health impact of the COVID-19 lockdown:
Evidence from the Netherlands.
A thesis presented for the degree of MSc Econometrics,
Operations Research and Actuarial Studies
1 Introduction 2
2 Literature review 3
3 Data 6
3.1 Summary characteristics 8
3.2 Gender, age and SES 10
3.3 Other mental health measures 12
3.4 Selection bias 12
3.4.1 Sample selection bias 12
3.4.2 Attrition bias 13 4 Methodology 14 4.1 Cross-sectional model 14 4.2 Base model 15 4.3 Probit model 15 4.4 Extended model 16
4.5 COVID-19 data only 16
5 Results 16
5.1 Cross-sectional analysis 16
5.2 Panel analysis base model 17
5.2.1 Gender 17
5.2.2 Time, age and SES 20
5.3 Other results 20
5.3.1 Three-way gender interactions 20
5.3.2 Probit model 22
5.3.3 Extended model 22
5.3.4 COVID-19 data only 24
6 Conclusion 24
7 Discussion 25
8 Acknowledgments 26
9 Appendix 32
9.1 Sample characteristics 32
9.5 Results cross-sectional model 37
9.6 Results base model 39
health impact of the COVID-19 lockdown:
Evidence from the Netherlands.
Author: Alinde Vloo
Abstract
Background: Recent contributions highlighted gender differences in mental health con-sequences of COVID-19 related lockdown measures. However, their cross-sectional designs cannot differentiate between pre-existing gender differences and differences induced by the lockdown. Using longitudinal data with repeated mental health measurements throughout the lockdown, we overcome this caveat.
Methods: Data from the Lifelines biobank and cohort study in Northern Netherlands was used, including the MINI for diagnosing psychiatric disorders of depression and anxiety using DSM-5 criteria. Models were estimated using fixed effects and probit regression. Results: Significant gender differences in mental health during the lockdown were found, where women experience more depression issues and men experience more anxiety issues stemming from the lockdown. Further significant gender differences are: young men ex-perience more anxiety disorders stemming from the lockdown than young women; and the effect of being married, highly educated, or having low income on depression is significantly worse for women in lockdown than for men.
Interpretation: Policymakers need to keep in mind that the COVID-19 lockdown has different effects on mental health for men and women, and, therefore, men and women might need different forms of help when dealing with depression and anxiety issues stem-ming from the lockdown. Furthermore, policymakers need to be aware of the fact that married, highly educated and low income women experience significantly more depression issues than men in these categories, in order to provide extra support for women in these categories, e.g. by providing free treatment or medication for mental health issues. Funding: ZonMW Corona Fast-Track Grant.
JEL classification: I10
1
Introduction
By October 15, 2020, worldwide 38,394,169 people were infected with the coronavirus disease 2019 (COVID-19). “Its high infectivity, combined with the susceptibility of unexposed popu-lations to a new virus, creates conditions for rapid community spread.” (Alwan et al. (2020), p.71). In the Netherlands, 195,933 confirmed COVID-19 cases existed with 6,654 deaths by Oc-tober 15, 2020. Over five months earlier, on March 31, 11,750 confirmed COVID-19 cases were present in the Netherlands, and 864 deaths (WHO (2020b)). To address this critical situation, various restrictions were imposed and the country was in lockdown.
Although these restrictions were necessary to limit the occurrence of new COVID-19 cases and deaths, prolonged home confinement impacts both physical and mental health (Wang et al. (2020); WHO (2020a)). Potential results of quarantine include depressive symptoms, anxiety, insomnia and acute stress disorders (Brooks et al. (2020)). Holmes et al. (2020) state: “It is already evident that the direct and indirect psychological and social effects of the coronavirus disease 2019 (COVID-19) pandemic are pervasive and could affect mental health now and in the future.” (p. 547).
occurrence of major depressive disorders and the occurrence of generalized anxiety disorders, which are defined using the Mini International Neuropsychiatric Interview (MINI) included in the Lifelines questionnaires, and using DSM-5 criteria (Kernberg (2013)). Next, we inspect how age, SES, marital status and children are related to gender and mental health, since evi-dence suggests that age and SES are important factors of risk during the COVID-19 pandemic (Ji Kang and In Jung (2020); Khazanchi et al. (2020)), and marital status and children are related to the state of mental health (Gualano et al. (2020)).
The first lockdown in the Netherlands started on March 16 when places including, but not limited to, schools, sporting facilities, restaurants and cafe’s were closed. The first COVID-19 questionnaire of Lifelines was sent out two weeks after the start of the lockdown (March 30). Until the week of May 18, questionnaires were sent out weekly, and, thereafter, bi-weekly. The first COVID-19 lockdown ended on the first of June when facilities as restaurants were allowed to reopen under strict conditions, elementary schools were allowed to reopen on May 11. During the lockdown, most people had to work from home and children did not go to school. This situation is likely to cause more mental health issues, especially for women (Hamadani et al. (2020)). Even though the first COVID-19 lockdown ended, strict conditions regarding the re-openings were still in play.
It is worth noticing that the Northern Netherlands had very few reported COVID-19 cases dur-ing the first lockdown. Therefore, our analysis is able to highlight the impact of the lockdown itself, i.e. without the presence of COVID-19. Note that the lockdown in the Netherlands is different from lockdown measures in other countries, since only those in contact with COVID-19 had to be quarantined. For example, the lockdown in Italy forced everyone to stay home with exceptions for necessity, work and health circumstances.
In the next section, we review the existing literature concerning mental health in the COVID-19 pandemic. In Section 3 we describe the data we used and address potential selection bias problems. Section 4 shows the empirical analysis, from which the corresponding results can be found in Section 5. Finally, in Section 6 we provide the conclusions drawn from the results and in Section 7 we discuss limitations of this study and recommendations for further research.
2
Literature review
Most of the current research studying mental health consequences of COVID-19 lockdown mea-sures is based on cross-sectional data settings. We summarized the main results of the following literature review for various countries in Table 1.
people are more prone to having depression and anxiety issues. Furthermore, women indicated to have significantly more problems regarding anxiety than men. A similar study to the effects of the lockdown on mental health in Italy was performed for the first weeks of the lockdown by Cellini et al. (2020). Their results indicated a similar percentage for depression symptoms, but higher percentages for anxiety symptoms. Mazza et al. (2020) provided evidence suggest-ing a higher percentage (32.4%) for people with depression symptoms, but a lower percentage (18.7%) for people with anxiety symptoms in Italy. They also found that women reported more problems with both depression and anxiety symptoms.
The results of these studies are in line with studies in other countries. Research in China indi-cates depression symptoms between 16.5% and 37.1% of the population and anxiety symptoms between 12.9% and 35.1% of the populations (Ahmed et al. (2020); Huang and Zhao (2020); Lei et al. (2020); Wang et al. (2020)). Ahmed et al. (2020) and Huang and Zhao (2020) pro-vided evidence showing that young people are more prone to mental health related issues. Both Ahmed et al. (2020) and Huang and Zhao (2020) reported no significant difference between men and women for depression and anxiety problems, whereas Wang et al. (2020) found that women reported significantly higher levels of depression and/or anxiety issues.
Research in the United Kingdom on mental health during the COVID-19 pandemic showed a significant difference between men and women in level of anxiety issues, whereas no significant difference was found in the level of depression symptoms (White and Van der Boor (2020)). It is worth noticing that the United Kingdom reports less occurrence of depression and anxiety than the other countries.
Spain provided evidence showing 18.7% indicated to have depression issues and 21.6% indi-cated to have anxiety issues (Gonz´alez-Sanguino et al. (2020)), and found that women as well as younger people (between age 18 and 39) were more prone to depression and anxiety issues than men and people older than age 39.
Similar research in Austria is in line with these results: Pieh et al. (2020) did research on mental health in Austria during the COVID-19 restrictions and found that women and young adults (under age 35) indicate lower mental health. Pieh et al. (2020) found that 21% of the partici-pants showed symptoms of depression, and 19% of the participartici-pants showed symptoms of anxiety.
