental health problems are more likely to discontinue participation in followup measures (da Graca et al., 2023).Thus, lack of information regarding participants mental health history and/or use of psychotropic medication is another limitation that should be acknowledged.
Although we controlled for several sociodemographic variables, our design was not optimized for stratified analyses, for example based on gender, age, psychiatric history, geographical location, ethnicity, or education, which could explain the low effect sizes observed.Several studies have identified potential sociodemographic risk factors that predict worse mental health outcomes during the COVID-19 pandemic, including being female, younger in age, having pre-existing mental health problems, lack of social support, previous trauma, and experiencing additional stressful events in the past month (O' Sullivan et al., 2021;Li and Wang, 2020;O'Connor et al., 2021;Varga et al., 2021;Olff et al., 2021).Given the age distribution of this sample, future work could look at the association of age with measured variables, such a social isolation, impact on loved ones, etc. (Minahan et al., 2021;Sojli et al., 2021).A final limitation is that our sample is self-selected, meaning that it may not be representative of the entire population.Thus, reported findings should be extrapolated with caution.

Conclusion
To conclude, this network-based study evaluated the potential psychological repercussions of the COVID-19 pandemic in the Netherlands and Belgium.We identified worries about COVID-19 to be temporally associated with the reciprocal interplay between distress and low positive mood.Increased distress and low positive mood, in particular, seem to be important factors that may possibly, in the long run, be associated with adverse mental health outcomes in the current health crisis.

Fig. 1 .Fig. 2 .
Fig. 1.A schematic overview of the daily confirmed COVID-19 cases and the measures taken to mitigate the spread of the virus in the Netherlands and Belgium, displayed in chronological order.All text labels refer to measures that were implemented in the Netherlands.

Fig. 3 .
Fig. 3. Network analyses.Top: Cross-sectional correlation analyses using all eligible first surveys (n surveys =1145).Bottom: Associations obtained from a vectorautoregressive model using all eligible repeated measures (n surveys =1038).Associations visualized in 3A-B represent Pearson product-moment and partial correlation coefficients, respectively.Associations visualized in 3C represent the estimated contemporaneous correlations among variables.Associations visualized in 3D represent slopes/B raw coefficients and are directional; they indicate how ratings of variable X at timepoint t are associated with ratings of variable Y at timepoint t + 1. Temporal autocorrelations in 3D are visualized as curved/circular arrows.Green = positive association; red = negative association.Line thickness and color intensity corresponds to the association strength.Only significant associations are shown (based on Bonferroni corrections for 21 unique correlation pairs in 3A-B-C and for 49 time-lagged associations in 3D).

Table 1
An overview of questionnaire items.

Table 2
Sample characteristics.
a Suspected or confirmed influenza/COVID-19 by a medical expert and/ or PCR test.b Based on N = 1145 participants.S.D. Voulgaropoulou et al.

Table 3
Associations among item ratings.