## Digitalization and Employment Gender Gaps during the COVID-19 Pandemic: Evidence from Latin America and the Caribbean

## Source details

**Canonical URL:** [Digitalization and Employment Gender Gaps during the COVID-19 Pandemic: Evidence from Latin America and the Caribbean](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024012-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024012-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024012-print-pdf.pdf.json)

---

### Overview and context
- LAC historically faces gaps in connectivity infrastructure, high informality, and a largely unbanked population.
- Pandemic-driven digitalization:
  - E-commerce increased by 36.7 percent relative to pre-pandemic levels.
  - Over 13 million people across LAC completed an online transaction for the first time in 2020.
- Internet use and inequality:
  - In 2018, 68 percent of the population used the internet regularly (almost twice the share in 2010; below OECD average of 84 percent).
  - Internet access: 75 percent of the richest population uses the internet versus 37 percent of the poorest population.
  - Rich-poor internet access gap in LAC is almost 40 percentage points (compared with below 25 percentage points in OECD countries).
- Structural constraints and labor market risks:
  - Over 20 percent of jobs in some countries are likely to undergo some kind of automation.
  - On average, only 20 percent of jobs can be done remotely in LAC, compared to 41 percent in the United States.
  - LAC is aging faster than any other region.
- Policy implication: substantial investments in education and training, including for women, are needed to equip workers with digital skills.

### Digitalization across sectors
- Rapid adoption during the pandemic extended beyond commerce to education, healthcare, and labor markets:
  - Education delivery shifted to digital technologies, enabling continuity of classes though with quality costs.
  - Telemedicine expanded, increasing access where health systems were saturated.
  - Behavioral changes (e.g., online purchasing, telemedicine) may be persistent.

### Key microeconomic findings (Introduction)
- Using high frequency World Bank and UNDP household surveys for 22 LAC countries:
  - Higher levels of digitalization are associated with increased female employment.
  - Digitalization is associated with reduced job losses for both men and women.
  - Digitalization mitigates the negative impact of household chores on job losses, especially for women.
  - Digitalization alleviates the burden of childcare responsibilities for women and promotes their labor market participation.
  - Digitalization mitigates the negative relationship between pandemic-type shocks and female employment.

### Where LAC stands in digital infrastructure and finance (selected statistics)
- Fixed broadband access: LAC 46 percent; Eastern Europe 57 percent; Western Europe 87 percent; Asia Pacific 59 percent.
- Digital public platforms: Argentina, Brazil, Chile, Uruguay among top 50 performers in 2018 (slightly below the OECD average); Belize, Cuba, Haiti, Nicaragua among the worst LAC performers.
- Digital financial transactions and mobile money (2019):
  - Registered mobile money accounts growth: LAC +2.5 percent (Western Africa 14.5 percent).
  - Transaction value growth: LAC +1.4 (Western Africa +34.9 percent).
- Venture capital: Brazil and Mexico led the region in the number of deals and volume in 2019; region trails Asia Pacific in attracting investment.
- Skills gap: IDC (2017) estimates more than 450 thousand unfilled jobs in the technology area due to lack of trained professionals.

### Evidence from the literature on digitalization and gender
- Mechanisms:
  - Digitalization improves job information, reduces recruiting costs, and improves job matching.
  - Digitalization increases flexibility, reorganizes tasks, and liberalizes employment forms.
- Empirical findings cited:
  - Klonner & Nolen (2010): network expansion raises employment by 15 percentage points in covered communities (majority driven by increased female employment).
  - Dettling (2016): residential broadband availability leads to a 4.1 percentage point increase in labor force participation for married women (linked to telework and time savings).
  - Bahia et al. (2020): mobile broadband rollout in Nigeria increased household consumption and decreased poverty, largely via higher labor force participation and employment, especially among women.
  - Kumar and others (2023): digital infrastructure coverage, internet price, and usability predict internet use; demand most price responsive in low-income developing countries.
- Macro estimates:
  - GSMA: a 10 percent increase in mobile internet penetration can increase GDP by 1.2 percent.
  - GSMA: a 10 percent increase in a country's digitalization can generate 1.9 percent in GDP growth.

### COVID-19, gender, and digital resilience
- Pandemic impacts and gender:
  - Digitally-enabled firms in Middle East and Central Asia faced a lower decline in sales by about 4 percentage points relative to digitally-constrained firms (Abidi and others, 2022).
  - COVID-19 disproportionately affected women's employment because women are overrepresented in hardest-hit sectors, more likely in informal jobs, and faced increased childcare needs from school/daycare closures.
- Disproportionate burden in LAC:
  - Women, young workers, low-education workers, and those with limited internet connectivity were disproportionately affected.
  - Almost one in five absent workers lost employment after two months (World Bank, 2021).
  - ATUS (2021): among married couples with children, husbands provide an average of 7.4 hours of childcare per week; wives provide an average of 13.3 hours.
- Duration: gender disparities in employment declines are usually temporary, lasting on average one or two quarters (Bluedorn et al., 2021).

### Empirical strategy and data (summary)
- Two complementary approaches:
  - Micro (household-level): World Bank and UNDP high frequency household surveys across 22 countries.
  - Macro (country-level): cross-country panel data to estimate effects of internet expansion on employment rates by gender in LAC.