Table 1: Summary of results of various papers.
Country: Paper Depression Anxiety
Frequency (%) Gender differences Age dif-ferences Frequency (%) Gender differences Age dif-ferences Italy:
Table 1 continued: Summary of results of various papers.
Cellini et al. 24.2 - - 32.6 -
-Mazza et al. 32.4 Yes - 18.7 Yes Yes
China:
Ahmed et al. 37.1 No Yes 29.0 No Yes
Huang and Zhao 20.1 No Yes 35.1 No Yes
Lei et al. 22.4 Yes Yes 12.9 Yes Yes
Wang et al. 16.5 Yes - 28.8 Yes
-UK:
White and Van der
Boor 7.6 No - 10.2 Yes
-Spain: Gonz´
alez-Sanguino et al. 18.7 Yes Yes 21.6 Yes Yes
Austria:
Pieh et al. 21.0 Yes Yes 19.0 Yes Yes
As for SES, due to the COVID-19 pandemic, unemployment is increasing and people in a low-income category are hardest hit by the pandemic (Burstr¨om and Tao (2020)). Ettman et al. (2020) investigated the prevalence of depression symptoms in the United States before and dur-ing the COVID-19 pandemic. They found that durdur-ing the COVID-19 pandemic, prevalence of depression symptoms was more than three times higher than before the COVID-19 pandemic. They show that individuals with lower income and less than $5000 savings had more depression symptoms during the COVID-19 pandemic than others. Research in Austria also shows that people without work and people with low income are more prone to mental health issues (Pieh et al. (2020)).
It is also worth noticing that Gualano et al. (2020) showed that marital status and children are related to the state of mental health.
More studies like the ones mentioned above can be found for other countries (e.g. Ahmad et al. (2020); Elmer et al. (2020); Moghanibashi-Mansourieh (2020); Moreira et al. (2020); Rossi et al. (2020); Ueda et al. (2020); Zhang et al. (2020)), but most researchers agree: Women, young people and people with low SES report more problems regarding mental health issues as depression and anxiety under COVID-19 restrictions.
SES are generally good predictors for a subject’s mental health outcome (Angelini et al. (2019); Callander (2016); De Graaf et al. (2012); Gao et al. (2020); Kiely et al. (2019); Ritchie and Roser (2018); Silbersdorff and Schneider (2019)). For example, Kiely et al. (2019) provided evidence that women generally have significantly worse mental health when assessed in terms of symptoms of depression and anxiety than men, and De Graaf et al. (2012) show that young subjects (18 to 24 years) have higher odds of getting mental health problems than older sub-jects. Furthermore, Callander (2016) researched pathways between health and education, and health and income, finding health to be significantly related with income and level of education achieved.
In order to overcome the caveats of existing literature, we use panel data to distinguish between gender effects stemming from the COVID-19 lockdown and gender differences in mental health outcomes in general. For both measures, i.e. depression and anxiety, we are interested not only in the continuous measure, but also in a binary measure, which indicates who would be classified as having serious depression and/or anxiety issues according to a professional. Rose et al. (2008) stresses the importance of using both measures, i.e. continuous and binary, since these are highly related to each other. Obviously, a shift in the whole distribution of the popu-lation implies that the extreme values of these popupopu-lation also change. Therefore, reducing the number of extremes, in our case the number of people classified as having serious depression and/or anxiety problems, can be achieved by reducing the population average. Rose et al. (2008) state: “Moderate and achievable change by the population as a whole might greatly reduce the number of people with conspicuous problems” (p. 103).
3
Data
Data from the Lifelines study will be used. Lifelines is a multi-disciplinary prospective population-based cohort study examining in a unique three-generation design the health and health-related behaviors of 167,729 persons living in the North of the Netherlands. It employs a broad range of investigative procedures in assessing the biomedical, sociodemographic, behavioral, physical and psychological factors which contribute to the health and disease of the general population, with a special focus on multi-morbidity and complex genetics.
During the COVID-19 pandemic, the participants of the Lifelines study, aged 18 or older, received (bi-)weekly questionnaires regarding, among other things, their physical and mental health, well-being and lifestyle. On March 30, 2020, adult Lifelines participants were invited to participate in the first round of the COVID-19 questionnaire, after this, questionnaires were sent out (bi-)weekly. For more details on the Lifelines COVID-19 cohort, we refer to Mc Intyre et al. (2020).
wave of data before COVID-19 as starting point.
The main variables of interest, i.e. the measures of mental health, are the number of depression and anxiety symptoms and the occurrence of major depressive and generalized anxiety disor-ders. These symptoms are asked by means of the Mini International Neuropsychiatric Interview (MINI) in the COVID-19 questionnaires. We use the DSM-5 specification to determine depres-sion and anxiety based on certain symptoms (Kernberg (2013)). Major depressive disorders require five or more of the following symptoms to be present during a two week period, from which at least one of the symptoms is either a depressed mood or loss of interest:
1. Depressed mood most of the day, nearly every day.
2. Markedly reduced interest in (almost) all activities most of the day, nearly every day. 3. Significant weight loss or gain when not dieting or increase or decrease in appetite nearly
every day.
4. Insomnia or hypersomnia nearly every day.
5. Psychomotor agitation or retardation nearly every day. 6. Loss of energy nearly every day.
7. Feelings of guilt or worthlessness nearly every day.
8. Diminished ability to concentrate or indecisiveness nearly every day.
9. Recurrent thoughts of death, recurrent suicidal thoughts or suicidal attempts.
For the number of depression symptoms we count the symptoms provided that one of the symp-toms is either a depressed mood or loss of interest.
Similar criteria apply to generalized anxiety disorders. Generalized anxiety disorders require excessive anxiety and worry occurring more days than not for at least six months and the indi-vidual finds it difficult to control this anxiety or worry. Additionally, the anxiety is associated with at least three of the following symptoms:
1. Restlessness.
2. Being easily fatigued. 3. Difficulty concentrating. 4. Irritability.
5. Muscle tension. 6. Sleep disturbance.
For the number of anxiety symptoms we count the symptoms provided that the individual has excessive anxiety and finds it hard to control this for at least six months. For more details on the DSM-5 criteria, we refer to Kernberg (2013).
Age is determined as current age of the individuals, i.e. the age categories are based on age during the COVID-19 pandemic. Note that age is a time invariant variable. We calculated current age as 2020 (current year) minus the year of birth of the individual. Age categories are: 18-30 years old, 31-50 years old, 51-65 years old, and 66 years old and older. It is worth noticing that students are underrepresented in our sample, meaning that only a small fraction of the sample belongs to category 18 to 30 years old.
For SES we used two different measures: the highest level of education achieved and net income per month. The categories for education are based on the last assessment before the COVID-19 pandemic started, since no such information is available during the COVID-COVID-19 pandemic. However, income is categorized according to the mean of two moments during the COVID-19 lockdown at which net income per month was available, time period 8 and 10. For highest education achieved we distinguish four categories: none or primary education; lower secondary vocational or junior general secondary education; secondary vocational or senior general sec-ondary education; and higher vocational or university education. For net income per month we distinguish five categories: less than e1000,-; between e1000,- and ; between e2000,-and e3000,-; between e3000,- and e4000,-; and more than e4000,-. Note again that highest education achieved and net income per month are time invariant variables.
Marital status is also a time invariant variable since it was only asked before the COVID-19 lockdown. Marital status categories are: married; registered partnership; in relationship, but living apart; single; and other. For the number of children at home we distinguish between children aged between zero and 12 years old, and between 13 and 18 years old. These variables are time varying since these were asked various times during the COVID-19 pandemic. We have categories: zero children; one or two children; three or four children; and five or more children.
3.1
Summary characteristics
We calculated summary statistics based on our sample as a whole, the sample of respondents and the sample of non-respondents of the COVID-19 questionnaires. These characteristics can be found in Table 4 in Appendix 9.1. From the last column, i.e. characteristics from the sam-ple as a whole, we observe that in total 134,434 individuals were invited to participate in the COVID-19 questionnaires. Almost 3% reported to have major depressive disorders and over 5% reported to have generalized anxiety disorders. The mean number of depression and anxiety symptoms are 0.27 and 0.32 before the COVID-19 outbreak, respectively.
