Covid-19 Lockdown in Ecuador: Are there Gender Differences in Unemployment?

Abstract To fight Covid-19, governments have imposed restrictions on personal mobility and social interactions which may have negative consequences in the labor market. These consequences may be different across demographic groups particularly for female workers. We examine whether the policy that restricted operations in some economic sectors affected formal employment for Ecuadorian female workers differently. We use a difference-in-differences-in-differences model to compare female employees working in restricted economic sectors with other workers, before and after the lockdown policy. The results show that the number of unemployment spells rose by approximately 15 per cent for women working in the restricted economic activities. We also document a decrease in the probability of being employed, which is particularly strong for the youngest women (15–24 years-old), oldest women (45–65 years-old), and less educated female workers. We conclude that the lockdown policy imposed in Ecuador is a plausible explanation for women’s job loss in the formal sector.


Introduction
The Covid-19 crisis is one of the most challenging events that the world has faced in recent decades. According to the World Health Organization (WHO), as of July 2022, Covid-19 is purported to have caused more than 547 millions infections and more than 6 million deaths worldwide. Under the pretext of reducing contagion, a large number of countries undertook non-pharmaceutical interventions (NPIs) to restrict personal mobility and social contact among individuals. These restrictions produced detrimental effects on the economy, which saw the labor market severely affected. For instance, for the Latin American and Caribbean region, according to the International Labor Organization (ILO), the unemployment rate rose from 8.1 per cent (in 2019) to 10.6 per cent (in 2020), which translates to 5.4 million additional job seekers and 30.1 million in total by the end of 2020. fatalities per million inhabitants during the first months of the crisis. According to Cu ellar et al. (2021), Ecuador had an excess death rate of around 64 per cent. 2 Figure A1 in the Supplementary Materials section shows (raw) excess deaths per month during the year 2020 in Ecuador. From here, we see that the peak was around April and May, two months after the onset of mandated home confinement.
According to the ILO (2020), this crisis affected Latin American labor markets in a nonnegligible way. The average unemployment rate for the region reached 10 per cent in 2020 (Comisi on Econ omica para Am erica Latina y el Caribe, 2020). In Ecuador, it was around 13 per cent during the second quarter of 2020 (INEC, 2020). One of the main problems behind job loss in Ecuador is the large number of micro, small, and medium-sized firms, which represent around 95 per cent of all formal companies and generates around 50 per cent of formal employment (Superintendencia de Compañ ıas Valores y Seguros, 2018). Micro and small companies are more vulnerable to bankruptcy than large firms (Beverinotti & Deza, 2020;Camino-Mogro, Ordeñana, & Portalanza, 2020;Carrillo-Maldonado, Deza, & Camino-Mogro, 2020), which makes the Ecuadorian labor market even more sensitive to negative shocks that affect formal employment.
Like many other countries in the region (and the world), Ecuador implemented a strict lockdown policy, which was legally established by means of the Presidential Decree Number 1017 of 2020. In this document, the authorities placed restrictions on personal mobility and closed establishments, beginning on 16 March 2020.
The measures taken by the government included a quarantine of the entire population and restricted mobility that was allowed only in justified cases such as medical emergencies or going to workplaces considered essential. The group of essential economic activities included public and private health services, national security, the financial and insurance sectors, food provision, agriculture and mining activities, electricity and water supply, information and communication, and the transport sector, which was allowed to operate only in relation to health and public security services. On the other hand, the activities affected by the lockdown included wholesale and retail trade (in the section not corresponding to food supply), construction, other types of transportation, accommodation and food services, some types of information and communication, professional services, administrative and support services, and both public and private education (all levels), among others. We will be more specific about affected and nonaffected economic sectors in the following section.

Main sources
Regarding flows from employment to unemployment, we use unique administrative data provided by the Social Security Administration of Ecuador (IESS, in Spanish) from January 2019 to December 2020. This data comprises the universe of new unemployment spells for individuals working in the formal sector, that is, those that contribute to the social security system. The unit of observation is the unemployment spell in an economic sector in a given province, for which there is information on the age, gender, and province of residence of the individual that lost the job. We also observe the type of contribution scheme of the employee, which can be private, public, autonomous or domestic workers. In addition, there is information on the economic activity sector of the firm or institution at the four-digit level according to the International Standard Industrial Classification (ISIC). Finally, we have the date on which the formal job loss occurred. With this, we are able to obtain the outcome of interest, which is the monthly (log) total number of new ended contracts.
