COVID-19 and the employment gender gap in Japan

Abstract

This paper examines how the COVID-19 pandemic affected female employment in Japan. Our estimates indicate that the employment rate of married women with children decreased by 3.5 percentage points, while that of those without children decreased by only 0.3 percentage points, implying that increased childcare responsibilities caused a sharp decline in mothers’ employment. Further, mothers who left or lost their jobs appear to have dropped out of the labor force even several months after school reopening. In contrast to women, the employment rate of married men with children was not affected, which hindered progress in narrowing the employment gender gap.

1. Introduction

The COVID-19 pandemic has significantly affected labor markets around the world. For example, in April 2020 in the U.S., the pandemic pushed the unemployment rate to a record high of 14.7% and caused millions to leave the workforce, causing the labor force participation rate to decline to 60.2%.1

One of the key differences of the pandemic recession from previous ones such as the Great Recession is that women’s employment has been more severely affected than men’s (see Alon et al. (2020)). The literature so far has offered two major explanations for this ‘shecession.’ First, the pandemic decreased the demand for labor in the hospitality and tourism industries in which the female employment share is high. Second, many schools and daycare facilities were closed to prevent the spread of COVID-19, which increased parents’ childcare responsibilities. Because women typically have more childcare responsibilities than men, the closures were likely to have caused many working mothers to leave their jobs at least temporarily.

The objective of this paper is to estimate the extent to which women’s employment and labor force participation have been affected during the COVID-19 pandemic in Japan between April and December in 2020, and how the effects depend on childcare responsibilities. Our approach is to construct a counterfactual outcome such as the employment rate in the absence of the pandemic by modeling labor force status as a linear trend with month effects, and estimating it using pre-pandemic data. We define the disemployment effect of the pandemic as the deviation of the observed employment rate from its counterfactual (or predicted) counterpart.

We find that the employment rate of married women with children, defined by married women who live in a household with children aged 0–15 years, began to decrease in March 2020 when a nation-wide school closure took place and decreased further in April 2020 when the government declared the first COVID-19 state of emergency. Although the government lifted the state of emergency on May 25 and reopened most schools by the beginning of June, the mothers’ employment rate had not recovered even by December 2020. In contrast, the employment rate of married women without children was only modestly affected during the same period in that they were temporarily on leave in April and May during the first state of emergency.

To isolate the effects of increased childcare responsibilities during the pandemic, we compare the disemployment effects of the pandemic between married women with and without gradeschoolers or younger children (aged 0–12 years). These children need more attention and care from adults than teenagers. We find that, in April 2020, the employment rate of married women with children aged 0–12 years decreased by 3.5 percentage points while that of married women without these children decreased by only 0.3 percentage points. Because we control for demographic characteristics and pre-COVID-19 occupation/industry, the difference between the two is explained by childcare responsibilities. Our estimates indicate that the unemployment rate of married women with children aged 0–12 years did not change, but the fraction of those out of labor force increased, which implies that married women with children who left or lost their jobs dropped out of the labor force rather than becoming unemployed.

These results suggest that the COVID-19 pandemic has dealt a blow to the momentum reducing the gender employment gap in Japan, which had been narrowing for more than three decades since the Equal Employment Opportunity Law was enacted in 1985. While the gender gap in the employment rate among those who are married with children was 33 percentage points in 2015 and steadily decreased to about 25 percentage points in 2019, this progress stalled during the pandemic because married men with children were not affected by the pandemic while married women with children were severely affected. The pandemic thus increased the gender employment gap by about 3 percentage points relative to what would have occurred with no pandemic.

We also analyze possible heterogeneous effects of the pandemic across subgroups, and find that the unemployment rate of single mothers rose while their labor force participation rate remained at a similar level, suggesting that single mothers remained in the labor force after a job separation while married mothers left the labor force. This different labor force participation between single and married mothers is likely to be accounted for by non-labor income through their spouses.