More interestingly, respondents have less depression and anxiety symptoms and less major de-pressive and generalized anxiety disorders, and rate their health better. This indicates that we face a selection problem, which we will discuss in more detail in section 3.4.
The mean number of depression and anxiety symptoms and disorders are represented graphi-cally in Figures 1 and 2, respectively. In these figures, and in the next figures, we use only the individuals who answered at least one of the COVID-19 questionnaires additional on having answered the last available questionnaire before the COVID-19 pandemic. Time point zero indicates the last wave before the lockdown started. It is worth noticing that this is gathered in years 2014 to 2017, i.e. between three and six years before the COVID-19 questionnaires. Since this is a long time ago, we also investigate results without this initial time period. After time zero, each COVID-19 questionnaire is considered a different point in time starting at time two. In these figures, as well as in the next figures, the vertical red line indicates the end of the lockdown, i.e. June 1, and the vertical gray line indicates when elementary schools reopened, i.e. May 11. It is worth noticing that the percentage reporting major depressive disorders or generalized anxiety disorders during the lockdown is much lower than what we observed for other countries in the literature review. This might be partly due to differences in forms of measures of depression and anxiety since multiple measures are available, examples include PHQ-2 and/or GAD-2 (Gualano et al. (2020)), DASS-21 (Cellini et al. (2020); Mazza et al. (2020); Wang et al. (2020)), SDS and SAS (Lei et al. (2020)) for depression and anxiety screening respectively. However, these screenings are not very different from the screening we used, meaning that this cannot fully explain the difference. Note, however, that our findings of depression and anxiety before the COVID-19 lockdown are in line with what has been found in the Netherlands by Ritchie and Roser (2018), and what has been found by Alonso et al. (2004) who used DSM-4 criteria.
Surprisingly, we observe that both the mean number of depression and anxiety symptoms and disorders decrease when the lockdown starts and never return to their initial high state before the lockdown. As mentioned above, time point zero is measured at least three years before the outbreak of COVID-19. In these years a lot might have happened that explains this sud-den drop of depression and anxiety symptoms and disorders. Furthermore, since self-rating instruments were used to measure depression and anxiety, the sudden drop might be, partly, explained by social desirability bias since the COVID-19 pandemic is widely discussed (Ahmed et al. (2020)). As said, we will also briefly investigate results without the initial time period and investigate the differences.
.1
.15
.2
.25
.3
Mean number of symptoms
0 4 8 12
Time
Depression Anxiety
Depression and anxiety symptoms
.01
.02
.03
.04
.05
Mean number of disorders
0 4 8 12
Time
Depression Anxiety
Depression and anxiety disorders
Figure 1: Depression and anxiety symptoms. Figure 2: Major depressive and generalized anxiety disorders.
Similar graphs were made using a balanced panel, i.e. with only those who answered each questionnaire, which resulted in similar patterns. Furthermore, for each different depression and anxiety symptom we also made similar graphs to observe if obvious differences between symptoms were present, which were not.
In the next section we will address how various subgroups experience the lockdown. Even though we are mainly interested in differences between gender, we also show graphs for age categories and SES categories since these variables will be included in our model and will also be used to find differences between men and women in age and SES categories.
3.2
Gender, age and SES
We can distinguish depression and anxiety symptoms and disorders for various subgroups of the population. We only consider the number of depression symptoms in this subsection, however, in Section 3.3 we will investigate similarities and dissimilarities in patterns of the subgroups between the number of depression symptoms and the other mental health measures: number of anxiety symptoms, major depressive disorders and generalized anxiety disorders.
Next to a distinction in gender, we can also distinguish mental health problems by age. As mentioned, we use age categories based on the current age of the subjects. Figure 4 shows the mean number of depression symptoms by various age categories. We clearly observe that the higher the age category, the lower the mean number of depression symptoms, as we would expect from the literature review.
.1
.15
.2
.25
Mean number of depression symptoms
0 4 8 12
Time
Male Female
Depression symptoms by gender
Figure 3: Depression symptoms by gender.
.1
.2
.3
.4
Mean number of depression symptoms
0 4 8 12
Time
Age 18-30 Age 31-50 Age 51-65 Age 66 and older
Depression symptoms by age
Figure 4: Depression symptoms by age.
The mean number of depression symptoms can also be distinguished by subjects’ SES. As mentioned, we use both the highest level of education achieved and net income per month as SES measures. Figure 5 and Figure 6 show the mean number of depression symptoms by education and income categories, respectively. In general, we observe that the higher the SES, the less depression symptoms are present. This is in line with our expectations obtained from the literature review. Interestingly, subjects with none or primary education have a much steeper decrease in the number of depression symptoms than higher educated at the start.
.1
.2
.3
.4
.5
Mean number of depression symptoms 0 4 8 12 Time
None or primary education
Lower secondary vocational or junior general secondary education Secondary vocational or senior general secondary education Higher vocational or university education
Depression symptoms by education
Figure 5: Depression symptoms by highest level of education achieved.
0
.1
.2
.3
.4
Mean number of depression symptoms
0 4 8 12
Time
Less than 1000 Between 1000 and 2000 Between 2000 and 3000 Between 3000 and 4000 More than 4000
Depression symptoms by income (in euros)
3.3
Other mental health measures
Up to this point, we only considered the number of depression symptoms as mental health measure. The same graphical analysis as in Section 3.2 has been performed for the number of anxiety symptoms, as well as the occurrence of major depressive disorders and generalized anxiety disorders. The figures regarding these measures can be found in Appendix 9.2. Sim-ilar patterns for each of the subgroups were found for these measures, but we observe certain differences between depression and anxiety. The most important difference concerns gender: For the number of anxiety symptoms as well as the occurrence of generalized anxiety disorders, we observe that the difference between men and women was smaller during and after the first COVID-19 lockdown than before, whereas this difference is larger during and after the lockdown for the number of depression symptoms and the occurrence of major depressive disorders. This finding suggests that, whereas the difference in depression symptoms and disorders between men and women becomes larger, the difference in anxiety symptoms and disorders between men and women becomes smaller. Another difference between depression and anxiety is that, whereas categories of education seem to behave similarly at the start for anxiety symptoms and disorders, categories of education behave differently at the start for depression symptoms and disorders where depression symptoms and disorders decrease faster for lowly educated subjects than for highly educated subjects.
Although the graphical analysis in Section 3.2 and this section provides compelling evidence that clear gender differences exist regarding the trend of mental health; some problems arise regarding the interpretation of this graphical analysis. In this analysis we do not account for potential bias, which we know exist since the mental health measures are significantly different for respondents and non-respondents as discussed before and as can be seen in Table 4. Further-more, the above analysis only considers mental health of various subgroups without accounting for other variables, i.e. when considering gender and mental health, we did not account for age and SES at the same time. Furthermore, variables such as whether an individual smokes, whether an individual had the COVID-19 infection, and so on, are also not accounted for. Fur-ther empirical research is needed to confirm that men and women react differently to the first COVID-19 lockdown in terms of mental health.
In the next subsection, we will consider potential problems stemming from selection bias.
3.4
Selection bias
3.4.1 Sample selection bias
generalized anxiety disorders are significantly different for respondents and non-respondents of the COVID-19 questionnaires. This is measured at time point zero, i.e. before the COVID-19 pandemic. We also observed that respondents were significantly more often ex-smokers and sig-nificantly less often current smokers, and that respondents have a sigsig-nificantly higher SES than non-respondents. Hence, we face a sample selection problem in the form of a healthy user bias, meaning that the respondents were healthier than non-respondents to begin with, i.e. before the lockdown even started. If sample selection bias is present, we do not have a random sample, meaning that the sample is not representative for the population and conclusions drawn from this sample are incorrect for the whole population. Furthermore, sample selection may also affect the estimation of relationships between variables which results in inaccurate estimates. A result of sample selection bias might be that we wrongly conclude for the population that mental health is not as bad as we expect during the COVID-19 pandemic, whereas, in fact, this conclusion is based on a sample with healthy subjects reporting less mental health problems than the overall population of which the sample is drawn.