### Household-level analysis — Data
- Data source: second phase of the High Frequency Phone Survey, wave 2 in LAC; collected between October and December of 2021.
- Coverage: 22 countries in Latin America and the Caribbean.
- Survey topics: individual and household demographics, income, employment, health, education.
- Limitations and adjustments:
  - Survey is not a panel; latest information from wave 2.
  - Phone surveys subject to coverage and non-response bias; sample reweighting applied.
- Outcome variables:
  - "job loss": binary = 1 if reported not currently working and having lost pre-pandemic job.
  - "employed": binary = 1 if reported having worked in the last week.
- Main independent variable:
  - "Women with internet": Female dummy * household internet connection dummy (value 1 only for women with an internet connection in their household).
- Controls and fixed effects: country, state, education level, marital status.
- Complementary binary sample splits:
  1. "income reduced" = 1 if household income decreased compared to previous wave (around June 2021).
  2. "High Covid-19 level" = 1 if country had COVID-19 cases rate higher than world median.
  3. "Increased household chores" = 1 if individual reported increase in household chores since start of pandemic.
  4. "Household with children" = 1 if at least one child in household.
- Additional covariates: smartphone ownership; presence of children.
- Income proxy: household self-report question about income change.

### Household-level descriptive statistics (Table 1 highlights)
- Sample totals:
  - No. Obs: 26.662
  - Job loss (%): 17,4
  - Employed (%): 64,4
  - Women (%): 51,7
  - Internet access (%): 62,4
  - Women with internet access (%): 31,0
  - Increased household chores (%): 28,7
  - Household with children (%): 60,5
- Country-level variation examples (No. Obs; Job loss (%); Employed (%); Women (%); Internet access (%); Women with internet access (%); Increased household chores (%); Household with children (%)):
  - Haiti: 2.361; 33,3; 46,2; 51,6; 9,0; 3,9; 35,4; 74,8
  - Dominica: 879; 15,1; 66,0; 49,6; 83,2; 43,8; 37,7; 51,6
  - Mexico: 2.511; 14,7; 66,0; 52,0; 68,6; 34,0; 30,4; 58,1
- Descriptive highlights:
  - Average job loss: around 17 percent.
  - Average employment: 64 percent.
  - Average percentage women in sample: 52 percent.
  - Average households with children: 60 percent.
  - Internet access average rate 61 percent driven by high rates in Dominica (83 percent), St. Lucia (81 prcent), Jamaica (71 percent), and extremely low rate in Haiti (9 percent).

### Household-level empirical strategy
- Baseline specification (textual):
  - Y_i = α + β1 (Female_i * Internet_i) + β2 Female_i + β3 Internet_i + δ_i + ε_i
  - Y_i is either "employment" or "job loss".
  - δ_i: fixed effects for country, state, marital status, education level.
  - Coefficient of interest: β1 (effect of being female and having internet access on labor market outcomes relative to women without internet access).
- Authors note: specification likely suffers bias from unobserved variables correlated with both outcome and internet access; coefficients interpreted as correlations, not causal effects.
- Sample-split analyses apply same specification to subsamples defined by complementary binary variables.
- Estimation details: clustered standard errors at the individual level; survey weights for country population.

### Household-level results — Main findings (Table 3 summary)
- General patterns:
  - Internet access increases probability of being employed and decreases probability of job loss (simple correlations significant).
  - Main specification finds positive correlation for "women with internet" on employment and negative correlation on job loss; holds controlling for children in household, increased chores, or smartphone ownership.
  - Digitalization appears more effective at mitigating job loss related to household chores than at promoting employment in some splits.
- Key coefficients (Panel A. Job loss; Panel B. Employment; Panel C. Hours worked — values preserved as reported):
  - Panel A. Job loss
    - Women with internet: -0.0515*; -0.0523**; -0.0510*; -0.0524** (standard errors: (0.0263), (0.0262), (0.0264), (0.0264))
    - Women: 0.185***; 0.187***; 0.178***; 0.186*** (standard errors: (0.0228), (0.0225), (0.0227), (0.0228))
    - Internet: -0.0798***; -0.0585***; -0.0290; -0.0281; -0.0288; -0.0250 (standard errors: (0.0153), (0.0152), (0.0186), (0.0185), (0.0188), (0.0189))
    - Increased household chores: 0.0589*** (0.0137)
  - Panel B. Employment
    - Women with internet: 0.0403*; 0.0425*; 0.0397; 0.0427* (standard errors: (0.0245), (0.0244), (0.0245), (0.0245))
    - Women: -0.241***; -0.248***; -0.236***; -0.244*** (standard errors: (0.0189), (0.0187), (0.0189), (0.0189))
    - Internet: 0.0612***; 0.0315**; 0.00223; -0.000671; 0.00331; -0.00793 (standard errors: (0.0144), (0.0141), (0.0190), (0.0189), (0.0190), (0.0190))
    - Increased household chores: -0.0533*** (0.0140)
  - Panel C. Hours worked
    - Values reported identical to Panel B in the table (preserved above).
- Sample and fit (Table 3):
  - Observations: 26,662; 26,507; 26,507; 26,507; 26,507; 14,217
  - R-squared: 0.012; 0.087; 0.135; 0.137; 0.137; 0.143
  - Country FE: YES for all columns; State FE, Education FE, Marital status FE included in many specifications.
  - Notes: Clustered standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

