1. Introduction

According to the International Labor Organization (ILO), care work is defined as all the activities conducted by households, government, and the private sector to provide services for domestic work and care for children, the elderly, or people with disabilities. It also includes professional services related to education and health. Cross-country comparisons made by ^{ILO (2016)} utilizing time-use surveys show that unpaid care work is mainly undertaken by women, which affects their ability to enter into the labor market and the quality of their jobs, further increasing gender inequality in the workspace (^{Addati et al., 2019}). In Mexico, according to 2021 estimations made by the National Institute of Statistics and Geography (Instituto Nacional de Estadística y Geografía, INEGI, 2022), almost three-quarters (73.9%) of the hours spent on unpaid care work hours are performed by women. According to ^{Pacheco (2018)} women devote on average more than twice as many hours per week (28 hours) than men (12 hours) caring for other household members and almost three times as many hours per week (29 hours) in domestic work than men (10 hours). In contrast, men spend an average of ten hours more per week in paid work (48 hours) compared to women (38 hours).

This paper develops a discrete choice model for mothers’ labor and care strategies and simulates fiscal policies’ effects on their labor and care decisions. The model focuses only on mothers with children under five years of age, modeling their labor and leisure decisions assuming that other household income sources are exogenous. Similar labor and discrete choice models have been developed by ^{Kornstad and Thoresen (2007)}, ^{Kalb and Thoresen (2010)}, ^{Averet et al. (1997)}, and ^{Gong and van Soest (2002)}.

The proposed model aims to contribute to the current literature by proposing that women choose the type of childcare based on the number of hours, the wage, and the labor market segmentation between formal and informal employment. This is relevant for most Latin American countries, particularly in Mexico, since labor status in Mexico determines access to social security. Workers with formal jobs have access to public childcare services financed by social security contributions, while workers in informal jobs can only access childcare services through private centers. The proposed utility function for working mothers depends on the household’s disposable income, leisure, and care options. Using the National Survey on Time Use (Encuesta Nacional sobre Uso del Tiempo, ENUT) conducted in 2019 by INEGI, a mixed logit model is estimated using a proposed parametric form of the utility function, as well as relevant demographic variables, such as the number of children in the household, geographic region, household size, and household assets.

The model uses a sample of 3,534 women to simulate the decision making process of working mothers, who can choose different combinations (or states) of labor options and childcare arrangements. The estimated distributions of the simulations are compared with the observed distribution of the sample for the different states, with the best model showing similar results for eight of the ten possible states. Using the best-fitted model, three policies are simulated: 1) a reduction of the marginal rates of the personal income tax; 2) a tax credit to the cost of private care; and 3) a conditional cash transfer to mothers with children under five who work in the formal sector and live in poor households. Our findings suggest that a personal income tax reduction increases women’s participation in the formal market, while the tax credit and the conditional cash transfers to women have more limited effects on increasing formal employment and reducing parental childcare.

2. Care work in Mexico and the labor market

As noted by UN Women and the Economic Commission for Latin America and the Caribbean (^{UN Women and ECLAC, 2020}) the demographic transition in the region, along with the changes in family structure and migration, have increased women’s participation in the Latin American labor markets during the last couple of decades. From 1990 to 2018, Latin America’s average female labor force participation rate increased from 42% to 52% (^{UN Women and ECLAC, 2020}). There have been gradual changes in the traditional breadwinner structure of families based on the sexual division of labor, in which the male partner was the sole provider and the women within the household took on exclusively the role of caregiver. However, even though the gender gap in the labor force has been closing during the first two decades of this century, male participation in the labor market remains, on average, 25 percentage points above female participation (^{UN Women and ECLAC, 2020}).

In 2018, Mexico had the lowest female labor force participation rate (43.5%) in Latin America and the Caribbean after Guatemala (39.2%) (^{ECLAL and ILO, 2019}). Despite many remaining challenges to include women in the workplace, the Mexican gender gap in the labor market decreased during the last ten years prior to the pandemic. From 2010 to the first quarter of 2020, the number of working women for every 100 males increased from 60 to 65. However, the COVID19 pandemic and its economic effects on the labor market were not gender-neutral, affecting women disproportionately during the recovery months after the pandemic. While general occupation, including men and women, fell 19% from the first to the second quarter of 2020, male occupation recovered to pre-pandemic levels during the first quarter of 2021, growing almost 20%, while female participation only grew 15% in the same period (^{INEGI, 2021b}).

