Policy Research Working Paper 11184 The Global Gender Distortions Index (GGDI) Penny Goldberg Charles Gottlieb Somik Lall Meet Mehta Michael Peters Aishwarya Lakshmi Ratan Development Economics Development Policy Team August 2025 Policy Research Working Paper 11184 Abstract The extent to which women participate in the labor market force participation) and can be computed using data on varies greatly across the globe. If such differences reflect labor income and job types. The methodology also high- distortions that women face in accessing good jobs, they lights an important distinction between welfare-relevant can reduce economic activity through a misallocation of misallocation and the consequences on aggregate GDP if talent. This paper builds on Hsieh et al. (2019) to provide misallocation arises between market work and non-market a methodology to quantify these productivity consequences. activities. To showcase the versatility of the index, the anal- The index proposed, the “Global Gender Distortions Index ysis examines gender misallocation within countries over (GGDI)”, measures the losses in aggregate productivity that time, across countries over the development spectrum, and gender-based misallocation imposes. The index allows for across local labor markets within countries. The findings separate identification of labor demand distortions (e.g., indicate that misallocation is substantial and that demand discrimination in hiring for formal jobs) from labor supply distortions account for most of the productivity losses. distortions (e.g., frictions that discourage women’s labor This paper is a product of the Development Policy Team, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at penny.goldberg@yale.edu, charles.gottlieb@unige.ch, slall1@worldbank.org, meet.mehta@yale.edu, m.peters@yale.edu, and aishwarya.ratan@yale.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Global Gender Distortions Index (GGDI)* , Charles Gottlieb ‡ Penny Goldberg† , Somik Lall§ , , Michael Peters| Meet Mehta¶ , | and Aishwarya Lakshmi Ratan** * We thank many seminar participants for useful comments and suggestions, the Yale Economic Growth Center for supporting the Gender and Growth Gaps project through which this work was initiated, and the Development Policy and Finance team at the Gates foundation for financial support. Gottlieb gratefully acknowledges financial support from Structural Transformation and Economic Growth (STEG) LRG Grant 943. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. † Yale University, BREAD, CEPR, CESifo, NBER. penny.goldberg@yale.edu ‡ University of Geneva and Aix-Marseille School of Economics. charles.gottlieb@unige.ch § World Bank. slall1@worldbank.org ¶ Yale University. meet.mehta@yale.edu || Yale University, CEPR, NBER. m.peters@yale.edu ** Yale University. aishwarya.ratan@yale.edu 1. I NTRODUCTION Labor market outcomes vary greatly along gender lines. Across most countries of the world, women have lower participation rates and earnings, are overrepresented in unpaid, informal jobs, and spend longer hours on home production and household chores. While lots of progress has been made among developed economies in the last decades, we are still far from equal representation along gender lines. In this paper, we propose a methodology to measure the economic costs of such gender differences at the macroeconomic level. Building on the recent work by Hsieh et al. (2019), we show how labor income and job type data can be used to infer distortions that women face in the labor market and quantify their impact on aggregate productivity. We introduce the ”Global Gender Distortions Index (GGDI)” a model-based, scalar measure of gender-based misallocation. The GGDI answers a simple question ”How much higher would economic activity be if women were to face the same playing field as men?” Grounded in economic theory, the GGDI has a clear cardinal interpretation and allows for straightforward comparisons across countries and over time. This makes it a valuable tool for assessing the macroeconomic impact of policies promoting gender equality – or the lack thereof. A key conceptual benefit of our approach is that it allows us to aggregate and meaning- fully compare different dimensions of gender inequality. As highlighted above, gender differences are salient across a variety of measures such as labor force participation, in- come, and sorting across job types. Our index provides a natural way to combine these differences into a single scalar: the aggregate productivity consequences. Moreover, our methodology is suitable to both capture distortions on labor demand (e.g. firms discriminating against women and paying them less than their marginal product) and labor supply (e.g. the presence of institutions that discourage women to partake in full-time, market work). These distortions arguably coexist and interact in non-trivial ways, and our index captures both. Another advantage of our index is its simplicity. Using data on labor income and job type choices for both men and women, the GGDI can be easily computed. Moreover, the definitions of job types can be flexibly adjusted to suit different contexts. In de- veloped economies such as the US, Sweden, Canada or France, gender differences in access to high-skill occupations such as lawyers, doctors, and programmers might be particularly important. In developing economies, gender disparities are, for example, more pronounced in terms of job formality and payment. Our index is amenable to these contexts, providing researchers and policymakers with the flexibility to study whichever labor market dimension their application requires. As part of this paper, we also provide a full set of easy-to-use codes to compute the GGDI. 1 Finally, our analysis highlights an important distinction between welfare-relevant economic activity and measured GDP per capita. In the context of the study of gender gaps, home work plays an important role since large gender gaps in market work, are mirror images of large gender gaps in the provision of domestic services. While these activities generate economic output, they are not captured in measures of GDP according to existing national accounting rules.1 The effect of distortions on aggregate GDP can therefore be larger than their effect on welfare-relevant economic activity if they prevent women to participate in market or unpaid work. Our methodology offers a straightforward-way to quantify such discrepancies and we do as part of our cross-country application. For the vast majority of countries we find that the effects of distortions on GDP exceed the one as implied by the GGDI. The reason is that the type of distortions we estimate, reduce female labor force participation and keep women in the home sector. Removing distortions therefore raises women’s participation in market work and leads to a large increase in measured GDP. These features of our index, of course, come with a cost. As mentioned above, the GGDI is a model-based measure and, as such, relies on our specific modeling assumptions and two structural parameters: the labor supply elasticity and the elasticity of occupational demand. We therefore view our index as a useful complement to existing, atheoretical measures of gender equality such as the World Bank “Women, Business and Law Index” (World Bank, 2024), that provides a de-jure measure of gender-based restrictions and regulations. To showcase the applicability of our methodology in practice, we consider three ap- plications. First, we compute the GGDI for selected countries over time. This exercise answers the question whether gender-based misallocation has shrunk over time and quantifies its productivity consequences. We find that misallocation indeed dropped markedly in the US between 1970 and 2020. At the same time, we also highlight that a fall in misallocation is by no means an automatic corollary of economic develop- ment. For example, despite its remarkable growth experience, we find no changes in gender-misallocation in India. Second, we perform a cross-country exercise using microdata for 51 countries across the development spectrum from the Harmonized World Labor Force Survey (HWLFS). We find that gender-based misallocation is large and quantitatively important: in some countries productivity could be increased by 15%-20% if labor market distortions for women were abolished. We also find a systematic negative correlation between economic development (as measured by GDP per capita) and the GGDI, indicating that 1 Some but only few countries report GDP figures based on the extended system of national accounts, which includes the valuation of unpaid domestic work. For instance, the Bureau of Economic Analysis provides data on Household Production Satellite Account, see Bridgman et al. (2022). 