1 Introduction

In 2020 an unexpected shock dramatically affected the lives and jobs of everyone in the world: the COVID-19 pandemic. Latin America was not an exception. In an attempt to contain the spread of the disease, governments imposed national lockdowns, school closures, travel restrictions, and social-distancing measures. These measures contributed to saving lives but at the same time they inevitably brought negative consequences in the labor market. The economies of the region experienced an unprecedented increase in job losses, unemployment and income reduction.

One of the asymmetries that were early noticed is that of gender: women were hit harder by the crisis than men (Alon et al. 2022a, 2022b, Viollaz et al. 2022). In particular, job losses were significantly larger among women, implying a setback with respect to the advances of recent decades in terms of gender equality, and also limiting the possibilities of recovery for the economies of the region. The strong negative shock suffered by women and the difficulties in resuming their past work trajectories threatens the United Nations’ Sustainable Development Goals (SDGs) of achieving gender equality and promoting sustained economic growth and full employment–goals 5 and 8 of the SDGs.

At least two factors could account for the gap between women and men in the impact of the crisis. The first one is the presence of children at home and the associated childcare activities. In a context of lockdowns and school closures, if social norms assign childcare mostly to women, only mothers with the ability to work from home could have been able to reconcile their labor market activities with their family responsibilities. A second factor that could contribute to explain the gender asymmetry is occupational segregation: if women were employed in sectors more affected by the pandemic and worked in occupations that are more difficult to perform from home, the impact would be understandably harsher.

By analyzing behaviors and reactions to an unexpected strong negative shock–the COVID-19 pandemic–this paper contributes to the understanding of some fundamental development issues, such as the hindrances to women’s labor force participation, the role of social norms and the opportunities opened by new technologies. Our analysis is focused on Latin America, a region where labor gender gaps are among the widest in the world, and where the impact of the pandemic has been particularly strong in terms of lives, jobs, incomes and welfare.

We study these issues with the help of one of the most ambitious data sets collected immediately after the onset of the pandemic: the World Bank’s High-Frequency Phone Surveys (HFPS). These surveys were carried out in three waves between May and August 2020 in over 100 countries around the globe, including 13 in Latin America. Variables in the HFPS were harmonized by the World Bank, which helped foster a growing literature (Ballon et al. 2021, Cucagna and Romero 2021, Khamis et al. 2021, Kugler et al. 2021, Mejia-Mantilla et al. 2021).

We contribute to this initiative by carefully codifying the occupation variables in the HFPS, which allows us to construct measures of potential for working from home, a crucial factor to cope with the shock generated by the pandemic and the containment measures. We compute a work-from-home measure following the proposal of Dingel and Neiman (2020) based on occupational characteristics from O*NET. Given that in some occupations the possibility of teleworking depends on having internet access, which is far from universal in Latin America, we adjust our individual work-from-home measures by taking into account home internet access reported in the HFPS (Garrote-Sanchez et al. 2021).

Our main analysis is based on regression models of job losses. We focus the analysis on the effect of two factors that may account for the severity of the shock: the individual’s potential for working from home and some sociodemographic characteristics such as gender and number of children at home. We control for several factors and include fixed effects by occupation and by country. We also explore the role of differences in the stringency of social-distancing measures across countries and over time.

We first confirm two important results: (1) the impact of the COVID-19 shock was strongly decreasing in the possibility of working from home, and (2) women were more likely to lose their jobs than men. Importantly, we also find that the alleviating effect of working from home on the severity of job losses was especially relevant for women with children. In particular, the coefficient for the interaction between the work-from-home measure and the dummy for woman is large and statistically significant in the sample of households with children but not among those without children. The result holds when we control for occupation fixed effects, implying that even among women in the same occupation, the ability to work from home was pivotal among those with children. This evidence is consistent with a plausible mechanism: given the traditional intrahousehold distribution of childcare responsibilities in Latin America, women with children were more likely to stay home due to school closures, and therefore the possibility of working from home became very relevant for them to keep their jobs.

We organize this paper as follows. Section 2 briefly reviews the literature. Section 3 introduces the data used in the analysis, with special emphasis on the treatment of the occupation variables, and discusses our work-from-home. The acronym is introduced later in the text measure. Section 4 presents the main results of the paper by assessing the gender asymmetries in the relationship between job losses, the ability to work from home and childcare responsibilities in a regression analysis setting. Section 5 explores potential heterogeneities in the results according to the level of gender inequality and the stringency of social-distancing measures applied to cope with the pandemic in each country. We conclude in Section 6 with a discussion of the main findings and the policy implications.