We need to examine if this selection problem is still present when controlling for other character-istics. We perform a regression on these variables before the lockdown using the characteristics mentioned in Table 4 as independent variables (except, of course, the variables of depression and anxiety, which we use as dependent variables, and self-rated health). We include a variable indicating whether the individual responded to at least one of the COVID-19 questionnaires or not (Respondent). For each of the above mentioned dependent variables, we find insignificant coefficients for Respondent on a 5% significance level, meaning that we cannot reject the null hypothesis that the coefficient for Respondent is equal zero, indicating that the selection prob-lem vanishes when controlling for other characteristics. The results of variable Respondent can be found in Appendix 9.3 where we excluded all other independent variables for convenience. 3.4.2 Attrition bias
Other than sample selection bias, we might suffer from a different selection bias, namely attri-tion bias. Whereas sample selecattri-tion bias occurs before the study starts, attriattri-tion bias occurs when we have loss of participants during the study, i.e. individuals stop filling in COVID-19 questionnaires. Even if no selection bias is present at the first questionnaire, after some ques-tionnaires the respondents may differ significantly from subjects no longer responding. Attrition bias is present when differences exist between individuals leaving the study at some point and individuals who continue the study. In our case, attrition bias is present when, for example, individuals are too depressed to fill in the COVID-19 questionnaires on how depressed they are. Consequences are similar to the consequences of sample selection bias: The sample is no longer representative for the population and, hence, conclusions drawn using the sample cannot be applied to the population as a whole, i.e. the results are biased.
constructing dummy variables for each subject, equal to one if the individual filled out the COVID-19 questionnaire in this time period and the individual filled out a COVID-19 ques-tionnaire in a next period; and the dummy variable is equal to zero if the individual filled out the COVID-19 questionnaire in this time period, but then drops out of the sample. We include these dummy variables in Model 2 as described in Section 4. Next, we use an F -test to test joint significance of these variables. If these variables are jointly significant, we have evidence of attrition bias. Performing F -tests we obtain the P -values as given in Table 6 in Appendix 9.4. Fortunately, we do not suffer from attrition bias.
4
Methodology
In order to confirm the graphical analysis, which showed that women seem to experience neg-ative effects on depression from prolonged lockdown whereas the opposite is true for anxiety, we need to define appropriate models. The dependent variables of our models are the number of depression symptoms, the number of anxiety symptoms, the occurrence of major depressive disorders, and the occurrence of generalized anxiety disorders. For both the continuous and binary measures, we start by using a fixed effects approach. Note that time invariant variables such as gender, age category, education category and income category, will be omitted in the fixed effects model. Before diving into the main focus of this paper, i.e. effects of the COVID-19 lockdown on mental health differences in gender, we want to confirm what has already been found as discussed in the literature review by means of a cross-sectional analysis during the COVID-19 pandemic.
4.1
Cross-sectional model
Let us start by defining the model that we will use for the cross-sectional analysis for i = 1, . . . , N and t = 2, . . . , 12:
yit = x0itβ + ci+ uit, (1)
where i denotes individual subjects and t denotes the time period. Note that we only use the periods during the lockdown and discard the period before the lockdown in order to compare with findings in the literature review. We estimate Model 1 by means of pooled OLS. In Model 1, ci is the individual specific effect capturing unobserved heterogeneity. The vector xit
contains the regressors which are variables including time dummies and variables on gender, age categories, education categories and income categories. For the cross-sectional analysis we assume that ci + uit is the composite error term orthogonal to xit, with uit the idiosyncratic
4.2
Base model
Let us now consider the base model for the panel data approach. We define for i = 1, . . . , N and t = 1, . . . , 12:
yit = x0itβ + ci+ uit, (2)
where i denotes individual subjects, t denotes the time period, ci is the individual specific effect
which captures time invariant regressors (such as gender, age and SES) and is allowed to be correlated with xit, and uit is the idiosyncratic error. Again, the standard errors are clustered
at the individual level. The vector xit contains variables including time dummies and variables
on time periods interacted with gender, age categories, education categories and income cate-gories. We are mostly interested in the coefficients for time interacted with gender, from which we observe the additional gender differences in mental health due to the COVID-19 lockdown. It is worth noticing that t = 1 corresponds to the time period before the lockdown; t = 2 until t = 9 correspond to the time period during the first COVID-19 lockdown; and t = 10 until t = 12 correspond to the time period after the first lockdown ended. Note that t = 1 will be omitted to avoid multicollinearity.
We are interested in how differences in mental health between men and women evolve over time. For this reason, we are interested in the parameters corresponding to gender interacted with time. After estimating this base model, we expand the model by adding three-way inter-actions with gender, time and age; gender, time and SES; gender, time and marital status; and gender, time and number of children at home. In doing this, we can make further distinctions between men and women, i.e. we can answer questions like “Does a significant mental health difference between men and women exist in different age categories when a prolonged lockdown is present?” or “Have married women more mental health problems than married men stemming from the lockdown?”.
4.3
Probit model
correlation between ci and the independent variables. We define
ci = ¯xiγ + ωi, (3)
which we substitute in model 2, where ωi is the unobserved individual effect assumed to be
uncorrelated with xi = (xi1, . . . , xiT), xit is the vector of independent variables, and ¯xi is the
vector of means of time varying independent variables. Note again that we use standard errors clustered at the individual level.
4.4
Extended model
Next to the base model, we also estimate a model in which we account for other characteristics: marital status, the number of other adults living in the house, the number of children living in the house, the smoke situation of an individual (current, ex, or never), COVID-19 infection and COVID-19 infection in household, and the work situation (student, paid work, retired, unfit for work, unemployed but looking for work, or other such as maternity leave). Note that marital status is fixed since it was only asked before the COVID-19 lockdown and will therefore be omitted in a fixed effects regression. Again, we use standard errors clustered at the individual level.
4.5
COVID-19 data only
As also discussed in Section 3.1, data from time point zero, i.e. the last available wave before the lockdown started, is gathered between three and six years before the COVID-19 questionnaires. A lot happened in these years which might explain the sudden dive of mental health as seen in Figures 1 and 2, at the first time point during the lockdown. Results without this time period might be different from results including this period. Therefore, we use the base model to perform a similar analysis using only data gathered during the COVID-19 pandemic.
5
Results
5.1
Cross-sectional analysis
mental health status. Note again that these differences are not necessarily stemming from the COVID-19 lockdown, but are likely to exist even without the lockdown.
5.2
Panel analysis base model
Let us now consider the panel approach from which we can extract gender differences induced by the lockdown. We discuss the results obtained for model 2. From the previous section we already know that women face significantly more anxiety issues than men, which is in line with the literature review. Next, we want to determine if the COVID-19 lockdown affects these existing gender differences in mental health.
5.2.1 Gender
The results regarding the interaction terms between gender and time, as can be seen in Table 8 in Appendix 9.6, are represented graphically in Figures 7 and 8 for the number of depression symptoms. From Figure 7 it is clear that the effect of the time period on mental health differs by gender, as we also observed in Section 3. Let us now assess these differences, the results can be found in Figure 8, where the reference line at zero is added since we are looking for effects different from zero. Note again that the end of lockdown is at T ime = 9 and elementary schools reopened at T ime = 7.5. We observe that, compared to being in time period one, the effect of being in time period two on the number of depression symptoms is significantly larger for women than for men, ceteris paribus. Similarly for time periods five, six, eight and nine. Therefore, women have significantly more depression symptoms stemming from the COVID-19 lockdown than men. Note that this widens the existing structural gender gap.