### Heterogeneity — Household chores, children, income, COVID splits (selected coefficients)
- By household chores and children (Table 4):
  - Job loss (Women with internet)
    - Household chores increased: -0.0904* (0.0481)
    - Household chores NOT increased: -0.0527* (0.0298)
    - Household with children: -0.0631** (0.0292)
    - Household without children: -0.0440 (0.0424)
  - Employment (Women with internet)
    - Household chores increased: 0.0555 (0.0441)
    - Household chores NOT increased: 0.0352 (0.0277)
    - Household with children: -0.000933 (0.0281)
    - Household without children: 0.0974** (0.0389)
  - Note: difference between groups only statistically significant for employment when comparing by children in the household.
- By income reduction and COVID-19 severity (Table 5):
  - Job loss (Women with internet)
    - Income reduced: -0.0902** (0.0428)
    - Income NOT reduced: -0.0525* (0.0295)
    - High Covid-19: -0.0527* (0.0275)
    - Low Covid-19: -0.0472 (0.0362)
  - Employment (Women with internet)
    - Income reduced: 0.0401 (0.0395)
    - Income NOT reduced: 0.0534* (0.0284)
    - High Covid-19: 0.0441* (0.0260)
    - Low Covid-19: 0.0135 (0.0337)
  - Notes: differences between groups are not statistically significant in most comparisons.

### Robustness and alternative specifications
- Logistic regressions (Table 12) confirm main household-level correlations:
  - Job loss (selected): Internet coefficient -0.543*** (0.0993) with odds ratio 0.581*** (0.0577) in one specification.
  - Employment (selected): Internet coefficient 0.268*** (0.0630) with odds ratio 1.307*** (0.0824) in one specification.
  - Women coefficients show large, significant magnitudes in both job loss and employment logistic models in specified cells.
- Country-level alternative measures (Table 13) remain robust:
  - Percent of Internet Users: 0.080*** (0.013) for Female Employment Ratio; 0.032*** (0.011) for Male Employment Ratio.
  - Number of Internet Users: 0.027*** (0.007) for Number of Female Employees; 0.037*** (0.005) for Number of Male Employees.
  - Observations: 4,036 for employment ratios; 4,029 for number of employees.
- GMM endogeneity checks (Table 14):
  - Percent of Internet Users: 0.002* (0.001) and 0.004* (0.002) for Female LFPR specifications.
  - Results remain statistically significant and consistent with a positive relation between digitalization and female labor force participation.
  - Observations across GMM specifications: 3,987; 3,816; 3,904; 3,785.

### Interpretive conclusions and policy-relevant implications
- Core conclusion: Higher levels of digitalization are associated with higher employment and lower job loss in LAC, and digitalization helps mitigate the COVID-19 crisis’s negative effects on female employment.
- Additional findings:
  - Digitalization’s effects are stronger where childcare resources are scarce, suggesting digital tools can alleviate childcare burdens and promote women’s participation.
  - Positive effects can be limited by institutional weaknesses and aggravated structural vulnerabilities.
- Policy recommendations and implications:
  - Improve institutional capacity to fully realize digitalization benefits in weak-institution contexts.
  - Invest in education and infrastructure to enable digitalization to deliver its full positive impact on female employment.
- Research caveat and future direction:
  - Findings from the COVID-19 shock are informative but it is premature to generalize broadly.
  - Future research should estimate an all-encompassing expected vulnerabilities of women to digitalization to inform social and economic policy discussions.

*IMF Working Paper — "Digitalization and Employment Gender Gaps during the COVID-19 Pandemic: Evidence from Latin America and the Caribbean" (Introduction; Household-level analysis; Results; Robustness; Conclusions).*

### Introduction ...........................................................................................................

### Introduction

### Overview and context
- Latin America and the Caribbean (LAC) historically face gaps in connectivity infrastructure, high informality, and a largely unbanked population.
- The COVID-19 pandemic accelerated digitalization in the region:
  - E-commerce increased by 36.7 percent relative to pre-pandemic levels.
  - Over 13 million people across LAC completed an online transaction for the first time in 2020.
- Internet use expanded prior to the pandemic:
  - In 2018, 68 percent of the population used the internet regularly, almost twice the share in 2010, but below the OECD average of 84 percent.
  - Internet access is unequal: 75 percent of the richest population uses the internet versus 37 percent of the poorest population.
  - The rich-poor internet access gap in LAC is almost 40 percentage points, compared with below 25 percentage points in OECD countries.
- Structural constraints and risks:
  - Over 20 percent of jobs in some countries are likely to undergo some kind of automation.
  - High informality and labor composition: on average, only 20 percent of jobs can be done remotely in LAC, compared to 41 percent in the United States.
  - Aging population: LAC is aging faster than any other region.
- Implication: substantial investments in education and training, including for women, are needed to equip workers with digital skills.

### Digitalization across sectors
- Rapid adoption during the pandemic extended beyond commerce to education, healthcare, and labor markets:
  - Education delivery shifted to digital technologies, enabling continuity of classes though with quality costs.
  - Telemedicine boomed, increasing access where health systems were saturated.
  - Behavioral changes (e.g., online purchasing, telemedicine) may be persistent in a region that lagged in digital adoption.