Some studies on the COVID-19 pandemic have focused on women in the Japanese labor market. Using data from an ad hoc survey,2 Dang and Nguyen (2021) find more permanent job loss for women than men in late April, 2020, while Kikuchi et al. (2021), using the LFS, find that more women aged 25–64 lost employment than men in April and May 2020. Neither of these papers examine the role of increased childcare responsibilities during the pandemic on the labor supply of Japanese women. Further, Yamamura and Tsutsui (2021) find that mothers are more likely to work from home than fathers when both parents work, and while they do not estimate the disemployment effect of the pandemic, they show that mothers shoulder the burden of working remotely and caring for children at home.3

The strength of our research compared to some other studies that use ad hoc surveys is our use of the Japanese Labor Force Survey (LFS), a nation-wide survey conducted monthly since 1947 and with a large sample size of about 40,000 randomly chosen households. In addition, we use data from that LFS for the period from the several years prior to COVID-19 through December 2020. This enables us to predict counterfactual outcomes with reasonable precision, based on the previous five years before the COVID-19 pandemic. Because the employment and labor force participation rates of married women with children have been steadily increasing (Kawaguchi et al., 2021), merely using 2019 outcomes as the comparison would underestimate the effects of the pandemic. We account for the pre-pandemic trend and month effects to address this issue. Second, the LFS has a short panel structure which, unlike ad hoc surveys, enables us to control for the previous industry, occupation, and hours of work before the pandemic without relying on retrospective questions.4

Our contribution to the literature is to offer another piece of evidence from Japan that increased childcare responsibilities decreased women’s employment and labor force participation rates disproportionately during the pandemic. A growing body of studies is reporting a gender gap in the disemployment effects of the COVID-19 pandemic. Alon et al. (2020) find that women’s unemployment in the U.S. increased by 12.8 percentage points between February and April 2020, while that of men increased by a smaller 9.9 percentage points. While some countries did not experience a ‘shecession’ like the U.S,5 (Alon et al., 2021) find that women’s employment was disproportionately hurt in 18 out of 28 countries, which is different from previous recessions in which men lost more jobs than women. Similarly, Bluedorn et al. (2021) also find that two thirds of 38 developed and developing countries suffered a ‘shecession.’ The literature shed some light on the sources of heterogeneity in ‘shesession’ and point to two factors, the difference in industry/occupation structure and the difference in the gender gaps in family responsibilities across countries.6

In light of the development of the literature, our findings from Japan provide additional evidence that the gender imbalance in childcare responsibilities widened the gender gap in the labor market effects of COVID-19.7 Although the increase in the unemployment rate due to COVID-19 was limited in Japan compared with other developed countries,8 we find a sharp contrast in the employment responses between men and women with children. This substantial gender gap reflects well the pre-existing gender gap in family responsibilities, measured by the time spent for household chores and child care, in Japan that is the largest among OECD countries, according to OECD (2020) and Cabinet Office of Japan (2020). Examining Japanese mothers’ labor supply thus provides an interesting case study from a country with the largest gender division of labor (Fuwa, 2004), which may have exacerbated the pandemic’s negative effects for women.

The rest of the paper is structured as follows. In Section 2, we lay out the econometric model to evaluate the effect of the COVID-19 pandemic on employment outcomes. In Sections 3, 4, we describe our data from the LFS and present our estimation results. In Section 5, we examine how the pandemic affected the gender employment gap, and we conclude the paper in Section 6.

2. Econometric model

In this study, we estimate the effect of the COVID-19 pandemic on employment by gender, particularly focusing on the role of parenthood. The challenge is that because the pandemic has affected the entire labor market, it is hard to find an appropriate comparison group that is unaffected by the pandemic. Our approach is to instead construct a counterfactual employment rate in the absence of the pandemic by modeling and estimating an employment process using pre-COVID-19 data from the LFS. We then estimate the impact of the pandemic by comparing the actual employment rate and the counterfactual one. This approach is based on the standard method for estimating excess mortality.9