.12 .14 .16 .18 .2 .22 Linear Prediction 1 2 3 4 5 6 7 8 9 10 11 12 Time Males Females
Predictive margins of Time#Gender
-.05
0
.05
.1
Effects on Linear Prediction
1 2 3 4 5 6 7 8 9 10 11 12
Time
Average Marginal Effects of female with 95% CIs
Figure 7: Average marginal effects of time on the number of depression symptoms by gender.
Similar graphs for the occurrence of major depressive disorders, the number of anxiety symp-toms and the occurrence of generalized anxiety disorders can be found in Figures 9 to 14. These results show that, firstly, no significant gender differences regarding the occurrence of major depressive disorders exist, and secondly that, compared to being in time period one, men show significantly more anxiety symptoms and generalized anxiety disorders in most periods after this than women, ceteris paribus. The above results indicate that women experience signifi-cantly more depression issues stemming from the COVID-19 lockdown, whereas men experience significantly more anxiety issues stemming from the lockdown, i.e. on top of the pre-existing gender differences in mental health.
.012 .014 .016 .018 .02 Linear Prediction 1 2 3 4 5 6 7 8 9 10 11 12 Time Males Females
Predictive margins of Time#Gender
-.01
-.005
0
.005
.01
Effects on Linear Prediction
1 2 3 4 5 6 7 8 9 10 11 12
Time
Average Marginal Effects of female with 95% CIs
Figure 9: Average marginal effects of time on the occurrence of major depressive disorders by gender.
Figure 10: Difference in gender of average marginal effects of time on the occurrence of major depressive disorders.
.08 .1 .12 .14 .16 Linear Prediction 1 2 3 4 5 6 7 8 9 10 11 12 Time Males Females
Predictive margins of Time#Gender
-.1 -.08 -.06 -.04 -.02 0
Effects on Linear Prediction
1 2 3 4 5 6 7 8 9 10 11 12
Time
Average Marginal Effects of female with 95% CIs
Figure 11: Average marginal effects of time on the number of anxiety symptoms by gender.
.015 .02 .025 .03 Linear Prediction 1 2 3 4 5 6 7 8 9 10 11 12 Time Males Females
Predictive margins of Time#Gender
-.02
-.015
-.01
-.005
0
Effects on Linear Prediction
1 2 3 4 5 6 7 8 9 10 11 12
Time
Average Marginal Effects of female with 95% CIs
Figure 13: Average marginal effects of time on the occurrence of generalized anxiety disorders by gender.
Figure 14: Difference in gender of average marginal effects of time on the occurrence of generalized anxiety disorders.
For each mental health measure, we also tested joint significance of the coefficients for gender interacted with time. We tested if the interactions are jointly significant in time periods two to nine (during the first COVID-19 lockdown), time periods 10 to 12 (after the lockdown), and time periods two to 12 (during and after the lockdown). The results can be found in Table 2, where we observe that whereas the coefficients for gender interacted with time are jointly significant on a 5% level for the number of depression symptoms and the occurrence of major depressive disorders during the lockdown and during and after the lockdown, the coefficients of gender interacted with time are jointly significant for the number of anxiety symptoms and the occurrence of generalized anxiety disorders after the lockdown ended.1 The
joint significance of gender interacted with time on depression symptoms during the lockdown means that women respond differently in terms of the number of depression symptoms than men during the COVID-19 lockdown. Similar conclusions apply to the other variables.
Table 2: Joint significance of interaction between gender and time dummies (on 5% level).
During After During and after
Depression symptoms Yes No Yes
Major depressive disorders Yes No Yes
Anxiety symptoms No Yes No
Generalized anxiety disorders No Yes No
1We also performed a fixed effects Poisson regression for the number of depression and anxiety symptoms,
5.2.2 Time, age and SES
Shortly considering the time dummies included in the model, we observe that none of them has a coefficient significantly different from zero on a 5% level, underlining our expectation from Section 3.1 that the lockdown duration itself has no significant impact on mental health. Let us now consider the variables regarding the age categories interacted with time. From Table 8, we barely find evidence regarding the effect of age on the number of depression symptoms and the occurrence of major depressive disorders. Slightly more evidence exist regarding the effect of age on the number of anxiety symptoms and disorders when inspecting joint significance of time and age during and after the first COVID-19 lockdown: Younger subjects react significantly worse in terms of mental health during and after the COVID-19 lockdown than older subjects. As for SES, we consider both the highest level of education achieved and the net income per month. SES measured by education provides no significant effects on mental health, but SES measured by income does provide some evidence suggesting that subjects in higher income categories react significantly worse in terms of mental health during and after the first COVID-19 lockdown, ceteris paribus, meaning that they get closer to the lowest income category in terms of mental health during the pandemic.
5.3
Other results
5.3.1 Three-way gender interactions
As discussed in the methodology section, we first expand the base model by adding three-way interactions between gender, time and age; gender, time and SES; gender, time and marital status; and gender, time and the number of children at home. The results of these three-way gender interactions are graphically represented in Appendix 9.7. Note again that the end of the lockdown is at T ime = 9 and elementary schools reopened at T ime = 7.5.
Firstly, let us consider the interaction terms of gender, time and age in Figures 27 to 30. Whereas significant differences between gender for the effect of age on the number of depression symptoms or disorders are not found for any time period, Figure 30 provides evidence that being men is related to more generalized anxiety disorders from time period 6 onwards in age category 18 to 30 years old, ceteris paribus. Hence, young men experience more anxiety dis-orders stemming from the COVID-19 lockdown than young women, which might be explained by the fact that young people and men tend to react worse to the lockdown than older people and women in general (Gebhard et al. (2020); Ji Kang and In Jung (2020); Khazanchi et al. (2020)).
women with higher achieved education (especially higher vocational or university education) experience more depression symptoms and disorders than men with this education in almost all periods during and after the first COVID-19 lockdown, ceteris paribus. Hence, highly educated women experience more depression symptoms and disorders from the COVID-19 lockdown than highly educated men. A potential explanation might be that highly educated women experience more pressure to find and keep a job than lower educated women (Bussemakers et al. (2017); Samarakoon and Parinduri (2015)), whereas, on top of the fact that the COVID-19 pandemic limits job opportunities, the odds of finding and keeping a job is smaller for women than for men in general (Batz-Barbarich et al. (2018); McGinn and Oh (2017); Petrongolo (2019)). This situation, i.e. wanting to find and keep a job with the knowledge that women are less likely to do this than men, might cause more mental health problems.
Inspecting gender interacted with time and income, we observe from Figures 35 to 38 that women in low income categories have significantly more depression symptoms and disorders than men in a low income category when in lockdown, ceteris paribus. Again, an explanation might be that, especially in low income categories, it is more difficult for women to find and keep a job than that it is for men (Batz-Barbarich et al. (2018); McGinn and Oh (2017); Petron-golo (2019)). For anxiety symptoms and disorders we find that women in the middle income category have significantly less anxiety symptoms and disorders than men in this category in all periods, ceteris paribus, i.e. prolonged lockdown is worse in terms of anxiety for men in the middle income class than for women in this class.
Investigating the interaction between gender, time and marital status in Figures 39 to 42, we observe that married women experience significantly more depression symptoms than married men from prolonged lockdown, ceteris paribus. An explanation might be that opposed to mar-ried men, marmar-ried women are generally mostly in charge of housekeeping and taking care of the children, besides any full-time or part-time job they might have, which might cause extra mental health problems when children are at home during the COVID-19 lockdown (Alon et al. (2020)).
Lastly, we inspect the interaction between gender, time and the number of children in the house, where we look at number of children aged between zero and 12, and at the number of children aged between 13 and 18. Both do not provide significant differences in mental health of gender over time, therefore, we omit these results.
Continuing with education in Figures 47 to 50 we observe that men with no or primary edu-cation are at greater risk of depression and anxiety issues during the pandemic compared to before, ceteris paribus. Furthermore, women with no or primary education are at greater risk of having anxiety issues, whereas women with higher vocational or university education are at greater risk of depression problems during the COVID-19 pandemic compared to before, ceteris paribus.
Inspecting income in Figures 51 to 54 we find that mostly those in high income categories are at greater risk of having bad mental health during the COVID-19 pandemic than before, ceteris paribus. Especially men in the lowest income category seem to bear minimal risk of bad mental health during the pandemic compared to before.