2 gender-based misallocation is more prevalent in developing countries. At the same time, there is substantial variation in the GGDI across countries at similar levels of development, indicating the importance of differences in culture and institutions as predictors of gender-misallocation. We also document that, reassuringly, the cross- country variation in the GGDI is highly correlated with the World Bank’s “Women, Business and Law Index” (WBL) suggesting that our model-based measure of de-facto misallocation is broadly consistent with the de-jure measure of the WBL. Finally, we conduct a within-country analysis by applying our methodology to the experience of Indian states in 2018. As with the cross-country data, we document large potential gains from reducing gender-based misallocation, which are negatively correlated with state-level GDP per capita. While the poorest states in India could increase GDP by up to 15%, we estimate that richer states only “lose” 5% of GDP through gender-based misallocation. In terms of the relative importance between demand and supply distortions, our results suggest a larger role for distortions operating on the demand side. Interestingly, we find important complementarities: dismantling only demand distortions generates welfare gains that are quite close to the full gains when reducing both demand and supply distortions. The intuition for why this is the case is reminiscent of the ”theory of the second best”. Introducing a distortion in an efficient economy, does not have any welfare consequence to first order, because all marginal products are equalized in the initial allocation. Vice versa, starting from an allocation where supply distortions already generate a wedge between marginal products, a reallocation of resources induced by falling demand distortions can have large welfare consequences. We think that this result is important for policy-makers: to the extent that demand distortions are easier amenable through policies (e.g., because supply distortions reflect social norms), policies aimed at demand distortions might be able to capture a large share of the welfare benefits of reduced gender misallocation. Related Literature Our paper builds on a large literature documenting the pervasive- ness of gender gaps, in particular in developing countries (Jayachandran, 2015; Fletcher et al., 2019; Jayachandran, 2021; Duflo, 2012; Klasen, 2019; Agte et al., 2024; Gottlieb et al., 2024). A particular focus of that literature has been the participation gap and how female labor force participation varies with the course of economic development. Goldin (1994) uses cross-country data to highlight a U-shaped pattern in female la- bor force participation. Relatedly, Ngai et al. (2024) documents a U-shaped trend in women’s hours worked in the US throughout the 20th century. A recent literature highlights that such differences in labor market outcomes are the result of gender-specific frictions. Hsieh et al. (2019) document that such forms of misallocation are important and that improvements in allocative efficiency account 3 for a sizable share of US economic growth in the post-war period. Chiplunkar and Kleineberg (2022) use cross-country data and show that frictions are important to explain gender gaps even when other determinants of gender-specific labor demand, such as structural change, are taken into account. For the case of India, Chiplunkar and Goldberg (2021) focus on female entrepreneurship and emphasize - consistent with the results of this paper - the importance of simultaneously addressing supply and demand-side distortions. An alternative measure of the aggregate implications of gender gaps is proposed in Pennings (2022). Relative to us, Pennings (2022) focuses on the gap between male and female employment as a share of total employment, as opposed to the (mis)allocation of talent across different activities. The paper by Hsieh et al. (2019) is particularly related because we closely build on their methodology. In essence, our model is a simplified version of theirs in that we abstract from human capital accumulation. In doing so, our model is easier to implement and requires less data than they do. We hope that this increases the applicability of our method for researchers and policy-makers in the future. A complementary literature highlights that gender gaps can also be the result of struc- tural change (Ngai and Petrongolo, 2017; Rendall, 2018; Ngai et al., 2024), evolving social norms (Fern´ andez, 2013; Fogli and Veldkamp, 2011; Jayachandran, 2015; Olivetti et al., 2024), technological change in household appliances (Greenwood et al., 2005), or the organization of production (Bandiera et al., 2022). Gottlieb et al. (2024) combine time-use data from 50 countries with a macroeconomic model of household time use and show that gender norms, more than home technology or income, drive the division of market and domestic work. For our analysis we take technologies and institutions as given when conducting our counterfactuals. In our applications, we rely heavily on the cross-country data from the Harmonized World Labor Force Survey (HWLFS). This micro dataset has the benefit that it provides harmonized data on worker characteristics, employment and wages for a broad set of countries across the development spectrum. The latter information is particularly important for us because information on labor income is crucial to achieve identification in our model. These data, together with harmonized data on time-use, have also been used in Gottlieb et al. (2024), who document cross-country patterns of the gender division of market, domestic and care work. Organization of the paper The remainder of the paper is organized as follows. Section 2 introduces our data. Section 3 empirically documents the pervasiveness of gender gaps around the world. Section 4 contains the theory and develops our index, the GGDI. There we also outline the distinction between welfare-relevant productivity and measured GDP. Section 5 discusses the identification of distortions, both on the supply 4 and the demand side. Section 6 contains our quantitative results. In Section 7, we relate our results to the WBL Index. Section 8 considers the robustness of our results with respect to our parametric assumptions. Section 9 concludes. An appendix contains details of the data and technical results. There we also include a full replication package, including MATLAB codes for other researchers to use. 2. D ATA , M EASUREMENT AND E VIDENCE For this paper, we construct two datasets: a country level dataset with 51 countries and a Indian States dataset in 2018. We use these data to compute the GGDI index for many countries, its evolution over time, and across Indian states. 2.1 Data sources Our cross-country dataset draws from the Harmonized World Labor Force Survey (HWLFS), a large scale micro-dataset of labor force and household surveys, that har- monizes thousands of surveys from any countries across the globe.2 For the purpose of this paper, we use surveys of the HWLFS that are (i) nationally representative, (ii) provide information on wages and hours worked, and (iii) on the industry of the main job for all workers.3 See Appendix Section B-1.1 for more details. For our application on Indian states, we rely on the publicly available data from Periodic Labour Force Survey (PLFS). The PLFS is a nationally representative household survey and covers about 100,000 households in India—see Appendix Section B-1.1 for more details. Finally, to compute the GGDI along the development spectrum, we use data in GDP per capita at the country-level (measured in PPP) from the Penn World Tables (Feenstra et al., 2015). 2.2 Measurement For all the data sources we mention, we use the micro data to measure a set of labor market and education outcomes for men and women. Also, we use the wage data to measure gender gaps in labor income. Throughout, we focus on individuals that are between 25 and 60 years old. Job type. In our theory, individuals’ labor market decisions are modeled as discrete choices: each individual chooses their preferred job type. Given our focus on develop- ing countries, we focus on four mutually exclusive activities: wage jobs, unpaid work, 2 It encompasses previously harmonized cross-country datasets such as EU-SILC (Survey Income and Living Conditions) dataset or the IPUMS amongst others. 3 Some surveys provide industry and occupation codes for wage workers only. 5 self-employment and not employed.4 Following the International Conference for Labor Statisticians (ICLS), self-employment includes unpaid workers, own-account workers and employers. Unpaid workers work for the family farm or business and are unpaid because they don’t derive income or profits from their activity.5 They are employed since their work contributes to the production of goods and market services that fall within the production boundary as defined by the System of National Accounts (SNA).6 We choose to make a distinction between self-employment and unpaid work, recognizing the significant gender impli- cations of each. Therefore, when referring to self-employed workers, we specifically mean own-account workers and employers. We use data on the main job to assign workers to these three job types. We refer to individuals who don’t have a job as not-employed, and consider that they contribute to home production by providing (unpaid) care and domestic services. Education. To measure educational attainment, we use data on years of schooling and highest completed degree. If information on years of schooling is not available, we impute years of education based on the completed degree using country level information on the education system. For example, if an individual in the US reports high-school degree as the highest completed degree, we consider that they have 12 years of schooling. Labor income. To measure the return to different activities, we use data on labor earnings. We construct weekly income by combining information on hours worked and reported labor income. We prioritize data on actual hours worked; when unavailable, we use usual hours worked as a substitute. Reported earnings for wage workers are adjusted based on their reporting period and combined with hours data to compute weekly income. While this measurement is conceptually straightforward for individuals with wage employment, we also need to assign economic returns to the other three activities. Ideally, we would directly measure the monetary value of providing labor as an unpaid family worker or in home-production. Because this information is, by construction, not observed, we impute the labor income for individuals that are not working for a wage based on their individual and job characteristics. In doing so, we follow the previous work of Young (1995); Gollin (2002) and Lagakos (2016), who uses similar imputation 4 In principle, our theory is amenable to also allow for part-time work whereby individuals spend part of their time in wage work and part of their time in subsistence agriculture. If one could empirically measure the employment share and implied wage income in such activities, our theory would extend in a straight-forward way to this case. 5 See paragraph 9b of the 13th ICLS (ILO, 1982) for the official definition of unpaid workers. 