Specifically, we estimate the following regression equation;

$$y_{it}={\beta }_0+\sum _{\genfrac{}{}{0ex}{}{l\in \{Jan20,}{\dots ,Dec20\}}}{\delta }_{l}D(t=l)+x_{it}^{\prime }\beta +\sum _{\genfrac{}{}{0ex}{}{m\in \{Jan,}{\dots ,Dec\}}}{\mu }_{m}D(t=m)+f\left(t\right)+u_{it},$$ (1)

for each individual $i$ and month $t$. The dependent variable $y_{it}$ is an outcome indicator variable of employment that takes one if individual $i$ was employed in period $t$ and takes zero otherwise,10 and the indicator variable $D\left(\cdot \right)$ takes one if the condition in the bracket is satisfied and zero otherwise. The control variable $x_{it}$ includes education, age, the number of adults in the household, and industry and occupation 12 months ago. The month fixed effects ${\mu }_m$ account for seasonality in employment status, the function $f\left(t\right)$ is a linear time trend, and $u_{it}$ is the error term. Note that the intercept (${\beta }_0$), the coefficients for the control variables ($\beta$), the month fixed effects (${\mu }_m$), and the linear trend ($f\left(t\right)$) are estimated from observations up to 2019 because the parameter ${\delta }_l$ is dedicated to observations in 2020. Deliberately, we do not allow for a nonlinear time trend to avoid overfitting. We construct the counterfactual employment rate in 2020 in the absence of the pandemic by setting $D(t=l)=0$, and hence, the parameter ${\delta }_l$ captures the difference between the actual and counterfactual employment rates, which we interpret as the effect of the pandemic.11

The disemployment effect of the pandemic is likely to be different between women with children and those without children because many schools closed for an extended period, increasing mothers’ childcare responsibilities as they spent additional time at home with their children. To examine the heterogeneous effects of the pandemic by childcare responsibilities, we test whether the disemployment effects differ by the presence of school-aged children. To do this, we extend the estimation equation to allow for heterogeneous effects of the pandemic on employment between married women with and without children. Let $K_{it}$ be an indicator variable that takes one if an individual has a child aged between 0 and 12 and takes zero otherwise. We estimate the following regression;

$$y_{it}={\beta }_0+\pi K_{it}+\sum _{\genfrac{}{}{0ex}{}{l\in \{Jan20,}{\dots ,Dec20\}}}{\delta }_{l}D(t=l)\times K_{it}+\sum _{\genfrac{}{}{0ex}{}{l\in \{Jan20,}{\dots ,Dec20\}}}{\gamma }_{l}D(t=l)+\sum _{k\in \{0,1\}}(x_{it}^{\prime }{\beta }_k+\sum _{\genfrac{}{}{0ex}{}{m\in \{Jan,}{\dots ,Dec\}}}{\mu }_{km}D(t=m)+f_k\left(t\right))D(K_{it}=k)+u_{it}.$$ (2)

Our parameter of interest is ${\delta }_l$ that captures the difference in the disemployment effect of the pandemic between women with and without children. Because women with children were more affected by increased childcare responsibilities, ${\delta }_l$ is expected to be negative from March 2020 on.

Note that our approach differs from a difference-in-differences (DID) estimation. When estimating the effects of increased childcare responsibilities by DID, a researcher might choose women with children as those treated and women without children as the control. This choice leads to a common trend assumption between women with and without children which we doubt is satisfied because the children’s school calendar only affects women with children. The Japanese fiscal year begins in April, and people often enter the workforce and children begin attending schools and childcare centers in April. Because of this practice in Japan, women with children tend to return to work in April because they start using childcare centers in April. In addition, during holidays and long vacations in summer and winter, when children are out of school, the labor supply may differ between women with and without children. Indeed, we reject the null hypothesis that seasonality is the same between women with and without children (see Figures 21–22 in Appendix C).

Addressing this problem of using a DID approach, our method estimates the seasonality effects for each group separately, drawing upon the regression discontinuity design (RDD) studies of the effects of a policy change in parental leave (e.g. Lalive et al. (2014)). Because the female employment rate exhibits seasonality, the parental leave studies use data from years before the policy change as a comparison to control for seasonality. However, as seasonality can bias estimates even in the RDD analysis unless the bin width is very narrow, a hybrid method known as DID-RDD is often adopted in policy evaluations of parental leave, and this is the basic idea behind our analysis as well.12

3. Data

For our analysis, we use the monthly Labor Force Survey (LFS), a household survey conducted by the Statistics Bureau of the Ministry of Internal Affairs and Communications (MIC) and which is the Japanese counterpart of the Current Population Survey (CPS) in the United States. The target population of the LFS is all residents of Japan,13 and about 40,000 households are randomly surveyed each month. We use the LFS from January 2015 to December 2020.

