Mothers of Young Children during the COVID-19 Pandemic: Depressive Symptoms and Perceived Stress

The COVID-19 pandemic may be described as a long-term stressor for the people and public health care systems worldwide [1, 2]. Parents of younger children and especially mothers seemed to be particularly affected by the pandemic situation [3‒5], especially in terms of higher stress levels and depressive symptomatology [6].

While even before the pandemic, the peripartum period per se had already been a vulnerable time for a woman to develop mental health issues [7, 8], this vulnerability may have even increased in the context of the COVID-19 pandemic [9]. Systematic reviews and meta-analyses indeed reported rising prevalences in maternal peripartum depressive symptomatology in the context of the pandemic [10‒14]. Sorting by regional factors, a recent meta-analysis demonstrated that studies conducted in Asia showed an increase of prevalence rates during the COVID-19 period compared to pre-pandemic prevalences, while studies conducted in Europe and North America showed no significant difference [15]. Regarding perceived stress, studies showed higher levels than before the pandemic [16] and a significant augmentation of stress levels during the pandemic [17‒19]. Higher levels of perceived stress were strongly associated with higher levels of depressive symptoms in a cross-sectional online assessment during the early phase of the pandemic [20].

Investigating the bidirectional link between stress and depression has been of particular interest for some time now [21‒23]. Not only stress increases the risk for depression, i.e., a stress exposure model of depression [24], but depression, or depressogenic vulnerabilities, in turn, enhances the susceptibility to stressful events which are at least partly influenced by the individual, i.e., a stress generation model [22, 25]. Only few studies exist which examine the interplay and different trajectories of perceived stress and depressive symptoms in the peripartum period [26‒28]. These studies emphasized the concurrent comorbidity of depressive symptoms and perceived stress during the process of pregnancy and the early years of motherhood. To the best of the authors’ knowledge, only one study examined the extent to which the chronic stress of the COVID-19 pandemic alters the trajectory of depression, anxiety, and stress symptoms in peripartum women [29]. The study focused on the prepartum and very early postpartum period and found that during COVID-19, symptoms of depression rose from early to mid-pregnancy but then lowered slightly while stress levels remained elevated.

As far as we are aware, no studies have investigated the reciprocal, i.e., two-way or bidirectional, relationship between the severity of depressive symptoms and perceived stress in the early years of motherhood during the COVID-19 pandemic. This study, therefore, aimed at (1) assessing the overall severity of depressive symptoms and perceived stress in Germany at two time points during the pandemic (i.e., summer/fall 2020 and early spring 2021) and reporting prevalence rates for both measures, (2) comparing the respective overall severity of depressive symptoms and perceived stress between the 2 time points, and (3) examining the reciprocal relation between these two constructs.

The study sample is part of a larger longitudinal online survey (CoviFam), conducted in Germany, Switzerland, and Austria during the COVID-19 pandemic; for a more detailed description of the procedure, see Woll [30]. In the present study, we only included mothers living in Germany as restrictions were quite heterogeneous across countries. The first measurement point (T1) was between July 15 and November 15, 2020, and the second measurement point (T2) was between February 19 and March 22, 2021. T1 represents a time of more lenient confinement measures due to a lower infection rate after the first wave of the pandemic in spring 2020 (e.g., reopening of day care facilities and public spaces as well as more relaxed social distancing measures). In contrary, T2 represents a time of stricter confinement measures due to a higher infection rate (e.g., closures of day care facilities and public spaces as well as stricter social distancing measures).

We aimed to assess the psychosocial well-being of parents with infants and toddlers aged 0 to 3 years during the COVID-19 pandemic. Parents were recruited with flyers that were distributed as printout to medical institutions (e.g., birth clinics, pediatricians, gynecologists), and digitally via professional regional and national networks (e.g., midwifes, nurses). Additionally, we used different social media networks (i.e., Instagram, Twitter, Facebook) where we presented our study inviting “mothers and fathers (regardless of being a couple) of infants and toddlers between 0 and 3 years of age.” As over half of the participants answered soon after posting the study on social media, we assume that most were recruited this way. There were no other eligibility criteria. Both parents could participate in the survey. In case both self-identified as female, we would have included the mother who had filled out the questionnaire first. However, there was no all-female couple in this longitudinal sample.

