Theoretical considerations

The FSM emphasizes that child adjustment problems are exacerbated when economic hardships are placed on family processes. Economic hardships can be manifold and defined objectively as well as subjectively [18]. Objective measures are factual reports of life circumstances (such as employment, and income). Subjective measures, in contrast, describe one’s perception of life circumstances and economic situation [18, 19]. Together, hardships (such as job losses, reduced income, income instability, and unstable work conditions) can result in unstable conditions for families with impacts on family functioning [16]. A recent meta-analysis found that both objective and subjective economic hardships showed associations with subjective wellbeing. Importantly, the links were stronger for subjective measures [18]. In the context of the COVID-19 pandemic, many families experienced economic hardships, particularly job loss and financial insecurity [6, 7, 20]. For example, a study from India found that these two aspects of economic hardship were associated with poorer mental health outcomes [20]. Reflecting both the assumptions of the FSM and the pandemic-specific sources of disruptions, we used COVID-19-related job loss as objective indicator and perceived economic pressure (i.e., perception of potential economic consequences for one’s family) as subjective indicator in our conceptual model (Fig 1). Addressing the specifics of the global health crisis, perceived pandemic-related health risks were included as additional indicator of hardship.

In accordance with the FSM, we assumed that objective and subjective hardships experienced by families during the pandemic drained parents’ mental and emotional resources and contributed to higher parental stress (Fig 1). Research shows that parents’ risk of experiencing stress increased substantially after the COVID-19 outbreak began in early 2020 [12, 21–23]. It has been reported that parents feel “far less energetic, more stressed, tired, angry, listless, nervous, bored, and quite worried” [23] (p. 165). Common stressors were changes in daily routines, anxiety around COVID-19 and health risks, and demands related to at-home care of children [21]. Importantly, since the start of the pandemic, parents were more likely to feel stressed than adults without children [24, 25]. According to Sprang and Silman [26], pandemic disasters create conditions that families and children find traumatic. Reports from 398 parents with varying disease-containment experiences following the H1N1 outbreak in 2009 indicated that 25% of parents met criteria for PTSD. In addition, significant associations between PTSD symptoms in parents and children were reported [26], suggesting that pandemic stressors disrupt family wellbeing [6].

The FSM further assumes that economically-influenced parents’ stress can cause problems in the partner relationship by increasing partner conflict and negative perceptions of the relationship and decreasing satisfaction and perceived spousal support [16]. Similar processes have been used to explain relationship stress in the context of COVID-19 [6, 27]. Previous literature suggests that stressors associated with disasters interfere with adaptive dyadic relationship processes. The unknown duration of the current pandemic and the pervasive uncertainty and fear regarding its impacts may destabilize relationships [27]. A growing body of evidence supports the conclusion that parents’ stress during the COVID-19 pandemic is adversely affecting spousal relationships (e.g., Singapore [28]; Israel [29]; Iran [30]; US/Canada [31]).

In the context of the FSM, parents’ stress and spousal relationship problems are assumed to increase disrupted parenting, leading to insensitive and unresponsive parenting behaviors, which is thought to result in lower quality of parent-child relationships [16]. The conceptual framework of family wellbeing builds on these assumptions and argues that heightened parents’ stress due to the COVID-19 pandemic changes family relationships due to the interconnections among family members [5]. Similarly, in life course perspectives, the construct of linked lives is used to argue that the COVID-19 pandemic is not experienced in a vacuum, but children navigate it together with important others, in particular their parents, and that relationships of family members, including parent-child relationships, can be undermined by parents’ stress [17]. The theoretical assumptions are substantiated by growing empirical evidence testing the impact of the COVID-19 pandemic on family functioning. For example, one study showed a decrease in family cohesion and an increase in family conflict which both were associated with child maladjustment [32]. Other studies also found that parents’ stress during the COVID-19 pandemic changed family relationships. Parents reported to have more arguments with their child and found it more difficult to enjoy parent-child interactions [12, 23]. As such, disruptions caused by the COVID-19 pandemic “get inside the family by altering interpersonal relations within the family” [6] (p. 632).