6The economic value of unpaid work is included in GDP through imputation methods, typically by using market equivalent prices or wages (see chapters 6 and 7 of United Nations (2008)). 6 methods to estimate labor earnings in economies with large self-employment shares. More specifically, as we outline in more detail in Appendix Section B-1.2, we impute labor income for the self-employed, unpaid, and non-employed population, by running regressions of overall earnings of wage workers on their gender, age, marital status, education, and sector of employment.7 We then use the estimated parameters to predict counterfactual earnings for non-wage workers.8 Having to rely on such imputed earnings is, of course, not particularly attractive. However, given the ubiquity of self-employment in developing countries, one cannot proceed without estimates of the economic returns to these activities. By associating earnings of self-employed or unpaid workers with the typical earnings an observa- tionally equivalent individual receives in the market sector, we might understate the importance of misallocation, if frictions drive a wedge between market earnings and the opportunity costs of non-wage work. Improving the measurement of such opportunity costs is, in our opinion, an important avenue of future research. 2.3 Datasets We measure the above mentioned outcomes for 581 surveys. We combine these data into three datasets, a cross-country dataset, a growth experience dataset and a state-level dataset for India in 2018. See Table B-I in the Appendix for a full list of surveys. Cross-country dataset. The dataset provides data for 51 countries at all stages of development. For each country, we use data from the year 2015 or the nearest available year. The countries in our sample cover 69 percent of the World Population in 2015. The poorest country in our sample is Niger in 2005 with an GDP per capita level of USD 624 . The richest country in our sample is Luxembourg in 2005 with an GDP per capita level of USD 97973 .9 Growth experience dataset. In addition, we used our dataset to study the evolution of gender gaps in market work for a broad set of countries. This dataset is an unbalanced panel at the country level, and the country with the longest time series in our sample is the USA (1965-2022). For each country in this dataset, we have on average 12 years of data. 7 Forindividuals that are not employed, we assign agriculture as their sector of employment. In doing so, we implicitly assume that agricultural work acts as the closest measure for the opportunity cost of being out of the labor force. 8We rely on this approach also for the self-employed, because self-employment earnings are recorded inconsistently across countries, making cross-country harmonization particularly challenging. Even when such data are available, additional measurement issues arise in both high- and low-income settings (Bhandari et al., 2024; Gollin, 2002). Given these limitations, we rely on wage data, which provide more comparable information across countries. 9We use the GDP per capita numbers provided by the Penn World Tables (Feenstra et al., 2015). 7 Indian states dataset. We apply the same approach to measure these outcomes for 27 states in India using the Periodic Labour Force Survey (PLFS) of 2018-19. 3. G LOBAL G ENDER G APS We use these datasets to document gender gaps in employment, labor income and educational attainment, both across countries at different income levels, as well as over time. 3.1 Gender Gaps across Countries We first describe gender gaps across countries around the year 2015 using the ”Cross- country dataset”. We focus on differences in the type of jobs that men and women perform, differences in education, and differences in wage income. Gender Work Gaps. We begin by documenting cross-country patterns in the nature of work conducted by men and women. To highlight differences across the development path, we aggregate the 51 countries in our dataset into three income groups: low (below $10,000 of GDPpc), middle (between $10,000 and $30,000), and high (above $30,000). Table B-IV in the Appendix contains a list of all countries in the respective groups. Table 1 contains the results. The data reveals significant gender disparities in employment patterns, particularly in different types of work. Across all countries, wage work is the most common form of employment, but men have consistently higher participation rates in wage work than women. In low-income countries, only 16% of women work for a wage, compared to 37% of men. This gap is much narrower in high-income countries, where 61% of women and 69% of men work for a wage. This pattern suggests that high country income levels are associated with more opportunities for women to enter formal wage employment. It is well known that in low-income countries, self-employment is the most prevalent form of work. Table 1 shows that this is particularly true for men, where 42% of them are self-employed in contrast to 28% of women. Comparing self-employment rates of men and women across country income groups reveals that the gender gap in self-employment rates is larger in high-income (2.2 = 0.13/0.06) than in low-income countries (1.5=0.42/0.28). Additionally, these data shows a large gender gap in unpaid work, with women being more likely to pursue this type of work across all country income groups. This is particularly true in low-income countries, where 15% of women do unpaid work, compared to only 5% of men. Arguably, this reflects barriers that limit women’s access to income-generating opportunities, even when they are employed. Table 1 further shows that the labor force participation varies greatly by gender and 8 Country Income Group Activity Low Middle High All Wage work Men 0.37 0.56 0.69 0.53 Women 0.16 0.41 0.61 0.38 Self-employed Men 0.42 0.23 0.13 0.27 Women 0.28 0.12 0.06 0.16 Unpaid work Men 0.05 0.03 0.00 0.03 Women 0.15 0.06 0.01 0.08 Not employed Men 0.16 0.19 0.18 0.17 Women 0.41 0.41 0.32 0.38 Number of countries 18 19 14 51 TABLE 1: Activities of Men and Women by country income group. Notes. This table reports the average share of men and women that are 25 to 60 and pursue either of the four activities. The income brackets we use to classify countries into income groups are [0, 10,000), [10,000, 30,000), [30,000, ∞), which we refer to as “Low income”, “Middle income”, and “High income”. We assign to each country an income group based on the GDP per capita (PPP) reported by PWT (Feenstra et al., 2015) for that particular year. income level. Across all countries, the share of women not working (38%) is twice higher than that of men (18%) (column 4). This gender gap varies strongly across country income levels. While in low-income countries, the share of women not working is 2.6 (= 0.41/0.16) times higher than for men, the gender gap is 1.9 (=0.32/0.17) in high- income countries. This suggests that while high-income countries are associated with a smaller gender gap in labor force participation, gender gaps persist in high-income countries. Gender Education Gaps. A key determinant of both labor supply and demand is worker skill. The observed differences in the gender division of work and their allocation between work types may reflect gender gaps in educational attainment. A first-order measure to study the role of skills in explaining gender work gaps is the gender gap in years of schooling in each type of work. We measure the gender education gap, the ratio of women’s years of schooling relative to men, for individuals in each activity. We report the average of these ratios for each country income group in Table 2. The data suggest that the gender education gap across work types varies considerably in low-income countries. While women that work for a wage have similar education levels than men, women working in self-employment or as unpaid worker have a level of schooling that is 20 p.p. and 43 p.p lower than men, respectively. In middle-income countries, the gender education gaps are smaller. Men and women who are self-employment have on average the same level of education, while women working for a wage have 9 p.p. more years of schooling. In high-income 9 countries, the gender education gap is larger than one and of similar magnitude across all activities, meaning that, across the board, women have higher levels of schooling than men. The fact that in high-income countries, fewer women work although they have higher levels of schooling is suggestive evidence for barriers to women’s market work. Country Income Group Activity Low Middle High All Wage work 1.00 1.09 1.02 1.04 Self-employed 0.80 1.00 1.04 0.94 Unpaid work 0.57 0.72 0.61 Not employed 0.73 0.96 1.01 0.89 Number of countries 18 19 14 51 TABLE 2: G ENDER E DUCATION G APS BY COUNTRY INCOME GROUP. Notes. This table reports the average ratio of women and men’s years of education for each country income group, and across all countries. Note that the in high-income countries, we don’t report the gender education gap, since there are very few (if any) unpaid workers in those countries. The income brackets we use to classify countries into income groups are [0, 10,000), [10,000, 30,000), [30,000, ∞), which we refer to as “Low income”, “Middle income”, and “High income”. We assign to each country an income group based on the GDP per capita (PPP) reported by PWT (Feenstra et al., 2015) for that particular year. Gender Income Gaps. Finally, we use our micro data on wage income to measure gender income gaps in each activity. Recall that the gender income gap is the ratio of the weekly labor income of women to that of men working for a wage. Weekly income being the product of the hourly wage and weekly hours worked, a low gender income gap can be due to a gender wage gap or a gender gap in weekly hours worked. Figure B-1 in the Appendix displays gender income gaps for all countries in our sample. We find that the gender income gap does not vary much across countries. In low-income countries, women who work for a wage earn 35 p.p. less than men, while it amounts to 20 p.p. and 30 p.p. in middle- and high-income countries, respectively. 