### Key microeconomic findings reported in the Introduction
- Using high frequency World Bank and UNDP household surveys for 22 countries in LAC, the paper finds:
  - Higher levels of digitalization are associated with increased female employment.
  - Digitalization is associated with reduced job losses for both men and women.
  - Digitalization mitigates the negative impact of household chores on job losses, especially for women.
  - Digitalization alleviates the burden of childcare responsibilities for women and promotes their labor market participation.
  - Digitalization mitigates the negative relationship between pandemic-type shocks and female employment.

### Where LAC stands in digital infrastructure and finance (selected statistics)
- Fixed broadband access: LAC 46 percent; Eastern Europe 57 percent; Western Europe 87 percent; Asia Pacific 59 percent.
- Digital public platforms: Argentina, Brazil, Chile, Uruguay among top 50 performers in 2018, performing slightly below the OECD average; Belize, Cuba, Haiti, Nicaragua among the worst LAC performers.
- Digital financial transactions and mobile money:
  - In 2019 LAC had the lowest growth rate in registered mobile money accounts: +2.5 percent (Western Africa 14.5 percent).
  - In 2019 LAC had the lowest growth rate in transaction value: +1.4 (Western Africa +34.9 percent).
- Venture capital: Brazil and Mexico led the region in the number of deals and volume in 2019 but the region trails the Asia Pacific in attracting investment.
- Skills gap: IDC (2017) estimates more than 450 thousand unfilled jobs in the technology area due to lack of trained professionals.

### Evidence on digitalization and gender from the literature
- Mechanisms:
  - Digitalization can improve labor market efficiency through better job information, reduced recruiting costs, and improved job matching.
  - Digitalization can increase flexibility and deregulation, reorganize tasks, and liberalize employment forms.
- Empirical findings:
  - Klonner & Nolen (2010): network expansion raises employment by 15 percentage points in covered communities, with the majority of the impact driven by increased female employment.
  - Dettling (2016): residential broadband availability leads to a 4.1 percentage point increase in labor force participation for married women; effects are linked to telework and time savings in household activities.
  - Bahia et al. (2020): mobile broadband rollout in Nigeria increased household consumption and decreased poverty, largely via higher labor force participation and employment, especially among women.
  - Kumar and others (2023): digital infrastructure coverage, internet price, and usability are robust predictors of internet use; demand is most price responsive in low-income developing countries.
- Macro estimates of digitalization on GDP:
  - GSMA: a 10 percent increase in mobile internet penetration can increase GDP by 1.2 percent.
  - GSMA: a 10 percent increase in a country's digitalization can generate 1.9 percent in GDP growth.

### COVID-19, gender, and digital resilience
- Pandemic impacts:
  - Abidi and others (2022): digitally-enabled firms in Middle East and Central Asia faced a lower decline in sales by about 4 percentage points relative to digitally-constrained firms, suggesting digitalization acted as a hedge.
  - Unlike typical recessions where male employment is more affected, COVID-19 disproportionately affected women's employment due to:
    - Higher representation of women in sectors hit hard by the pandemic (hospitality, retail, healthcare).
    - Higher likelihood of women working in informal jobs with less flexibility and benefits.
    - Increased childcare needs from school and daycare closures, pushing many mothers to reduce labor market participation.
- Disproportionate burden:
  - Women, young workers, low-education workers, and those with limited internet connectivity were disproportionately affected in LAC.
  - Absences from work were historically high, especially for women and independent workers; almost one in five absent workers lost employment after two months (World Bank, 2021).
  - Time use (ATUS, 2021): among married couples with children, husbands provide an average of 7.4 hours of childcare per week, whereas wives provide an average of 13.3 hours.
- Duration: Bluedorn et al. (2021) find cross-country diversity, with greater decreases in women's employment rates in more than half to two-thirds of countries; these gender disparities are usually temporary, lasting on average one or two quarters.

### Empirical strategy and data (summary)
- The paper uses two complementary approaches:
  - Micro (household-level) approach: exploits World Bank and UNDP high frequency household surveys across households and countries to identify effects of digital availability on employment and job losses.
  - Macro (country-level) approach: employs cross-country panel data to estimate effects of internet expansion on employment rates by gender in LAC.

*Italic source: IMF Working Paper — "Digitalization and Employment Gender Gaps during the COVID-19 Pandemic: Evidence from Latin America and the Caribbean" (Introduction).*

### 1. Household-level analysis

### 1. Household-level analysis

### 1.1 Data
- Data source: second phase of the High Frequency Phone Survey, wave 2 in LAC; collected between October and December of 2021.
- Coverage: 22 countries in Latin America and the Caribbean.
- Survey topics: individual and household demographics, income, employment, health, education.
- Survey limitations and adjustments:
  - The survey is not a panel; latest information is from wave 2.
  - Phone surveys are subject to coverage and non-response bias; sample reweighting is applied to alleviate these biases.
- Outcome variables:
  - "job loss": binary variable = 1 for individuals who reported not currently working and having lost their pre-pandemic job.
  - "employed": binary variable = 1 if the individual reported having worked in the last week.
- Main independent variable:
  - "Women with internet": interaction of female dummy (1 if woman, 0 otherwise) and household internet connection dummy (1 if yes, 0 otherwise). Takes value 1 only for women with an internet connection in their household.
- Controls and fixed effects: country, state, education level, marital status included as fixed effects.
- Complementary binary variables used to split samples:
  1. "income reduced" = 1 if individual reported a decrease in household income compared to the previous wave (around June 2021).
  2. "High Covid-19 level" = 1 if country had a rate of COVID-19 cases higher than the world median.
  3. "Increased household chores" = 1 if individual reported an increase in household chores since the beginning of the pandemic.
  4. "Household with children" = 1 if there is at least one child in the household.
- Additional covariates in regressions include whether the woman has a smartphone and whether there are children in the household.
- Income proxy: question "Your household income has stayed the same, decreased or increased compared to the previous wave (around June 2021)?" used to assign "decrease" = 1.