Participants were asked for their informed consent to participate at the beginning of the survey and, at the end of the survey, for consent to be contacted again via e-mail. For T2, a random ID code was created which was sent to participants together with the link to our follow-up survey.

Consent to be contacted again was given by 67.13% (i.e., 1,109 out of 1,653 others). Of these mothers, 60.05% (i.e., 666 out of 1,109) took part at T2. Sociodemographic data of the 666 mothers and their children (51% male, n = 340) are presented in Tables 1 and 2. Notably, according to their date of birth, 63 children were above 3 years of age, which was probably due to parents misinterpreting the targeted age range. Three children were even slightly above 4 years of age. As we focus on the dynamics of two maternal outcomes and not on any child-specific outcomes, we decided not to exclude these mothers. Maternal age ranged from 19.7 to 47.7 years at T1 and from 20.2 to 48.2 years at T2. The longitudinal sample (i.e., 666 mothers participating at both T1 and T2) did not meaningfully differ from the sample of mothers who only completed the survey at T1 (n = 987) on any of the sociodemographic or outcome scales, except for the mothers’ age and the change in partners’ job situation. However, these differences were small or close to a negligible effect size. Moreover, the longitudinal sample (n = 666) did not meaningfully differ from the sample of mothers who initially consented to be contacted again, but did not respond at T2 (n = 443), except for the mothers’ age. Here again, the difference was very small and close to a negligible effect size. Additional information regarding our extensive drop-out or attrition analyses can be found in the online supplementary material (for all online suppl. material, see https://doi.org/10.1159/000543715).

To assess the mothers’ sociodemographic and COVID-related data, such as educational background, job and living situation, contact with the coronavirus, and child care situation, a self-developed questionnaire was employed.

Severity of depressive symptoms was assessed using the German version [31] of the Edinburgh Postnatal Depression Scale (EPDS [32]). It has been validated in numerous studies as a ten-item self-rating scale, all items coded from 0 to 3, to detect prenatal and postnatal depression [33]. Greater sum scores indicate greater severity of depressive symptoms during the last 7 days. In the German version, a score of 10 or higher is regarded as an indicator for a depressive disorder [31]. The scale additionally distinguishes between a minor (≥10) and a major depressive disorder (≥13) according to DSM-IV-TR [34]. In other words, participants below a score of 10 carry a low risk for depression. The scale’s internal consistencies were above 0.85 at all time points (Cronbach’s α = 0.86 at T1, α = 0.87 at T2).

To assess individuals’ self-reported stress levels, the German version of the Perceived Stress Scale (PSS [21]) in its 10-item version was used, as published by Reis et al. [35] and translated by Büssing and Recchia [36]. Each item is answered on a 5-point Likert scale and coded from 0 to 4 (“never” to “very often”). A greater overall sum score indicates a greater level of perceived stress. To be able to compare our findings to other studies which applied a categorization of the score (e.g., [37]), we reported low (0–13), moderate (14–26), or high (27–40) levels. The scale’s internal consistencies were above 0.90 at both time points (Cronbach’s α = 0.90 at T1, α = 0.90 at T2).

We hypothesized that the severity of depressive symptoms as well as the level of perceived stress increased across T1 and T2. We also expected the severity of depressive symptoms and perceived stress to be positively and reciprocally related to each other.

Paired t tests were performed to test the increase across time points. As additional analyses, we, first, calculated prevalences of a risk for depression as well as of low, moderate, and high levels of perceived stress as percentages of mothers in the respective categories, complemented by their 95% confidence intervals (CIs). Second, Cohen’s d [38] was calculated to obtain the effect size of the changes between both measurement points for both of our outcome measures. Third, to assess if mothers changed from being at a low risk for depression to being at a high risk for depression from T1 to T2, a McNemar’s χ² test was performed.

The reciprocal relationship was tested using structural equation modeling in a cross-lagged panel design. A cross-lagged panel model (CLPM) is more specific compared to traditional repeated measures methods, as it examines causal relationships and directionality between variables over time. Parameter estimates of the CLPM were obtained by a maximum-likelihood estimation with robust estimators of model fit (MLR) with robust (Huber-White) standard errors and a scaled test statistic that is (asymptotically) equal to the Yuan-Bentler test statistic as it is robust against the violation of normal distribution [39]. We did not need to compensate for missing data via the preregistered full information maximum-likelihood approach as no missing data occurred in our outcome measures.