3.2 Gender Gaps over time In Section 3.1 we analyzed the patterns of gender gaps across countries. We now examine gender gaps over time. To do so, we regress each gender gap on a time trend and country fixed effects. Specifically, let yct denote a particular gender gap, we consider a regression of the form yct = δc + β × t + uct . (1) 10 Note that we run (1) separately for each gender gap y. The coefficient β, reported in Table 3, therefore, reflects the average annual change in gender gaps within countries in our sample. In the first column of Table 3, we focus gender differences in each activity. Our data suggests that over time, the share of women working for a wage or as self-employed has increased relative to men. On average, the gender gap in self-employment and wage work increased every year by 0.4 and 0.34 p.p relative to men. Since men are overrepresented in wage work and self-employment across countries, these increases for women imply a narrowing of gender gaps in these activities. Taken together these numbers suggest that over the course of a decade, the share of women worker for a wage or in self-employment increases by 3.5 and 4.1 p.p. relative to men, respectively. With regard to unpaid work and not working (row 3 and 4 in Table 3), the estimated co- efficients are negative, suggesting that the share of women in these activities decreased relative to men. Since women are historically overrepresented in these activities, the gender gap in unpaid work and non-work have declined. The coefficient of the gender gap in not-employed is precisely estimated, and suggests that the ratio of non-working women relative to men has gone down by 3.1 percentage point per year relative to men. Note that since the gender gaps in non-employment and unpaid work are above one changes in the ratio cannot be directly interpreted as changes in relative group shares. Gender Gap Activity Education Income Wage work 0.0034 0.0021 0.0052 (0.0011) (0.0014) (0.0010) Self-employed 0.0040 0.0041 (0.0008) (0.0011) Unpaid work -0.1898 0.0029 (0.1070) (0.0013) Not employed -0.0310 0.0021 (0.0092) (0.0011) TABLE 3: G ENDER G APS OVER TIME . Notes. This table presents coefficients from linear regressions of gender gaps on year and country fixed effects. Standard errors, reported in parentheses, are clustered at the country level. For example, the coefficient in the top-left cell indicates that, on average, the gender gap increased by 0.003 per year. A positive coefficient suggests women’s increasing relative participation in that activity. However, it does not necessarily imply convergence toward gender parity, as this depends on whether women were initially underrepresented. Each regression is based on 530 observations, corresponding to an average of 10 observations per country. Column 2 in Table 3 reports the estimated changes in the gender education gap by activity. Across all activities, the gender education gap has declined over time, as 11 women’s years of schooling have increased relative to men’s. Since historically, women had fewer years of schooling, the gender gap in education narrowed over time. This is true for all women independently of their activity. Our findings suggest that the closure of the gender education gap has been particularly pronounced in the self-employed population, where on average women’s years of schooling increased by 0.41 percentage point every year relative to men. Finally, as seen in column 3 of Table 3, we find that the gender income gap has shrunk over time. The coefficient 0.0052 indicates that women’s labor income has increased by 0.52 percentage points per year relative to men. Based on this trend, a country with an initial gender income gap of 0.79, the average gap we measure, would reach income parity in approximately 48 years. 4. T HEORY: T HE G LOBAL G ENDER D ISTORTIONS I NDEX Motivated by these gendered differences in labor market outcomes, we now develop our methodology to quantify their consequences for aggregate productivity. The main result of this section is the ”Global Gender Distortions Index (GGDI)”, an easy-to- compute, scalar measure of the gender-based misallocation. To construct the GGDI, we build heavily on the work by Hsieh et al. (2019). We consider a simplified version of their framework where we have two groups, men and women, and four types of jobs: wage work, self-employment, unpaid work and home production. Implicitly, we assume that not-employed individuals do home production. These choices, in particular the distinction between wage work on the one hand and unpaid work and self-employment on the other hand reflect that reality of most developing countries, where such informal work arrangements are the norm rather than the exception. An important implication, which is absent in Hsieh et al. (2019), is that our model features a distinction between market-based GDP and overall welfare-relevant economic activity; non-market home-production is not part of GDP, but generates economic output. In our model, men and women differ in three dimensions. First, women might face a demand distortion, which we model as an exogenous tax wedge on their labor earn- ings. Second, women might face a supply distortion, whereby choosing a particular job reduces their utility. We refer to such gender-specific utility differences as ”distortions”, because our premise is that intrinsic preferences do not vary by gender, but that the utility derived from a specific employment choice may depend on existing norms and institutions that are beyond the control of individuals. Third, men and women can differ in their job-specific human capital, which we, for simplicity, take as exogenous. Hence, relative to Hsieh et al. (2019) we abstract from dynamic choices of human capital and only consider a static model of labor supply. 12 4.1 Labor Supply: Distortions vs Skills We model the utility of an individual i of group g = m, w, i.e. man or woman, choosing a job j as i i ln Ujg = ln Cjg + ln z jg . (2) Here Cjg denotes consumption of individual i and z jg the utility of working in job j. Note that the utility consequences of working in a particular job are identical across individuals of a particular group. These utility consequences can be interpreted as job-specific amenities—that is, non-monetary attributes such as job-flexibility or the work environment. By contrast, consumption is individual-specific, because it is linked directly to total earnings, which in turn is given by individual’s human capital, the equilibrium wage rate (per efficiency unit), and the prevailing demand distortion by the budget constraint i Cjg = (1 − τjg ) w j h jg ϵij . (3) Here w j is the prevailing wage rate in job type j, h jg denotes the job-specific human cap- ital of an individual of group g, and τjg parametrizes the demand distortion. Finally, ϵij is an individual-specific, idiosyncratic productivity draw for job j that allows individuals to differ in their comparative and absolute advantage in different job types. Substituting (3) into (2), an individual’s utility is given by i Ujg = w jg ϵij where w jg ≡ w j (1 − τjg )h jg z jg . (4) Hence, w jg summarizes the systematic attractiveness of a job j for group g. It depends on skill prices w j , labor demand distortions τjg , the average human capital endowment h jg , and labor supply distortions z jg . Note that skill prices w j per efficiency unit of labor are common across groups g and determined in equilibrium. Individuals choose a single job j to maximize Ujg . For tractability, we assume indepen- echet-distributed ϵij :10 dent Fr´ −(ϵij )−θ F (ϵij ) = e . Standard arguments then allow us derive our first result: Proposition 1. Let p jg denote the fraction of people from group g who choose job j and let 10The fact that this distribution has the same mean for all groups g and jobs j is without loss of generality, because we allow for unrestricted heterogeneity in group-job specific human capital h jg . 