Most of these results regarding differences between categories are in line with what we observed in Section 5.2.2, since most results do not depend on gender and are therefore in line with the two-way interactions of time with age and SES. Next, we focus on marital status and number of children. For each of these variables, we do not observe clear differences for each category in mental health, therefore, the results are omitted.
5.3.2 Probit model
As mentioned, for the binary outcome variables, we also estimated a correlated random effects probit model because linear fixed effects might be unreliable since only a small fraction of the sample indicated to have a major depressive disorder or generalized anxiety disorder. We estimated both the average marginal effects of the dummy variables indicating time and the pairwise comparisons of the average marginal effects of the subgroups. Even though some small differences exist, overall, the average marginal effects and standard errors are very close to what we obtained under the fixed effects model. An exception is the time dummies, which have similar average marginal effects, but lower standard errors and are now significant. Since other results are very similar, we omit the results.
The tests for joint significance of time and gender produce similar results as with the fixed effects base model, underlining the conclusion that men and women differ significantly in how they react to the lockdown in terms of mental health.
5.3.3 Extended model
similar results as we found for the base model regarding anxiety symptoms and disorders, but the coefficients for gender interacted with time are no longer jointly significant for depression symptoms and disorders.
Table 3: Results extended model.
(1) (2) (3) (4)
Depr. symptoms Depr. disorder Anx. symptoms Anx. disorder
Number of children in house between 0 and 12 years old
-0.00274 -0.000848 0.0177 0.00317
(0.00960) (0.00121) (0.00997) (0.00190)
Number of children in house between 13 and 18 years old
0.0131 0.00197 0.00407 0.000211 (0.0129) (0.00197) (0.0124) (0.00220) Number of adults in house 0.00428 0.000428 0.00748 0.00131 (0.00530) (0.000604) (0.00806) (0.00130) Work situation Paid work 0.0664 -0.00178 0.0217 0.00323 (0.128) (0.0197) (0.131) (0.0207) Retired 0.0352 -0.00492 -0.0223 -0.00592 (0.130) (0.0199) (0.133) (0.0210)
Unfit for work 0.425∗∗ 0.0465∗ 0.338∗ 0.0528∗
Table 3 continued: Results extended model. Someone in house-hold has/had ... COVID-19 0.0226 0.00781 0.0257 0.00155 (0.0382) (0.00605) (0.0394) (0.00679) cons 0.141 0.0204 0.244 0.0381 (0.136) (0.0207) (0.139) (0.0218) N 223197 223197 223278 223340
Standard errors in parentheses, clustered at individual level.
∗p < 0.05, ∗∗ p < 0.01,∗∗∗ p < 0.001
5.3.4 COVID-19 data only
As mentioned, we also estimated the base model without time period one, i.e. the time period long before the lockdown. Although some changes occurred in the estimated coefficients, we obtain similar results as in Table 2 when testing joint significance for the coefficients for gen-der interacted with time. Hence, this confirms our previous results that the lockdown causes additional mental health differences between men and women.
6
Conclusion
In this panel data study, we investigated gender differences in the mental health impact of the COVID-19 lockdown, where mental health was measured by the number of depression symp-toms, the occurrence of major depressive disorders, the number of anxiety symptoms and the occurrence of generalized anxiety disorders using DSM-5 criteria. Our results indicate that women react significantly different in terms of mental health to the COVID-19 lockdown than men: Women react significantly worse to the lockdown in terms of depression symptoms and disorders, but significantly better in terms of anxiety symptoms and disorders than men. Hence, the effects of the first COVID-19 lockdown on mental health are worse for women regarding de-pression, but worse for men regarding anxiety. The fact that women experience more depression symptoms and disorders due to the lockdown, contributes to widening the existing structural gender gap (World Economic Forum (2020)). Ways to try to close this structural gender gap is to diminish (work) pressure for women by, for example, providing child day care near to or in the building that they work in, or equal wages for men and women who carry out the exact same job in order to close the gender wage gap (Burnette and Zhang (2019); Gharehgozli and Atal (2020)).
stemming from prolonged lockdown, whereas married women, highly educated women and women with low income, experience more depression problems from prolonged lockdown than married men, highly educated men and men with low income, respectively. An example to support low-income women who face mental health issues, is for policymakers to consider pro-viding free accessible treatment or medication for depression and anxiety, such that low-income women can increase their engagement in treatment to reduce depression and anxiety symptoms and disorders, which was also suggested by Levy and O’Hara (2010).
Considering that the COVID-19 lockdown is likely to continue for some time, the results can be used to better understand how men and women react to prolonged lockdown in terms of mental health, and these findings can be used to support decision making, i.e. the findings can be taken into account when issuing new restrictions and/or lockdown measures.
7
Discussion
When investigating mental health, we found that, as mentioned in Section 3.1, our sample shows a much lower percentage indicating to have a major depressive disorder or a generalized anxiety disorder during the COVID-19 lockdown than other countries we reviewed in Section 2. Even though this difference might, partly, be due to differences in forms of screening of depres-sion and anxiety, the difference is quite substantial. It is also worth noticing that self-rating scales were used to assess depression and anxiety according to DSM-5 criteria, this bears not only a risk of wrong judgment, but also the risk of social desirability bias since the COVID-19 pandemic is broadly discussed (Ahmed et al. (2020)). Further research might be necessary to identify exactly where these differences stem from and how these affect the results.
mental health to social restrictions and home confinement, and will provide insight in how to provide support to men and women.
8
Acknowledgments
References
Ahmad, A., I. Rahman, and M. Agarwal (2020). Early psychosocial predictors of mental health among Indians during coronavirus disease 2019 outbreak. Journal of Health Sciences. Ahmed, M.Z., O. Ahmed, Z. Aibao, S. Hanbin, L. Siyu, and A Ahmad (2020). Epidemic of
COVID-19 in China and associated Psychological Problems. Asian Journal of Psychiatry 51. Alon, T., M. Doepke, J. Olmstead-Rumsey, and M. Tertilt (2020). The Impact of COVID-19
on Gender Equality. National Bureau of Economic Research. Working Paper 26947.
Alonso, J., M. C. Angermeyer, S. Bernert, R. Bruffaerts, T. S. Brugha, H. Bryson, G. De Giro-lamo, R. De Graaf, K. Demyttenaere, I. Gasquet, J. M. Haro, S. J. Katz, R. C. Kessler, V. Kovess, J. P. L´epine, J. Ormel, G. Polidori, L. J. Russo, G. Vilagut, J. Almansa, S. Arbabzadeh-Bouchez, J. Autonell, M. Bernal, M. A. Buist-Bouwman, M. Codony, A. Domingo-Salvany, M. Ferrer, S. S. Joo, M. Mart´ınez-Alonso, H. Matschinger, F. Mazzi, Z. Morgan, P. Morosini, C. Palac´ın, B. Romera, N. Taub, and W. A. M. Vollebergh (2004). Prevalence of mental disorders in Europe: results from the European Study of the Epidemi-ology of Mental Disorders (ESEMeD) project. Acta psychiatrica Scandinavica 109 (420), 21–27.
Alwan, N.A., R.A. Burgess, S. Ashworth, R. Beale, N. Bhadelia, D. Bogaert, J. Dowd, I. Eck-erle, L.R. Goldman, T. Greenhalgh, D. Gurdasani, A. Hamdy, W.P. Hanage, E.B. Hodcroft, Z. Hyde, P. Kellam, M. Kelly-Irving, F. Krammer, M. Lipsitch, A. McNally, M. McKee, A. Nouri, D. Pimenta, V. Priesemann, H. Rutter, J. Silver, D. Sridhar, C. Swanton, R.P. Walensky, G. Yamey, and H. Ziauddeen (2020). Scientific consensus on the COVID-19 pan-demic: we need to act now. The Lancet 396 (10260), E71–E72.