To evaluate the quality of model fits, a range of fit indices were inspected, including the comparative fit index (CFI), the Tucker-Lewis index (TLI), the root-mean-square error of approximation (RMSEA), and the standardized root-mean-square residual (SRMR). In line with Hu and Bentler [40] and Little and Kline [41], cut-off values for good model fit were CFI >0.90, TLI >0.90, SRMR <0.09, and RMSEA <0.08. Cut-offs for excellent model fit were CFI >0.95, TLI >0.95, SRMR <0.08, and RMSEA <0.06.

Our global alpha level was 0.05. We did not need to correct p values for multiple testing using the preregistered Hochberg’s false discovery rate ([42]) as our p values were close to zero.

We first assessed configural and metric measurement invariance (for details, see online suppl. material). We may cautiously assume a certain extent of metric invariance and further addressed this issue in the discussion.

As this study focused on the relation and overall scores of the two constructs under study rather than the measurement procedure, pathway models were tested, comprising four continuous manifest variables: (1) the sum scores of the EPDS at T1 and T2, as well as (2) the sum scores of the PSS at T1 and T2. These four sum scores represent the severity of depressive symptoms and levels of perceived stress, respectively. As main analysis, a cross-lagged panel design was employed. Following Burić et al. [43], four competing models to assess unidirectional and/or bidirectional relations among the severity of depressive symptoms and perceived stress were tested and compared [44]:

• 1.

  Stability model, including auto-regressive effects (β₃ and β₄)

• 2.

  Pathway model, including auto-regressive effects and the cross-lagged effect from the severity of depressive symptoms to perceived stress (β₁, β₃, and β₄)

• 3.

  Reverse pathway model, including auto-regressive effects and the cross-lagged effect from perceived stress to the severity of depressive symptoms (β₂, β₃, and β₄)

• 4.

  Reciprocal model, including auto-regressive and all cross-lagged effects (β_1–4).

The correlations of the two manifest variables assessed at the same time point (r₁ and r₂) were specified in models 2–4. The reciprocal model is a saturated model, inherently producing a perfect model fit which is of little statistical use. Still, it may be interpreted due to our strong theoretical foundation and can be compared to the other models. To determine the best-fitting model, models 2–4 were compared to the baseline model 1 by applying the Akaike information criterion (AIC). Models were regarded as considerably different with a difference of the AIC values between 4 and 7 and as essentially different with a difference of higher than 7, with the lower value model among models 2, 3, and 4 being accepted in such a case [45]. The first model was merely considered as a baseline model for the purpose of model comparison. Given that the “anova” function only allows to compare nested models, we compared models 1, 2, and 4 in a first step, and models 1, 3, and 4 in a second step.

To explore general and pandemic-specific covariates or control variables and to check the robustness of our model, we examined a range of variables, including the number of children, children’s and mother’s age, etc. (for a full list, see online suppl. material). All analyses were conducted using R, version 4.2.1 [46], applying the following main packages: “lavaan” [47], “semPlot” [48], “tidyverse” [49], “ggplot2” [50], and “raincloudplots” [51]. Moreover, we follow the Journal Article Reporting Standards [52].

According to the established cut-off sum score of 10 on the EPDS, 33.8% (95% CI: 30.3–37.5%) of mothers carried a high risk and 66.2% (95% CI: 62.5–69.7%) a low risk for depression at T1. At T2, 55.1% (95% CI: 51.3–58.8%) carried a high risk and 44.9% (95% CI: 41.2–48.7%) a low risk (χ² = 60.449, degrees of freedom [df] = 1, p < 0.001). At T1, 19.5% (95% CI: 16.7–22.7%) of mothers even reached or exceeded the cut-off value of 13, whereas 36.5% (95% CI: 32.9–40.2%) did so at T2. Assessing the numbers of mothers who changed their risk status from T1 to T2, we found that a greater proportion of mothers changed from the non-risk to the risk group (27.0%) than from the risk to the non-risk group (5.7%). Moreover, 39.2% carried a low risk for depression at both time points, whereas 28.1% carried a high risk at both time points, respectively. The change in proportions is corroborated by a significant result of the McNemar’s χ² test (McNemar’s χ² = 92.495, df = 1, p < 0.001).