13 wage jg denote the geometric average of earnings in job j by group g. Then wθ jg p jg = (5) ∑ j wθ jg 1 θ wage jg = Γθ ∑ wθ jg z− 1 jg . (6) j where w jg is given in (4) and Γθ ≡ Γ 1 − 1 θ is the gamma function. Equations (5) and (6) are at the heart of our identification strategy. Equation (5) high- lights that job-specific employment shares, p jg , reflect a job’s relative attractiveness w jg . This attractiveness in turn is fully determined by market prices and human capital en- dowments on the one hand and distortions (either from the supply or the demand side) on the other hand. In this environment, if a particular group g has a high employment share in job type j, it could be that (i) their human capital h jg is large, (ii) skill prices w j are high, (iii) distortions τjg are low, or (iv) the utility of working in this job, z jg , is high. Using data on individuals’ job choices alone is therefore not sufficient to disentangle these different objects. Equation (6) highlights the information that is contained in group g’s earnings and consumption. First of all, for a given group g, the only variation of average earnings across jobs is due to differences in supply distortions z jg : The lower z jg , the higher the average wage in job j, because earnings play the role of a compensating differential. By contrast, if z jg was equalized across jobs, average earnings would also be equalized across jobs, i.e. neither differences in human capital, nor differences in skill prices or demand distortions, affect average earnings across jobs. This stark result, which is due to the selection of individuals into different jobs, is a particular property of our assumptions of ϵij to be Fr´ echet distributed. However, the underlying intuition is more general: if there was a large demand distortion in job j for group g or if their systematic human capital h jg was low, less people of group g would work in that job. As a consequence, the selection of individuals into that job would improve. This form of selection increases the average talent in job j and decreases the average talent in all other jobs. With the Fr´ echet distribution, these effects exactly cancel out, leaving overall earnings unaffected. The variation in earnings across different jobs alone is thus a poor measure of distortions that individuals face. It is only in conjunction with information on “quantities” (that is the job-specific employment shares by by group g) that the presence and the size of demand distortions can be identified. To see this more directly, use equation (6) to substitute wage jg for the labor supply 14 distortions z jg in w jg (see equation (4)). We can then express the share of women in job j relative to men as θ θ −θ p jw 1 − τjw h jw wage jw = × × . (7) p jm 1 − τjm h jm wage jm Equation (7) highlights that relative employment shares in different jobs are driven by three considerations: women can be underrepresented in a particular activity j if (i) they face a higher labor demand distortion τjw > τjm , (ii) they have a lower human capital endowment h jw < h jm , and (iii) their average earnings are relatively high (wage jw > wage jm ). This last effect of the wage operates through the labor supply distortion as high earnings reflect a compensating differential. Under our assumptions on labor supply, we can also derive a closed form expression for the aggregate supply of efficiency units in activity j, Hj . In particular, Hj is given by θ −1 Hj = ∑ L g p jgθ h jg Γθ , (8) g θ −1 where L g denotes the aggregate population of group g. The term p jgθ accounts for selection: while overall human capital supply is increasing in p jg , the elasticity is θ − 1 θ < 1, reflecting the fact that average efficiency declines as more people sort into a particular job type. Finally, average human h jg is, of course, a determinant of overall labor supply because it governs the average amount of efficiency units of members of group g in activity j. Equation (8) fully determines the labor supply side. In particular, the expression for p jg in (5) shows that Hj can be computed directly as a function of {w jg }. And given that everything in w jg except for the skill prices w j are exogenous, (8) determines labor supply as a function of {w j }. 4.2 Labor Demand: Technology To close the model in general equilibrium, we need to specify overall labor demand. To do so, we follow again Hsieh et al. (2019) and assume a representative “firm” that produces final output according to σ σ −1 σ −1 Y= ∑( A j Hj ) σ . (9) j 15 Hence, overall economic output is a CES aggregator of the output produced in different activities and such activities differ by their total factor productivity, A j .11 The reason we put “firm” in quotes is that the representation in (9) is, of course, an abstraction. As highlighted above, in our context individuals’ labor supply takes place both in the market and in informal activities outside of formal market arrangements. This distinction becomes important when thinking about the relationship between welfare-relevant economic output Y in (9) and measured GDP, which only pertains to activities tied to the production of goods and services exchanged on the market. A richer model would explicitly model the distinction between goods and market services (produced according to (9)) and home services, such as care and domestic work, which enter utility directly and, by construction, are not exchanged on the market. For simplicity, in this paper, we assumed that overall labor demand for different activities stems from the representation in (9). 4.3 Equilibrium A competitive equilibrium consists of a sequence of occupational choices, total efficiency units of labor in each group Hjg , final output Y , and a wage per efficiency unit w j in each activity such that 1. Each individual’s job choice maximizes utility taking as given skill prices w j , job-specific human capital h jg , demand distortions τjg , and supply distortions z jg , 2. The set of skill prices w j clear each occupational labor market, 3. Total output is given by the production function in equation (9). 4.4 The Global Gender Distortion Index (GGDI) Given this environment, we can now formally define the Global Gender Distortion Index (GGDI). From above we can compute overall economic output as a function of supply and demand distortions (for all activities j) in each country c and year t: D Yct ≡ Yct ( τjgct , z jgct j ), where Yct ( τjgct , z jgct j ) denotes equilibrium output in the presence of distortions, hence the superscript D. By contrast, economic output in the absence of distortions is given by E Yct ≡ Yct ( τjwct = τjmct = 0, z jwct = z jmct = 1 j ), 11 Because the aggregator Y in (9) has constant returns to scale, differences in population size (across countries) or population growth (over time) do not affect any allocations and keep income per capita unchanged. 16 where we use the mnemonic E to indicate that it is the equilibrium output under gender equality. Hence, gender equality requires that women do not face any demand distortions relative to men τjwct = τjmct = 0 and that the utility of men and women to enter a particular activity is equalized z jwct = z jmct = 1 . The Global Gender Distortion Index is then given by D E GGDIct ≡ ln Yct /Yct . (10) In words, our index is given by the percentage loss in overall output due to gender-based distortions: the higher the GGDI , the higher the extent of gender-based talent misallo- cation. Because our model is efficient in the absence of distortions, Yct E ≥ Y D , which ct implies that GGDIct < 0. 4.5 GDP versus Welfare-Relevant Output The GGDI captures changes in total economic activity due to gender-based distortions, including home production. Since home production, i.e. the production of domestic services, such as household chores and care work for household members, falls outside the production boundary defined by the System of National Accounts (SNA 2008), the GGDI does not amount to changes in GDP. By incorporating home production, the GGDI also accounts for how supply and demand distortions influence the economy at large. In doing so, it provides a welfare-relevant measure of gender-based misallocation. While total economic activity is, by construction, higher than measured GDP, the relationship between the GGDI and GDP changes due to gender-based distortions is unclear. It will depend on the relative magnitudes of supply and demand gender distortions across different activities. If, for example, women only faced demand distortions for wage jobs, a fall in distortions would both raise overall output (and hence the GGDI) and market GDP because more women would join the market sector. By contrast, if women were to only face distortions to provide home work, falling distortions would reduce misallocation, increase the GGDI, but market GDP could, in principle, contract if women leave the market sector and reallocation toward the home-sector. To highlight the quantitative importance of the differences between the GGDI and induced changes in GDP, we will compute both in our model. Letting w j denote the equilibrium skill price in activity j, GDP is given by the implicit total factor payments in all activities of employed workers. In the context of our model, wage work, self- employment, and unpaid work are labor inputs to measured GDP. Unpaid work, despite the fact that it does not entail a monetary compensation, should be part of the GDP, while home production should not. In practice, efforts are taken to adjust 17 official GDP measurement for the existence of unpaid work.12 By contrast, while home production, i.e. domestic services, generates economic value its output is not captured by GDP. Given a system of distortions {τjgct , z jgct }, GDP can therefore be computed as GDP Yct ({τjgct , z jgct }) ≡ ∑ ∑ D w j Hjg = Yct − ∑ wh Hhg , (11) j∈W ,SE,U g∈w,m g ∈ w,m where the first summation sums over wage (W ), self-employment (SE), and unpaid (U ) activities, and the second equality highlights that measured GDP is equal to overall, economic output minus the aggregate output of the home sector. Given this definition of GDP, in the same vain like we compute the GGDI, we compute the change in GDP if distortions were abolished as GDP ( { τ GDP Yct jgct , z jgct } ) ∆ct ≡ ln GDP ( { τ . (12) Yct jgct = 0, z jgct = 1} ) In words, ∆ct GDP reflects the percentage change in aggregate GDP that would occur if all distortions, including the ones for the home sector, were to be eliminated. 