Event-study impacts on men and women
Figures 3 and 4 present the main event-study results from Eq. (1). The sample includes all adult men and women (separately) from the ENOE who are 18–64. The green diamond plotted points reflect men, and the purple triangle plotted points show the impact on women. The gray shaded area represents survey waves within 2020 (potentially the worst months of the pandemic). The excluded reference period, 2020Q1, is represented by the vertical line.
Figure 3 displays the main impact of the COVID-19 pandemic on labor supply and time use. Focusing on the labor supply consequences of the pandemic (at the top of Fig. 3), both men and women experience substantial negative impacts on their labor supply. Employment falls by roughly five percentage points in the first (available) quarter (2020Q3), with women experiencing more considerable employment losses than men. Men also recover their employment slightly faster than women and are almost back to baseline levels by 2021Q2. Despite the more evident impact on women’s extensive probability of employment, men’s and women’s hours worked decline (and recover) similarly. Hours spent working are nearly back to baseline level by 2021Q2 for both men and women.
Because Mexico has a robust informal sector, and Mexican women are more likely to be participating in informal and unpaid work (Ortega-Díaz, 2020), we separate the impact on employment by the informal and formal sectors. The results in the second two panels of Fig. 3 show a more apparent and persistent impact on formal employment than informal employment. Formal employment declines by 2–3.5 percentage points and fails to recover by 2021Q2. Instead, workers appear to shift into informal work, where both men’s and women’s employment recovers relatively quickly. By the end of 2020, men have entirely recovered their initial levels of informal employment, and by quarter two of 2021, they have surpassed the initial levels of informal employment. These graphs suggest that informal employment leads Mexico’s labor market recovery. Despite the benefits of the quick employment gains in the informal sector, the loss of formal sector jobs has potential costs to workers. Informal laborers have few employment protections; there is no minimum wage, access to unions is nonexistent, and there are few protections over firings (Busso et al., 2012, Levy, 2010). Thus, while labor markets have recovered, the majority of the gains are in positions with few guarantees and protections for workers.
We then present measures of unemployment in addition to measures of employment. While there are clear job losses during the pandemic, some workers may choose to leave the labor market entirely. Parents may exit the labor force to care for children, and other workers may decide that the risk of infection is too high to justify a job search. The fifth graph of Fig. 3 shows that unemployment rises for both men and women, but slightly more for men. For both groups, the rise in unemployment is at most 1.5 percentage points, less than the employment loss of 5 percentage points. The unemployment results indicate that some individuals may exit the labor force instead of searching for a new position.
To better understand the job loss by the economic sector, Fig. 4 breaks out employment into sectors, including service, construction and manufacturing, trade, and agriculture. The service sector shows the most considerable employment reductions for men and women, with greater employment losses for women. Construction and manufacturing employment also dip in the third quarter of 2020 but begin to recover by the end of 2020, with the recovery more apparent for men. Trade and agricultural sectors show less evidence of employment declines. The importance of employment losses in the service sector reflects similar findings in the United States (Alon et al., 2020b). Our results suggest that the pandemic’s impact on the service sector is important for both men’s and women’s employment, with women more affected than men. While the service sector in Mexico is smaller than the United States (64.5% versus 80%, Agency (2017)) service jobs still show significant reductions by 3–4 percentage points.
Then, moving back to Fig. 3, we consider the impact of the pandemic on individual income. We take the inverse hyperbolic sine of income to approximate a natural log distribution while maintaining zero earners (Bellemare and Wichman, 2020). Income declines substantially in the first available quarter (for both men and women) but almost wholly recovers by the end of the data series. The pattern of income loss is similar across both men and women.
Then, in the bottom two panels of Fig. 3, we show measures of time use. Men exhibit the clearest adjustment as they spend more time on household chores after the pandemic begins. Despite the increase in time spent on chores, men allocate no extra time toward caring for others (including children). For women, there is no significant reallocation of time toward household chores or time toward caring for others. Though time spent on household chores appears to be on a secular decline before the start of the pandemic, making the precise impact of the pandemic difficult to isolate.
Overall, the findings suggest that both men and women experienced considerable reductions in employment, hours worked, and income, but men recover their employment faster, especially in the informal sector of the economy. The only puzzling fact appears in Fig. 3, where women fail to reallocate their time to household chores or toward caring for others. However, due to the preexisting decline in time on household chores, the actual effect of the COVID-19 pandemic is difficult to disentangle. We next explore whether this response differs for households with children.
Impact on households with children
We next focus on households with children, especially those with school-aged children and younger children. Of particular interest is whether the primary sample of households fails to reallocate their time toward caring for others due to heterogeneous impacts by the presence of children. Households with children should be the most affected by school closures, and parents likely bear the main burden of this change. We split the sample into household heads and their spouses with children of different ages. We categorize children into under five, from five to nine, and ten to fourteen. Figure 5 shows results for mothers by the child age categories, and Fig. 6 presents the same for fathers.