Family processes as mediators

It is well established in the literature that acute and chronic stressors contribute to child adjustment problems. Knowledge about the mechanisms underlying these associations is crucial to inform policies and interventions aimed at supporting families. A central assumption of the FSM is that pathways to negative child outcomes are mediated by family processes [16, 33]. Importantly, empirical evidence suggests that family processes are malleable and programs can be effective in changing maladaptive family processes [34–36]. As such, testing family processes as mediators can provide useful information for optimizing and implementing policies and interventions.

However, empirical evidence regarding mediation pathways is mixed. For example, Mistry et al. [37] found that associations between children’s exposure to cumulative risk and their school readiness skills were (partially) mediated through family processes, while Shelleby et al. [38] found that family income led to higher levels of children’s conduct problems, observed through family processes. Yet, other studies did not find evidence for significant mediations via family processes, even though the individual paths included in the mediation model were significant [39]. An additional limitation of previous research is that mediations within the FSM framework was often tested using cross-sectional data [39]. However, cross-sectional mediator analysis can be biased in that indirect effects may be detected that do not hold in longitudinal analysis [40].

The present study

This research reflects experiences of parents of young children (ages 0–8 years) living in the UAE during the COVID-19 pandemic. The overarching aim was to test the assumptions of the FSM in the pandemic context (Fig 1). We considered recent literature reviews to reflect the specifics of the COVID-19 pandemic when adapting the FSM which we based on frameworks of family wellbeing [6], relationship science [27], and life course perspectives [17].

The child-specific impacts of the COVID-19 pandemic depend on the developmental timing, with younger children being particularly vulnerable [17]. Based on the literature suggesting that COVID-19-related hardships and pressures may have similar effects on parents’ wellbeing as economic hardships and pressures, we expected to find associations with higher levels of parents’ stress. We also expected that higher levels of parents’ stress would show direct negative links with family relationships (i.e., higher spousal relationship problems and lower quality of parent-child relationships) and child wellbeing.

Because parents’ wellbeing “may serve as a funnel through which social disruptions due to COVID-19 infiltrate family functioning via changes to marital [and] parent-child […] relations” [6] (p. 632), we assumed that the associations between parents’ stress and child wellbeing would be mediated through family relationships. Specifically, we expected higher levels of parents’ stress to be related with lower quality family relationships which, in turn, would be linked to lower child wellbeing.

The COVID-19 pandemic was not a one-time shock but an ongoing stressor. We thus explored associations between COVID-19-related hardships and pressures, parents’ stress, family relationships, and child wellbeing over time. Because it is possible that time may either buffer or exacerbate the pandemic’s negative effects on families, we treated this aspect as an exploratory part of our analyses.

The study has several contributions to the literature. It extends research on the FSM beyond studies with school-aged children and adolescents [16] to younger children (birth to 8 years) who may be particularly vulnerable to developmental shocks resulting from the pandemic [17]. In addition, it extends research exploring the temporal dynamics between parent and child wellbeing during a period of unexpected and drastic societal change due to the COVID-19 pandemic [5, 9, 15]. By collecting data twice during the pandemic (t1, t2), the study helps to uncover immediate and lagged effects of the pandemic on parent and child wellbeing.

Participants

The Life During COVID project, a university-government collaboration exploring the experiences of families with young children during the COVID-19 pandemic, targeted parents of children aged 0–8 years in the Emirate of Abu Dhabi, UAE. After cases of COVID-19 started to rise in the UAE in February 2020, the government mandated people to avoid family gatherings and social events to reduce the risk of infections, a restriction particularly challenging within the cultural context of the UAE, where an important cultural pillar is close social bonds with the immediate and extended family [41, 42]. In early March 2020, businesses, places of worship, daycare centers, and schools were closed nationwide.