### 1.1 Descriptive statistics (Table 1: Summary Statistics for Household-level Analysis)
- Sample totals:
  - No. Obs: 26.662
  - Job loss (%): 17,4
  - Employed (%): 64,4
  - Women (%): 51,7
  - Internet access (%): 62,4
  - Women with internet access (%): 31,0
  - Increased household chores (%): 28,7
  - Household with children (%): 60,5
- Country-level means (No. Obs; Job loss (%); Employed (%); Women (%); Internet access (%); Women with internet access (%); Increased household chores (%); Household with children (%)):
  - Belize: 898; 22,9; 57,7; 50,3; 69,3; 34,3; 31,3; 67,1
  - Guatemala: 1.521; 15,9; 66,9; 51,9; 33,3; 17,1; 20,6; 72,7
  - El Salvador: 812; 13,7; 66,1; 55,3; 47,3; 24,8; 20,3; 63,5
  - Honduras: 1.004; 21,2; 57,8; 52,5; 42,1; 20,1; 20,3; 76,2
  - Nicaragua: 865; 13,2; 67,5; 51,8; 32,6; 15,9; 23,4; 72,5
  - Costa Rica: 905; 18,7; 60,8; 49,9; 62,2; 31,1; 28,2; 50,2
  - Panama: 986; 25,6; 53,1; 50,3; 58,3; 29,8; 26,4; 60,3
  - Haiti: 2.361; 33,3; 46,2; 51,6; 9,0; 3,9; 35,4; 74,8
  - Peru: 1.302; 21,9; 67,7; 50,7; 48,9; 24,0; 33,2; 69,0
  - Mexico: 2.511; 14,7; 66,0; 52,0; 68,6; 34,0; 30,4; 58,1
  - Argentina: 1.321; 13,1; 65,6; 51,8; 76,9; 39,3; 25,8; 49,4
  - Chile: 1.329; 13,1; 62,5; 51,1; 73,6; 32,9; 37,9; 45,2
  - Colombia: 1.376; 26,5; 58,9; 52,1; 58,5; 30,1; 27,9; 64,5
  - Bolivia: 1.183; 13,7; 74,0; 50,3; 59,9; 30,6; 22,9; 67,8
  - Guyana: 875; 16,3; 63,1; 49,8; 70,3; 34,2; 37,7; 61,6
  - Ecuador: 1.615; 17,3; 64,7; 50,6; 74,2; 36,6; 25,8; 71,8
  - Paraguay: 1.061; 10,9; 77,4; 50,0; 51,9; 22,2; 20,7; 61,8
  - Uruguay: 930; 16,8; 61,0; 52,3; 72,4; 35,9; 22,6; 42,8
  - St Lucia: 860; 13,2; 69,1; 50,7; 81,5; 42,4; 31,0; 47,2
  - Dominica: 879; 15,1; 66,0; 49,6; 83,2; 43,8; 37,7; 51,6
  - Dominican Republic: 1.197; 20,8; 61,4; 50,5; 57,0; 29,1; 28,8; 63,5
  - Jamaica: 871; 14,4; 65,7; 51,0; 71,0; 36,2; 35,4; 57,2
- Descriptive highlights from Table 1:
  - Average job loss (pre-pandemic job lost and not currently working): around 17 percent.
  - Average employment: 64 percent.
  - Average percentage women in sample: 52 percent.
  - Average households with children: 60 percent.
  - Internet access shows significant variation: average access rate 61 percent driven by high rates in Dominica (83 percent), St. Lucia (81 prcent), Jamaica (71 percent), and extremely low rate in Haiti (9 percent).

### 1.2 Empirical strategy
- Baseline individual-level fixed effects specification (textual representation):
  - Y_i = α + β1 (Female_i * Internet_i) + β2 Female_i + β3 Internet_i + δ_i + ε_i
  - Y_i is either "employment" or "job loss".
  - δ_i denotes fixed effects for country, state, marital status, and education level.
  - Coefficient of interest: β1, effect of being female and having internet access on labor market outcomes compared to women without internet access.
  - Authors note: specification likely suffers from bias due to unobserved variables correlated with both outcome and internet access; coefficients should be interpreted as correlations, not causal effects.
- Sample-split analyses:
  - Same specification applied to subsamples defined by group_i = 1 or 0 for the complementary binary variables ("High Covid-19 level", "Decreased income", "Increased household chores", "Household with children").
  - For all estimations: clustered standard errors at the individual level and survey weights for country population are used.