From the universe of new unemployment spells, we restrict the sample to unemployment episodes that occurred only in private firms (private sector of the economy) as this was the group that was (officially) targeted with the lockdown policy. In addition, we focus on individuals of legal working age (from 15 to 65 years old) and contributing to the general social security system. 3 This definition leaves us with a total of 2,105,368 new ended contracts in the period of analysis.
With respect to the Covid-19 data, we use official administrative data released by the Ecuadorian institution in charge of registering births and deaths 'Registro Civil'. From here, we construct a variable that contains excess deaths per province and month between January 2019 and December 2020. We use excess deaths, instead of Covid-19 cases as there might be concerns regarding the accuracy of confirmed cases, especially during the so-called first wave of the pandemic in the country. In Figure A1 of the Supplementary Materials section, we show monthly (absolute) excess deaths for the year 2020. We can see that the peak is between April and May 2020 (which is compatible with the peak of Covid-19 cases), and, as mentioned in the introduction, the lockdown restrictions were imposed several weeks before the peak of the crisis.

Treatment and control groups
As pointed out in the introduction, our main approach compares differences across affected versus non-affected economic activities (first difference) and women versus men (second difference), before and after the implementation of the lockdown (third difference).
Regarding the first difference, we rely on what was established in the Decree 1017 (signed by the president of the country) with respect to the restricted economic activities. From here, using the ISIC classification at the four-digit level, we derive both groups. The treated group consists of activities that were ordered (by law) to cease their functioning, such as construction, accommodation and food service activities, education, wholesale and retail trade, and manufacturing, among others, which totals 1,394,708 unemployment spells. In the control group, the economic activities that were not forced to close, we have groups mainly in wholesale and retail sectors (in the section that corresponds to food provision), human health and social work activities, financial and insurance activities, and manufacturing (in activities that corresponds to vital goods), which adds up to 710,660 new ended contracts. In total, we have 21 groups of economic activities that we categorize into affected or non-affected sectors depending on whether the economic activity at the 4-digit ISIC level was restricted by law. 4 An extensive list with the treated economic activities is shown in Table A1 of the Supplementary Materials section. In addition, in Table 1, we present the total number of unemployment spells across gender and affected and non-affected economic sectors. For the second difference, we construct the female indicator variable using the reported sex of the employee. We generate a dummy variable which takes the value of one for women and zero for men.

Descriptive evidence
We start by exploring the monthly evolution of new unemployment spells in 2019 and 2020 across treated and control groups. Panel (a) of Figure 1 shows raw trends for women working Covid-19 lockdown in Ecuador 837 in restricted sectors versus women in non-restricted sectors. We see that the first group seems to experience a steeper increase in the number of unemployment spells right after the implementation of the lockdown. When we compare unemployment spells for women working in affected sectors with their male counterparts (men working in restricted sectors), both groups seem to have experienced increases in the total number of new ended contracts as depicted in panel (b). In any case, this is only descriptive evidence that we will contrast in a later regression framework. We also present in Figure 1 raw trends for monthly unemployment spells separately by gender, that is, female vs male. As one can observe from panel (c), both groups seem to have followed very similar trends, though men, in absolute numbers, show more unemployment spells. Once more, this is a first bit of descriptive evidence that we need to explore further. Moreover, panel (d) depicts the evolution of unemployment spells comparing restricted sectors vs nonrestricted sectors, where the affected group shows a steeper positive slope. Finally, in Figure A2 of the Supplementary Materials section, we present the monthly evolution of unemployment spells for the entire economy.