Initially, school closures were planned to last four weeks, but facilities remained closed for the remaining months of the academic year with distance learning implemented. Schools reopened in September 2020, but throughout the 2020–21 academic year, they were closed several times as cases climbed. Similarly, work-from-home policies were implemented to reduce the spread of the virus, forcing employees from various sectors to work remotely.

At the time of recruitment for the Life During COVID project from which we draw the data, in-person human subject research was not permitted due to the COVID-19 situation in the country. Thus, data was collected through an online survey. The study protocol was reviewed and approved by the Institutional Review Board of the authors’ university (protocol HRPP-2020-65). This study was not preregistered. Prior to completing the survey, all participants provided informed consent. In Fall 2020 (between September-November), the survey link was disseminated on social media (Facebook and Twitter), through short message service (SMS), and via a panel management service. The survey link dissemination strategy was similar to other studies conducted with parents during the pandemic (e.g., [12]). One recruitment goal of the present study was to particularly target UAE citizens who make up approx. 19% of the population in the Emirate of Abu Dhabi [43]. The government partner facilitated a targeted recruitment to reach that group of the population.

Initially, 5,165 survey link clicks were registered; 3,109 (60.2%) did not complete more than the first survey block with demographic questions. Entries were further checked for inclusion criteria (parents of children aged 0–8 years; residents in the Emirate of Abu Dhabi), which were not met by 316 respondents. Furthermore, 71 duplicates were identified and deleted. Disqualified entries (n = 3,496, 67.7% of all clicks) were removed.

The resulting sample included 1,669 participants in Fall 2020 (t1). The majority of participants (n = 1,096, 65.7%) were from countries other than the UAE; the remaining participants were UAE citizens (n = 573, 34.3%). While the proportion of UAE citizens in the sample was larger than that of the UAE’s total population (11.6%) [44], the sample reflects diversity of the population in the Emirate of Abu Dhabi. In an attempt to verify participants, we estimated their age at the child’s birth. Taking local traditions and customs into account (such as early marriage), no unreasonable responses were found. Participants were re-contacted for a follow-up survey between February-April 2021 (t2), when 1,144 survey link clicks were registered; 282 (24.7%) did not complete more than the first few demographic questions. These entries were again deleted. After removing cases that could not be merged with responses from the first data collection (9%), the final sample included 783 parents who completed the survey twice. Post-hoc power analysis for structural equation modeling yielded a minimum sample size of 589 participants (using the following assumptions: effect size = 0.2; statistical power level = 0.8; probability level = 0.05). Of these parents, 96 completed the survey for two children in the same 0–8 age range. We verified that this subsample was representative of the full sample. A series of t-tests showed that the subsample, for which data about two children in the same family was available, was a close representation of the full sample. There was no evidence that the two samples were different in terms of parent age, being a UAE national or not, and region of residence (S1 Table).

The mean age of parents (75% female) was 34.61 years (SD = 6.05) (see Table 1 for demographic details). Reflecting the diversity of the population in the UAE, most parents (69%) were from other countries (including Egypt, India, Jordan, Pakistan, the Philippines, Syria, and the United Kingdom); 31% were UAE citizens. Half of the sample resided in the city of Abu Dhabi (urban region) and half in suburban and rural regions which is representative of the general population in the Emirate of Abu Dhabi where 49.9% of the population live in the city of Abu Dhabi [45]. In terms of education, 44% of the sample held a bachelor’s degree or higher.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic information about the sample.

https://doi.org/10.1371/journal.pone.0292292.t001

Measures

The survey was available in both Arabic and English and used Qualtrics. Arabic is the official language in the UAE; English is used in conjunction with Arabic in most sectors. The survey was translated and back-translated by native Arabic speakers who were fluent in English to ensure the quality of the translation and consistency of meaning. The same survey questions were asked at both measurements (t1 and t2). The first block focused on demographic information of the family, COVID-related indicators of hardships and pressures, parents’ stress, and spousal relationships problems. The second part asked about the wellbeing of their child and parents’ perceived quality of the relationship with them. If parents indicated that they had two children aged 0–8 years, they received a third part of the survey that asked questions regarding the second child’s wellbeing and their relationship quality with that child. In the demographic part of the survey, parents were asked to provide the name(s) of their child(ren). The questions regarding child 1 and child 2 were personalized and used the name for that child to make it easier for the parent to respond to these questions for each specific child.