2 Literature review

Despite the convergence of roles of men and women in labor markets over the last century, gaps remain considerable (Goldin 1995, 2006, 2014, Blau and Kahn 2017) and motherhood and the associated care needs stand out as the key factors in explaining these remaining gender gaps.

Recent literature has documented the existence of large and persistent child penalties on women’s labor market outcomes worldwide (e.g., Angelov et al. 2016, Kleven et al. 2019a, 2019b, Kuziemko et al. 2018, Berniell et al. 2020). The presence of children at home decreases women’s labor supply and employment, while those who remain in the labor market tend to choose more flexible jobs–part-time jobs, self-employment, informal jobs–to balance family and work (Kleven et al. 2019b, Bertrand et al. 2010, Goldin and Katz 2011, 2016 Berniell et al. 2021a, 2021b, Berniell et al. 2022b). The evidence points to childcare responsibilities, rather than biology, pregnancy, or marriage, as responsible for these effects (Berniell et al. 2022a, Kleven et al. 2021).

The effect of this extra burden on mothers due to childcare responsibilities became even more apparent during the COVID-19 pandemic, when schools closed and childcare had to be resolved at home. Our paper is related to the strand of the economic literature on the COVID-19 pandemic that focuses on gender differences in the impact of the pandemic (Adams-Prassl et al. 2020, Alon et al. 2022a, Alon et al. 2022b, Bluedorn et al. 2022, Copley et al. 2020, Costoya et al. 2021, Cucagna and Romero 2021, del Boca et al. 2020, 2022, de Paz et al. 2020, Farré et al. 2022, Goldin 2022, Sevilla and Smith 2020, Zamarro and Prados 2021). These papers typically find that female workers were among the most negatively impacted in the early stages of the pandemic, and suggest two drivers of this asymmetry: an increase in caregiving responsibilities and a particularly large fall in the activities in which women were predominantly employed.

Our paper is also related to another strand of the economic literature on the COVID-19 pandemic that looks at the issue of working from home and the ability to teleworking (Adams-Prassl et al. 2022, Dingel and Neiman 2020, Garrote-Sanchez et al. 2021, Gottlieb et al. 2021, Saltiel 2020, Berniell and Fernandez 2021, Delaporte et al. 2021). This line of research highlights the role of working from home in alleviating the impact of the shock and stresses the asymmetries in the ability to teleworking among socioeconomic groups.

This paper also relates to the regional studies that examine the impact of the shock in Latin America (Busso et al. 2020, de la Flor et al. 2021, Delaporte et al. 2021, ECLAC 2020, Lustig et al. 2020). These studies find that Latin America was one of the regions hardest hit by the COVID-19 pandemic and the containment measures, with staggering costs in terms of lives, jobs, incomes and welfare.

Finally, this paper is related to the literature that studies the impact of the shock exploiting the World Bank’s HFPS surveys both around the globe, and specifically in Latin America (Ballon et al. 2021, Cucagna and Romero 2021, Khamis et al. 2021, Kugler et al. 2021, Mejia-Mantilla et al. 2021). The richness of these surveys allows exploring the impact of the shock in the labor market as well as in education, food security and other outcomes.

Our results contribute to this literature by shedding light on how the possibility of teleworking alleviated the initial costs of the crisis, and on the role of working from home and childcare behind the gender asymmetries generated by the pandemic in Latin America. More generally, our paper contributes to a better understanding of the labor market impact of the pandemic and the social-distancing measures in the specific context of developing countries.

3 Data

Our main source of data is the High-Frequency Phone Surveys (HFPS) conducted by the World Bank in 2020 to assess the impact of the COVID-19 pandemic. In this section we present this data set, explain how we codify the occupation variables, present basic descriptive statistics of job losses, and discuss an index of the individual’s potential for working from home.