*IMF WORKING PAPERS Digitalization and Employment Gender Gaps during the COVID-19 Pandemic: Evidence from Latin America and the Caribbean*

### 3.1 Household-level results

### 3.1 Household-level results

### Summary of main household-level findings
- Regressions examine two primary outcomes: job loss (Panel A) and employment (Panel B), with hours worked per week as a secondary outcome (Panel C).
- Simple correlations (columns (1)–(2) of Table 3) show that having access to the internet increases the probability of being employed and decreases the probability of losing one's job; coefficients have the expected signs and are statistically significant.
- Main specification (column (3) of Table 3, from Equation (1)) finds a positive correlation between being a woman with internet access and being employed, and a negative correlation with job loss. This relationship holds when controlling for living in a household with children, having increased household chores, or having a smartphone.
- Heterogeneity analyses split the sample by household chores, presence of children, income reduction, and COVID-19 severity (Tables 4 and 5). Key interpretations:
  - Digitalization may be more effective in mitigating the negative impact of household chores on job loss than in promoting employment.
  - Interaction coefficients are more negative for job loss among those living in households with children, suggesting internet access may allow more flexibility (e.g., working from home).
  - Coefficients of the interaction between female and internet are larger for individuals whose household income was reduced, indicating digitalization can be particularly beneficial for women in households experiencing income shocks.
  - A similar pattern is observed for individuals living in countries with higher rates of COVID-19 cases.

### Key estimates from Table 3 (Household Level Results)
- Panel A. Job loss
  - Women with internet: -0.0515*; -0.0523**; -0.0510*; -0.0524** (standard errors in parentheses: (0.0263), (0.0262), (0.0264), (0.0264))
  - Women: 0.185***; 0.187***; 0.178***; 0.186*** (standard errors: (0.0228), (0.0225), (0.0227), (0.0228))
  - Internet: -0.0798***; -0.0585***; -0.0290; -0.0281; -0.0288; -0.0250 (standard errors: (0.0153), (0.0152), (0.0186), (0.0185), (0.0188), (0.0189))
  - Household with children: -0.0164 (0.0123)
  - Increased household chores: 0.0589*** (0.0137)
  - Has an smartphone: -0.0642 (0.0442)
  - Constant: 0.225***; 0.212***; 0.125***; 0.134***; 0.111***; 0.184*** (standard errors: (0.0130), (0.0122), (0.0162), (0.0191), (0.0173), (0.0442))

- Panel B. Employment
  - Women with internet: 0.0403*; 0.0425*; 0.0397; 0.0427* (standard errors: (0.0245), (0.0244), (0.0245), (0.0245))
  - Women: -0.241***; -0.248***; -0.236***; -0.244*** (standard errors: (0.0189), (0.0187), (0.0189), (0.0189))
  - Internet: 0.0612***; 0.0315**; 0.00223; -0.000671; 0.00331; -0.00793 (standard errors: (0.0144), (0.0141), (0.0190), (0.0189), (0.0190), (0.0190))
  - Household with children: 0.0558*** (0.0124)
  - Increased household chores: -0.0533*** (0.0140)
  - Has an smartphone: 0.143*** (0.0383)
  - Constant: 0.605***; 0.623***; 0.754***; 0.725***; 0.766***; 0.623*** (standard errors: (0.0105), (0.0105), (0.0150), (0.0169), (0.0157), (0.0390))

- Panel C. Hours worked
  - Values reported are identical to Panel B in the table:
    - Women with internet: 0.0403*; 0.0425*; 0.0397; 0.0427* (standard errors: (0.0245), (0.0244), (0.0245), (0.0245))
    - Women: -0.241***; -0.248***; -0.236***; -0.244*** (standard errors: (0.0189), (0.0187), (0.0189), (0.0189))
    - Internet: 0.0612***; 0.0315**; 0.00223; -0.000671; 0.00331; -0.00793 (standard errors: (0.0144), (0.0141), (0.0190), (0.0189), (0.0190), (0.0190))
    - Household with children: 0.0558*** (0.0124)
    - Increased household chores: -0.0533*** (0.0140)
    - Has an smartphone: 0.143*** (0.0383)
    - Constant: 0.605***; 0.623***; 0.754***; 0.725***; 0.766***; 0.623*** (standard errors: (0.0105), (0.0105), (0.0150), (0.0169), (0.0157), (0.0390))

- Sample and fit statistics (Table 3)
  - Observations: 26,662; 26,507; 26,507; 26,507; 26,507; 14,217
  - R-squared: 0.012; 0.087; 0.135; 0.137; 0.137; 0.143
  - Country FE: YES for all columns
  - State FE: NO; YES; YES; YES; YES; YES
  - Education FE: NO; YES; YES; YES; YES; YES
  - Marital status FE: NO; YES; YES; YES; YES; YES
  - Notes: Clustered standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

### Heterogeneity results from Tables 4 and 5 (selected coefficients)
- Table 4. Household Level Results by Household Chores and Children in the Households
  - Panel A. Job loss (Women with internet)
    - Household chores increased: -0.0904* (0.0481)
    - Household chores NOT increased: -0.0527* (0.0298)
    - Household with children: -0.0631** (0.0292)
    - Household without children: -0.0440 (0.0424)
    - Observations: 5,402; 11,731; 10,982; 6,164
    - R-squared: 0.191; 0.163; 0.173; 0.181
  - Panel B. Employment (Women with internet)
    - Household chores increased: 0.0555 (0.0441)
    - Household chores NOT increased: 0.0352 (0.0277)
    - Household with children: -0.000933 (0.0281)
    - Household without children: 0.0974** (0.0389)
    - Observations: 8,147; 18,334; 16,594; 9,886
    - R-squared: 0.187; 0.159; 0.169; 0.179
  - Notes: The difference between groups is only statistically significant for women with internet on employment when comparing by children in the household.