According to applied cut-off sum scores of <14 (low), 14 to 26 (moderate), and ≥27 (high) on the PSS, 15.2% (95% CI: 12.6–18.1%) of mothers showed a high level, 55.9% (95% CI: 52.1–59.6%) a moderate level, and 29.0% (95% CI: 25.7–32.5%) a low level of perceived stress at T1. At T2, 26.0% (95% CI: 22.8–29.4%) demonstrated a high level, 57.8% (95% CI: 54.0–61.5%) a moderate level, and 16.2% (95% CI: 13.6–19.2%) a low level of perceived stress. Thus, moderate stress levels remained roughly equal across both time points. The percentages of low and high levels of stress, on the other hand, almost switched to the opposite from T1 to T2 (χ² = 43.146, df = 2, p < 0.001). Assessing the numbers of mothers who changed their category from a low/moderate to a high level of perceived stress or vice versa, we found that a greater proportion of mothers changed from a low/moderate to a high level of stress (16.4%) than from a high to a low/moderate level of stress (5.6%), respectively. Additionally, 68.5% of mothers showed a low/moderate level of stress at both time points, whereas 9.6% showed a high level of stress at both time points. Again, the change in proportions is corroborated by a significant result of the McNemar’s χ² test (McNemar’s χ² = 35.507, df = 1, p < 0.001).

The paired t tests revealed a significant difference between the two time points for both outcome measures. As hypothesized, the overall severity of both depressive symptoms and perceived stress was significantly lower at T1 than at T2 (ps < 0.001). The overall mean values are presented in Table 2. The tests of normality according to Shapiro-Wilk [53] and Kolmogorov-Smirnov [54] do not support the assumption of normality. Thus, we additionally conducted Wilcoxon signed rank tests which revealed a significant difference for both outcomes as well (ps < 0.001). For the severity of depressive symptoms, the effect size was moderate (d = 0.57, 95% CI = [0.48, 0.65]). For perceived stress, it was small to borderline moderate (d = 0.49, 95% CI = [0.40, 0.57]). Figures 1 and 2 illustrate the distributions, box plots, as well as the single and the overall sum scores of both outcomes across time points.

Table 3 presents the model fit indices of all tested models. As intended and expected, testing the (1) stability model resulted in a poor model accuracy. The pathway and reverse pathway model, however, yielded an excellent fit according to CFI and SRMR, a good fit according to TLI, and a poor fit according to RMSEA, respectively.

The saturated reciprocal model can only produce a perfect model fit, but may still be compared to the other models via the AIC values. Comparing the (1) stability, (2) pathway, and (4) reciprocal model clearly identified the reciprocal model as the best-fitting model with the lowest AIC value (AIC_stability = 16,826, AIC_pathway = 15,790, AIC_reciprocal = 15,774; $χdiff,reciprocal−pathway2=14.57$⁠, p < 0.001; $χdiff,pathway−stability2=808.38$⁠, p < 0.001). Also, comparing the (1) stability, (3) reverse pathway, and (4) reciprocal model clearly identified the reciprocal model as the best-fitting model (AIC_stability = 16,826, AIC_pathway = 15,791, AIC_reciprocal = 15,774; $χdiff,reciprocal−reverse2=19.19$⁠, p < 0.001; $χdiff,reverse−stability2=770.99$⁠, p < 0.001). Thus, in comparison with the AIC value of the reciprocal model, all other AIC values were larger by a difference of at least 16 (i.e., clearly above the preregistered cut-off of 7). Hence, the reciprocal model may be regarded as the best-fitting model (see Fig. 3). The significant standardized partial regression weights β₁₋₄ point in the expected positive direction (ps < 0.001). Thus, a greater severity of depressive symptoms at T1 predicted greater levels of perceived stress at T2 and a greater severity of depressive symptoms at T2. Equally, greater levels of perceived stress at T1 predicted a greater severity of depressive symptoms at T2 as well as greater levels of perceived stress at T2. Both variables highly and positively correlated within time points. The significant cross-lagged effects of the same size (β₁, β₂) affirmed our assumption that the severity of depressive symptoms and perceived stress is reciprocally related over time, even when controlled for moderate auto-regressive effects (β₃, β₄) and large correlations within time points (r₁, r₂). The respective R² indicated that 38.9% of the variance of perceived stress at T2 and 40.6% of the variance of the severity of depressive symptoms at T2 were explained by the reciprocal model.