The pandemic increased the need for additional care work in the household, given the new health and educational challenges. This increase in care, paired with the structural barriers women face in the labor market, might hamper the modest progress made during the last decades. In Mexico, 75% of the total hours of unpaid care are done by women. Ninety percent of all the hours invested by women on care within the house are focused on caring for underage members of the household, including educational support. The pandemic dramatically increased this kind of care work. From 2020 to 2021, the monetary value of unpaid care done by households increased from 22.9% of gross domestic product (GDP) to 27.6% (^{INEGI, 2021a}). Considering the current family structure of care within the households, this will likely undermine the recovery of female labor in the workplace.

According to the National Survey of Occupation and Employment (Encuesta Nacional de Ocupación y Empleo, ENOE), in the fourth quarter of 2021, 4.9 million women were outside the formal workforce but available for employment.¹ Of these, 63% of them (3.1 million) reported unpaid care within their household as the main barrier to entering the labor market. Moreover, 1.5 million women were completely unavailable for work, reporting as the main reason their sole responsibility was to care for the children, the elderly and/or the sick within the household (^{INEGI, 2021c}).

Available care options are directly related to limited choices in the labor market. Almost 60% of the 22 million women currently employed (approximately 13.2 million) do not have direct access to social security in their jobs (^{INEGI, 2021c}). In Mexico, public care options for women with children under five years old are directly linked with their formal status. Therefore, working mothers with children under five without social security must rely on private (i.e., paid) childcare or care provided by members of the household, most of the times with help from grandparents or other relatives, to be able to manage their time between work and care. Even women with working male partners with access to social security cannot take their children to public childcare since the current Mexican legislation and institutions assume a breadwinner structure.

In most countries in the world, women with children under five years old have a lower probability of entering the labor market. In almost every region, the labor participation of women with children under five years old is significantly lower than in the case of men, independent of their civil status, and even lower compared to women without children. According to ^{UN Women and ECLAC (2020)}, in Latin America, the gap between women with children under five and women with a partner but without children is more than 10 percentage points and more than 30 percentage points, respectively, compared to single women. Early motherhood affects not only women’s probability of entering the market but also the quality of the job they are able to secure. According to ^{Addati et al. (2019)}, women with children under six tend to work less paid hours, have lower wages, and are more prone to work in the informal sector as self-employed, which increases their chances of falling into poverty and remaining in precarious conditions.

3. Previous studies

^{Kornstad and Thoresen (2007)} present a discrete choice model for Norway, where they model the decision-making process by women with children under five in the labor market, choosing between public and parental options. Every option in the labor market has a fixed number of hours, and every care option comes with a number of observable attributes (including costs); the authors simulate the different states in order to model the effect of the reforms to the care system in Norway and their potential effects in women’s choices. For the United States (US), ^{Averet et al. (1997)} present a structural model of labor supply where they took into consideration the cost of care options to estimate the impact of subsidies to care centers; similar to ^{Kornstad and Thoresen (2007)}, the authors estimate the model only for women with children under five years old, taking the income of the partner as exogenous. The authors find that the supply of labor of this group is sensitive to changes in net wages rather than gross since the first takes into account the cost of care subsidies. Therefore, there is evidence that government subsidies increase the labor supply of women under these conditions.

^{Del Boca (2015)} explores the relationship between different family structures in Europe and North America and the work arrangements made by working mothers. She finds an inverse relationship between the investment made by governments from different countries in care centers- through direct services or subsidies- and the involvement of grandparents raising children, which is the most traditional form of childcare. This last strategy is more prevalent in more traditional societies with bigger families and more social cohesion within the household. She also finds that the larger the number of children, the lower the participation rate of women in the market. The paper suggests that the use of public care options in almost every country has a positive effect on labor participation for women with children.