Analytical strategy

We first created latent variables for the constructs that were measured with three or more items (such as parents’ stress, spousal relationship problems, and child wellbeing) using confirmatory factor analyses (CFA). When necessary, modifications were introduced to reach the best fitting model. Because constructs were measured at two time points, multi-wave CFAs were used to test their measurement invariance over time at three levels: configural, metric, and scalar. Configural invariance assumed the same scale structure over the two waves without restricting factor loadings of the scale items. Metric invariance assumed the same scale structure over time by fixing the factor loadings of the items across the two waves.

In addition to these two conditions, scalar invariance additionally fixed the intercepts of the scales’ items across the two waves. Change in comparative fit index across the configural, metric, and scalar models (>±0.01 considered the cutoff) was used to determine measurement invariance [54]. In addition, CFA model fit was evaluated using the root mean square error of approximation (RMSEA; < 0.08), comparative fit index (CFI; > 0.90), the Tucker–Lewis index (TLI; > 0.90), and standardized root mean residual (SRMR; < 0.08) [55]. The results are presented in S2 Table.

Next, structural equation modeling (SEM) was then used to test the assumptions of the FSM in the context of the COVID-19 pandemic. The SEM analyses were conducted in RStudio (version 1.4.1106), using the following variables: COVID-19-related indicators of hardships and pressure, parents’ stress, spousal relationship problems, parent-child relationship, and child wellbeing. Control variables (t1: parent age [in years], level of education [binary; 1: BA degree or higher, 0: otherwise], nationality [binary; 1: UAE national, 0: otherwise], child age [in months], child sex [binary; 1: boy, 0: girl], and physical activity) were included. The model estimated associations concurrently (t1, t2) for one child in a family (the same child at t1 and t2) (Fig 1). In addition, autoregressive paths (such as the path between parents’ stress at t1 and parents’ stress at t2) and cross-lagged paths (for instance, the path between parents’ stress at t1 and spousal relationship problems at t2) were estimated to explore relationships over time (t1-t2).

In a supplemental analysis, a second SEM estimated concurrent (t1) associations between the variables of interest for families with data reported from two children (S1 Fig). The goal of the supplemental analysis was to assess whether similar associations are evident for two children of similar age (0–8 years) in the same family, and in particular, to examine child-specific parent-child relationship quality amid COVID-19 stressors. The results are presented in S2 Fig and S3 Table.

SEM models were fitted with both maximum likelihood parameter estimates (MLM) and maximum likelihood with robust standard errors (MLR) whereby the latter results are reported here (MLM results can be found in the supplementary information, S3 and S4 Tables). For both models, tests of direct and indirect effects were run. To test the full mediation effect, an inferential test of the entire specific indirect effect was conducted. More specifically, we tested whether the indirect effect of parents’ stress on child wellbeing via (a) spousal relationship problems or (b) parent-child relationship was significant, and (c) whether the overall indirect effect of parent stress on child wellbeing was significant. For each predictor-to-outcome effect, we used the lavaan package [56] to first compute the direct effect and then the indirect effect (along with standard errors and significance tests). Standard errors were computed using the delta method (producing what is known as the Sobel test for indirect effects). The standardized regression coefficients were used as effect size measures, with β ≈ 0.10 indicating a small effect, a β of ≈ 0.20 a typical-sized effect, and β>0.30 indicating a large effect [57]. In addition, we reported the R-squared for each latent variable in the direct effects models to estimate the proportion of variance explained by all predictors in the model.