3.1 The High-Frequency Phone Surveys

One of the reactions to understanding the impact of the pandemic was to collect new data, given that regular national household surveys were not well-suited to deal with this novel situation. One of the more ambitious initiatives was led by the World Bank, which implemented or supported several waves of High-Frequency Phone Surveys in over 100 countries, and harmonized the results. Thirteen Latin American countries participated in the survey: Argentina, Bolivia, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala, Honduras, Mexico, Paraguay, and Peru. These countries represent around 60 percent of the region’s population.

The HFPS have a panel structure over three rounds conducted between May and August 2020.Footnote 1 The surveys collected information on multiple dimensions, such as changes in employment, access to health and education services, and coping mechanisms to deal with job loss or other shocks related to the pandemic or the measures implemented by governments to mitigate the spread of the disease. The questionnaires also inquired about households’ knowledge of the disease and the degree of compliance with preventive measures.Footnote 2

In each selected household, only one adult aged 18 and above was interviewed, and he/she answered both individual and household-level questions. The same respondent was contacted and interviewed in all rounds. An average of 1011 interviews per country were completed in the first round, but average response rates declined to 72 percent in the second round and 69 percent in the third round. Attrition was addressed in the estimation of the survey sample weights for the second and third rounds. The weights were calibrated to incorporate population projections of the United Nations Economic Commission for Latin America and the Caribbean (Mejia-Mantilla et al. 2021).Footnote 3 Survey estimates for each country are representative of individuals aged 18 and above who have an active cellphone number or a landline at home.Footnote 4 Also, the survey is representative of households with a landline and for which at least one member has a cellphone.Footnote 5 Tables 1 and 2 present summary statistics of the main variables in the HFPS for the 13 Latin American countries taken together. Tables A.1 and A.2 in the Online Appendix report descriptive statistics by country. According to Mejia-Mantilla et al. (2021), compared to national household surveys, individuals in the HFPS sample are somewhat more educated, younger and more likely to reside in urban areas.

Table 1 Descriptive statistics of demographic variables
Full size table
Table 2 Descriptive statistics of labor market variables
Full size table

The HFPS were extremely useful in the midst of the crisis, when data from national household surveys (NHS) was not available. But, even when these larger surveys became available, the relevance of the HFPS remained high for various reasons. First, the questionnaire of the World Bank surveys was especially tailored to the crisis and includes a large number of specific questions. Second, unlike most NHS, the HFPS are multi-wave panel surveys that allow tracking short-run changes during the pandemic. Finally, the HFPS were uniform across countries (same questionnaire, same time window, etc.), so the comparability of the results and the possibility to construct aggregate regional measures are enhanced compared to the much more heterogeneous national household surveys.

3.2 Occupations

The HFPS dataset includes seven questions on the occupation of a worker: one for the pre-pandemic situation and two in each of its three rounds, one for people who worked the previous week and another for those who did not work but who still have a job. These questions on occupations capture important information on the activities performed by workers, and hence are very useful for the analysis of the labor market, as stressed by the task-based approach (Acemoglu and Autor 2011). Although the HFPS has been used as the main input by several recent papers on the impact of the COVID-19 crisis, to the best of our knowledge, the occupation variables were not considered by those studies. The most likely reason for this neglect is that, in contrast to the rest of the variables in the HFPS, the occupation variables had not been codified before. The responses to the occupation questions are open and hence difficult to assign to a few categories.

We find 9223 different answers in all the occupation variables included in the Latin American HFPS. In this study, we make an effort to codify these answers in a few categories. To that aim, we map the open individual responses to the occupation questions in the HFPS to the 41 groups in the 2-digit International Standard Classification of Occupations version 08 (ISCO-08). The ISCO categorization, proposed by the International Labour Organization (ILO), is currently adopted by most countries in the world, including many in Latin America, to classify the data on occupations collected in surveys, census and administrative records. We select the 2-digit grouping since it provides enough heterogeneity for the analysis, and at the same time avoids problems of lack of observations (Vosters 2018).

The codifying of the occupation variables in the HFPS implies a painstaking process of assigning each answer to an ISCO group, following the guidelines of the ISCO codebook, which includes specific information on the names, characteristics and activities of each occupation. In particular, we use the Spanish version of the codebook, consistent with the official language of the 13 Latin American countries included in the analysis. We were able to match 98 percent of the answers to the HFPS surveys to an ISCO group.Footnote 6

Figure A.1 in the Online Appendix shows the share of respondents in each of the ten occupation groups in the 1-digit ISCO classification before the start of the pandemic. On average, for all countries, the group with the largest share is services and sales workers (21 percent) followed by professionals (19 percent), elementary workers (14 percent), and craft and related trades workers (12 percent). There are some groups with a share between 6 and 9 percent (technicians and associate professionals; clerical support workers; skilled agricultural, forestry and fishery workers; and plant and machine operators and assemblers). Finally, the number of observations is small for managers (3 percent) and negligible for armed forces occupations (0.27 percent).