Motivating difference-in-differences estimates
The baseline specification is a difference-in-difference-in-differences (DDD) which compares differences across economic activities (first difference) and gender (second difference) before versus after the lockdown in Ecuador (third difference). Why might the COVID-19 lockdown in Ecuador be particularly harmful for the employment (measured through unemployment spells) of women working in restricted economic activities? First, as discussed in the introduction, there are a number of studies that find that this crisis negatively affected employment more for women than for men. According to Alon, Doepke, Olmstead-Rumsey, and Tertilt (2020), one reason for this may be that women's employment is concentrated in less cyclical sectors such as health care and education. In addition, the COVID-19 crisis had a bigger impact on activities requiring more in-person contact such as services, in which women represent an important share of the labor supply. Regarding the situation in Ecuador, according to the National Employment Survey (ENEMDU in Spanish) for the year 2019, 51.9 per cent of individuals employed in the wholesale and retail trade sectors, 66.49 per cent in accommodation and food service activities, 65 per cent in the education sector, and 62.30 per cent in other personal service activities are women. These sectors were hit the hardest by the closure of economic activity. Second, these activities are also less likely to be performed on-line or via telecommuting, which creates differences between male and female labor outcomes. Third, in Ecuador (as in many other countries), educational activities for children and adolescents moved online, which might have increased the demand for childcare at home, a task that is more likely to be carried out by women (Alon et al., 2020). 5 From Figure 1, we learned that women working in affected economic sectors seemed to experience a sharper increase in the number of unemployment spells after the lockdown compared with women working in non-affected sectors. We assess this change by performing a difference-in-differences regression, which compares female workers in restricted sectors with women in non-restricted sectors. To do this, we collapse female unemployment spells data at the province, economic sector, and time (month-year) level, and estimate an equation of the form: where TreatSector s is a binary indicator that takes the value of one for restricted economic sectors and zero otherwise. Post mt denotes the pre-post lockdown order. Y spmt is the natural logarithm of the number of new unemployment spells in the female sample. We also include controls Controls pmt to account for the regional variation in the spread of Covid-19, the share of excess deaths, in province p, month m, year t. We include a set of fixed effects: (i) month c m and year s t fixed effects to account for potential common time shocks across units, (ii) economic sector d s , and province fixed effects q p accounting for time-invariant heterogeneity in each labor market. There are 21 economic sectors and 24 provinces. We cluster standard errors at the economic sector level since this is the level at which the effect takes place (Cameron, Gelbach, & Miller, 2008;Wooldridge, 2003). As there are 21 clusters, we use wild bootstrapped clustered standard errors, applying 999 replications. 6 In Equation (1), b 3 captures the post-lockdown effect of being a woman working in affected economic sectors relative to women working in non-affected sectors. Column (1) of Table 2 shows the results. As expected, the coefficient of interest is positive and significant, which suggests that the number of unemployment spells for women working in restricted sectors increased by approximately 41 per cent compared with women working in other sectors after the implementation of the lockdown.
Furthermore, it is also relevant to analyze differences in unemployment spells between women and men working in restricted economic sectors. For this, we focus on new ended contracts that took place in the affected economic sectors and collapse the data set at the province, gender and time (month-year) levels. Formally, we estimate a DD model that compares women with men: In this specification, the outcome of interest, Y sgpmt , the variable Post mt , controls Control pmt , and the fixed effects are the same as in Equation (1). Female g is a binary variable that takes the value of one for female workers and zero for male workers. Standard errors are clustered at the economic sector level (12 clusters) and computed using wild bootstrapped replications. In this equation, b 3 captures the post-lockdown effect of being a woman relative to men (when both are) working in the affected economic sectors on the (ln) number of new unemployment spells. From Column (2) of Table 2, we see that the number of new unemployment spells after the lockdown was mandated increased by about 7.28 per cent.
We also perform the same specification as in Equation (1) for two additional sub-samples: unemployment spells (i) for males and (ii) in the whole economy. Columns (3) and (4) of Table 2 show the estimated coefficients. When we compare unemployment spells for male workers working in the affected sectors with those for male workers in non-affected sectors, we do not find a significant result. Moreover, when comparing unemployment spells in the Notes: Column (1) shows estimates for DiD that compare women working in restricted sectors (treated group) relative to women working in non-restricted sectors. Column (2) reports DiD estimates that compare women working in restricted sectors (treated group) with men working in restricted sectors beforeafter the mandated quarantine. Column (3) reports DiD estimates that compare men working in restricted sectors with men working in non-restricted sectors. Column (4) reports DiD estimates that compare all workers working in restricted sectors with workers in non-restricted sectors. p-Values refer to standard errors clustered at the economic activity level in square brackets, calculated using the wild cluster bootstrap with 999 replications. ÃÃ p < 0.05, Ã p < 0.10. 840 G. Armijos-Bravo and S. Camino-Mogro restricted economic sectors relative to the non-affected sectors for the whole economy, we do not find a significant coefficient, either.

Baseline equation
The differences found in the previous subsection motivate a DDD estimator which includes differences over time, gender, and economic sectors. Recall that we are interested in assessing whether the lockdown policy affected women differently than men in terms of the number of inflows from employment to unemployment in the formal sector (private sector). To formally assess this, we collapse the unemployment spells data at the province, economic sector, gender, and time (month-year) levels, and estimate a DDD equation of the form: where Y spgmt is the natural logarithm of the number of new unemployment spells in economic sector s, province p, gender group g, month m, year t. The Post mt , Treat s , Female g , and Controls pmt variables are defined in the same way as in Equations (1) and (2). The coefficient of interest is b 6 , which captures the effect of the lockdown on new ended contracts for women working in restricted economic sectors. We also include month c m , year s t , economic sector d s , and province fixed effects q p . 7 There are 21 economic sectors and 24 provinces. We compute wild bootstrapped clustered (at the economic sector level) standard errors, applying 999 replications.