In the case of Mexico, ^{Gong and van Soest (2002)} develop a discrete structural model of labor supply to estimate the effect of wages, autonomous income, and family structure on the labor supply of working mothers with children under three years old. As in the case of ^{Kornstad and Thoresen (2007)}, the authors estimate a discrete utility model using a multinomial logit, and their results suggest that: 1) the presence of older children and substitute mothers in the household increases the probability of women entering into the market; 2) as the number of total children increases, the supply of working hours decreases; and 3) similar to other developing countries, the traditional family structure in Mexico shapes the household’s production function, thereby strongly influencing women’s decisions, not only in terms of an economic rationale but also social and cultural norms.

Using data from the Mexican Census of Population (Censo Nacional de Población, CNPV) of 1990 and 2000, ^{Arceo-Gómez and Campos-Vázquez (2010)} estimate female labor supply elasticities for married women and find that married women with children under five years old are less engaged to the labor force than average married women. They hypothesize that these results might suggest a lack of attachment to the labor market caused by time constraints in caregiving responsibilities.

In terms of evaluating public care policies in Mexico, ^{Calderón (2014)} evaluates the policy “Estancias Infantiles” (childcare centers) carried out by the Mexican government from 2007 to 2018. This was a subsidy program for private care centers to help women with children under five years old without access to social security. The author uses a triple differences model to estimate the effect of the program on women’s labor supply. Her findings suggest that the program increased women’s insertion in the market, reduced their care burden, and even increased labor income for partners since the new redistribution of care liberated time for all the carers within the household, women and men alike.

^{López-Rodríguez and Orozco (2019)} explore the relationship between social capital, care work in the household, and financial inclusion in female labor force participation. Using data from Mexico’s National Survey of Household Income and Expenditure (Encuesta Nacional de Ingresos y Gastos de los Hogares, ENIGH), the National Council for the Evaluation of Social Policy (Consejo Nacional de Evaluación de la Política de Desarrollo Social, CONEVAL), administrative registries from the Ministry of Social Development, and the National Survey on Financial Inclusion, the authors estimate a probit model with Heckman correction. According to their findings, 1) higher wages increase the probability of women working in full-time jobs, 2) the higher the number of children, the lower the probability of working, and 3) the presence of childcare facilities in the municipality increases the probability of working in a full-time job.

^{Talamas (2023)} studies the role of grandmothers -who often play the role of a childcare provider in Mexican households- in mothers’ employment rates. Using Mexico’s ENOE, the author estimates a triple difference model using the death of a grandmother within a household with children as an exogenous variable to obtain causal effects and to control other effects that might influence the employment rate through other channels. The death of a co-inhabiting grandmother reduces the employment rate of mothers by 15 percentage points. In addition, this negative effect is smaller where public childcare or affordable private childcare is available.

Using data from the Mexico Social Mobility Survey 2017 and the National Statistical Directory of Economic Units (Directorio Estadístico Nacional de Unidades Económicas, DENUE), ^{Orozco et al. (2022b)} estimate through the rank-to-rank regressions the relationship between women’s social mobility, and childcare facilities, female labor force participation and social security of parents. Women who have access to childcare facilities in their immediate surroundings have higher upward social mobility than those without access, and this effect is larger for women who have lower socioeconomic status.

^{Orozco et al. (2022a)} estimate the effect of childcare services on female labor force participation at the municipal level in Mexico using a geospatial model with data for 2015 and 2020 from DENUE, CNPV, the Intercensal Survey, the Economic Census, and the Financial Inclusion Database. They find a negative relationship between the number of children under five years old and female labor participation and that the existence of childcare services in a municipality has an effect of 10.3% on female labor participation. This effect is larger for municipalities with high output in tertiary sector activities.

Finally, ^{Hoehn-Velasco et al. (2022)} analyze the impact of the COVID-19 recession on childcare and employment. According to their findings, men spent more time engaged in domestic care work during the pandemic, although the total distribution remained unchanged, and the hours spent on childcare also remained unchanged by the pandemic, mostly due to the fact that before the pandemic women already spent significant hours outside of the labor market doing domestic care work.