5. I DENTIFICATION : M EASURING M ISALLOCATION To compute the GGDI in (10) and ∆GDP in (12), we need to (i) measure the prevailing demand and supply distortions, (ii) compute Yct D and Y GDP , and then (iii) compute ct counterfactual output Yct E and GDP if these distortions were absent. To do so, one also needs to identify the job-specific demand shifters A j in the aggregate production function (9) and the two structural parameters, the elasticity of substitution σ and the dispersion of idiosyncratic skills θ . In this section, we describe these steps in detail. In Section 5.1 we focus on the estimation of demand distortions τjg and supply distortions z jg taking as given the structural parameters σ and θ . In Section 5.2 we turn to estimation of A j and the parameters σ and θ . Because the estimation of distortions and demand-shifters is done at the country-year level, for notational convenience we suppress the subscripts for country (c) and year (t). 12 Subsistence agriculture and unpaid family labor contributing to market output more broadly fall within the production boundary as defined by the System of National Accounts. The economic value of these activities is included in GDP through imputation methods, typically by using market equivalent prices or wages (see chapters 6 and 7 of United Nations (2008)). 18 5.1 Measuring Labor Market Distortions To measure the extent of misallocation we utilize our empirical measures of job-specific employment shares p jg , average earnings wage jg , and average human capital h jg . For the latter we follow our approach in Section 2 and take human capital as reflecting educational attainment. To translate educational attainment to human capital, we follow the usual approach of Mincerian returns, where we assign an annual return to schooling of 8% - see Appendix Section B-2 for details. Furthermore, we assume that the parameters θ and σ are known. Given these objects, we can infer supply and demand distortions directly from the data. Demand Distortions τjg To identify demand distortions, note first that all allocations only depend on the relative distortions between men and women. Hence, without loss of generality, we take men’s labor market choices to not be affected by demand distortions, that is τjm ≡ 0 for all j. The demand distortions for women, τjw can then be directly recovered from equa- tion (7): −1/θ −1 1 p jw h jw wage jw = × × . (13) 1 − τjw p jm h jm wage jm Holding relative earnings and human capital fixed, a lower relative employment share of women in activity j is indicative of higher distortions. Intuitively, a low employment share means that the few women that choose job j are very favorably selected. To rationalize this positive selection with a constant level relative earnings requires a large demand distortion. A similar intuition holds true for relative human capital: holding employment shares and earnings constant, a higher relative human capital goes hand in hand with higher distortions. Equation (13) is the first key equation for our measurement. As explained in detail in Section 2, our microdata from the HWLFS allows us to consistently measure em- ployment shares p jg and human capital attainment h jg . Furthermore, our imputation procedure for hourly wages also allows to compute average earnings wage jg for all activities j and groups g. Given an estimate of θ , these data directly imply a level of τjw from (13). While, in principle, we could use (13) for all demand distortions τjw , we do not use it for the home sector. The reason is that we do not feel comfortable in leaning too heavily on the relative home-sectors earnings for men and women, as these are both imputed using the observed earnings in agriculture. We therefore assume that women also do 19 not face demand distortions in the home-sector, that is τhw = 0. To ensure that our data is consistent with this assumption, we thus do not rely on mea- sured women’s home-sector earnings but require it to be consistent with equation (13): −1/θ phw hhw wagehw = × × wagehm . (14) phm hhm Given the imputed earnings for men in the home-sector and the observed employment shares and human capital, we compute wagehw according to (14). Supply Distortions z jg To identify z jg , we use the expression for relative earnings from equation (6). Note first that we can normalize the utility of one activity for each group because the expression for p jg and wage jg in Proposition 1 are homogeneous of degree zero: multiplying all z jg by a constant for a given group g keeps p jg and wage jg unchanged. We therefore impose the normalization zhw = zhm = 1. Hence, for both groups, z jg denotes group g’s utility in activity j relative to working at home. Equation (6) therefore implies that −1 wage jg z jg = (15) wagehg The higher average earnings in job j relative to home production, the lower the utility of choosing that activity has to be. This again highlights that earnings-premia play the role of compensating differentials for supply distortions. If, for example, traditional gender roles impose a high disutility for men to provide home-produced services, such as cooking, cleaning, or childcare, their utility in all other activities would be high. Equation (15) then implies that their earnings in non-home activities would be relatively low. 5.2 Other Parameters The dispersion of the talent distribution θ Equations (13) and (15) are contingents on estimates of the parameter θ , which determines the degree of selection, or alternatively the labor supply elasticity across activities. Hsieh et al. (2019) estimate θ = 1.52 after adjusting for the elasticity of human capital w.r.t. human capital expenditure. We follow their estimates and assume θ = 1.5. Furthermore, we assume θ to be constant 20 across countries and time. In Section 8 below we document that our results are robust to different choices of θ . Job-specific demand shifters { A j } and the elasticity of substitution σ So far, we only exploited implications of the labor supply side. To compute counterfactual allocations, which are required to compute the GGDI and the impact of distortions on GDP, we need to specify the demand side of the economy, that is, we need to identify { A j } and σ in (9). For the elasticity of substitution σ we follow our strategy for θ and set it exogenously. We assume σ = 3 and show in Section 8 how our results change with σ. To estimate A j , we use the equilibrium restrictions of our model together with informa- tion on aggregate GDP. As we discuss in detail in see Section A-1 in the Appendix this is sufficient to fully calibrate A j . Intuitively, there is a unique vector of relative demand shifters A j , that ensures that the observed relative employment shares and average earnings are consistent with labor market clearing. To identify the level of A j , we require our model to be consistent with the observed level of real GDP across countries and time. 6. T HE E CONOMIC L OSSES OF G ENDER M ISALLOCATION We now turn to our quantitative results. Given the estimated distortions, [τjgct , z jgct ], job-specific demand shifters [ A jct ], and the two parameters σ and θ we can compute D , counterfactual output, Y E , and the GGID according to (10). equilibrium output, Yct ct ct Similarly, we can use equation (11) to compute GDP in both the distorted and undis- torted economy and equation (12) to compute the implied change in GDP. In Section A-2 in the Appendix we outline the exact algorithm, including a sample code, to compute these objects. A key benefit of our methodology is its versatility. Given the necessary data on earnings and employment shares, the GGDI can be computed over time for a given country, across countries at a given point in time, or even across local labor markets within a country. In Section 6.1 we focus on time-variation of the GGDI within countries. This exercise highlights whether gender misallocation changes over time and quantifies the aggregate consequences of such changes. In Section 6.2 we turn to the cross-country variation and compare estimates of the GGDI across countries at different stages of development. Finally, in Section 6.3 we turn to a within-country exercise and compute the GGDI across Indian states. 6.1 The GGDI Within Countries We start by documenting changes in the GGDI over time within countries. As a starting point, we focus on the US. In Figure 1 we plot the GGDI for the US between 21 F IGURE 1: T HE GGDI IN THE US: 1970-2020 8 6 Aggregate GGDI GGDI 4 Demand Distortions (τ) Supply Distortions (z) 2 0 1970 1980 1990 2000 2010 2020 Year Notes: The figure shows the GGDI for the US for the period 1967-2022. Grey points show increase in output when both labor demand and labor supply distortions are removed. Red points show the increase in output when only labor supply distortions are removed. Blue points show increase in output when only labor demand distortions are removed. 1970 and 2020 as the dark solid line. The extent to which economic growth in the US was associated with reductions in gender-based misallocation is vividly apparent. In 1970, differences in distortions between men and women reduced overall GDP by about 7%. Since then, misallocation has continuously declined. Today, we estimate that misallocation losses in the US are below 2% of annual GDP. In the other two lines in Figure 1 we highlight the relative role of demand distortions (in blue) and supply distortions (in red). More specifically, we compute the GGDI in (10) when we either only set demand distortions to zero (keeping supply distortions at their estimated level) or only reduce supply distortions (keeping demand distortions fixed). Figure 1 shows that reductions in misallocation at both margins contributed economic growth. Quantitatively, we find demand distortions to be more important. Interestingly, note that the GGDI is not ”additive” in these two components: the overall gains from reducing both demand and supply distortions is smaller then the sum of the individual part. This reflects the fact that both margins are complementary. Intuitively, if the labor market is subject to demand distortions which raise the marginal product of labor of women relative to men, supply distortions that keep women out of the labor force have a first order impact on overall GDP. This complementarity is important when thinking about policies. If, for example, demand distortions are amenable to particular policies such as anti-discrimination laws or other regulations, these policies have large effects on allocative efficiency in an environment where supply distortions are also present. 