The LFS asks about the employment status of each member of the household aged 15 and over in the last week of each month. The employment status includes five categories: mainly work, work besides housework or study, absence from work, unemployed, and out of the labor force. The category “absent from work” includes leaves both with and without compensation.14 Other demographic variables such as age, gender, and marital status are also collected.

The sampling structure of the LFS enables us to construct a short panel data set that allows us to observe the change of labor force status of the same individual after the onset of COVID-19 pandemic relative to the status before the pandemic. Similar to the CPS, the LFS has a rotating sample structure in which the sample households are surveyed a total of four times: two consecutive months in the first year and the same two months in the following year.15 As long as the surveyed household members do not move or refuse to answer the questionnaire, the information on employment for the four surveys can be connected. Thus, for households first entered in the LFS in 2019, we can connect them to information for the year 2020, which is under COVID-19.16 With this capability, our analysis also examines the heterogeneous effects of the pandemic by workers’ pre-COVID-19 employment status.

We construct our main analysis sample as follows. First, we restrict our main sample to married men and women between ages 25 and 54 because our primary interest is in the labor supply of parents. Households with children are defined by whether or not there is a child in the household 15 years old or younger, the age just before entering high school. Second, we exclude observations with missing variables that are necessary for our analysis such as employment status and educational background. As educational background is only asked in the fourth survey, only households that continued to respond to the survey throughout the fourth wave are included in our sample. Third, as our analysis uses the employment status a year ago as a control variable, the analysis sample includes only the third and fourth waves. Finally, a sample of single mothers is also taken for an additional analysis.

Table 1 presents the descriptive statistics. About half of the sample is female, indicating that the couples mostly live together. The average age is 42.4 years, and about 65 percent of the sample have children. Looking at the presence of children by the age of the youngest child, 30.2 percent have a preschooler, 18.7 percent have a child attending elementary school, and about 16 percent have a child attending junior or senior high school. This latter group is smaller than those with preschool or elementary school children because our sample includes relatively young individuals between ages 25 and 54. The percentage of people with a university degree or higher is about 46 percent. The employment rate for the current year is about 85 percent, and the number of people who are absent from work or unemployed is very low at about 2 percent. The employment rate for the previous year was 84 percent, which is almost the same as that of the current year.17

Table 1.

Descriptive statistics.

Variable	Obs	Mean	St. dev.
Woman	1,459,145	0.527	0.499
Age	1,459,145	42.399	7.485
With Child in Pre-school	1,459,145	0.302	0.459
With Child in Elementary School	1,459,145	0.187	0.390
With Child in Jr. High School	1,459,145	0.075	0.263
With Child in High School	1,459,145	0.086	0.281
Employment	1,459,145	0.847	0.360
Temporary Leave	1,459,145	0.024	0.153
Unemployment	1,459,145	0.011	0.102
Out of Labor Force	1,459,145	0.142	0.349
Re-openings	1,459,145	0.539	0.401
High Education	1,069,369	0.459	0.498
Employment in Previous Year	544,967	0.844	0.363
Permanent in Previous Year	403,686	0.798	0.402

Notes: The sample is restricted to married individuals aged 25–54. Woman represents the fraction of women in the sample. With Child in Pre-school/Elementary School/Jr. and High School/High School represent the proportion of individuals in the household where the youngest child is in the respective educational category. Employment, Temporary Leave, Unemployment, and Out of Labor Force represents the fraction of individuals who are in that respective labor force category. Re-openings represents the fraction of individuals living in a prefecture where the proportion of elementary schools re-opening on June 1 is higher than 50%. To construct this variable, we use the information from government statistics: https://www.mext.go.jp/content/20200603-mxt_kouhou01-000004520_4.pdf, and merge it with the LFS using a prefecture identifier. High Education represents the fraction of individuals who have a bachelors degree or higher. Employment/Permanent in Previous Year represents the fraction of individuals who are under an employment/permanent contract when the first wave of the LFS was conducted.