Data preparation and analyses were performed using Stata and R. Missing data were handled by multiple imputation using the MICE package in R [58]. Predictive mean matching algorithm was selected for the multiple imputation as it is a versatile method that allows discrete variables in the data and the resulting imputed data are realistic, given that they fall within the range of the observed variables [59]. Data and code for all analyses are available (https://github.com/rali314/family_stress_during_covid_19_pandemic).

COVID-19-related hardships and pressures and parents’ stress

The first SEM model estimated associations concurrently and over time for one child in a family, using the full sample (N = 783) (Fig 1; S4 Table shows the correlation between variables). Parents’ perceived COVID-19-related health risks were significantly correlated to parents’ stress, however, the direction of association differed across time (Fig 2A and S5 Table). At t1, fears of getting infected with the coronavirus were positively related with parents’ stress (β = 0.14, p<0.01). The association was negative at t2 (β = -0.256p<0.001), but positive over time (t1-t2) (β = 0.15, p<0.01). No significant associations were found between COVID-19-related job loss and parents’ stress, neither concurrently at t1 (β = 0.05, n.s.) or t2 (β = 0.03, n.s.) nor over time (t1-t2) (β = 0.06, n.s.). COVID-19-related economic pressure was significantly associated with parents’ stress at t1 (β = 0.21, p<0.01), but not at t2 (β = 0.00, n.s.) or over time (t1-t2) (β = -0.03, n.s.).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Parameter estimates from SEM testing assumptions of the FSM.

(a) Associations between COVID-19-related hardships and pressures on parents’ stress as well as direct associations between parents’ stress (and family relationships, respectively) and child wellbeing cross-sectional at t1 (t2, respectively) and over time (t1-t2). (b) Cross-sectional (A paths) and lagged (B paths) mediation pathways. The model controlled for parent age, level of education, and nationality, and child sex, age, and physical activity. The analysis reported here used the MLR estimator. Model fit: CFI = 0.906; RMSEA = 0.048; SRMR = 0.062. * p ≤ 0.05; ** p ≤ 0.01; *** p ≤ 0.001. ^a Higher values reflect higher levels of spousal relationship problems.

https://doi.org/10.1371/journal.pone.0292292.g002

Parents’ stress and family relationships

Parents’ stress was positively and significantly associated with spousal relationship problems at t1 and t2 (t1: β = 0.28, p<0.01; t2: β = 0.40, p<0.001) and over time (t1-t2) (β = 0.21, p<0.001) (Fig 2B and S5 Table). Higher levels of parents’ stress were related to increased relationship problems between the caregiving partners. Similarly, higher levels of parents’ stress were related to lower perceived quality of the parent-child relationship. Associations were significant and negative at t1 (β = -0.31, p<0.001) but positive at t2 (β = 0.16, p<0.001). The association between parents’ stress at t1 and parent-child relationship quality at t2 was not significant (β = -0.09, n.s.).

Direct and indirect associations between parents’ stress and child wellbeing over time

The direct association between parents’ stress and child wellbeing was significant and negative at t1 (β = -0.17, p<0.01) and t2 (β = -0.30, p<0.001) (Fig 2A and S5 Table). No significant association was found between parents’ stress at t1 and child wellbeing at t2 (β = -0.05, n.s.). To explore indirect associations between parents’ stress and child wellbeing, we first examined the individual direct paths between the mediators (spousal relationship and parent-child relationship) and child wellbeing (Fig 2A and S5 Table). The association between spousal relationship problems and child wellbeing was significant and negative at t1 (β = -0.12, p<0.01) and t2 (β = -0.18, p<0.05), but the association between spousal relationship problems at t1 and child wellbeing at t2 was not significant (β = 0.07, n.s.). The association between parent-child relationship and child wellbeing was significant and positive at t1 (β = 0.22, p<0.001). Neither the association at t2 (β = 0.03, n.s.) nor the association over time (parent-child relationship at t1 and child wellbeing at t2; β = -0.03, n.s.) were significant. A total of eight mediation pathways from parents’ stress to child wellbeing were tested: four cross-sectional mediations (“A” paths; mediated by family relationships at t1 and t2) and four lagged mediations (“B” paths) exploring mediation pathways over time (Fig 2B and S5 Table). For the latter, we tested two conditions: (a) effect of family relationships at t1 on child wellbeing at t2 (paths B1 and B3) and (b) effect from parents’ stress at t1 on family relationships at t2, which were related to child wellbeing at t2 (paths B2 and B4).