Andrews, B. (2020). Second wave in the Netherlands may last through March, says RIVM. Dutch Review . Retrieved from: https://dutchreview.com/news/coronavirus-news/second-wave-in-the-netherlands-may-last-through-march-says-rivm/.
Angelini, V., D.D.H. Howdon, and J.O. Mierau (2019). Childhood Socioeconomic Status and Late-Adulthood Mental Health: Results From the Survey on Health, Ageing and Retirement in Europe. The Journals of Gerontology: series B 74 (1), 95–104.
Batz-Barbarich, C., L. Tay, L. Kuykendall, and H.K. Cheung (2018). A Meta-Analysis of Gender Differences in Subjective Well-Being: Estimating Effect Sizes and Associations With Gender Inequality. Psychological Science 29 (9), 1491–1503.
Burnette, J. and W. Zhang (2019). Distributional Differences and the Native American Gender Wage Gap. Economies 7 (2), 46.
Burstr¨om, B and W. Tao (2020). Social determinants of health and inequalities in COVID-19. European Journal of Public Health 30 (4), 617–618.
Bussemakers, C., K. Van Oosterhout, G. Kraaykamp, and N. Spierings (2017). Women’s World-wide Education–employment Connection: A Multilevel Analysis of the Moderating Impact of Economic, Political, and Cultural Contexts. World Development 99, 28–41.
Callander, E.J. (2016). Pathways between health, education and income in adolescence and adulthood. Archives of Disease in Chilhood 101 (9), 825–831.
Cellini, N., N. Canale, G. Mioni, and S. Costa (2020). Changes in sleep pattern, sense of time and digital media use during COVID-19 lockdown in Italy. Journal of Sleep Research. De Graaf, R., M. Ten Have, C. Van Gool, and S. Van Dorsselaer (2012). Prevalence of mental
disorders and trends from 1996 to 2009. Results from the Netherlands Mental Health Survey and Incidence Study-2. Social Psychiatry and Psychiatric Epidemiology 47, 203–213.
Elmer, T., K. Mepham, and C. Stadtfeld (2020). Students under lockdown: Comparisons of students’ social networks and mental health before and during the COVID-19 crisis in Switzerland. PLoS One 15 (7).
Ettman, C.K., S.M. Abdalla, G.H. Cohen, L. Sampson, P.M. Vivier, and S. Galea (2020). Preva-lence of Depression Symptoms in US Adults Before and During the COVID-19 Pandemic. JAMA Netw Open 3 (9).
Gao, W., S. Ping, and X. Liu (2020). Gender differences in depression, anxiety, and stress among college students: A longitudinal study from China. Journal of affective disorders 263, 292–300.
Gebhard, C., V. Regitz-Zagrosek, H.K. Neuhauser, R. Morgan, and S.L. Klein (2020). Impact of sex and gender on COVID-19 outcome in Europe. Biology of sex differences 11 (1). Gharehgozli, O. and V. Atal (2020). Revisiting the gender wage gap in the United States.
Economic Analysis and Policy 66, 207–216.
Gonz´alez-Sanguino, C., B. Aus´ın, M. ´A. Castellanos, J. Saiz, A. L´opez-G´omez, C. Ugidos, and M. Mu˜noz (2020). Mental health consequences during the initial stage of the 2020 Coronavirus pandemic (COVID-19) in Spain. Brain, Behavior, and Immunity.
Hamadani, J. D., M. I. Hasan, A. J. Baldi, S. J. Hossain, S. Shiraji, M. Bhuiyan, S. F. Mehrin, J. Fisher, F. Tofail, S. Tipu, S.and Grantham-McGregor, B. A. Biggs, S. Braat, and S. R. Pasricha (2020). Immediate impact of stay-at-home orders to control COVID-19 transmission on socioeconomic conditions, food insecurity, mental health, and intimate partner violence in Bangladeshi women and their families: an interrupted time series. The Lancet. Global Health 8 (11), e1380–e1389.
Holmes, E.A., R.C. O’Connor, V.H. Perry, I. Tracey, S. Wessely, L. Arseneault, C. Ballard, H. Christensen, R.C. Silver, I. Everall, T. Ford, A. John, T. Kabir, K. King, I. Madan, S. Michie, A.K. Przybylski, R. Shafran, A. Sweeney, C.M. Worthman, L. Yardley, K. Cowan, C. Cope, and E. Hotopf, M. andBullmor (2020). Multidisciplinary research priorities for the COVID-19 pandemic: a call for action for mental health science. Lancet Psychiatry 7 (6), 547–560.
Huang, Y. and N. Zhao (2020). Generalized anxiety disorder, depressive symptoms and sleep quality during COVID-19 outbreak in China: A web-based cross-sectional survey. Psychiatry Research 288.
Ji Kang, S. and S. In Jung (2020). Age-Related Morbidity and Mortality among Patients with COVID-19. Journal of infection and chemotherapy 52 (2), 154–164.
Kapteyn, A., R. Alessie, and A. Lusardi (2005). Explaining the wealth holdings of different cohorts: Productivity growth and Social Security. European Economic Review 49 (4), 1361– 1391.
Kernberg, American Psychiatric Association (2013). Diagnostic and Statistical Manual of Men-tal Disorders. American Psychiatric Association Publishing.
Khazanchi, R., E.R. Beiter, S. Gondi, A.L. Beckman, A. Bilinski, and I. Ganguli (2020). County-Level Association of Social Vulnerability with COVID-19 Cases and Deaths in the USA. Journal of General Internal Medicine 35, 2784–2787.
Kiely, K.M., B. Brady, and J. Byles (2019). Gender, mental health and ageing. Maturitas 129, 76–84.
Lei, L., X. Huang, S. Zhang, J. Yang, L. Yang, and M. Xu (2020). Comparison of prevalence and associated factors of anxiety and depression among people affected by versus people unaffected by quarantine during the COVID-19 epidemic in Southwestern China. Medical Science Monitor 26.
Levy, L.B. and M.W. O’Hara (2010). Psychotherapeutic interventions for depressed, low-income women: A review of the literature. Clinical Psychology Review 30, 934–950.
A Nationwide Survey of Psychological Distress among Italian People during the COVID-19 Pandemic: Immediate Psychological Responses and Associated Factors. International Journal of Environmental Research and Public Health 17 (9).
Mc Intyre, K., P. Lanting, P. Deelen, H. Wiersma, J.M. Vonk, A. Ori, S.A. Jankipersadsing, R. Warmerdam, I. Van Blokland, F. Boulogne, M. Dijkema, J.C. Herkert, A. Claringbould, O. Bakker, E.A. Lopera Maya, U. Bultmann, A. Zhernakova, S.A. Reijneveld, E. Zijlstra, M. Swertz, S. Brouwer, R. Van Ooijen, V. Angelini, L. Dekker, A. Sijtsma, S.A. Scherjon, J. Dekens, J.O. Mierau, H.R. Boezen, and L. Franke (2020). The Lifelines COVID-19 Cohort: a questionnaire-based study to investigate COVID-19 infection and its health and societal impacts in a Dutch population-based cohort. medRxiv .
McGinn, K.L. and E. Oh (2017). Gender, social class, and women’s employment. Current opinion in psychology 18, 84–88.
Moghanibashi-Mansourieh, A. (2020). Assessing the anxiety level of Iranian general population during COVID-19 outbreak. Asian Journal of Psychiatry 51.
Moreira, P. S., S. Ferreira, B. Couto, M. Machado-Sousa, M. Fernandez, C. Raposo-Lima, N. Sousa, M. Pico-Perez, and P. Morgado (2020). Protective Elements of Mental HealthSta-tus during the COVID-19 Outbreak in the Portuguese Population. Psychiatry and Clinical Psychology.
Petrongolo, B. (2019). The gender gap in employment and wages. Nature Human Be-haviour 3 (4), 316–318.
Pieh, C., S. Budimir, and T. Probst (2020). The effect of age, gender, income, work, and phys-ical activity on mental health during coronavirus disease (COVID-19) lockdown in Austria. Journal of Psychosomatic Research 136.