22 In Figure 2 we display the evolution of our estimated demand and supply distortions. For parsimony, we focus on the distortions for wage work.13 In the left panel, we plot the time paths for τwt . Recall that τ measures the implicit tax a women has to pay relative to a men when working for a wage or in self-employment. In 1970, we estimate a tax of almost 70%, indicating that women only kept 30 cents of a dollar. This observations reflect the fact that few women worked but that the wage premium of these very positively selected women was not particularly large. Over time, this implicit tax declined substantively. Today, we estimate a tax of about 30%. In the right panel we plot the path of relative supply distortions in the wage sector, that is zww /zwm . Recall that we normalized the utility of home work to unity for both men and women. Hence, zww /zwm measures the utility loss of women, relative to home production, when working, relative to men. A number of 0.7 means that the typical woman has to be paid 30% more than a man to compensate her for the higher utility in home production. Figure 2 shows this relative disutility of market work declined sharply. In 1970, women faced a utility loss of 50%. 50 years later, we estimate that overall supply distortions are very small, below 10%. This pattern reflects the well-known fact of rising female labor force participation, especially between 1970 and 2000. F IGURE 2: D EMAND AND S UPPLY D ISTORTIONS IN THE US ( A ) Demand Distortions ( B ) Supply Distortions .9 1.1 .8 1 Supply distortions (zw relative to zm) .7 Demand Distortions (τ) .9 .6 .8 .5 .7 .4 .6 .3 .2 .5 1970 1980 1990 2000 2010 2020 1970 1980 1990 2000 2010 2020 Year Year Notes: Panel A plots demand distortions for market sector (τ ) for USA for the period 1967-2022. Panel B zw plots relative disutility of working in market sector ( z m ) for USA for the period 1967. Figures 1 and 2 paint an optimistic picture in that they suggest that rising gender equality might be a natural corollary of economic growth. This view, however, is only 13The patterns for self-employment are quite similar. And unpaid work does not play a large quantita- tive role in the US economy. 23 F IGURE 3: T HE GGDI IN D IFFERENT C OUNTRIES 15 10 USA India GGDI Brazil Korea, Rep. Chile 5 Mexico 0 1980 1990 2000 2010 2020 Year partly borne out in the data. To analyze the relationship between the GGDI and economic development more sys- tematically, we now compute the GGDI for four countries along the whole development spectrum: Brazil, India, South Korea, and Chile. For comparison, we also include the US experience that we already depicted in Figure 1. To see that the systematic decline of gender-based misallocation along the development path observed in the US is not a natural law, consider Figure 3, where we depict the evolution of the GGDI in different countries. Two patterns are apparent. First of all, relative to the US, gender-based misallocation in much higher in India, Chile, Brazil, and South Korea. While these level differences might reflect differences in economic development (i.e. among the countries depicted in Figure 3, the US is the richest), the GGDI is not declining in all countries, despite the fact that these countries grew substantially. While allocative efficiency in Chile declined from about 8% in the early 1990s to around 3% in 2018, misallocation along gender lines, if anything, increased in India. 6.2 The GGDI Across Countries We now move beyond the ”case-studies” shown in Figure 3 and study the relationship between gender misallocation and economic development more systematically. Figure 4 plots the GGDI for all countries against log GDP per capita. While there is a system- atic negative relationship, the cross-country variation holding economic development constant (that is for a given GDP pc) is also very large. While GDP per capita in Egypt could be increased by 24% if gender-distortions were abolished, this number falls to 5% in Peru, even though Peru is only slightly richer. This highlights the importance of country-specific characteristics in shaping gender-based misallocation and the potential 24 F IGURE 4: T HE GGDI A ND E CONOMIC D EVELOPMENT 25 EGY 20 PAK Sub-Saharan Africa PSE South Asia 15 JOR Middle East & North Africa IND GGDI Latin America & Caribbean LKA High income: OECD 10 Europe & Central Asia NER ETH East Asia & Pacific ECU RWA BEN BOL Other BTN PHL PER 5 UGA MEX CRI GHA BRA KOR KEN NGA COL CHLGRC ITA KHM NAM URY CHN GEO JPN BEL IRL ZAF ESP SRB GBRAUT USA AUS ARM BGR HRV HUN POL FRACAN CHE LUX 0 1000 3000 10000 30000 100000 GDP per capita Notes: The figure plots GGDI for all country observations near the year 2014 against their GDP per capita. The points are colored according to the continent they belong to. GGDI measure the increase in output from removal of labor demand and labor supply distortions. productivity gains of leveling the playing field. To see this underlying heterogeneity more directly, in Figure 4 we also indicate the broad geographical regions of the different economies. The pattern is stark. Countries in the Middle East and North Africa experience annual productivity losses of about 20% through gender-based misallocation in the labor market. Countries in Sub-Saharan Africa or Latin America have substantially smaller losses given their low GDP per capita. Interestingly, the relationship between gender-based misallocation and GDP pc capita is systematically negative within region but much smaller than the unconditional relationship. Despite these differences for the overall GGDI , the relative importance between demand and supply distortions is in fact rather similar to the case of the US. As we have shown in Figure 1 above, demand distortions played a much more important role for the over GGDI in the US as opposed to supply distortions. In Figure B-3 in the Appendix, we show that the same is true along the development spectrum: independent of countries’ GDP pc, we find that demand distortions have a bigger on aggregate productivity than supply distortions. The economic gains of falling misallocation are achieved through reallocation whereby men and women resort in the labor market. To illustrate one margin of such reallo- cation, in Figure 5 we plot the changes in female labor force participation that were 25 to ensue if demand distortions (blue markers) and supply distortions (red markers) were dismantled. Figure 5 shows distortions are major drags on female labor force participation. Demand distortions reduce the private benefits of working and therefore contribute to keep women out of the labor force. If women were not disadvantaged by the implicit taxes that we estimate, labor force participation would increase by 40% in India, by 20% in Brazil, and by 10% in the US. Hence, a key reason why demand distortions lower overall productivity is that they keep talented women out of the labor force altogether. For comparison, Figure 5 also displays the change in female labor force participation (FLFP) in the absence of supply distortions. Again, we see that the presence of supply distortions is an important aspect of low participation. However, their overall impact of smaller than for demand distortions, especially in rich countries. For the US, for example, supply distortions only account for a small share of female participation.14 The reason why falling supply distortions have a smaller impact on overall productivity as measured by the GGDI despite the fact that they trigger large changes in labor market participation is again the complementarity between supply and demand distortions. In the presence of demand distortions on the labor market, changes in the relative utility of market work might trigger large changes in participation, but only smaller changes in overall productivity, precisely because women are still subject to distortionary taxes that affects their relative marginal product across different activities. GGDI versus GDP The fact that distortions affect women’s labor supply between market activities and home production highlights that they reduce both overall, welfare-relevant productivity and measured GDP. However, as discussed in Section 4.5, the economic gains in total economic activity captured by the GGDI are distinct from those measured by GDP. Whether the GGDI exceeds the impact on GDP, ∆GDP , is theoretically ambiguous. If distortions keep a large number women in home activities and dismantling distortions encourages more market participation, the impact on GDP tends to be higher because the opportunity cost of market work, the forgone output of home production, is not counted. At the same time, in equilibrium, an increase in female labor supply crowds out male employment and male participation in home production increases. This form of reallocation will reduce GDP substantially but have smaller effects on the GGDI. 14 Note that, for a small number of countries, labor supply distortions increase FLFP. Recall that we infer the labor supply distortions for a given group using the wage in job type j relative to wage in the home sector, which we proxy as agricultural earnings, adjusted for individual characteristics. For some countries, this ”earnings premium” is higher for men compared to women, leading us to infer higher higher labor supply distortions in the home sector for men. And because our counterfactual equalizes distortions across groups, the home sector becomes relatively more attractive for women. 