International literature and previous studies about Mexico find evidence that: 1) women with children under five years old decrease the probability of working; 2) the existence of childcare services available for women with children increases their probability of working, and 3) government subsidies of childcare and social programs are likely to have a positive effect on female labor force participation. This paper builds on previous discrete choice models to simulate the labor and care work decisions of women in Mexico with children under five years old to study the potential outcomes of fiscal incentives, direct subsidies, and social programs. The model considers the structure of the Mexican labor market and its effects on workers’ access to childcare.

Model

5. Data

As said earlier, most data come from ENUT 2019. The sample for that survey is representative at the national level and measures different characteristics at the individual and household level, such as the time spent on different activities- including care, labor conditions, access to social security, income sources, assets of the household, and whether people take their children to childcare centers. Only women with children under five years old but still not in elementary school will be considered in the model. The total sample of women in ENUT with these characteristics is 3,536 women.

For the consideration of labor options in B, the definition of an employed worker is any woman from the sample who declared on the survey that they worked during the week and received a wage or other type of payment (tips or commissions). For the selection of the nonemployed, the sample only considered women that explicitly declared care as the main reason for not having a paid job. Other working women excluded from the model were the boss/employer of the business, as well as self-employed women since the working decision for the latter would require other specifications and assumptions. For example, the model could work better in a continuous environment if women had the flexibility to allocate as much time as they wanted if they were to own the business or have direct control of their work. Finally, the model also excludes women who received some form of permanent injury pension or who have some disease or permanent disability that precludes them from entering the labor market.

As previously mentioned, all women who reported working 30 hours a week or less would be considered to work part-time, and all women with more than 30 hours full-time. Additionally, a variable of access to public social security was created. All working women who received medical insurance as part of their job benefits job were considered to be registered in IMSS or ISSSTE. All women who did not receive this benefit were considered without access to social security.

For the care options in S, ENUT reports whether a child attends a care center or kindergarten and identifies if the mother of that child has access to any public healthcare system. The model considers users of public care centers as children that attend a care center, and the mother has direct access to social security through her job, which provides the benefit of care services. On the other hand, the option for private care centers applies to all women with children who enter into one of three cases: 1) attend a care center but the mother does not have direct access to social security; 2) attend a care center and the mother has basic access to social security but not explicitly access to public care; this can happen for independent workers paying in a special regime in IMSS; or 3) children who have paid private care at home (like tutors or nannies).

It should be noted that in Mexico, male workers in a household-father, brothers, or grandparents- can grant access to social security to the rest of the family, but only in very few cases to care centers. This happens only if the male worker can prove that he is a widower or divorced, given the original design of social protection that assumes a breadwinner structure with very clear gender roles.

Parental care can be defined as women who declare that they do the majority of the hours designated to childcare, and their children do not attend any care center nor have any paid private care at home. Following previous studies on childcare, in order to consider the effect of having other adults take care of children in the labor market, a dummy variable for grandparents in the household was created. While exploring the data, cases were found in which the male head of the household and working women both had full-time jobs and still reported not taking their children to a care center. In order to avoid inconsistencies, those cases were excluded, assuming that another member of the family does the full care of those cases; therefore, the structure of the incentives was different from the model proposed in this paper.

To calculate disposable income, there is information on the individual level of labor income, plus other sources of income at the household level, such as cash transfers from social programs, rent from property, rent from physical assets, rent from interests, family transfers, pensions, etc. To use only current income, the following items were considered and adjusted to monthly quantities: 1) labor income from women considered in the model, 2) labor income from the rest of the members of the household, 3) per capita current nonlabor income at household level from pensions, cash transfers from social programs, private transfers- including remittances, rents from financial and physical assets.

To calculate the amount of taxes paid, the Mexican income tax (Impuesto Sobre la Renta, ISR) is considered labor income, and social security contributions are also considered in the case of having direct access to social security. Other sources of income were taken out from the tax scheme since 1) the model assumes that this source is exogenous and, therefore, will not be considered in the policy simulations, and 2) there is no detailed data at the individual level to calculate the effective tax rate. In order to derive the tax rate from disposable income reported in ENUT, the model uses the procedure followed by ^{Absalón and Urzúa (2012)} to recover the tax base from ISR and social security contributions for the labor income of women. It is assumed that women with no social security do not pay social security contributions nor ISR, and they are considered to be informal since there is no available data to consider if those individuals actually pay any taxes.