4. Results

This section presents the estimation results for the effects of the COVID-19 pandemic on labor force status, which is comprised of employment, temporary leave, unemployment, or out of the labor force. We begin by examining the effects on married women with children in Section 4.1. Then, to isolate the effects of increased childcare responsibilities, in Section 4.2 we compare the disemployment effects of the pandemic between married women with and those without children. We then analyze how the pandemic affected single mothers and married mothers differently in Section 4.3.

4.1. Effects of the COVID-19 pandemic

Fig. 1 and Figure 12 in Appendix A show the employment rate and the rate of temporary leave from 2015 to 2020 for married women with children. The solid line in Fig. 1 shows the observed employment rate for each month from 2015 to 2020, while the dashed line shows the counterfactual employment rate in the absence of the pandemic. Figure 12 in Appendix A presents the observed and counterfactual rates of temporary leave. As detailed in Section 2, we construct the counterfactuals based on a function of individual attributes, the linear time trend, and month effects using the observations until 2019. To be more precise, we use the estimates for Eq. (1) under the assumption that $D(t=l)=0$. As expected, the predicted outcomes closely trace the observed outcomes until 2019 in both figures.

Fig. 1.

The predicted and observed monthly employment rates for married women with children.

Notes: The solid line represents the employment rate of married women with children (aged 0–15 years) from 2015 to 2020. The dashed line represents the predicted employment rate from 2015 to 2020 calculated from Eq. (1) with $D(t=l)=0$. The vertical line indicates March 2020 when the nationwide school closure took place. The estimation sample is restricted to married women aged 25–54 with children whose information on their education and working status in the previous year was available.

We interpret the differences between the observed and counterfactual employment rates in Fig. 1 as the effect of the pandemic, which is shown in Fig. 2. Note that these are estimates for parameter ${\delta }_l$ in Eq. (1). The estimates indicate that the pandemic decreased the employment rate of married women with children by 4.4 percentage points in April and that the disemployment effects continued to be around −3 to −4 percentage points. This is a significant drop relative to the mid-70 percent employment rate of the group before the pandemic. Because the COVID-19 outbreak occurred at the end of February 2020 in Japan, the estimates for January and February 2020 can be used for placebo tests for the effects of the pandemic. As expected, the effects in January and February are zero, which provides evidence in support of the validity of our method.

Fig. 2.

The effect of the COVID-19 pandemic: Employment rate

Notes: The solid line represents the effect of the COVID-19 pandemic on the employment rate, defined by the difference between the actual and counterfactual employment rates of married women with children (aged 0–15 years) in 2020. The shaded area represents the 90% confidence intervals, which are computed from standard errors clustered at the individual level. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to married women aged 25–54 with children (aged 0–15 years) whose information on their education and working status in the previous year were available.

The effect in March 2020 is modest at −1.3 percentage points, which is consistent with a finding by Kikuchi et al. (2021). As discussed earlier, the nationwide school closure was initiated at the beginning of March 2020. However, we may not necessarily expect a large and immediate effect for at least two reasons. Firstly, about two weeks of spring vacation in March is scheduled annually regardless of the COVID-19 pandemic. Thus, married women with children may be used to having children at home during this period anyway. Secondly, married women with children may expect the March school closure to be temporary. While it is true that the prime minister declared the first state of emergency in April and the school closure continued until the end of May, this may not have been fully anticipated by working parents in March.18

Figure 13 in Appendix A shows the effect on the rate of temporary leave, which rose sharply in April by 7.6 percentage points but completely disappeared from June onward.19 This pattern sharply contrasts with the persistent effects on the employment rate. Note that the temporary leave in our data includes leaves from the workplace due to various reasons, including furloughs, sickness, and caregiving. With the economic downturn under the state of emergency in April, we expect that furloughs accounted for a large portion of the spike in Figure 13 in Appendix A.

In the LFS, those not in employment are categorized as either unemployed or out of the labor force. Next, we examine to which of these two possibilities those who lost or left their employment moved. Fig. 3, Fig. 4 show the effect on the unemployment rate and the rate of being out of labor force, respectively. The estimates show that the pandemic substantially affected only the rate of being out of labor force, but not the unemployment rate. Specifically, increases in the rate of being out of labor force account for 83.0% to 98.7% of the decrease in the employment rate after the outbreak of COVID-19.20

Fig. 3.

