Costs of maternity leave to support breastfeeding; Brazil, Ghana and Mexico

Abstract Objective To develop a method to assess the cost of extending the duration of maternity leave for formally-employed women at the national level and apply it in Brazil, Ghana and Mexico. Methods We adapted a World Bank costing method into a five-step method to estimate the costs of extending the length of maternity leave mandates. Our method used the unit cost of maternity leave based on working women’s weekly wages; the number of additional weeks of maternity leave to be analysed for a given year; and the weighted population of women of reproductive and legal working age in a given country in that year. We weighted the population by the probability of having a baby that year among women in formal employment, according to individual characteristics. We applied nationally representative cross-sectional data from fertility, employment and population surveys to estimate the costs of maternity leave for mothers employed in the formal sector in Brazil, Ghana and Mexico for periods from 12 weeks up to 26 weeks, the WHO target for exclusive breastfeeding. Findings We estimated that 640 742 women in Brazil, 33 869 in Ghana and 288 655 in Mexico would require formal maternity leave annually. The median weekly cost of extending maternity leave for formally working women was purchasing power parity international dollars (PPP$) 195.07 per woman in Brazil, PPP$ 109.68 in Ghana and PPP$ 168.83 in Mexico. Conclusion Our costing method could facilitate evidence-based policy decisions across countries to improve maternity protection benefits and support breastfeeding.


Introduction
Creating an enabling environment for women to successfully breastfeed has wide-reaching health, economic and environmental benefits. 1,2 Improving breastfeeding outcomes globally could prevent an estimated 823 000 child deaths and 20 000 breast cancer deaths every year. 1 However, the prevalence of exclusive breastfeeding among infants younger than 6 months remains low, around 37% globally. 3 Breastfeeding practices are affected by a wide range of factors, including sociocultural and economic contexts, health systems, families and communities, employment, and individual attributes of the mother, the infant and their relationship. 2 Interventions in these areas can potentially promote a more enabling environment, and in turn, achieve the global World Health Organization (WHO) target of 70% of babies exclusively breastfed up to 6 months by 2034. 4,5 Public policies are needed, especially for women such as working mothers who may be deterred from breastfeeding. Given the increase in women's participation in the labour market around the world, maternity protection policies are considered essential for improving breastfeeding practices. 6 Giving women a period of paid absence from work after childbirth provides social, developmental and health benefits for working mothers and their children and has been shown to be effective for increasing exclusive breastfeeding. 2,7,8 Evidence from Brazil, Canada, China, Sweden and the United States of America suggests that the duration of maternity leave has a positive association with exclusive breastfeeding and maintenance of breastfeeding. 6,[9][10][11][12][13][14] A study that assessed the expan-sion of the maternity and parental leave mandate in Canada from 25 to 50 weeks found a significant increase in exclusive breastfeeding rates at 6 months by 5.8 percentage points. 6,14 Evidence from Sweden reveals that long periods of mandated maternity leave promote higher rates of breastfeeding and a larger share of women returning to work: both important factors for social well-being and development. 6 Recent evidence from 38 low-and middle-income countries showed that the extension of maternity leave has the potential to reduce barriers to breastfeeding for working mothers. 8 In addition, the length of maternity leave is associated with improved mother's mental health, 15,16 and lower neonatal and postnatal mortality. 16 Previous studies have highlighted work-related issues as a major reason why mothers do not start breastfeeding or stop exclusive breastfeeding early. 10 The effects of work on women's decisions to breastfeed are multidimensional, including fatigue and financial stress. 2,6 Hence, labour protection policies have a strong potential to positively influence both breastfeeding and women's labour market participation. 13 Although many countries have maternity protection legislation, only 99 (out of 185) meet or exceed the minimal 14 weeks of paid maternity leave recommended by the International Labour Organization (ILO), 17 57 countries meet 14-17 weeks of leave, and just 42 countries meet or exceed 18 weeks leave. These numbers imply that employed women globally face inadequate maternity protection to enable them to achieve their breastfeeding goals. 2 Maternity leave can be financed in different ways: social security schemes that rely on a mix of contributions from employers, employees and government funds; public funds; or solely by the employer. To effectively scale up and sustain Research Mireya Vilar-Compte et al.
coverage of effective breastfeeding interventions, the costs must be considered, 2 specifically at the country level. 18 Identifying the economic implications of breastfeeding should be a priority, as increasing breastfeeding prevalence could have substantial economic effects, 19 for example, on a country's gross domestic product. Previous studies have highlighted the need for standardized breastfeeding costing frameworks at the national level. 18,20,21 Global costing frameworks for breastfeeding have helped highlight the need for further investment and resources. 22,23 However, these methods have seldom been adopted at the national level to estimate the costs of maternity leave policies that could be used by local breastfeeding advocates and policy-makers.
Previous studies have estimated the costs of extending the duration of maternity leave for women employed in the formal sector in Chile, 24 Indonesia 25 and Norway 26 and the cost of implementing new maternity schemes in the USA. 20 Despite the relevance of these specific costing studies, there is a need for pragmatic, standardized algorithms for establishing the costs of incrementally expanding the duration of mandates at the country level. Governments can then assess the financial feasibility of implementing or expanding programmes. Given that the cost of extending maternity leave can vary greatly across countries due to differences in policies and wages, it is important to develop a method that uses data commonly available across countries. The aim of our study was to develop a method for estimating the cost of extending the duration of maternity leave for mothers employed in the formal sector at the national level using existing country-specific data and apply it in Brazil, Ghana and Mexico.