- Table 5. Household Level Results by Income Status and COVID-19 Levels
  - Panel A. Job loss (Women with internet)
    - Income reduced: -0.0902** (0.0428)
    - Income NOT reduced: -0.0525* (0.0295)
    - High Covid-19: -0.0527* (0.0275)
    - Low Covid-19: -0.0472 (0.0362)
    - Observations: 5,685; 11,273; 13,205; 3,995
    - R-squared: 0.210; 0.147; 0.128; 0.146
  - Panel B. Employment (Women with internet)
    - Income reduced: 0.0401 (0.0395)
    - Income NOT reduced: 0.0534* (0.0284)
    - High Covid-19: 0.0441* (0.0260)
    - Low Covid-19: 0.0135 (0.0337)
    - Observations: 8,527; 17,557; 19,744; 6,763
    - R-squared: 0.187; 0.154; 0.135; 0.126
  - Notes: The differences between groups are not statistically significant.

### Interpretive conclusions (text preserved from the source)
- Interaction coefficients are larger for individuals who experienced an increase in household chores when analyzing 'Job loss.' For 'Employed,' there is no significant effect with the split sample, although coefficients remain positive. This suggests digitalization may be more effective in mitigating the negative impact of household chores on job loss than in promoting employment.
- Splitting the sample by presence of children shows interaction coefficients for 'Job loss' are negative and larger for those living in households with children, possibly because the internet allows more important role for women with more responsibilities at home (flexibility, work from home).
- In Table 5, differential effects on income are reduced; coefficients of the interaction between female and internet are larger for individuals whose household income was reduced, suggesting digitalization can be particularly beneficial for women in households experiencing income shocks.
- A similar pattern is observed for individuals living in countries with higher rates of COVID-19 cases, highlighting digitalization’s potential to mitigate the pandemic’s negative impact on female employment.
- Overall: digitalization can play an important role in promoting gender equality in the labor market, particularly where women face significant household and childcare responsibilities. However, additional policies and investments may be necessary to fully realize the potential of digitalization in these contexts.

*IMF Working Papers — Digitalization and Employment Gender Gaps during the COVID-19 Pandemic: Evidence from Latin America and the Caribbean*

### 3.4 Did Digitalization Help Women Get Through COVID-19?

### 3.4 Did Digitalization Help Women Get Through COVID-19?

### Key empirical findings (interaction of digitalization and COVID-19)
- Regression evidence shows a positive and significant coefficient on the interaction term of digitalization and COVID-19 dummy, indicating larger benefits of digitalization in countries more adversely affected by COVID-19.
- Female labor force participation rate (LFPR) results (Table 11):
  - Percent of Internet Users: 0.054*** (0.015) and 0.041*** (0.014) in specifications (1) and (2).
  - Percent of Internet Users * COVID-19: 0.024* (0.014) and 0.029** (0.013) in specifications (1) and (2).
  - Male LFPR shows smaller and statistically insignificant coefficients: Percent of Internet Users 0.017 (0.012) and 0.014 (0.013); interaction terms 0.013 (0.010) and 0.014 (0.010).
- Observations and model fit (Table 11):
  - Observations: 4,199 and 4,025 across specifications.
  - R-squared: 0.970, 0.971 for female LFPR models; 0.947, 0.945 for male LFPR models.
  - Country FE YES; Year FE YES.
- Interpretation: results are consistent with the hypothesis that digitalization, to some extent, helps women participate in the labor force during the pandemic; no substantial digitalization benefits are observed for male workers during the COVID-19 period.

### Household-level robustness checks (logistic regressions)
- Logistic regression results confirm main findings: positive correlation between internet access and employment for women and negative correlation between internet access and job loss.
- Panel A. Job loss (selected estimates, Table 12):
  - Internet: -0.543*** (0.0993) with odds ratio 0.581*** (0.0577) in one specification; -0.416*** (0.103) odds ratio 0.659*** (0.0677); -0.384** (0.173) odds ratio 0.681** (0.118).
  - Women coefficient: 1.203*** (0.158) with odds ratio 3.331*** (0.527) in a specification.
  - Women with internet: -0.0235 (0.195) coefficient and odds ratio 0.977 (0.190) in one cell.
  - Constants reported, e.g., -0.849*** (0.132) with odds ratio 0.428*** (0.0566).
- Panel B. Employment (selected estimates, Table 12):
  - Internet: 0.268*** (0.0630) odds ratio 1.307*** (0.0824); 0.148** (0.0643) odds ratio 1.159** (0.0746); 0.0372 (0.105) odds ratio 1.038 (0.109).
  - Women: -1.135*** (0.0954) with odds ratio 0.321*** (0.0307) in a specification.
  - Women with internet: 0.134 (0.124) coefficient and odds ratio 1.144 (0.142) in one cell.
- Sample and fixed effects:
  - Obs: 26,662; 26,448 across specifications.
  - Country FE YES in all specifications; State FE, Education FE, Marital status FE included in several specifications as noted.