The effect of the COVID-19 pandemic: Unemployment rate.

Notes: The solid line represents the effect of the COVID-19 pandemic on the unemployment rate, defined by the difference between the actual and counterfactual unemployment rates of married women with children (aged 0–15 years) in 2020. The shaded area represents the 90% confidence intervals, which are computed from standard errors clustered at the individual level. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to married women aged 25–54 with children (aged 0–15 years) whose information on their education and working status in the previous year were available.

Fig. 4.

The effect of the COVID-19 pandemic: Rate of being out of the labor force.

Notes: The solid line represents the effect of the COVID-19 pandemic on the rate of being out of labor force, defined by the difference between the actual and counterfactual rates of being out of labor force for married women with children (aged 0–15 years) in 2020. The shaded area represents the 90% confidence intervals, which are computed from standard errors clustered at the individual level. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to married women aged 25–54 with children (aged 0–15 years) whose information on their education and working status in the previous year were available.

Although we cannot directly examine why mothers exited the labor force, we consider three possible explanations. The first is that mothers may have decided not to seek a job because they believed it was difficult to find one given the weak labor demand under the pandemic. The second is that mothers chose to stay at home to take special care of their children under the continuing stress of the pandemic. The third is that mothers chose not to seek a job to prepare for future closures of schools and childcare facilities. The last two channels are related to childcare responsibilities and can explain the persistent increase in the rate of being out of labor force. We explore the role of increased childcare responsibilities in accounting for the disemployment effect of the pandemic in the next subsection.

While our analysis focuses on the extensive margin of labor supply, we also estimate the effects on hours worked, or the intensive margin. Figure 16 in Appendix A shows the effect of the pandemic on weekly hours of work for those who did work positive hours, and we see that they decreased by about two hours in April and May, but the effects are negligibly small from June to November. More results on hours of work are available in Appendix A.

In sum, for married women with children, both the intensive and extensive margins of the labor supply were negatively affected in April and May when the first state of emergency was declared. The negative effects on the extensive margin remained throughout 2020, but the intensive margin was not affected from June onward.

4.2. Heterogeneity by childcare responsibilities

In this subsection, we examine the extent to which childcare responsibilities account for the disemployment effects of the pandemic on married women. The literature offers two major explanations. The first one is a decrease in the labor demand for female-dominated industries such as hospitality and tourism. The second one is an increase in childcare responsibilities. We expect that the latter plays a particularly important role in Japan because the traditional view of gender roles demands mothers to bear the burden of childcare responsibilities disproportionately.

We apply Eq. (2) to isolate the effects of increased childcare responsibilities by comparing the disemployment effects of the pandemic between married women with children aged 0–12 (i.e. elementary school students or younger) and married women with no children. In our estimation, we control for individual characteristics such as education, age, the number of adults in the household, and industry and occupation one year prior.

Fig. 5 shows the estimated effects on each subgroup.21 For married women with children, the negative effect appears in March when school closure began and continues throughout the year. In contrast, the effects on those without children remain at zero in nearly every month in 2020. Fig. 6 shows the differences between the two, which are the estimates for the coefficient ${\delta }_l$ in Eq. (2). Our estimates indicate that the effects of the pandemic on the employment rate differ between married women with and without children by −3.3 percentage points in April, which is statistically significant. Although the estimates are bumpy and noisy, the results suggest that the negative effects of increased childcare responsibilities persisted throughout 2020.

Fig. 5.

The effect of the pandemic on the employment rate: Married women with and without children (aged 0–12 years).

Notes: The solid and dashed lines represent the effect of the COVID-19 pandemic on the employment rate for married women with and without children (aged 0–12 years) in 2020, respectively. The effect of the COVID-19 pandemic on the employment rate is defined by the difference between the actual and counterfactual employment rates in 2020. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to married women aged 25–54 whose information on their education and working status in the previous year were available.

Fig. 6.

The effect of the pandemic through childcare responsibilities: Employment rate.