A detailed description of our correlational analyses and robustness check of our main model may be found in the online supplementary material. As the following variables correlated with either one or both of our outcome measures, we included them as additional predictors in our exploratory reciprocal models: (1) number of children, (2) school degree, (3) income, and (4) change in mothers’ job situation. The regression weights of these additional predictors, however, were very small or even negligible in all exploratory models.

The present study investigated depressive symptomatology and perceived stress of mothers with young children (0–5 years, most of them being under 3 years of age) during the COVID-19 pandemic. We had three major goals: first, we assessed the overall severity of maternal depressive symptoms and perceived stress at two time points during the pandemic and reported prevalence rates for these two constructs. Second, we compared the respective severity of depressive symptoms and perceived stress between the two time points, and, third, we examined the reciprocal relation between these two constructs.

Regarding our first and second goal, we may summarize that about a third of the mothers in the present study carried a high risk for depression during a time of more lenient confinement measures from May to November 2020 and that more than half of the mothers showed a high risk for depression during a time of stricter confinement measures in February/March 2021. Based on the stricter EPDS cut-off of 13, 19.5% at T1 showed a high risk of depression as compared to 36.5% at T2. Concerning perceived stress, about 15% of the mothers at T1 and about a quarter of the mothers at T2 showed high levels. Our analyses revealed that these prevalence rates significantly differed between the two time points for both measures, respectively. Examining the mean difference of both measures revealed that the severity of depressive symptoms and perceived stress significantly worsened from T1 to T2. The effect size of this change was moderate for the increase of depressive symptom severity (d = 0.57) and small (or borderline moderate) for the increase of maternal perceived stress (d = 0.49). Furthermore, a larger proportion of mothers changed their risk status from being at a low risk for depression at T1 to being at a high risk at T2 (27.0%) than vice versa (5.7%) and the same pattern was observed for the levels of perceived stress (16.4% vs. 5.6%, respectively).

When comparing our prevalence rates of a risk for depression during early motherhood to other studies conducted during the COVID-19 pandemic, our rates are very much in line with the meta-analysis by Chen et al. [12] in which they included studies on postpartum depressive symptomatology from April 2020 to November 2021. They reported a pooled prevalence rate of 34%, only including studies that used the stricter EPDS cut-off of equal or above 13. Applying the same cut-off, we obtained a very similar prevalence of 36.5% in February/March 2021 (T2). On a descriptive level, studies included in the meta-analysis which were published in 2020 tended to have smaller prevalence rates than studies published in 2021. Another meta-analysis by Yan et al. [13] investigated postpartum depressive symptoms from the beginning of the pandemic up to September 2020, which mostly reflects our T1 time point. This meta-analysis revealed a similar pooled prevalence rate of 22% compared to our finding of 19.5% at T1. Thus, our findings regarding depressive symptoms in a German sample of mothers are in line with prevalence rates in other countries at different stages of the pandemic.

Applying the more common cut-off of equal or above 10 reveals the disturbing finding that our prevalence rate is almost twice as high at T2 (55.1%) and by 10.2 percentage points higher at T1 (33.8%) than the prevalence of 23.6% found in a pre-pandemic, representative German sample by Reck et al. [55]. Their 95% CI of 20.8–26.5% does not overlap with neither one of ours at T2 (51.3–58.8%) nor at T1 (30.3–37.5%), rendering a significant difference between our prevalence rates of both measurement points during the pandemic versus pre-pandemic prevalence rates quite likely. Another pre-pandemic prevalence rate of 17% found in a similarly large German study [56] was even lower than the one by Reck et al. [55], highlighting the troubling number of mothers at risk for depression during a time of pandemic crisis in the present study.