3.3 Comparison with national household surveys

In this section, we compare the structure of occupations in the HFPS before the pandemic with the one constructed with microdata drawn from national household surveys (NHS). To that aim we take advantage of the Socioeconomic Database for Latin America and the Caribbean (SEDLAC), a large project of harmonization of national household surveys that has been carried out by CEDLAS-UNLP and The World Bank since 2002. In particular, we use the circa 2019 NHS of the 13 Latin American countries included in the HFPS project.

The occupation variables are already codified in the household surveys by the national statistical offices although, unfortunately, the codification is not uniform across countries.Footnote 7 Latin American countries use different systems of occupation codes: they use different versions of the ISCO classification or even their own codes (e.g., Argentina). In order to have a unique classification we converted the occupation codes of each country to the 2-digit ISCO 08 using official crosswalks.

Once we have the occupations in the two data sets (HFPS and NHS) grouped following the same classification (2-digit ISCO 08), we proceed to compare the resulting structure of occupations. In particular, we compare the structure according to the pre-pandemic occupation question in the HFPS with the structure in the 2019 NHS. Figure A.1 shows the results for the 1-digit grouping. The shares of workers in each group are similar across data sets, with two noticeable exceptions. Compared to the NHS, there is an over-representation of professionals and an under-representation of workers in skilled agricultural and elementary occupations in the HFPS. Figure A.2 shows that the under-representation in the latter group occurs mainly among the agricultural, forestry and fishery laborers.Footnote 8 These differences are consistent with the findings of Mejia-Mantilla et al. (2021) regarding the bias of the HFPS sample towards more educated and urban households: a typical limitation in phone surveys given the difficulties in reaching rural workers. Naturally, these biases should be kept in mind when looking at aggregate statistics, but they are not necessarily a significant problem when we carry out the analysis at the individual level, as in our regression analysis.

3.4 Job losses

Our main variable of interest is job losses during the COVID-19 pandemic. We focus on permanent job losses, defined as situations where the respondent was working before the pandemic but is not working at the time of the interview and does not have a job to return to.

For simplicity, in this section, we construct a binary indicator for “any job loss” that equals one if the worker suffered a job loss in any of the 3 waves compared to the pre-pandemic situation. For those who had a job prior to the pandemic, column (1) in Table 3 shows the percentage who lost their job in waves 1, 2, or 3 of the HFPS. Job losses were very large; the share of workers who lost their job in this very short time window was around 30 percent on average in the region. There are heterogeneities across countries, from 18 percent in Argentina to 47 percent in Colombia.

Table 3 Any job loss by country and characteristics of workers and their pre-pandemic jobs
Full size table

We are particularly interested in the heterogeneities of the impact of the crisis across socioeconomic characteristics, such as gender. Columns 2 and 3 of Table 3 show that job losses were larger for women than for men in all countries, and the gender gaps were as high as 22 percentage points in Paraguay. Also, job losses were larger for the youth (columns 4 and 5 of Table 3) and the unskilled (columns 6 to 8 of Table 3) in all the countries. Taking into account the pre-pandemic occupation, non-salaried workers experienced larger job losses in most countries (columns 9 and 10 of Table 3).

Figure A.3 shows job losses by occupations defined at the 1-digit ISCO. Job losses varied from 15 and 16 percent for managers and professionals, respectively, to 37 percent for machine operators and assemblers, and up to more than 50 percent for workers in elementary jobs. Since the ISCO classification sorts occupations by job complexity, from managers to elementary workers (Vosters 2018), Figure A.3 suggests that job losses were decreasing in the complexity of the job. In the following section, we work with the 41 groups of the 2-digit ISCO classification, which implies a richer analysis.