Notes: The solid line represents the effect of the COVID-19 pandemic through childcare responsibilities on the employment rate for married women with children (aged 0–12 years) in 2020. This effect is defined as the difference between the effect of the pandemic on the employment rates of married women with and without children (aged 0–12 years) in 2020. The shaded area represents the 90% confidence intervals, which are computed from standard errors clustered at the individual level. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to married women aged 25–54 whose information on their education and working status in the previous year were available.

Figure 17 in Appendix B shows the effect of the pandemic on the rate of temporary leave, which increased sharply in April, but faded out by July for both groups. Figure 18 in Appendix B presents the difference between the two groups, and it is significant at 3.3 percentage points in April. This implies that the effect through childcare responsibility alone accounts for 45%22 of the overall effect on temporary leave of married women with children in April.

To see whether or not those who left or lost their employment stayed in the labor force, we show the effect of increased childcare responsibilities separately on the unemployment rate and the rate of being out of labor force in Fig. 7, Fig. 8, respectively. As mentioned earlier, the overall effect of the pandemic on the employment rate is accounted for by changes in the rate of being out of the labor force. The point estimates for the unemployment rate are close to zero.

Fig. 7.

The effect of the pandemic through childcare responsibilities: Unemployment rate.

Notes: The solid line represents the effect of the COVID-19 pandemic through childcare responsibilities on the unemployment rate for married women with children (aged 0–12 years) in 2020, which is defined as the difference between the effects of the pandemic on the unemployment rate for married women with and without children (aged 0–12 years) in 2020. The shaded area represents the 90% confidence intervals, which are computed from standard errors clustered at the individual level. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to married women aged 25–54 whose information on their education and working status in the previous year were available.

Fig. 8.

The effect of the pandemic through childcare responsibilities: Rate of being out of the labor force.

Notes: The solid line represents the effect of the COVID-19 pandemic through childcare responsibilities on the rate of being out of labor force for married women with children (aged 0–12 years) in 2020, which is defined as the difference between the effects of the pandemic on the rate of being out of labor force for married women with and without children (aged 0–12 years) in 2020. The shaded area represents the 90% confidence intervals, which are computed from standard errors clustered at the individual level. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to married women aged 25–54 whose information on their education and working status in the previous year were available.

To further analyze the effects of increased childcare responsibilities, we compare the effects of the pandemic by the number of children (i.e., one child or two or more children). We expect the effect is stronger for those with two or more children because they bear a greater childcare burden. As shown in Figures 19–20 in Appendix B, the employment effect of the pandemic tends to be stronger for those with two or more children than those with only one child, although the estimates are not statistically significant.

Finally, we examine whether increased childcare responsibilities widened the existing employment disparities among mothers across education, employment contract, and region. To answer this question, we test whether the disemployment effects of increased childcare responsibilities differ across the subgroups. As extensively discussed in Appendix D, our estimates do not indicate the effects are heterogeneous, which implies that the pandemic has not widened the existing gaps among married women with children.

In summary, we find that increased childcare responsibilities significantly decreased mothers’ employment. These effects account for most of the decline in their employment rate in the pandemic, and the weakened labor demand does not seem to be a primary reason. Our results suggest that the negative effects were slightly stronger for those with more children.

4.3. Heterogeneity by marital status

In this section, we examine the effects of the pandemic on the labor force status of single mothers as compared with married mothers. Figure 14 in Appendix A shows the effect of the COVID-19 pandemic on the employment rate of single mothers, which was 5.2 percentage points in the third quarter, but it disappeared in the fourth quarter. We also present the effects on the unemployment rate and the rate of being out of labor force in Fig. 9, Fig. 10, respectively. While we find no effect on the rate of being out of the labor force, we find a significant increase in the unemployment rate in the third quarter of 2020.

Fig. 9.

The effect of the COVID-19 pandemic: Single mother unemployment rate.

Notes: The solid line represents the effect of the COVID-19 pandemic on the unemployment rate for single mothers, defined as the difference between the actual and counterfactual unemployment rates of single mothers in 2020. The shaded area represents the 90% confidence intervals, which are computed from standard errors clustered at the individual level. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to single mother aged 25–54 whose information on their education and working status in the previous year were available.

Fig. 10.

The effect of the COVID-19 pandemic: The rate of being out of the labor force for single mothers.