When comparing our prevalences of high perceived stress levels to the current literature, we found a similar pattern as Adams et al. [37] and Hiraoka and Tomoda [57]: stress levels differed significantly between the 2 time points, with higher stress levels at the time point when child care facilities and schools were closed. On a descriptive level, the prevalences of high stress levels are slightly higher than those reported by Adams et al. [37]. In their study, they assessed parents (94.5% female) of children aged 5–18 years in April and May 2020 and September 2022. They reported a prevalence of 15.2% versus 12.2% during a time of more lenient confinement measures and 26.0% versus 22.4% during a time of stricter confinement measures and stay-at-home rules. Thus, the mothers of younger children in the present study appear to be even more burdened than those of older children which has also been suggested by Hübener et al. [3] and Kowal et al. [4]. Moreover, during a lockdown in the second year of the pandemic mothers might have been even more worn out than during the first lockdown as reflected by increasingly higher rates of depressive symptoms [12]. The meta-analysis by Mahmud et al. [58] investigated stress levels in the general population during the first months of the pandemic in 2020. They report a pooled prevalence of 29.4% during the lookdown period of March and April 2020 and of 5.1% during the time of fewer confinement measures in May and June 2020. Our prevalence rate of 26.0% during a lockdown period in 2021 (T2) appears to be in a comparable range. For a time of fewer confinement measures, our prevalence of 15.2%, however, is considerably higher which might be explained by the generally higher challenges for mothers connected to the demands of child-rearing [7]. Comparing our perceived stress levels to pre-pandemic stress levels of mothers living in Germany [59] revealed a difference of more than one standard deviation (SD): M = 21.32 (SD = 7.27) versus M = 13.07 (SD = 6.08). Our mean at T1 lies between both values (M = 18.21, SD = 7.24). We might confidently conclude that the mothers’ levels of perceived stress increased significantly from T1 to T2 during the pandemic (d = 0.49) and very probably assume an increase compared to pre-pandemic levels of stress as highlighted by retrospective comparisons in Calvano et al. [18] and Adams et al. [37].

Concerning the increase in both depressive symptoms and perceived stress in our study, we may assume that it is due to the contextual differences between our two time points, i.e., the prevailing confinement measures. For mothers, however, these differing confinement measures translated into vastly different demands for child-rearing. Hence, the mothers’ individual experience of these additional demands and stressors is very likely the reason behind the observed increase in depression and stress levels over time [7].

Regarding our third goal, we examined the reciprocal relation between the severity of depressive symptoms and perceived stress of mothers with young children. The substantial size of the correlations within time points (r₁ = 0.77, r₂ = 0.69) is in line with previous findings (rs = 0.63–0.81 [21, 28, 35]), suggesting that the PSS may be regarded as capturing a state which puts people at risk for a clinical symptomatology although this state also belongs to a manifold set of feelings and states which characterizes clinical symptomatology. Despite this strong overlap of constructs, they still independently predicted each other as demonstrated by their significant standardized partial regression weights of the cross-lagged effects (see Fig. 3). We, thus, generated evidence for both the stress exposure model [24], i.e., preceding perceived stress predicted later severity of depressive symptoms, and the stress generation model [25], i.e., the preceding severity of depressive symptoms predicted later perceived stress. The strongest predictor of each depressive symptoms and perceived stress was each past depressive symptoms and perceived stress themselves as depicted by our auto-regressive effects, which is in accordance with past research [60]. Still, one of our cross-lagged effects (β₂ = 0.21) showed that perceived stress provided an additional contribution such that higher levels of preceding perceived stress predicted a higher severity of subsequent depressive symptoms. And the other cross-lagged effect (β₁ = 0.21) demonstrated that a higher severity of preceding depressive symptoms enhances the latter perception or appraisal of situations as stressful, i.e., unpredictable, uncontrollable, and overloading. To sum up, the size of our auto-regressive effects implies a certain stability of depressive symptoms and perceived stress across our two measurement points during the pandemic or a certain influence from the previous time point. The same size of the cross-lagged effects means that no potential causal predominance of either one of the variables is indicated and that no single variable may be clearly described as the source or the effect variable [61]. These findings, however, should be regarded as preliminary and interpreted cautiously as some methodological limitations need to be addressed (see limitations). We, therefore, cautiously conclude that the association between maternal depressive symptomatology and perceived stress may be best described as an evenly bidirectional or reciprocal link in which each may potentially cause increases in the other. Our exploratory analyses revealed that depressive symptomatology significantly correlated with school degree, income, and a change in the job situation. Perceived stress was significantly associated with the number of children, income, and a change in the job situation. Including these variables in exploratory reciprocal models led either to an unacceptable model fit or to only a minimal change in explained variance. These variables may, therefore, be valuable when looking into models of risk and protective factors, but not when looking into the dynamics between depressive symptoms and perceived stress.