Setting
We used nationally representative, publicly available, cross-sectional data from each country. While the data were comparable across countries, the dates of data collection were different; data for Brazil were collected in 2015, Ghana in 2017 and Mexico in 2013-2014. These countries were selected because they are diverse across several domains: economic development, labour market structure, women's participation in the labour force, fertility rate and breastfeeding indicators (Table 1). Furthermore, regulations on maternity leave differ. In Brazil, female employees receive mandatory maternity leave at full pay for about 4 months, paid by the social security agency, while employers have the option of offering an additional 2 months and deducting the amount paid from its corporate income tax. 29 In Ghana, female workers are entitled to a full period of paid maternity leave of at least 12 weeks, which is paid by the employer. 31 Mexico has extended the maternity leave mandate at full pay from 12 to 14 weeks, financed by the social security system. 29

Costing method
We adapted a costing method from the World Bank, 18,23 which estimates the financial needs for scaling up a nutrition intervention to achieve World Health Assembly global nutrition targets. 32 The method is based on the following equation:

FN UC IC Pop
where FN y is the annual financial need for a given intervention in year y, UC the unit cost, IC y is the incremental coverage (IC), assumed for year y and Pop y is the target population in year y.
We modified this costing approach to make it more precise and suitable to maternity leave mandates. We weighted the population by α, which is the probability of having given birth among formally employed women according to the following characteristics: age, marital status, educational level and locality (urban or rural). Hence, we estimated the cost of extending the maternity leave for women working in the formal sector as: Where ML y is the maternity leave cost needed for a given year of intervention, W is the maternity leave unit cost, IC y is the weekly incremental coverage for maternity leave assumed for year y and α × Pop y is the population of women of reproductive and legal working ages in a given country in year y weighted by α (probability of having given birth according to women's characteristics).  For each group of women (combinations) identify the mean or median weekly wage.
To decide whether to use the mean or the median, plot a density function graph of weekly wages to see if its distribution is symmetrical (see Fig. 1  The assumption for the three countries was that maternity leave benefits would cover 100% of the salaries Step A key aspect behind this modelling approach is that it is based on five clearly delineated steps that could be replicated across countries (Table 2). To apply this method, nationally representative surveys with data on employment and fertility should be available, and demographic data are required to adequately calibrate to the population size. These are data sources commonly available in different countries.