With respect to the outcome variable, which is the natural logarithm of the number of new unemployment spells, it would be ideal to use instead the ratio of work separations over the number of employees per province, time and economic sector. However, because of limitations on data availability, the working population disaggregated at the province and economic sector levels is not available for use by the authors. 8 Using a ratio instead of the number of work separations, may help to have a relative measure that takes into account the size of the workforce. Despite this drawback of the data available, one should keep in mind that our estimates are still a first step in quantifying gender differences in unemployment spells. Using a relative measure such as the proportion of unemployment spells over the number of employed individuals will provide a more accurate picture of the relation between the lockdown and unemployment spells.
The difference-in-differences estimator intends to provide an unbiased estimate of the treatment effect in a situation in which, in the absence of the treatment, the outcome in the two groups would have followed the same trend. This is more critical for the pre-treatment period; that is, one needs to add evidence that supports the parallel trend assumption before the implementation of the policy. On this point, Figure 1 provides preliminary descriptive evidence that adds validity to this assumption. In panel (c), we show the evolution of unemployment spells (in the private sector) for female and male workers. Panel (d) depicts the evolution of the outcome variable for restricted and non-restricted economic activities. From here, we can see that, in general, pairwise groups follow similar trends before the lockdown policy. We also see a sharp increase in the number of new ended contracts right after the restriction for the affected economic activities. Though the same increase is seen in the non-affected sectors, the increase seems to be sharper for restricted sectors. 9 Our identification strategy is supported by several facts. The first is the exogenous characteristic of Covid-19. Second, as mentioned before, the lockdown was established before the peak of contagion in Ecuador. Third, we analyse private sector workers as there might be important differences between public and private employees. Fourth, the DDD specification allows us to Covid-19 lockdown in Ecuador 841 control for two kinds of potentially confounding trends: changes that affect the number of unemployment spells of women unrelated to the lockdown, and changes that affect unemployment spells of all women and men working in the affected economic sectors. Finally, we provide, in the following sections, several robustness checks to support our results. For instance, we perform an event study to check for potential pre-trends. We also use placebo lockdown dates and restrict the sample by excluding the most populated provinces in the country. Table 3 shows the estimates of Equation (3). We report results for the outcome of interest, first without any disease-spread control variable, and then including excess deaths. Both regressions include province fixed effects, month and year fixed effects, and economic activity fixed effects. Standard errors are clustered at the economic activity level.

Results on unemployment spells
From the results of Column (1), we see that the restriction policy is associated with an increase of approximately 15.06 per cent (estimated coefficient 0.1506) in the number of new unemployment spells for women working in the affected economic sectors relative to other workers. This difference is statistically significant at the 5 per cent level. In Column (2), when we include the share of excess deaths per province and month, the estimated coefficient is very similar in magnitude and significance and points to a detrimental result for female employees. In this column, we also observe that the disease-related control variable is negative though very close to zero and slightly significant. 10 The results above suggest that female workers might have been differently hit by the lockdown. In fact, several studies indicate that the demographic composition of (un)employment varies between low and high education level, race, age group, and gender. For instance, Hoynes, Miller, and Schaller (2012) find that general recessions are worse for men. However, literature on Covid-19 and labor market outcomes for the United States of America finds that women seem to have been more affected in terms of unemployment and hours worked (Beland et al., 2020;Cajner et al., 2020;Forsythe, Kahn, Lange, & Wiczer, 2020;Mongey, Pilossoph, & Weinberg, 2021). In this line, there are several studies conducted (under the Covid-19 framework) for the U.S. that show larger impacts on women (Beland et al., 2020;Forsythe et al., 2020;Mongey et al., 2021;Montenovo et al., 2022), but there is also some evidence that suggests that during the first stages of the lockdown, male employees were more affected (B eland, Brodeur, & Wright, 2020). With respect to research on Europe, Fana, P erez, and Fern andez-Mac ıas (2020) find mixed results in terms of gender differences across countries like Spain, Poland, Germany and the U.K. On the contrary, Farr e et al. (2022), using household survey data from Spain, find that women were slightly more likely to lose their job than men.
Regarding results for the Latin American scenario, to the knowledge of the authors, there are no documents that explore gender differences in labor market outcomes (under the Covid-19 context). The closest work is that of Morales et al. (2022), who estimate the impact of Covid-19 restrictions for Colombia, where they find an estimate equivalent to 13.9 per cent more jobs lost in the excluded sectors relative to the non-excluded sectors. As seen, there is mixed evidence on whether women or men were more affected, so there is a need to explore the source of potential differences as Montenovo et al. (2022) establish in their research.