To our best knowledge, this was the first study to assess the reciprocal relations between depressive symptomatology and perceived stress of mothers with infants and young children in the course of the COVID-19 pandemic. Regarding our prevalence rates, it is noteworthy that they were assessed in Germany where nearly all individuals, regardless of their income, have health care coverage and access to state-funded child care. These circumstances may probably have resulted in an underestimation of the prevalence rates, given that many pre-pandemic and pandemic studies have found higher prevalence rates in less economically developed countries [12, 55, 62]. Additionally, our sample was overall highly educated, living in a relationship, and rather well-off. Mothers in more dire economic circumstances and/or single mothers might have been even more burdened but perhaps were less likely to find the additional time to participate in the survey even if they got the invitation. Therefore, the “total” prevalence in the general population may be even larger as our sample does not fully represent the population of mothers of young children in Germany. On the other hand, our study may carry a bias by self-selection, implying that individuals suffering from a greater severity of depressive symptoms or higher levels of stress were more interested in taking part in the online survey. This may have eventuated in an overestimation of our prevalence rates. However, regarding the fact that no systematic drop-out was observed in the sense that less affected individuals lost interest in participating at T2 more often, this seems less likely. Furthermore, it should be noted that the assessment durations for T1 (4 months) and T2 (1 month) differed, which may have introduced variability in participant experiences due to changes in external conditions during these periods. This discrepancy could potentially affect the temporal consistency of the observed relationships in the CLPM but would very likely not affect the size of our auto-regressive and cross-lagged effects in any major way. Finally, some critical arguments against traditional CLPMs remain to be discussed. It is recommended to apply a model which can separate within-person effects from between-person associations by introducing a random-intercept for each construct which are allowed to correlate, i.e., a random-intercept cross-lagged panel model [63]. This model, however, requires at least three measurement points in order to be identified [64], rendering it technically not feasible for this study.

This study lent support to the hypotheses that the severity of depressive symptoms and perceived stress of mothers with young children (mostly aged 0–3) increased in the course of the COVID-19 pandemic, which is in line with current meta-analyses [11, 12]. A total of 33.8% of mothers were at risk for depression in Germany between May and November 2020, whereas a disturbingly high percentage of 55.1% were at risk in February/March 2021. Here, 15.2% of mothers between May and November 2020 and 26.0% in February/March 2021 suffered from high perceived stress levels. Altogether, this may put their children at risk for developmental issues, abuse, neglect, and domestic violence [19, 65]. As the severity of depressive symptoms and perceived stress seems to linearly depend on their own previous values and to reciprocally predict each other, psychosocial support should (1) focus on screening and treating mothers as early as possible to mitigate the risk for subsequent depressive symptoms and perceived stress, and (2) aim attention at both depressive symptom reduction and stress relief to most successfully promote maternal mental health.

We would like to thank all volunteers who participated in the online survey and all colleagues who contributed to this study.

The study had been approved by the independent Ethics Committee of the medical faculty, Ruprecht-Karls University, Heidelberg, in agreement with the Ludwig Maximilian University, Munich (vote: S-446/2017). Written informed consent was obtained from participants to participate in the study.

The authors have no conflicts of interest to declare.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

C.F.J.W.-W. and C.R. contributed to the conceptualization, methodology, writing – original draft, review, and editing, and visualization. C.F.J.W.-W., A.K.G.M., and S.M.K. ran the formal analysis. A.M., M.M., A.T., N.N., and A.-L.Z. contributed to the writing – review and editing. All authors critically revised the manuscript and approved the submitted version.

Christian F.J. Woll-Weber and Corinna Reck contributed equally to the paper.

The study design, hypotheses, and analysis plan were preregistered on the platform “AsPredicted” prior to data analysis (see https://aspredicted.org/h4zz5.pdf). All anonymized data, statistical analysis scripts, and research materials are publicly available on PsychArchives (see https://doi.org/10.23668/psycharchives.16062 and https://doi.org/10.23668/psycharchives.16061). An additional online supplement providing very detailed descriptions of hypotheses building and the analyses is available at the Open Science Framework (see https://osf.io/y2ths/). Further inquiries can be directed to the corresponding author.