Regarding the cross-sectional mediations (Fig 2B), we found evidence for significant, albeit small indirect associations between parents’ stress and child wellbeing via spousal relationship problems at t1 (path A1: β = -0.03, p<0.05) and t2 (path A2: β = -0.07, p<0.05), and via parent-child relationship at t1 (path A3: β = -0.07, p<0.001). The indirect association via parent-child relationship at t2 was not significant (path A4: β = 0.00, n.s.). Only one of the cross-sectional mediations was replicated longitudinally. We found a similar small significant indirect association between parents’ stress at t1 and child wellbeing at t2 via spousal relationship problems at t2 (second condition/path B2: β = -0.05, p<0.05) but not via spousal relationship problems at t1 (first condition/path B1: β = 0.02, n.s.). None of the lagged mediations via parent-child relationship was significant (via parent-child relationship at t1/path B3: β = 0.01, n.s.; via parent-child relationship at t2/path B4: β = 0.00, n.s.).

Limitations

Several additional limitations need to be considered. Most importantly, variables were not assessed before the pandemic and, as such, we cannot rule out that pre-existing family vulnerabilities such as marginalization, mental health conditions, or family relational dysfunction [6], contributed to family processes and how pandemic adversities were navigated. Methodological challenges, in particular potential sampling bias and shared method bias, call for caution in interpreting the findings. The reliance on an online convenience sample with self-selected participants could have led to biased estimates, thus limiting the generalizability of findings [67]. Similar to other studies during the COVID-19 pandemic that used online surveys [12, 68], we do not know if and in what ways participants in our study differ from parents who chose not to participate in the study. As in prior studies [14], some of the observed associations might have been because the study relied on parents as the only informants. Parents reported information on their own psychosocial functioning, the functioning of family relationships, and their child’s psychosocial functioning. In-person human subject research was restricted during the time of data collection, limiting our ability to include other sources of data. Future research should include multiple sources of data, such as data from children and other informants (like teachers), and observations of family interactions and parent and child behaviors and outcomes.

In addition, one item assessing child wellbeing (social activities) did not show sufficient consistency with the construct across the two measurements and was therefore dropped from the analyses. Prolonged stay-at-home policies limited children’s social activities and parents might have had few or no opportunities to observe their child engaging in such activities as the pandemic continued. Thus, they might have used a different frame of reference when responding to this item. Furthermore, most respondents were mothers (75%). Caregiving has fallen disproportionally to mothers during the pandemic [69]. However, because our interest was not on levels of wellbeing but rather on patterns of associations between COVID-19-related disruptions and wellbeing, we think the potential bias introduced to the results is minimal.

Conclusion

The consistency of our findings with earlier tests of the FSM suggests family-based pathways that may alleviate or exacerbate child adjustment problems as consequences of adverse events. Although stressors might have been different during the COVID-19 pandemic as compared to pre-pandemic economic stressors, our findings suggest that the disruptions due to COVID-19 put parents at risk for experiencing stress reactions that may reduce caregivers’ ability to protect, support, and promote children’s wellbeing. This study contributes to understanding the mechanisms through which COVID-19-pandemic stressors may impact family processes and children’s development. As the pandemic continues, it will be important to develop and implement interventions that support families in maintaining or regaining a positive family climate that enables supportive relationships to prevent negative consequences on wellbeing.