Notes: The solid line represents the effect of the COVID-19 pandemic on the rate of being out of labor force for single mothers, defined as the difference between the actual and counterfactual rates of being out of labor force for single mothers in 2020. The shaded area represents the 90% confidence intervals, which are computed from standard errors clustered at the individual level. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to single mother aged 25–54 whose information on their education and working status in the previous year were available.

This result contrasts with that of married women with children who experienced a persistent decline in the employment rate of 2.6 and 3.3 percentage points in the third and fourth quarters.23 Another key difference is that while single mothers became unemployed but remained in the labor market, married mothers left the labor force. We suspect that non-labor income from spouses accounts for this difference in labor force participation.

5. Consequences for the gender gap in employment

In this section we examine how the COVID-19 pandemic influenced the gender gap in employment between married men and women with children. The gender employment gap is a relevant issue because Japan has more pronounced gender inequality than other developed countries. In 2016, the time per day spent on unpaid domestic and care work by women was 4.6 times longer than that by men.24 This number is strikingly larger than, for instance, that of the U.S., where women spend 1.6 times more than men in home production.

Before we look into the gender employment gap itself, we highlight the difference in the effects of the pandemic between married men and women with children. Figure 15 in Appendix A shows how the pandemic affected the employment rate of married men with children, and in contrast to married women with children (see Fig. 1), the pandemic did not affect the employment rate of married men with children. It is true that women were more negatively affected than men in many other countries including the US as well (Alon et al., 2021), this gender difference in Japan stands out.

Finally, Fig. 11 shows the gender employment gap from 2015 to 2020, which is defined as the difference in the employment rate between married men and women with children. Before the pandemic, the gender employment gap had steadily narrowed over the last several years in Japan from 33 percentage points in 2015 to about 25 percentage points in 2019. However, this progress stalled during the pandemic, with the gender employment gap remaining at around 25 percentage points until the end of 2020. Our estimates indicate that in the absence of the pandemic, the gender employment gap would have been 21.6 percentage points in the fourth quarter of 2020, while the actual gender employment gap was 24.9 percentage points.

Fig. 11.

The gender employment gap between married men and women with children.

Notes: The solid line represents the gender employment gap between married men and women with children (aged 0–15 years) from 2015 to 2020. The employment gap is defined as the difference in the employment rate between married men and women with children (aged 0–15 years), or the difference between the solid lines in Figure 15 and Fig. 1. The vertical line in the graph represents the outbreak of the COVID-19 pandemic in Japan at the beginning of March 2020. The estimation sample is restricted to married men and women aged 25–54 with children (aged 0–15 years) whose information on their education and working status in the previous year were available.

6. Conclusion

Using the LFS, we estimated the effects of the COVID-19 pandemic on Japanese mother’s employment and labor force participation. Our estimates indicate that the pandemic decreased the mothers’ employment rate by about four percentage points from 75 percent before the pandemic. Most of this decrease is accounted for by increased childcare responsibilities. Our estimates suggest that most mothers who left or lost their jobs dropped out of the labor force rather than becoming unemployed. These negative effects are persistent even six months after school closures ended.

We also find that the employment effect of the pandemic is strikingly different between men and women, as the labor force status of married men with children was unaffected by the pandemic. Consequently, the progress in narrowing the gender employment gap has stalled completely. In fact, the pandemic increased the gender employment gap by three percentage points relative to what would have occurred in the absence of the pandemic.

Our finding provides a lesson to policy makers that goes beyond the current context. The empirical results indicate that schools offer two important practical functions: educating children and freeing up parents from child care at home. Thus, school closures not only deprived children of learning opportunities, but also increased the child care burden of parents. While the temporary loss of learning opportunities due to school closures can be made up to a certain extent through online learning (see Ikeda and Yamaguchi (2021)), the extra child care burden is a substantial blow to mothers’ labor force participation in a society such as Japan in which the responsibility for child care is placed disproportionately on mothers. Especially during crises such as COVID-19, policy makers should pay careful attention to the often overlooked child care function of schools to avoid unintended consequences of school policies.

Appendix A. Supplementary data

The following is the Supplementary material related to this article.

MMC S1

Appendix includes additional results for the effect of COVID-19 and evidence on seasonality in labor force status.

Data availability

The data that has been used is confidential.