3.5 Work-from-home index

To analyze whether the differential impact of the pandemic between women and men was related to the potential for remote work in the pre-pandemic situation, we construct a variable for the potential for working from home. We proceed in two steps; first, we construct a work-from-home measure (WFH measure thereafter) at the occupation level (Dingel and Neiman 2020), and then adjust this measure for home internet access using the answers to the HFPS (Garrote-Sanchez et al. 2021).

The first step follows the widely used methodology of Dingel and Neiman (2020) (DN thereafter). They construct a WFH variable at the occupation level using data on 17 characteristics of more than 900 occupations drawn from the US Department of Labor, Employment and Training Administration’s Occupational Information Network (O*NET). The selected characteristics are suggestive of the difficulties of working from home (e.g., working directly with the public, outdoor activity, exposed to the weather). When at least one of these characteristics is assessed as “important” or “very important”, the occupation is classified as not compatible with working from home.

By following DN to construct the WFH variable, we implicitly assume that the US task content of occupations is representative of the Latin American context. This is potentially problematic since the task content of occupations varies across countries with different levels of development (Lo Bello et al. 2019). Motivated by that concern, researchers have used skills surveys with data from developing countries (Saltiel 2020, Delaporte and Peña 2020, Gottlieb et al. 2020, Gottlieb et al. 2021, Hatayama et al. 2023). These papers typically find differences in the task content of occupations across countries, mostly due to differences in technology adoption and organization of production. In particular, papers report that the ability to telework is lower in developing countries compared to the US results obtained from O*NET. Still, researchers find a very strong positive correlation between WFH measures computed with DN and other methodologies (Hatayama et al. 2023). In our paper we use DN estimates of WFH due to the combination of three reasons: (i) DN estimates are the most precise, given the richness of the information in O*NET; (ii) alternative indicators are unavailable for most Latin American countries,Footnote 9 and (iii) the correlation across measures is very high. Although we are aware that using DN surely implies an overestimation of the level of teleworking, our paper is more concerned about differences across occupations, which are probably better captured in the richer O*NET data.

Occupations in O*NET are classified according to the Standard Occupational Classification (SOC) System. The next step in the methodology is to map the results in the SOC-8 digits classification to the 2-digit ISCO classification, the one used in this paper. We follow official crosswalks and Bonavida-Foschiatti and Gasparini (2020) for this step. The number of groups in the 2-digit ISCO classification is smaller, so for each ISCO category we typically have several SOC occupations. As a result, our WFH variable at the 2-digit occupation level is not a binary variable but a share: the share of SOC occupations included in an ISCO category that are compatible with working from home.Footnote 10

This WFH measure is defined at the occupation level, which implies that all workers in a given 2-digit ISCO occupation have the same capacity of working from home. However, some of the occupations that can be carried out at home require access to the internet, a service with no universal access in Latin America. Given this concern, we adjust the measure of WFH to account for home internet access. We proceed in two steps. First, we follow Garrote-Sanchez et al. (2021) and define the share of SOC occupations within each 2-digit ISCO occupation that require internet access to be executed (ShrY) and the share of occupations that do not (ShrN). This step uses information from O*NET on the importance and frequency of computer and email use. In the second step we adjust the share that requires internet by home internet access at the individual level, taking advantage of a relevant question in wave 1 of the HFPS. Formally,

$$WF{H}_{ij}=WF{H}_{j}\times Shr{N}_{j}+WF{H}_{j}\times Shr{Y}_{j}\times Interne{t}_{i},$$
(1)

where i labels the individual and j her occupation, and Interneti is a binary variable that equals one if individual i has access to internet and zero otherwise. Suppose for example that all SOC categories included in occupation j require internet to be performed (ShrNj = 0; ShrYj = 1). If worker i has access to the internet (Interneti = 1) the individual WFH variable will be equivalent to the unadjusted indicator for her occupation (WFHij = WFHj). Instead, if the worker has no access to the internet (Interneti = 0), the individual WFH variable will be zero (WFHij = 0). In this case, the worker cannot perform the tasks associated with her occupation at home because she lacks internet access. Notice that by adjusting for home internet access the WFH index at the occupation level becomes an index defined at the individual level (WFHij instead of WFHj). This adjustment introduces more heterogeneity across workers and countries. Figure A.4 in the Online Appendix shows the distribution of the WFH index at the individual level for each country before the pandemic and Table A.3 provides descriptive statistics of the WFH measure by country and occupation. The distributions differ across countries because of differences in their occupational structures or home internet access. For instance, the occupational structure in Ecuador or Guatemala is much less compatible with remote work than that of Argentina or Chile.