Finally, some imputations were made to account for the cost of private care since ENUT does not have any data on expenditures. For all women in the state S=2 who have children that attend private care centers, the cost of private care is taken into account in the disposable income equation, assigning the average monthly cost per child under five years old by geographic regions obtained from the 2017 National Employment and Social Security Survey (Encuesta Nacional de Empleo y Seguridad Social, ENESS) (^{INEGI, 2023}).

The disposable income is then divided by the number of members in the household adjusted by the equivalence scale used in the methodology of the official poverty measures by ^{CONEVAL (2023)}.

The resulting sample for the model was 3,534 women with children under five years old, which represents roughly 4.8 million women at the national level. To improve the accuracy of the model, demographic variables from the household have been included as casespecific: household size, access to health, assets of the family, and the number of children within the household using age ranges, following the approximation made by ^{Conelly (1992)}, who shows that the age of other children within the household influences the decisions of women in the labor market and their care decisions. Table 2 shows the descriptive statistics used in the model.

Since we can only see the women’s wages of the selected state, a fundamental challenge is to estimate wages for women who are not employed and their expected wages if they change from the formal to the informal sector, or from part to full-time jobs. In order to derive the alternative-specific wages, a wage estimation for part-time and full-time jobs is made from an ordinary least square (OLS) model for the log of wages following ^{Kornstad and Thoresen (2007)}, using as regressors the level of education, education squared, experience, experience squared, a dummy variable to calculate if women are in the formal or informal sector, another dummy variable for being in a part-time or full-time job, and four dummies for the region of the country.³ These estimates are used to create an alternative-specific disposable income function. Finally, the leisure and childcare options are set as categorical variables dependent on state B, as mentioned in the previous section.

6. Results

To estimate the different wages for any given state, an OLS model is estimated using only data for working women with positive wages reported. The wage for women outside the market for the estimated states where women are not employed is zero since women who report not working in ENUT 2019 do not answer questions about wages. The estimation adjusts the dummy variables for informal and parttime jobs for the different states. Table 3 shows the results of that regression.

All the models except one in Table 3 show that three variables -education years squared, being in the informal sector, and having a part-time job- are statistically significant. Adjusted R squared, and the root of the mean squared error were estimated to select the best model for estimated wages. For the models that used the log of wages, the root of the mean squared error was estimated after transforming to wages the predicted results. The selected model to estimate wages for different states was the first model, which uses the level of wages rather than the logs and includes dummies for all the regions of the country. Even when the first model with levels has a smaller adjusted R squared (which is expected considering that the log of wages reduces variability), it has the highest root squared mean error amongst all the models. Wages for the tenth states are estimated using these coefficients, replacing the observed wages for all the cases to use as much observed data as possible. The mixed logit model will take as a baseline the scenario of women not employed with parental care since it is the most prevalent scenario. For the purposes of policy-making, it is relevant to analyze the factors that would increase the probability of a woman in that state becoming employed. Table A2 shows the model results for all the states, and Table 4 shows the predicted wages by employment status.

The selected model shows all the alternative-specific variables as statistically significant at the 99% confidence level. There is significant variation of case-specific alternatives depending on the specific states and the sub-sample for each state: states of part-time formal workers have very few observations, so standard errors tend to be comparatively large: state (b₄,s₁) has 51 observations, state (b₄,s₂) has 11, and state (b₄,s₃) only 36. However, some general insights can be drawn from Tables A1 and A2 in Appendix A. As one could expect, having children aged two or younger reduces the probability of a woman transitioning from being not employed and using parental care to almost any other option, even part-time jobs. This could be because the burden of care is more intense during a child’s first two years of life. On the contrary, having children aged three to five increases the probability of being employed in almost any other state. This could be due to the availability of private and public centers for this age group in preschool. In addition, having grandparents in the household increases the probability of being employed in several states since other members of the household can contribute to childcare.