In Fig. 5, mothers of school-aged children experience a one-quarter increase in their time caring for others. This increase in time spent caring occurs across mothers with children aged five to nine and ten to fourteen. The increase in time spent caring for others does not increase significantly for mothers of children under age five, even though the coefficient indicates a rise in time spent caring for others. For fathers in Fig. 6 hours on the house increase significantly and persistently, while time spent caring for others does not significantly increase.
While these results align more with our initial expectation than the baseline results, the failure of mothers to persistently increase time spent caring for others continues to be surprising. We attribute the relatively flat caring burden to women’s lower initial labor supply (OECD, 2020). In Mexico, women have lower pre-pandemic employment rates than high-income settings (Arceo-Gomez and Campos-Vazquez, 2010), indicating that prior to the pandemic, mothers already likely spent a substantial amount of their time on the household rather than in the labor market. There is, therefore, less ability for women’s time use to change during the pandemic relative to the United States and other high-income countries. Due to women’s lower labor supply generally, mothers in Mexico may be better positioned to absorb the school closures without an adjustment in time allocation. Extending from this primary explanation, past research has shown that women face barriers to childcare in Mexico (Ángeles et al., 2011, Calderon, 2014, Mateo Díaz and Rodriguez Chamussy, 2013), suggesting that the majority of households with young children may have already been caring for children at home. This theory is consistent with Fig. 5, where women with children under five experience less clear reductions in employment and hours worked as compared to the baseline findings.
To analyze other heterogeneous effects that may exist, we further explore households with children in Section C. First, we consider the impact on nuclear households where time spent caring for others is likely to be most representative of hours spent caring for children. In non-nuclear families, caring may be allocated toward elderly or ill family members, which is especially relevant during a pandemic. Because we cannot separate the time individuals spend caring for sick, elderly, or children, nuclear households with children will be our closest approximation to actual time spent caring for children. Restricting the sample to nuclear families also ensures that extended family members within the household are not sharing the caring responsibilities, which would change the interpretation of the main results. To consider nuclear families, we limit the sample of households to those composed of a head, the head’s spouse, and children of the household head, with mothers shown in Fig. C.1 and fathers presented in Fig. C.2.
The findings with nuclear families in Figs. C.1 and C.2 align with the results by child age, especially the findings for children 6–15. In households with school-aged children, women briefly increase their time caring for others, with the rise in time spent caring appearing in 2020Q4 and then dissipating. The effect only appears in households with school-aged children, and women do not significantly increase their time spent caring in households with very young children (under age five). Women with young children (under five) may be less likely to work before the pandemic began.
Second, we examine the impact on single-parent-headed households. We define single-parent headed households as households with only the household head and no spouse, where the household also has children under the age of 15. Figure C.3 shows the results for these households. In this sample of single parents, fathers fail to reallocate their time toward the household. Single mothers briefly increase their time spent caring for others, but the effect is only weakly significant.
Third, we show the impact on extended family members (non-household heads or spouses) who are present in households with children under the age of 15 in Fig. C.4. Here extended family members show no clear reallocation of time toward caring. Though, as in the baseline results, men increase their time on household chores. Fourth, we show the effects on households with children under 15 for all family members in Fig. C.5. Here the results are similar to the baseline, where men increase their time spent on the house, and there is little clear increase in time spent caring for others.
Overall, these results suggest that mothers with school-aged children increase their time caring for others for one period, 2020Q4, but the effect is temporary and does not persist. There is less evidence that mothers with young children reallocate their time toward caring, likely due to a high caring burden for this sample before the start of the pandemic. In almost all samples, aside from single-parent households, men increase their household chores but not their allocation of time toward caring for others.
Heterogeneous effects
We also investigate heterogeneity in the treatment effects by marital status, urban status, states with a high HDI, age, and education in Section D. First, we consider whether married and unmarried adults respond differently to the onset of the pandemic in Fig. D.1. For men, the results mostly do not vary by marital status. The one exception to this finding is that married men experience a reduction in formal sector employment. Single men and women suffer comparable losses in formal employment, with married women experiencing the smallest loss.
In Fig. D.1, we see that differences between married and single women are larger relative to differences between married and single men. Stated differently, marital status is more pertinent to the labor market responses of women relative to men. For example, married women show smaller employment losses than single women. Married women also fail to reallocate their time to the household, while single women temporarily increase time on household chores. Married women also show a temporary spike in time caring for others. This last finding aligns with the results from households with children, suggesting that married women with children show a slight temporary increase in time spent caring for others. However, this effect is muted in the full sample of women.
Second, we examine differences by gender across urban and rural areas in Fig. D.2. For the urban/rural divide, the results are similar, except for two points. First, urban men experience larger employment losses on the extensive margin than rural men. Second, urban men and women are more likely to be unemployed than their rural counterparts.
Third, we divide Mexican states by their Human Development Index (HDI), where we separate states into above and below Mexico’s national HDI. We expect states with a higher HDI to be better equipped to respond to the pandemic and potentially have more opportunities for remote work. The results are presented in Fig. D.3. The results are similar across the sample, except that women in high HDI states increase their time spent caring for others in 2020Q4. These findings suggest that women in high-income states reallocate their time toward caring for others, potentially reflecting distinct labor supply norms.