As we pointed out in previous sections, women tend to be more concentrated in sectors that require more in-person contact, but the composition between men and women might also be different within restricted and non-restricted sectors. Understanding these differences may be useful to interpret the results we find. To examine whether women and men are similar within affected and non-affected sectors, we provide in Table A3 of the Supplementary Materials section some descriptive statistics for selected characteristics comparing female and male workers within treated and untreated economic sectors, separately. For this, we use survey data from the 'National Survey of Employment' (ENEMDU) of Ecuador and restrict the sample using the same considerations as in Equation (3). Table A3 (Supplementary Materials) shows the results for the share of low educated workers, defined as those who have less than 10 years of education; the monthly payment, which corresponds to average monthly employment income; and the share of blue-collar workers, which corresponds workers performing manual labor.
The results show that in the treated sectors, the share of low educated female workers is lower than the share of male workers, and the difference is statistically significant. With respect to the monthly payment, on average women earn more compared to men, being the difference statistically significant. Likewise, women represent a lower share of blue-collar workers compared to men, the difference between men and women is also statistically significant. This may be because less educated women in Ecuador tend to do unpaid/paid domestic work, activity that is not considered in this analysis. Therefore, women who work in the treated sectors are probably more educated, which is compatible with a higher salary, and less probability of having a blue-collar position. The World Bank (2019) in a report for gender gaps in Ecuador find that almost 25 percent of employed women are unpaid workers (in or outside the household) or perform domestic chores, as compared to 6.6 percent of men performing unpaid work. They also find that women dominate in sectors of domestic service, hotel and restaurants, education, and services, while the representation of women in services like construction, manufacturing, transportation, and infrastructure is rather low. They also highlight that as a reflection of increased education, there are more women than men performing 'Professional' occupations.
We find something similar in the untreated sectors, in which we see that the share of low educated workers is lower for women, women earn on average a higher salary, and represent a lower share of blue-collar positions. The differences between men and women are all statistically significant, except for the average monthly payment.
In general, for the characteristics and sample analyzed, we might think that women working in the affected sectors are less likely to lose their jobs, because they are more educated, with higher average salaries and less blue-collar positions compared to men. This descriptive evidence might support the statement that the lockdown policy implemented in Ecuador is highly associated with job losses for women working in the restricted sectors.

Covid-19 lockdown in Ecuador 843
Overall, the results of this section suggest that the lock down policy is negatively associated with an important loss in the extensive margin of formal jobs for women working in the restricted economic sectors. This finding is relevant because these individuals might transition from the formal to the informal labor market. Informality has several disadvantages for workers. For instance, it is related to lack of health insurance, contributory pensions, limited career advancement, and to higher levels of inequality (Busso, Camacho, Messina, & Montenegro, 2021;Messina & Silva, 2018). Finally, it would also be interesting to assess differences in the number of hours worked and wages. However, this information from administrative records is not available for external use.

Probability of being employed and cause of job loss
So far, we find evidence that supports an increase in unemployment spells for women working in the restricted sectors relative to other groups of workers. Now, we go a step further and explore whether there are any changes in other labor market outcomes such as the probability of being employed, full time jobs, and cause of job loss.
We use survey data taken from the 'National Survey of Employment' of Ecuador (ENEMDU, for its name in Spanish), released by the National Institute of Statistics (INEC) for the years 2019 and 2020. The data is a pooled cross-section at the individual and quarter level. Each cross section consists of individual level data. The ENEMDU constitutes the official source for calculating labor market indicators in the country and is nationally representative, including both urban and rural areas (INEC, 2022). Regarding labor market information, the survey offers data on employment, job searches, hours worked, and cause of job loss, among others.
For this analysis, we use three labor market outcomes: (i) the probability of being employed, (ii) the probability of having a full-time job, and (iii) the cause of job loss (conditional on having moved from employment to unemployment). For the first outcome, we construct a dummy variable that takes the value of one when the individual reports that he/she worked the previous week. In this question, all types of jobs such as full-or part-time are included, and zero corresponds to unemployed respondents. For the second outcome, we construct a binary variable when the individual declares to have worked 40 hours or more (which corresponds to full-time jobs) and zero otherwise. In this way, we measure the probability of having worked 40 or more hours the previous week. With the last outcome, we intend to assess whether women working in the restricted economic sectors voluntarily left their jobs 11 or not. We construct a dummy variable that takes the value of one when leaving work corresponds to a voluntary act and zero when it corresponds to being fired.