Application of costing method
Following the steps of the costing method (Table 2), we estimated the annual costs of extending maternity leave for formally employed women in Brazil, Ghana and Mexico.
Step 1 was determining the number of women of reproductive and legal working age who reported having a child in the previous year; this number is necessary for computing α. Table 2 summarizes the data sources on fertility for each country. We categorized women of reproductive age according to their age bracket, marital status, educational level and urban or rural residential locality. While the goal was to have a process as standardized as possible, the definitions of the variables slightly differed across countries due to differences in definitions attributable to each country. This led to a different number of possible combinations of women's characteristics, which derived from the demographic features of each country. For each combination, we assessed the proportion of women who reported having given birth in the previous year. For example, in Brazil the proportion of women aged 30-34 years, who had completed high school, lived in an urban locality and were married, and who had a baby in the previous year, was 8.1%.
Step 2 was to determine the probability of a woman working in the formal sector having had a baby in the previous year (α). This step required defining formal employment (Table 2 presents country definitions). Then, using the combinations generated in Step 1, employment information was applied to estimate the probability of having had a baby only among formally employed women. This step required linking fertility and employment data for each of the combinations estimated in Step 1. Hence, the probability of having a baby and working in the formal sector was estimated for each of the combinations.
Step 3 was to identify the target population Pop y (women of reproductive and legal working ages) through national population estimates (census data and population projections). The national population of women of reproductive age was then weighted (multiplied) by each of the values of α estimated in Step 2, expressed as α × Pop y .
Step 4 was to identify the weekly wages of women working in the formal sector (W). We estimated W for each of the women's subgroups (based on combinations of their personal characteristics) and operationalized through the weekly wage in United States dollars (US$). The value of W was then multiplied by the weighted popula-  week. An important assumption in this step is that maternity leave covers 100% of the salary, but this can be tailored to country's specific context ( Table 2). The weekly mean and median costs per woman were calculated by dividing cost per week by the estimated number of women expected to receive the maternity leave. In Step 5 we determined the number of weeks of maternity leave to be assessed (IC). We assessed four relevant cut-off points: (i) 12 weeks, which is the number of weeks covered by the formal sector maternity leave in Ghana and Mexico (up to September 2019); 40 (ii) 14 weeks, which is the minimum duration recommended by the ILO; 41 (iii) 18 weeks, which is the length of maternity leave coverage currently being discussed by key stakeholders in Ghana; and (iv) 26 weeks, which is consistent with the WHO recommendation of exclusive breastfeeding for the first 6 months of life. 4 We present estimates for these proposed durations, but the method can be applied for any number of weeks.
All costing calculations were estimated in US$ and PPP$ using 2018 as the reference year using Stata, version 15 (StataCorp, College Station, USA).

Assessing validity and affordability
To assess the validity of our estimates, we compared our values with those obtained from the administrative records of the Mexican Institute of Social Security. These records represent the real costs incurred for the current maternity leave of working mothers in the formal sector. We restricted the Mexican sample to women affiliated with the social security system, which covers 77.8% (111 838 of 143 797) of formally employed women. We then applied the costing method using the selected population and compared the mean costs obtained with those reported from the Institute's public registries, corresponding to a maternity leave of 12 weeks in 2014. 42 In addition, to assess the feasibility of extending maternity leave for women working in the formal sector, we accessed supplementary data for Mexico. We compared the estimated mean cost of one additional week per woman with the weekly cost per child of the social security system's day-care services and with the weekly cost of feeding an infant with formula milk, if the woman is not breastfeeding.

Results
The unweighted survey estimates of the total numbers of women in formal em-ployment in Brazil, Ghana and Mexico were 31 665 725 and 143 798, respectively in the relevant year. Table 3 presents the characteristics of these women and the estimated numbers and proportions who gave birth in the previous year. Table 4 summarizes the population of women who would receive maternity leave benefits. According to estimates from our model, the numbers vary due to differences between countries in the population, share of women in the labour force and proportion of women in formal employment. For example, we estimated that 640 742 women in Brazil, 33 869 in Ghana and 288 655 in Mexico would have been granted maternity leave annually.