4 Job losses, work from home, and childcare

Our main interest is to assess the relationship between job losses and the potential for working from home of the pre-pandemic occupation and whether this relationship differs by gender. Figure 1 shows the correlation across countries between the share of workers who lost their jobs and the WFH index. In the first panel–wave 1–a clear negative association between the WFH measure and job losses is observed for both women and men.Footnote 11 In other words, the initial impact of the crisis implied more job losses for countries where the occupational structure was less compatible with remote work, both for male and female workers. While this pattern persists for women in the following waves of the survey, for men the relationship between job losses and the WFH measure loses strength and virtually disappears in the third wave. Figure 2 shows that a similar pattern arises from an individual-level analysis, i.e., the correlation between the probability of job loss and the individual WFH index is stronger for women, suggesting that the possibility of working from home was more decisive in preventing job losses for women than for men.

Fig. 1
figure 1

Job loss and the work-from-home index (WFH) by wave and gender. Share reporting a job loss in wave 1, wave 2, and wave 3. Job loss defined with respect to the pre-pandemic situation

Full size image
Fig. 2
figure 2

Job loss and the work-from-home index (WFH) by wave

Full size image

At least two mechanisms can account for the differential importance of working from home between men and women. One is related to gender differences in childcare responsibilities, which fall mostly on women as we mentioned above. Given the allocation of these responsibilities and the school closures during the pandemic,Footnote 12 the only way for women with young children to continue working was to do it from home. Therefore the possibility of working from home would be more relevant to avoid job losses for women with young children than for men or women without children.

Another mechanism arises from the occupational segregation by gender that characterizes labor markets. Then, if the possibility of doing remote work was key to keep a job during the lockdown, the fact that men and women work in occupations with different potential for working from home could also explain gender differences in job losses.

We explore these hypotheses using the following regression model:

$$\begin{array}{ll}JobLos{t}_{icw}=&\alpha +\gamma Wome{n}_{icw}+\psi WF{H}_{ic0}\\ &+\,\phi WF{H}_{ic0}\times Wome{n}_{icw}+\delta {X}_{icw}^{{\prime} }+{\lambda }_{c}+{\theta }_{w}+{\epsilon }_{icw}.\end{array}$$
(2)

JobLosticw is an indicator that takes the value 1 if individual i from country c lost her job in wave w relative to the pre-pandemic period; Womenicw indicates whether individual i is a woman and WFHic0 is the WFH index adjusted by home internet access corresponding to the job that individual i had before the pandemic;Footnote 13 the interaction between Womenicw and WFHic0 allows for gender differences in the effect of the WFH measure on job losses; vector \({X}_{icw}^{{\prime} }\) includes education attainment, age and squared age, and whether the individual was a salaried worker or self-employed before the pandemic, which allows controlling for job stability and other characteristics not related with the content of tasks, such as time flexibility;Footnote 14 the model also controls for household size and includes wave dummies (θw) as well as country-fixed effects (λc).

We estimate model (2) using the sample of individuals who were employed before the pandemic. Column 1 in Table 4 reports the results.Footnote 15 The estimated coefficient for the WFH index is negative and significant, indicating that, even after controlling for all other variables, there is a negative association between the possibilities of doing remote work and the probability of losing a job during the pandemic. The coefficient of the dummy Women is positive and statistically significant, while the interaction of this variable and variable WFH is negative and also highly significant. This implies that while women were more likely to lose their jobs during the pandemic than men, having a job that can be done from home largely offsets that disadvantage. For instance, women with jobs incompatible with remote work (WFH = 0) were almost 11 percentage points more likely to lose their jobs during the pandemic than men with similar characteristics including similar jobs in terms of work-from-home possibilities. However, if we compare similar women and men in jobs fully compatible with remote work, women were only 2.8 percentage points more likely than men to lose their jobs during the pandemic. This result is in line with our previous discussion: the possibility of working from home was more important for women than for men in helping avoid job losses.