As in our baseline estimates, we focus on individuals who are between the ages of 15 and 65 and work for the private sector. We exclude the agricultural and mining sectors, domestic workers, and self-employed individuals.
To formally explore these labor market outcomes, we rely on the same econometric approach as in Equation (3). We use a DDD approach which compares restricted and non-restricted economic sectors (first difference), women and men (second difference), before vs after the lockdown policy (third difference). We estimate an equation of the form: where Y igst represents one of the three labor market outcomes. Female g takes the value of one for women and zero for men. Post t is a dummy variable for the pre-post lockdown, and Treat s takes the value of one if the economic activity in which the individual works was restricted. In addition, we include a set of control variables such as the share of excess deaths (at the country level) Deaths t . At the individual level, we control for education attainment, marital status, age group, and urban/rural location, d 0 X i : 12 Control and treatment groups are defined and constructed as in Equation (3). We cluster standard errors at the economic activity level, and compute them using the wild cluster bootstrap with 999 replications. We use probability weights to account for the survey design. Table 4 presents the results for the exercise described above. Column (1) shows a decrease of 5.76 percentage points in the probability of being employed for women in restricted economic activities relative to other groups of employees after the lockdown policy. This represents a decrease of 6.35 per cent with respect to the pre-policy mean. With this result, we support the statement that the lockdown policy might have disproportionately affected women compared with men. In addition, we find evidence that points to a deterioration in employment and not only an increase in unemployment spells for our group of interest.
Column (2) presents the results for the probability of working 40 hours or more the previous week. Here, even we do not find a significant result, it points to the expected side, that is, a decrease in the share of women (in the restricted economic sectors) working full-time. Finally, Column (3) shows that the probability of a voluntary termination of the work contract is lower for female employees (working in the restricted economic sectors) compared with other groups of workers (however, as we will show in the following paragraph, we do not make conclusions based on this outcome because of the existence of a pre-trend). The difference represents a decrease of around 29.14 per cent in the share of voluntary job loss with respect to the pre-lockdown mean. In spite of the slightly significant result of outcome 1, we still consider it is interesting to further explore it across socioeconomic characteristics, an exercise that we will perform later in this document. Notes: Estimates from Equation (4). All outcome variables are dummy indicators. The time horizon includes the years 2019 and 2020 at the quarter level.
The year 2020 has no information during the first trimester. p-Values refer to standard errors clustered at the economic activity level in squared brackets, calculated using the wild cluster bootstrap with 999 replications. ÃÃÃ p < 0.01, ÃÃ p < 0.05, Ã p < 0.10.

Covid-19 lockdown in Ecuador 845
To add support to the validity of the results of this section, we perform an event study version of Equation (4) for the three outcomes. In Figure A3 of the Supplementary Materials section, we plot the set of DDD coefficients along with their 90 percent confidence intervals. We include four pre-and three post-lockdown periods at the quarter level for each outcome. The post lockdown periods are the second, third and fourth quarters of year 2020. 13 In these regressions, we use the same control variables, fixed effects, and method for computation of standard errors as used to obtain the estimates in Table 4. Overall, we do not find evidence of pre-trends in any of the outcomes, except for voluntary job loss, outcome for which we cannot make more conclusions. For the probability of being employed, in panel (a), none of the pre-policy coefficients are significant and are of a very small size (close to zero). For the post period, we find one coefficient significant at the 10 per cent level, which is compatible with the corresponding result of column (1) of Table 4. A similar pattern is seen for the probability of working fulltime. All pre-event estimates are very close to zero and not significant. With respect to post-policy estimates, we do not find significant results, but signs for the last periods are as expected.
Overall, we find evidence that the COVID-19 lockdown is associated with an increase in job loss for women working in the affected sectors compared to other workers. Furthermore, we find a significant increase in the number of unemployment spells, a result that is supported by a general decrease in the probability of being employed. According to the International Labor Organization ILO (2021), women suffered a disproportionate loss of employment and income because of their prominent presence in the most affected sectors, such as accommodation and catering services. In addition, the ILO states that The Americas region experienced the greatest loss of female employment as a result of the crisis (À9.4%).

Heterogeneity analysis: age groups and education level
In the previous subsection, we found evidence that suggests a decrease in the probability of being employed for women working in the restricted economic sectors. As a consequence, one may wonder whether there are differences across socioeconomic characteristics. To answer this question, we now focus only on the first outcome (probability of being employed) because the 'cause of job loss' showed a pre-trend. We begin by splitting the sample into age group categories and construct four groups: (i) 15-24 y/o, (ii) 25-34 y/o, (iii) 35-44 y/o, and (iv) 45-65 y/o, where each group corresponds to 20 per cent, 33 per cent, 23 per cent, and 22 per cent of the total number of respondents, respectively. Formally, we estimate Equation (4) for the four different (age) sub-samples. Table 5 shows the results of this exercise.