Discussion
This study fills a research gap by developing a replicable method to estimate the annual costs of extending maternity leave for women employed in the formal economy. Our approach built upon and extended the application of an accepted and widely used World Bank costing method. 23 The analysis suggests that estimates from the five-step method were feasible in three different countries from two different regions (Latin America and sub-Saharan Africa) and different income levels (lower-middle and uppermiddle income). The replicability of the method is important, as it suggests that costing a maternity benefit for women employed in the formal economy is feasible using data commonly available across countries through existing national sociodemographic and employment surveys, as well as census data. In each country the data sources were different, but the variables for estimation were comparable. It is important to highlight that the accuracy of the costing method will depend on the quality of the survey data of each country and so it is relevant to perform calculations of data quality before embarking on cost estimates. If the data are of adequate quality, we expect that our costing method will facilitate evidence-informed policy decisions across countries to improve maternity protection benefits and potentially improve breastfeeding and other maternal, child and family health outcomes. Our method was validated by comparing our estimates with actual expenditures observed in Mexico. Similar validations could not be performed for the other two countries due to limitations of the available data. Investigators applying our method in other countries should make comparisons with observed expenditures, as we did in Mexico, to further validate the method in additional settings.
The current research has some limitations. First, despite our efforts to standardize the costing method, there were differences in the national-level surveys, such as different time periods of data collection and the way surveys were structured. We therefore used slightly different data sources in each country. However, nationally representative data were available to estimate the relevant parameters. Another limitation in the standardization was that the difference between countries in definitions of some variables (such as education) led to different categorizations across countries. While the specific categories for each group are not strictly comparable across the three countries, the method leads to estimates that are applicable and valid to each particular context.
Due to the scope of the costing method, we aimed to estimate aggregate national level costs. Every country will need to do further adaptations in using the costing method to the institutional nature of national maternity leave schemes (such as contributory or tax-funded) and this calls for future research in this area. Similarly, although our analyses did not compare women employed in the public and private sector, our method can easily be extended to conduct such comparative analyses. This analysis would require cutting part of the data to the sub-population of interest; hence it is important to understand how dropping part of the data would affect the statistical power of the sub-analyses. Our analysis estimates the cost of extending maternity leave at a country level based on observed salaries and based on the assumption that the opportunity cost of women is similar between sectors.
Finally, the analysis was based on countries from the Latin American and sub-Saharan Africa regions and needs to be tested in additional areas including Asia, Europe and North America. While the current analyses focused on costing the extension of maternity leave mandates for women employed in the formal sector, in many low-and middleincome countries women are more likely to work in the informal economy. It is important to also develop costing methods to provide maternity benefits to these women. 47 While maternity leave protection is a key policy to promote and support breastfeeding for working women, there are other fundamental areas that should also be addressed, such as workplace policies, child care and paternal involvement. Protecting and supporting breastfeeding working mothers requires an integral strategy of which maternity leave mandates are a fundamental part. Supportive labour market policies, such as maternity leave, are essential in high-, middle-and low-income countries if increased breastfeeding rates are to be achieved alongside the participation of women in the labour force.
Further economic evaluations are needed to estimate the cost savings of expanding the duration of maternity leave through its impact on breastfeeding and long-term health outcomes. These evaluations could help advocates to strengthen their country's political will for the extension of maternity leave legislation. ■ Funding: This work was supported by the Family Larsson-Rosenquist Foundation through a grant to Yale University (PI Rafael Pérez-Escamilla; grant number R14001).
Competing interests: None declared.

Costos de la licencia de maternidad para apoyar la lactancia materna en Brasil, Ghana y México
Objetivo Elaborar un método para evaluar el costo que supone ampliar la duración de la licencia de maternidad de las mujeres empleadas oficialmente a nivel nacional con el fin de aplicarlo en Brasil, Ghana y México.