Table 4 Probability of job loss, potential for working from home, and gender
Full size table

Next, we explore the role of the presence of children at home. We, therefore, add to the model an indicator of the presence of children aged between 5 and 18 in the household, its interaction with the dummy Women, and the number of children in the household.Footnote 16 Column 2 in Table 4 shows the results that indicate that women with children at home were more likely to lose their jobs than men with similar characteristics who also have children at home.

In order to explore whether the differential effect of work-from-home possibilities by gender is associated with the presence of children at home, we run model (2) separately for households with and without children and report the results in columns 3 and 4 in Table 4. Notice that the coefficient of the interaction between the WFH measure and gender is negative, large in absolute value and very significant only in households with children, while in households without children the coefficient is still negative but smaller in absolute value and not statistically significant. This implies that the role of working from home in avoiding female job losses occurs mainly in households with children, possibly due to childcare needs. For instance, in households with children, a change from a job that is not compatible with remote work (WFH = 0) to a job fully compatible with remote work (WFH = 1) is associated with a reduction in the probability of job loss 11 percentage points larger for women than for men. In contrast, in households without children the differential effect of the WFH measure between genders is less than 3 percentage points.

To check the robustness of the results, Table A.6 in the Online Appendix reports the results for a model that includes fixed effects by region. The estimates confirm our previous findings.Footnote 17

Finally, we explore whether our results are consistent with another potential mechanism, i.e., that the gender differences in the impact of work-from-home possibilities on job losses are related to occupational segregation by gender. For this, we estimate a variant of model (2) that includes fixed effects by pre-pandemic occupation using the 2-digit ISCO classification. Table 5 reports these results. The coefficient associated with the interaction between gender and the WFH measure is still negative and significant in the sample of households with children, which means that even among women in the same occupation, those with children were the ones for whom the ability to work from home was pivotal. This suggests that our results are probably not driven by gender differences in occupational structure but more likely by childcare needs.

Table 5 Probability of job loss, potential for working from home, and gender. Including 2-digit ISCO fixed effects
Full size table

In a companion paper (Berniell et al. 2021c) we explore the role of work-from-home possibilities in other outcomes beyond job losses. We find that the probability of suffering an income fall was lower among workers whose pre-pandemic jobs were more compatible with working from home. The estimated coefficient of the interaction between gender and the WFH measure is negative but small and not statistically significant, implying that the possibility of working from home did not have a differential effect on incomes of men and women who continued to work during the pandemic. We also explore job entries during the pandemic and find that women who entered employment did so especially in jobs compatible with remote work. Lastly, we analyze changes across occupations for those who were employed before the pandemic and remained employed during the pandemic. We find that women were more prone to move into occupations more compatible with remote work. An increase in the WFH index (from the old to the new occupation) is associated with the presence of children at home.

5 Heterogeneities

This section explores potential heterogeneities in the previous results depending on the countries’ pre-pandemic gender inequality level and the stringency of social-distancing measures adopted during the pandemic.

5.1 Heterogeneities by country’s gender inequality level

We explore whether results differ across countries according to their pre-pandemic status regarding gender inequality using the Gender Inequality Index (GII) from the United Nations Development Programme (UNDP). The GII reflects gender-based disadvantage in three dimensions: reproductive health, empowerment and the labor market.Footnote 18 The index ranges from 0, where women and men fare equally, to 1, where one gender fares as poorly as possible in all dimensions.

We classify all the countries in our analysis based on the 2019 GII and separate the countries into two groups depending on whether the value of the GII is below or above the median for Latin America. The low gender inequality group includes Argentina, Chile, Peru, El Salvador, Costa Rica, Mexico and Ecuador. The average GII for this group is 0.33; the country with the lowest value (lower level of gender inequality compared to the other countries) is Chile (0.25) and the country with the highest value is Peru (0.40). The high gender inequality group includes Colombia, Guatemala, Honduras, Bolivia, Paraguay and Dominican Republic. The average GII for this group is 0.44; the country with the lowest value within this group is Bolivia (0.42) and the country with the highest value is Guatemala (0.48).

Table 6 shows the results of estimating the job-loss probability model for the two groups of countries separately. Interestingly, the results suggest that the protection from job loss for women with children arising from the ability to work from home was only relevant in low gender-inequality societies. A possible conjecture behind this result is that in more conservative societies, social norms are so rigid that the possibility of doing remote work is not enough to prevent the loss of employment of women who had to stay at home to care for their children. Of course, this is just a conjecture and more research is needed to understand what is behind these results.