This model is used in the next section to simulate changes in disposable income due to changes in different policies. In addition to McFadden’s pseudo R squared, in order to evaluate the fitness of the model, Table 5 compares the mean of observed state probabilities with the mean of the simulated probabilities and their 95% level confidence intervals. Considering the fact that the model was estimated without a constant (i.e., probabilities are only considered using the observable variables), the model fits relatively well for almost all states. It slightly overestimates public care for full-time job mothers in the formal sector.

7. Policy simulations

There is a wide array of public policies to reduce and redistribute unpaid childcare that disproportionately affects women- maternity and paternity leave, increasing school hours, more flexible working schemes for mothers, fiscal incentives for early child education (ECE), universal provision of public childcare, care-friendly social protection systems, among others (^{Addati et al., 2019}). Considering the available survey data for Mexico and the environment and assumptions of the model proposed in early sections, this paper focuses on fiscal and social policies that directly affect childcare costs through an increase in the disposable income of the household.

Countries like the United States, which rely on private ECE, have adopted direct subsidies like portable vouchers for low-income families to incentivize the use of childcare centers. Some of them, like the Child Care Development and Block Grant, require parental employment, while others, like the Early Head Start, do not require parental employment since the policy priority is focused on the positive effects of ECE on young children and toddlers and not on increasing employment (^{Morrisey, 2017}).

^{Morrisey (2017)} presents the main research findings for several countries with policies to reduce ECE costs (either public or private). Countries like the United Kingdom use instruments like the Working Families Tax Credits for low-income families, which can cover up to 70% of childcare costs up to L70 per week and show an increase of five percentage points in female labor participation amongst single mothers with children under five years old. In the case of Finland, a voucher program was introduced in several municipalities in 1994, showing an increase in the use of private childcare but not showing effects on female labor participation.

Using the estimates from the model, three policy simulations are conducted to evaluate whether changes in fiscal and social policy increase female labor participation for women with children under five years old and, therefore, change the choices of women with regards to their selected labor and care options.

The three options are 1) a reduction of marginal rates in Mexico’s income tax (ISR), 2) a tax credit for working women who choose private sector childcare, 3) a conditional cash transfer for working women living in households with a per capita income below the poverty line. Table 6 shows the change from the baseline predicted by the model to a specific simulation, and Table A3 in the Appendix shows changes through aggregate categories of labor status and childcare options.

8. Conclusions

Most women with children under five choose not to enter the labor market. As shown by the latest data from ENOE, this situation was aggravated further by the COVID-19 pandemic since the recovery for women has been slower than for men. Research from ^{Nieves and Robles (2016)} on policies in Latin America suggests that labor regulation, fiscal policies, and social protection systems can effectively reduce childcare work and increase labor participation in the market. This paper focused on the fiscal and social policy components and analyzed three policy simulations that explored the possible increase of labor force participation and childcare centers through increasing the disposable income of women with children under five years old. However, the options have different effects on labor markets and childcare options.

The most effective policy simulation, and the one with the highest direct cost, was the reduction of income taxes through adjusting the marginal tax rate for lower income brackets. The reduction of personal income tax marginal rates increases formal employment and reduces informal work and parental childcare for women who are not employed. In contrast, the tax credit and the conditional cash transfer have very limited effects. The former increases the percentage of women working in the formal sector with private childcare but has only a marginal effect on women not employed or childcare centers. The latter increases part-time formal employment but also has marginal effects on other states.

Fiscal and social policies targeted at formal workers have considerable limitations. In the analyzed sample, only 17% of the women are working in the formal sector. The other 15% of employed women are in the informal sector. It is also worth noting that even tax reduction has limited effects since the majority of the sample is concentrated in the lower brackets of the distribution and, therefore, the effective taxes paid -both in relative and absolute terms- are already low. An important area for future research is to analyze policies that could support women in the informal sector who tend to have lower wages without creating perverse incentives in the labor market. Most of these women are self-employed and have more flexibility in terms of working hours, a potential area of research outside the assumptions and design of this discrete choice model but important in terms of policymaking.