Ritchie, H. and M. Roser (2018). Mental health. Our World in Data. Retrieved from: https://ourworldindata.org/mental-health.
Rose, G., K. Khaw, and M. Marmot (2008). Rose’s Strategy of Preventive Medicine. Oxford University Press.
Rossi, R., V. Socci, D. Talevi, S. Mensi, C. Niolu, F. Pacitti, A. Di Marco, A. Rossi, A. Sira-cusano, and G. Di Lorenzo (2020). COVID-19 pandemic and lockdown measures impacton mental health among the general population in Italy. An N=18147 web-basedsurvey. Psychi-atry and Clinical Psychology.
Silbersdorff, A. and K.S. Schneider (2019). Distributional Regression Techniques in Socioeco-nomic Research on the Inequality of Health with an Application on the Relationship between Mental Health and Income. International Journal of Environmental Research and Public Health 16 (20), 4009.
Ueda, M., A. Stickley, H. Sueki, and T. Matsubayashi (2020). Mental Health Status of the-General Population during the COVID-19 Pandemic: A Cross-Sectional NationalSurvey in Japan. Psychiatry and Clinical Psychology.
Wang, C., R. Pan, X. Wan, Y. Tan, L. Xu, C.S. Ho, and R.C. Ho (2020). Immediate Psycho-logical Responses and Associated Factors during the Initial Stage of the 2019 Coronavirus Disease (COVID-19) Epidemic among the General Population in China. International Jour-nal of Environmental Research and Public Health 17.
White, R.G. and C. Van der Boor (2020). Impact of the COVID-19 pandemic and initial period of lockdown on the mental health and well-being of adults in the UK). BJPsych Open 6 (5). WHO (2020a). Mental health and psychosocial considerations during the COVID-19 outbreak.
WHO (2020b). WHO Coronavirus Disease (COVID-19) Dashboard. Retrieved from:
https:/covid19.who.int/.
Wooldridge, J.M. (2010). Econometric analysis of cross section and panel data. Massachusetts Institute of Technology.
World Economic Forum (2020). Global Gender Gap Report 2020. World Economic Forum. Zhang, S.X., Y. Wang, A. Afshar Jahanshahi, J. Jia, and V.G. Haensel Schmitt (2020). First
9
Appendix
9.1
Sample characteristics
Table 4: Characteristics of respondents and non-respondents to the COVID-19 questionnaires.
Variable Respondents Non-respondents P-value Total
Number of observations,
(%) 72,098 (53.63) 62,336 (46.37) 134,434
Number of
depres-sion symptoms before
COVID-19, mean (sd)
0.25 (1.11) 0.32 (1.26) P < 0.0001 0.27 (1.17)
Major depressive
disor-der before COVID-19, % 2.54 3.38 P < 0.0001 2.83
Number of anxiety symp-toms before COVID-19, mean (sd)
0.30 (1.24) 0.36 (1.37) P < 0.0001 0.32 (1.28)
Generalized anxiety dis-order before COVID-19, %
4.95 5.98 P < 0.0001 5.31
Self-rated health (1=ex-cellent, . . . , 5=bad), mean (sd)
2.63 (0.80) 2.70 (0.80) P < 0.0001 2.66 (0.80)
Men, % 39.31 43.99 P < 0.0001 41.48
Current age, mean (sd) 56.57 (12.19) 50.76 (12.27) P < 0.0001 55.22 (12.50) Marital status at last
visit, % Married 66.53 61.89 P < 0.0001 64.87 Registered partner-ship 16.54 20.24 P < 0.0001 17.86 In relationship but living apart 4.94 5.60 P < 0.0001 5.18 Single 11.09 11.30 P = 0.3409 11.16 Other 0.90 0.97 P = 0.3697 0.93
BMI at last visit, mean
Table 4 continued: Characteristics of respondents and non-respondents to the COVID-19 ques-tionnaires.
Smoking at last visit, %
never 46.52 47.06 P = 0.1686 46.68
ex 40.02 34.18 P < 0.0001 38.30
current 13.46 18.75 P < 0.0001 15.06
Number of people in
household, mean (sd) 2.76 (1.23) 3.03 (1.30) P < 0.0001 2.85 (1.26)
Highest education at last visit, %
None 0.29 0.65 P < 0.0001 0.42
Primary 0.92 1.62 P < 0.0001 1.17
Lower or prepara-tory secondary vo-cational
10.84 13.74 P < 0.0001 11.88
Junior general
sec-ondary 14.13 13.54 P = 0.0180 13.92
Secondary vocatio-nal or work-based learning pathway
28.53 31.44 P < 0.0001 29.57
Senior general sec-...,ondary education or pre-university secondary 8.12 7.08 P < 0.0001 7.75 Higher vocational 28.12 24.21 P < 0.0001 26.72 University 7.44 6.22 P < 0.0001 7.00 Other 1.62 1.49 P = 0.1397 1.57
Net income per month at last visit (1=Less than e750, . . . , 11=More than e5000), mean (sd)
5.23 (1.76) 4.98 (1.85) P = 0.1340 5.23 (1.76)
Work situation at last visit (before COVID-19), %
student 0.48 0.64 P < 0.0001 0.54
Table 4 continued: Characteristics of respondents and non-respondents to the COVID-19 ques-tionnaires.
retired 13.38 8.07 P < 0.0001 11.48
unfit for work 2.92 3.06 P = 0.7771 2.97
unemployed/look-ing for work 3.54 3.37 P < 0.0001 3.48
other 7.64 7.60 P = 0.0069 7.63
9.2
Mental health by gender, age and SES
Gender Age .01 .015 .02 .025 .03
Mean number of major depressive disorders
0 4 8 12
Time
Male Female
Major depressive disorders by gender
.01
.02
.03
.04
.05
Mean number of major depressive disorders
0 4 8 12
Time
Age 18-30 Age 31-50 Age 51-65 Age 66 and older
Major depressive disorders by age
Figure 15: Major depressive disorders by
gen-der. Figure 16: Major depressive disorders by age.
.1 .15 .2 .25 .3 .35
Mean number of anxiety symptoms
0 4 8 12
Time
Male Female
Anxiety symptoms by gender
Figure 17: Anxiety symptoms by gender.
0 .1 .2 .3 .4 .5
Mean number of anxiety symptoms
0 4 8 12
Time
Age 18-30 Age 31-50 Age 51-65 Age 66 and older
Anxiety symptoms by age
.01 .02 .03 .04 .05 .06
Mean number of generalized anxiety disorders
0 4 8 12
Time
Male Female
Generalized anxiety disorders by gender
Figure 19: Generalized anxiety disorders by gender. 0 .02 .04 .06 .08
Mean number of generalized anxiety disorders 0 4 8 12 Time
Age 18-30 Age 31-50 Age 51-65 Age 66 and older
Generalized anxiety disorders by age
Figure 20: Generalized anxiety disorders by age. Education .01 .02 .03 .04 .05
Mean number of major depressive disorders 0 4 Time 8 12
None or primary education
Lower secondary vocational or junior general secondary education Secondary vocational or senior general secondary education Higher vocational or university education
Major depressive disorders by education
Figure 21: Major depressive disorders by edu-cation. Income 0 .01 .02 .03 .04
Mean number of major depressive disorders 0 4 8 12 Time
Less than 1000 Between 1000 and 2000 Between 2000 and 3000 Between 3000 and 4000 More than 4000
Major depressive disorders by income (in euros)
Figure 22: Major depressive disorders by in-come. .1 .15 .2 .25 .3 .35
Mean number of anxiety symptoms
0 4 8 12
Time None or primary education
Lower secondary vocational or junior general secondary education Secondary vocational or senior general secondary education Higher vocational or university education
Anxiety symptoms by education
Figure 23: Anxiety symptoms by education.
0
.1
.2
.3
.4
Mean number of anxiety symptoms
0 4 8 12
Time
Less than 1000 Between 1000 and 2000 Between 2000 and 3000 Between 3000 and 4000 More than 4000
Anxiety symptoms by income (in euros)