Table 6 Probability of job loss by pre-pandemic Gender Inequality Index
Full size table

5.2 Heterogeneities according to social-distancing measures

Immediately after the onset of the pandemic governments around the world imposed strong measures aimed at containing the spread of the disease, including lockdowns, travel restrictions, school closures, and various social-distancing measures. In this subsection we explore potential heterogeneous effects according to the stringency of these measures, based on the Stringency Index of the Oxford COVID-19 Government Response Tracker (OxCGRT), which records the strictness of lockdown-style policies that restricted people’s behavior during the pandemic.Footnote 19

The Stringency Index of the OxCGRT is computed as a simple average of the nine following indicators rescaled to a maximum of 100: school closing, workplace closing, cancel of public events, restrictions on gatherings, close of public transport, stay-at-home requirements, restrictions on internal movement, international travel controls, and public information campaigns. The index and each component reflect the presence and intensity of the measures in each country over time. However, since our data source is restricted to the peak of the pandemic (May to August 2020), the stringency index has very little variability across countries and over time, a fact that imposes a limitation on this heterogeneity analysis.

To assess the role of the policy reactions to the pandemic we split our sample into two groups depending on whether the stringency index is below or above the median across Latin American countries over the period of time of our data. The high stringency group includes Argentina, Chile, Colombia, Guatemala, El Salvador, Honduras, and Bolivia, the first two waves of data for Peru, and the first wave for the Dominican Republic and Ecuador. The average of the stringency index for this group is 89.4. The low stringency group includes Costa Rica, Mexico, and Paraguay, the second and third waves of data for the Dominican Republic and Ecuador, and the third wave for Peru. The average of the stringency index is 75.0 in this case. Table 7 reports the results of estimating the job-loss probability model for the two groups separately. The results confirm that the role of work-from-home possibilities in reducing the threat of job losses was more important for women than for men, particularly in countries where high stringency measures were in place (column 1) and in households with children (column 3). We do not find differential effects between the high and low stringency groups in the sample of households with children, most likely because schools were closed in all countries during the time period analyzed, so the variability of the stringency index would not arise from differences in school closure policies.Footnote 20

Table 7 Probability of job loss by intensity of social-distancing measures. Stringency Index
Full size table

6 Concluding remarks

The COVID-19 pandemic was a huge unexpected shock on the lives of people around the world. Many lost their jobs, suffered income reductions, or had to change occupations to cope with the new situation. In this paper, we explore these issues in Latin America by exploiting an unusual rich survey carried out immediately after the onset of the pandemic: the World Bank’s High-Frequency Phone Surveys. In particular, by codifying the occupation variables in these surveys we are able to construct a variable for the potential for working from home and analyze the role of this factor in explaining the heterogeneous impacts of the shock, in particular in the asymmetries between women and men.

Our analysis leads to some interesting results. First, we confirm that the impact of the COVID-19 shock was (i) harder for women and (ii) strongly decreasing in the ability to work from home. More important, we find that the mitigating effect of work-from-home possibilities on the severity of the impact was especially relevant for women with children. Our results are consistent with a plausible mechanism: due to the traditional distribution of childcare responsibilities within the household, women with children were more likely to stay home during school closures, and therefore the ability to work from home was crucial for them to keep their jobs.

In most Latin American families, caregiving and home-production responsibilities still rely disproportionately on women. Our results suggest that traditional gender roles may have played a key role on the impact of the pandemic on women’s labor market outcomes. Given the long school closures during the pandemic and the traditional distribution of responsibilities across genders within households, women carried most of the extra burden imposed by the pandemic, and then suffered the consequences in the labor market. By drawing attention to the asymmetric impacts of the pandemic, this paper contributes to making visible the consequences of these deeply rooted behaviors in our societies, and highlights the need of policies that help modify traditional gender roles and ease the burden of care responsibilities on families, including, for instance, public childcare centers and longer and gender-balanced parental leaves.

Finally, our results are also important in highlighting the critical role of the ability to work from home. The pandemic showed the relevance of expanding connectivity and promoting digital skills. In fact, the demand for workers with technological skills has likely undergone a permanent shift, increasing the need for policies that facilitate or promote the acquisition of digital skills that help improve the employability of workers and their chances of advancing to better jobs.