Fourth, we show the impact by age in Fig. D.4 for women and Fig. D.5 for men. For women, the oldest group (55–70) shows the least recovery of employment and no increase in unemployment. For men, younger men (18–25) recover the fastest on both the extensive and intensive margins, mostly due to gains in the informal sector. Those who are 55–70 show the least gains in employment, similar to women.
Fifth, we present the results by education levels of women in Fig. D.6 and men in Fig. D.7. For men, the impact is similar across education levels. Women with higher education experience the least loss of employment overall and show almost no increase in reported unemployment.
Does the ENOE capture the full impact of the pandemic? Alternative findings using the ETOE
ETOE event-study specification
Due to the fact that the ENOE fails to follow individuals through the entire course of the pandemic, we supplement our main findings from the ENOE using the ETOE. The ETOE has two main limitations. First, it was administered via a telephone survey and only represents households with access to a telephone. Second, the ETOE is a smaller sample than the traditional ENOE. While the documentation associated with the ETOE emphasizes that the ETOE is representative at the national level, the ETOE may exhibit bias for particular subsamples and outcomes. Still, despite these limitations, the ETOE also has two advantages. First, the ETOE provides the sole source of information on the impact of the pandemic lockdown on households. Second, the ETOE follows the same individuals over time, allowing us to consider the effect on individuals while controlling for time-invariant characteristics of individuals (individual fixed effects). We thus consider the ETOE to complement our main analysis using the traditional ENOE.
We present the ETOE findings using an event-study design similar to our baseline in Eq. (1). The primary adjustment is that the ETOE follows individuals over months rather than quarters in the post-pandemic period. Thus our focus is on 2019Q2–2020Q1 (four quarters before) and 2020M4–2020M11 (8 months after the pandemic). More formally, our ETOE event-study specification appears as:
$${Y}_{it}={\alpha }_{i}+\mathop{\sum }\limits_{Q=-4}^{7}{\beta }_{Q}\,{{{\mbox{COVID}}}}_{Q}+{{{\bf{X}}}}^{\prime} \gamma +{\epsilon }_{it}$$
(2)
where we follow individuals before and after the pandemic, with the event-study indicator variable, COVIDQ. In this event study, Q represents the period relative to Q = 0, which captures the onset of the pandemic in 2020M4. Q ranges from four quarters before to 8 months after the start of the pandemic. We exclude the quarter before the beginning of the pandemic, Q = −1, which represents 2020Q1, as the baseline period. We also include individual fixed effects as αi, as the ETOE follows the same individuals over time. Due to the individual fixed effects and smaller sample size, we revise our controls to include age and age-squared. ϵit represents the standard error, which we cluster at the individual level. As with Eq. (1), we do not include time fixed effects as there is no variation in timing within each event-study indicator.
ETOE results
Figure 7 shows the results for men (navy triangles), women (purple circles) as well as in aggregate (light blue diamonds). The vertical line represents the excluded period (2020Q1). The gray shaded area represents the three months in the lockdown quarter of the pandemic (2020Q2). The lockdown quarter is of particular interest, as these 3 months are not included in the traditional ENOE.
As anticipated (based on Fig. 2), the results using the ETOE suggest a more substantial impact on labor supply than the traditional ENOE. Beginning with extensive margin employment, the initial reduction in employment appears similar for men and women, a reduction in the probability of working by 17 percentage points. Both men’s and women’s employment starts to rebound by the 3 month of the pandemic, which continues to month seven, where the recovery stalls. Men’s employment recovers faster than women’s employment during months five through seven, similar to the results shown in the traditional ENOE.
Despite the similar changes on the extensive margin, the intensive margin hours worked declines by more for men than women. Men’s hours worked falls by 15 h, while women’s drops by 10 h. However, men’s hours worked quickly starts to rebound, and by the end of the ETOE series, men and women experience similar losses in intensive margin employment.
Turning to employment losses by sector, formal employment fails to recover, similar to the traditional ENOE. By contrast, informal employment declines substantially in the first quarter of the pandemic for both men and women, but men began to recover their informal employment faster than women. As seen in the main results, the informal sector in Mexico leads the recovery during the pandemic.
In the bottom two panels of Fig. 7, men compensate for their employment losses by spending more time on household chores at the start of the pandemic. This spike in household chores is apparent during the lockdown phase of the pandemic but then starts to decline by month three as men return to work. Women also briefly increase their time on household chores in the second month of the pandemic. There is no clear reallocation toward time spent caring for others, for men or women.
Put together, the dynamic effects displayed in the ETOE results show a similar pattern to the traditional ENOE. Two exceptions exist. First, the ETOE results suggest a much larger magnitude of employment loss for households, likely because the ETOE sample includes the lockdown phase of the pandemic. Second, in the ETOE, both men and women briefly increase their time spent on household chores. In the main findings, men’s reallocation of time toward household chores continues for the entire sample, while women show no apparent response. The differences for men’s time use may reflect sample differences or methodological differences due to the use of individual fixed effects in the ETOE analysis.