### Country-level alternative measures and robustness
- Alternative country-level digitalization and female employment measures remain robust (Table 13):
  - Percent of Internet Users: 0.080*** (0.013) for Female Employment Ratio; 0.032*** (0.011) for Male Employment Ratio.
  - Number of Internet Users: 0.027*** (0.007) for Number of Female Employees; 0.037*** (0.005) for Number of Male Employees.
  - GDP Per Capita: -1.615 (1.220) for Female Employment Ratio; 0.802 (1.209) for Male Employment Ratio; -0.302*** (0.082) and -0.256*** (0.061) for number of female and male employees respectively.
  - Observations: 4,036 for employment ratios; 4,029 for number of employees.
  - R-squared: 0.965, 0.940 for employment ratios; 0.996, 0.997 for number of employees.
  - Country FE YES; Year FE YES.

### Endogeneity checks (GMM)
- GMM estimations control for endogeneity using internal transformation and lagged dependent variables; results remain statistically significant and consistent with a positive relation between digitalization and female labor force participation (Table 14).
  - Percent of Internet Users: 0.002* (0.001) and 0.004* (0.002) for Female LFPR specifications; 0.004*** (0.001) and 0.005** (0.002) for Female Employment Ratio specifications.
  - GDP Per Capita: -0.358* (0.194) and -0.222 (0.212) in two specifications.
  - Schooling: 0.119* (0.063) and 0.119* (0.066).
  - Women Law: -0.008* (0.004) and -0.010** (0.005).
  - Observations: 3,987; 3,816; 3,904; 3,785 across specifications.
  - Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

### Heterogeneity analyses (household chores, children, income, COVID levels)
- By household chores (Appendix Table A1):
  - Panel A. Job loss: Women with internet coefficients: -0.0904* (0.0481) when household chores increased; -0.0527* (0.0298) when not increased.
  - Internet: -0.0577** (0.0288) and -0.0532*** (0.0178) in job loss models; small/insignificant interaction terms in some columns.
  - Panel B. Employment: Women with internet coefficients 0.0555 (0.0441) and 0.0352 (0.0277) depending on chores increase.
  - Observations vary: 5,402 and 11,731 in job loss panels; 8,147 and 18,334 in employment panels.
  - R-squared values reported, e.g., 0.160, 0.129 for job loss.
- By children in household (Appendix Table A2):
  - Panel A. Job loss: Women with internet -0.0631** (0.0292) in households with children; -0.0440 (0.0424) without children.
  - Internet: -0.0510*** (0.0173) and -0.0542** (0.0265) in job loss models by children presence.
  - Panel B. Employment: Women with internet -0.000933 (0.0281) in households with children; 0.0974** (0.0389) in households without children.
  - Observations: 10,982 and 6,164 for job loss panels; 16,594 and 9,886 for employment panels.
- By income status (Appendix Table A3):
  - Panel A. Job loss: Women with internet -0.0902** (0.0428) when income reduced; -0.0525* (0.0295) when income not reduced.
  - Internet: -0.0867*** (0.0250) when income reduced; -0.0315* (0.0176) when income not reduced.
  - Increased household chores: 0.0535** (0.0224) and 0.0595*** (0.0163) in some specifications.
  - Panel B. Employment: Women with internet 0.0401 (0.0395) when income reduced; 0.0534* (0.0284) when income not reduced.
  - Observations: 8,527 and 17,557 across panels; R-squared reported (e.g., 0.139, 0.107).
- By COVID-19 levels (Appendix Table A4):
  - Panel A. Job loss and Panel B. Employment replicate patterns shown above with sample split by High COVID-19 (19,744 observations) and Low COVID-19 (6,763 observations).
  - Coefficients for Women with internet and Internet variables mirror those in income-reduced panels (e.g., Women with internet -0.0902** (0.0428) in High COVID-19).

### Conclusions and policy-relevant implications
- Main conclusion: Higher levels of digitalization are associated with higher employment and lower job loss in the LAC region, including mitigating the negative impact of the COVID-19 crisis on female employment.
- Additional findings:
  - Digitalization’s effect is stronger in countries with fewer childcare resources, suggesting digitalization can help alleviate childcare burdens and promote women's labor market participation.
  - Positive effects of digitalization can be limited by institutional weaknesses and aggravated structural vulnerabilities.
- Policy recommendations and implications noted in the text:
  - Policies aimed at improving institutional capacity may be necessary to fully realize digitalization benefits in contexts with weak institutions.
  - Investments in education and infrastructure are highlighted as potentially necessary to enable digitalization to deliver its full positive impact on female employment.
- Research caveat and future direction:
  - While COVID-19 provides a useful example of an economic shock, it is premature to generalize findings broadly.
  - Future research should estimate an all-encompassing expected vulnerabilities of women to digitalization to inform social and economic policy discussions.

*Digitalization and Employment Gender Gaps During the COVID-19 Pandemic: Evidence from Latin America and the Caribbean — Working Paper No. WP/2024/12*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024012-print-pdf.pdf_