For all four age groups, we obtain the expected signs, that is, a decrease in the probability of being employed the previous week. Interestingly, we find that the biggest coefficient corresponds to the oldest group (45-65 years old) of women working in the restricted economic sectors. We find a decrease of 15.69 percentage points in the probability of being employed. Moreover, we also find a statistically significant decrease for the youngest group, of around 11.87 percentage points.
Under the Covid-19 economic shock context, older groups of workers seem to be more negatively affected (Bui et al., 2020). This group might have more difficulties adapting to new technologies (Pit et al., 2021), which might place them in a more vulnerable position and therefore increase their probability of job loss in a situation such as the Covid-19. A plausible explanation for these findings may be that, in Ecuador, older individuals are concentrated in the economic sectors that were restricted because of the lockdown. For instance, according to data from ENEMDU year 2019, 35.6 per cent of employees working in these sectors were between 45 and 65 years old. Likewise, older women represent around 51.38 per cent of the workforce in the wholesale sector, 71.84 per cent in accommodation and food service activities, and 63.78 per cent in the education sector. These sectors were the most severely affected because of the nature of their activities.
Younger groups of workers are also affected by the lockdown. For example, Abraham et al. (2022), using data from India, find that women were seven times more likely to lose work during the nationwide lockdown, and in a sample stratified by gender, found that young workers, whether men or women, were more likely to face job loss. Gustafsson (2020) found, for the United Kingdom, that both younger and older workers experienced the brunt of the hit to jobs during the Covid-19 crisis. In the same line, Bui et al. (2020), using 'Current Population Survey' data from the U.S., found that the recession generated by the Covid-19 crisis (during its 'first wave') disproportionately affected older workers. In addition, these authors also found that women reached higher unemployment rates than men across all age groups. Finally, in research conducted for Mexico, Hoehn-Velasco et al. (2021) found that the youngest workers, the oldest workers (over 60 years old), and low-income earners were the most impacted in terms of formal employment.
The result for older workers is of particular interest because they may face more difficulties to reenter the labor market. For instance, they may encounter more age discrimination in hiring. It may also take longer for older workers to find a job (Neumark & Button, 2014), which is particularly problematic as this could contribute to a decline in earnings, less savings, less access to health care use, expanding working years before retirement, and reductions in quality of life (Bui et al., 2020).
We continue to explore whether there are differences across workers' education levels. To do this, we split the sample into three groups: (i) basic education, which corresponds to up to 10 years of studies, (ii) secondary education, which corresponds to 13 years of studies and the end of non-tertiary education, and (iii) tertiary education, which corresponds to a professional degree. The composition of the sample is as follows: 26.20 per cent of individuals have a basic education, 46.21 per cent have a high school diploma, and 26.84 per cent have a professional degree. As we did for the age group estimates, we use the same econometric specification as in Equation (4) for the three sub-samples where the outcome is the probability of being employed. Table 6 shows the heterogeneity across education level groups. For all groups, we get the expected signs, that is, an average decrease in the probability of being employed. However, less educated workers seem to be driving the results. In particular, we find that the group of individuals with basic education experienced a decrease of 12.78 percentage points in the probability of being employed. These results provide evidence that supports that less educated women working in the restricted economic sectors might be disproportionately affected by the lockdown. One explanation for this might be that less educated women may be more concentrated in economic sectors that require more in-person interactions and therefore would be more affected by mobility restrictions. Our results are in line with others in the literature. For instance, Escalante and Maisonnave (2022), using data from Bolivia, find that women suffered more than men from the negative impacts of the Covid-19 crisis in terms of employment. In particular, they find that unskilled women were the most affected, as they experienced the greatest losses of formal employment. Likewise, Adams-Prassl et al. (2020) found that women and less educated workers were more affected by the crisis in the German context. Holder et al. (2021) found evidence that supports greater job losses for less educated workers in the U.S. Likewise, Lee et al. (2021) show that the initial negative impact on employment was larger for women, minorities, and the less educated (also for the U.S.). Finally, Aldan et al. (2021), using quarterly Turkish household labor force surveys, found that the pandemic decreased employment and labor force participation of almost all groups, with women more affected than men, along with the least educated.

Conclusions
The Covid-19 crisis has seriously affected the world in various dimensions. Governments isolated their citizens by imposing a strict quarantine referred to as a 'lockdown', which produced an economic crisis whose effects on labor markets were particularly damaging. Ecuador also applied a lockdown and a number of activities considered 'non-essential' were forced to stop their operations.
Economic crises may affect demographic groups differently. In this line, several studies conducted by international organizations show that women are disproportionately affected in terms of job loss, declining earnings and hours worked. In addition, women may also face delays reentering the labor force, which reinforces already existing gender gaps. For these reasons, assessing whether the lockdown affected women more than men in terms of job loss becomes relevant.
In this paper, we evaluate the short-term relation of the lockdown imposed in Ecuador in March 2020 with labor market outcomes for women working in restricted economic sectors. Using a triple difference-in-differences approach, we find that the lockdown policy is associated with an increase in the number of unemployment spells for the population under study. We also show that the probability of being employed decreased. These results are consistent with others in the literature that find women more negatively affected in terms of labor market outcomes as a consequence of the Covid-19 (economic) shock (Bluedorn, Caselli, Hansen, Shibata, & Tavares, 2022).
We also add evidence that supports differences across age groups and educational level. We find that younger, older, and less educated women (in the restricted economic sectors) seem to have been more affected than their male counterparts. These results are consistent with Escalante and Maisonnave (2022), Hoehn-Velasco et al. (2021), and Serrano et al. (2019), who find that specific demographic groups were more affected, at least in the short term, after the Covid-19 lockdown.
The results are robust to a battery of robustness checks, which adds validity to the estimates. However, there are some limitations that stem mainly from data unavailability. First, we observe inflows into unemployment but not outflows. Because we are interested only in shortterm relations and, in the immediate post-lockdown period, the hiring of new employees was very unlikely, we believe our results are still of relevance. Second, it would be ideal to work, as an outcome, with the share of unemployment spells over the working population. Unfortunately, to date, we have not gotten access to this information, so our results should be interpreted taking this into account. Third, we acknowledge that there might be other factors that influence both supply and labor demand such as increases in transportation costs, reductions in production capacity, and decreases in job searches, among others. Fourth, this evidence should be interpreted with caution, as it is correlational. Unfortunately, the labor market is different for men and women in Ecuador and this causes men and women to have different characteristics, so their comparison leads us to associative conclusions.
Despite these drawbacks, we believe that the lockdown policy was a channel associated with the increase in unemployment spells for female workers in the formal sector of the Ecuadorian economy. We believe our results are of relevance in assessing the short-term consequences of the Covid-19 lockdown on the Ecuadorian labor market.
Finally, providing evidence with regards to (formal) job loss in a context such as the Ecuadorian becomes not only interesting but relevant. Ecuador is a country with high shares of informality, which might have increased because of the transition from formal employment to unemployment. Moreover, at the individual level, having no formal job in Ecuador is linked to losing access to social security benefits such as health care, personal loans, etc. Likewise, older groups may face more barriers to rejoining the labor market, which may negatively affect earnings, savings, and health care use and increases the chances of falling into poverty. In this sense, having a clearer panorama of the labor market consequences of the lockdown policy is an important tool for policy makers to design programs or strategies focused on the most affected groups to objectively target public policies.
Notes 1. For an extensive literature review of this topic, see Brodeur, Gray, Islam, and Bhuiyan (2021). 2. For an extensive review of statistics on COVID-19, see Roser and Hasell (2020). 4. We use this aggregation because using the economic sectors directly at the four-digit level may generate groups with too few observations, which in the estimates may raise concerns about few observations within clusters. 5. In Ecuador, education moved online on 13th March 2020, a situation which lasted for more than a year. 6. For computation, we use the boottest Stata command (Roodman, Nielsen, MacKinnon, & Webb, 2019). 7. Table A2 in the Supplementary Materials section presents the estimates after including gender-specific linear time trends and economic sector-specific linear time trends, separately. The results are very similar. 8. The official national statistics institution of the country, "Instituto Nacional de Estad ısticas y Censos" (INEC), does not publish disaggregated data on the workforce. This data is available only at the national level. In addition, to this date, we do not have access to the number of employees that contribute to social security for the periods of our analysis. 9. Though descriptive graphs are a valid tool, we have to keep in mind that they do not consider covariates or fixed effects as does a regression framework. In the supplementary materials section, we provide additional evidence that supports parallel trends. 10. As the pandemic is also very dynamic, as a robustness, we replace year and month fixed effects with month-year fixed effects. The results from this exercise are very similar and available upon request. 11. Voluntary termination occurs when an employee makes the decision to leave a job or end a contract early. 12. The ENEMDU for the period of analysis does not provide desegregation at the province level. Therefore, we can only control for the urban or rural place of residence of the respondents. 13. The periodicity of the ENEMDU survey is quarterly.