26 F IGURE 5: Distortions and Female Labor Force Participation 60 No Demand Distortions (τ) No Supply Distortions (z) PAK EGY PSE 40 EGY PAK JOR LKA Change in Female LFP IND IND ECU BEN MEX LKA BOL CRI CHN MEX PHL 20 NER JOR KOR NER COL BRA GEO CHL GRC JPN ITA PHL BOL BTN ZAF CRI ESP BEL AUS AUT IRL CHE ETH ARM ECU COL CHL URY GBR NGA GHA KHM PSE NGA NAM BTN PER BRA URY USA KEN PER SRB FRA LUX HRV ETH KHM HUN POL ITA KOR CAN RWA BGR GRC UGA KEN SRB HRV HUN LUX NAM BEL CAN USA 0 UGA ZAF POL ESP FRA AUT BEN JPN ARM BGR GBR AUS RWA GHA IRL GEO CHE CHN -20 1000 3000 10000 30000 100000 GDP per capita Notes: The figure shows the change in female labor force participation (FLFP) by removing all demand and supply distortions, respectively, for all countries, plotted against their GDP per capita. The size of points are proportional to the country’s population. The sample for the plot is all country observations near the year 2014. Hence, the effect on distortions on both the GGDI and overall GDP is, in the end, a quantitative question. In Figure 6, we report the GGDI and the change in measured in GDP in response to a dismantling of both demand and supply distortions. We find that the GGDI (shown in green) and the change in GDP (shown in yellow) are highly correlated and that the change in GDP is higher than the change in the GGDI for all countries in our sample. This pattern stems from the fact that distortions keep many women in home activities. If such distortions were abolished, overall efficiency would increase but measured GDP would increase even more.15 6.3 The GGDI Across Indian States We now turn to our last application: quantifying the losses of gender-based misalloca- tion across local labor markets within a country. This perspective might be particularly interesting because it potentially highlights the importance of specific institutions, laws, and regulations to the extent that other determinants of gender-specific differences such as social norms, cultural traditions, or the general functioning of the labor market are common across labor markets within the country. As a proof of concept of this exercise, we focus on the case of India and estimate the GGDI at the state-level. In Figure 7 we plot the GGDI among Indian States. In the left 15 In Appendix Figure B-5 we show a version of Figure 6 where we individually shut down demand and supply distortions. We find that the effect of GDP exceeds the GGDI for both sources of misallocation. 27 F IGURE 6: T HE GGDI A ND T HE C HANGE IN GDP 30 EGY Change in total output (GGDI) Change in measured GDP % change in measured GDP and total output PAK PSE JOR EGY 20 PAK IND PSE LKA JOR IND NER LKA ETH 10 NER ETH BOL ECU MEX BEN BTN PHL ECU RWA BOL CRI RWA BEN PER BTN GHA PHL PER COL BRA GRC KOR UGA UGA MEX CHL NGA CRI ITA KHM KEN GHA NGA NAM CHN BRA URY KOR KEN ZAF GEO COL SRB CHL GRC ESP BEL IRL KHM NAM ITA JPN URY HRV AUTUSA CHN ARM GEO ZAF HUN GBR AUS JPN ESP BEL FRA IRL CHE LUX SRB POL GBRCAN AUTUSA CHE AUS ARM BGR HUN HRV POL FRA CAN LUX BGR 0 1000 3000 10000 30000 100000 GDP per capita Notes: The figure plots GGDI and the change in measured GDP for all country x year observations in the sample against their GDP per capita. GGDI measure the increase in total economic activity (including home production) due to the removal of labor demand and labor supply gender distortions. Measured GDP is the increase in GDP following the removal of labor demand and labor supply gender distortions. panel we focus on the case of demand distortions, in the right panel we focus on the case of supply distortions. The results are interesting in three aspects. First, the productivity losses from gender-based misallocation are sizable; they range from 5% to 15%. Second, like in the cross-country data shown in Figure 4, there is a systematic correlation with state-level GDP: the richer the state, the lower gender-based misallocation. Third, Figure 7 shows that the productivity losses from supply distortions are if anything slightly lower than the ones from demand distortions. 28 F IGURE 7: T HE GGDI A CROSS I NDIAN S TATES ( A ) Demand Distortions ( B ) Supply Distortions 25 25 % change in Y from removal of demand distortions % change in Y from removal of supply distortions Arunachal Pradesh Bihar 20 Assam 20 Jammu and Kashmir Jharkhand Tripura Arunachal Pradesh Uttar Pradesh Assam Uttarakhand Bihar Jammu and Kashmir 15 Odisha Nagaland 15 Uttarakhand Punjab Tripura Andaman and Nicobar Islands Chandigarh Chandigarh Haryana Gujarat Delhi Rajasthan Andaman and Nicobar Islands Puducherry Madhya Pradesh Uttar Pradesh Punjab Kerala Manipur Jharkhand Gujarat Haryana West Bengal Manipur 10 Chhattisgarh Mizoram Karnataka Maharashtra Sikkim 10 Delhi Himachal Pradesh Madhya Pradesh Puducherry Telengana Kerala Tamil Nadu Odisha Nagaland Rajasthan Mizoram Andhra Pradesh Goa West Bengal Meghalaya Meghalaya Karnataka Sikkim Maharashtra Telengana Andhra Pradesh Tamil Nadu Chhattisgarh 5 5 Himachal Pradesh Goa 0 0 .2 .4 .7 1.2 .2 .4 .7 1.2 GDP per capita (in log) GDP per capita (in log) Notes: The figure plots the GGDI against the log of GDP per capita across India states. In panel A, we set τjw = 0. In panel B, we set z jw = z jm = 1. The x-axis measure GDP per capita of each state relative to that of Haryana. 7. VALIDATION : D IRECT M EASURES OF G ENDER E QUALITY The GGDI is a model-based measure of gender equality. This has the benefit that it has a clear cardinal interpretation and that it aggregates different aspects of women’s labor market outcomes such as participation gaps, wage gaps, and differences in occupational employment patterns into a scalar, that can be compared across countries, local labor markets, and time. In addition, it directly relates empirical gender gaps to aggregate productivity. The productivity consequences of gender-based misallocation might be of particular importance for policy makers. The disadvantage is that is is model-based and therefore relies on a particular theoretical structure and the assumptions on functional forms that we impose. In this section, we relate the demand and supply distortions and the GGDI to de-jure measures of gender discrimination. To do so, we use the Women, Business, and Law dataset (World Bank, 2024) (henceforth WBL) that provides measures of equality of economic opportunity under the law between men and women in 190 economies, for 50 years, from 1970 until today. The WBL Index is a summary statistic of specific restrictions women face to participate in the labor market and society at large. Figure 8 is a scatter plot of our GGDI and the WBL Index across countries. Reassuringly there is a strong negative relationship, indicating that countries with a higher index (a sign of more gender equality) are indeed inferred to suffer from less misallocation according to our index. The bivariate correlation between the two variables is 0.74. We find this strong relationship between the de-jure equality measures of the WBL and our de-facto measure according to the GGDI reassuring that the GGDI indeed captures gendered differences in labor market outcomes. 29 F IGURE 8: T HE GGDI AND THE W OMEN , B USINESS , AND L AW I NDEX 30 EGY 20 PAK JOR IND GGDI LKA 10 NER ETH ECU BOL BEN RWA BTN PHL PER UGA MEX GHA CRI NGA KEN BRA KOR NAM COL CHL GRC ITA KHM URY CHN JPN GEO ZAF IRL BEL ESP CHE SRB GBR USA AUT AUS ARM BGR LUX HUN POL HRV FRA CAN 0 -10 40 60 80 100 Women, Business and the Law (2024) Notes: The figure plots the GGDI that we measure for each country in our sample against the Women, Business and Law Index (World Bank, 2024) for that same year. 8. R OBUSTNESS In the section, we explore the sensitivity of our estimates to the values of the structural parameters σ and θ . For our baseline analysis we chose σ, the elasticity of substitution between the different activities for aggregate output, to be equal to three. In the left panel of Figure 9 we plot the GGDI for different values of σ from 2 to 5. Our results are robust to the values of σ. The estimates remain within 1% range of the baseline estimates. The second structural parameter, scale parameter of Fr´ echet distribution of idiosyncratic talent θ , is more consequential. Our baseline choice of θ = 1.5 is close to the estimate of 1.52 used in Hsieh et al. (2019). In the right panel of Figure 9, we plot the GGDI for F IGURE 9: R OBUSTNESS : T HE E FFECTS OF σ AND θ ( A ) Elasticity of Substitution σ ( B ) Talent Disperions θ 25 30 EGY EGY EGY EGY PAK 20 PAK PAK PAK GGDI (θ in {1.25, 1.75, 2}) PSE GGDI (σ in {2, 4, 5}) PSE 20 PSE PSE 15 JOR IND JOR IND JOR IND JOR IND EGY LKA LKA LKA PAK LKA 10 NER ETH NER PSE ETHNER NER ECU ETH 10 RWA ECU BOL ECU ETH IND JOR ECU BOL BEN BOL RWA BEN BOL RWA RWA BTN BEN LKA BTN BEN PER PHL BTN PER BTN PHL PER UGA PHL PER 5 UGA PHL NER MEX MEX MEX UGA ETH UGA CRI MEX CRI CRI GHA σ=2 θ = 1.25 GHA CRI ECU BRAGHA NGABRA KOR GHA BEN BOL KOR BRA KOR KENBRA RWA NGA KEN KOR NGA BTN GRCNGA KEN COL CHL COL GRC CHL ITA KEN PER COL CHL ITA GRC NAM COL CHL GRC MEX UGA PHL ITA σ=4 θ = 1.75 ITA NAM KHM NAM NAM KHM URY URY KHM KHM CRI GHA CHNURY CHN JPN GEOURY BRA KOR JPN GEO IRL CHN JPN GEO BEL IRL CHN BEL IRL ZAFJPN GEO KEN NGA ZAF BEL ZAF ESP IRL BEL ZAF COL CHL GRC ITA ESP ESP ESP AUT KHM NAM URY AUT USA USA σ=5 θ=2 SRB AUS AUS SRB USA SRB AUS GBR CHEAUT BEL IRL CHN GEO JPN GBR CHE LUX FRA HUN GBR CHE LUX HUN LUX FRA ZAF ESP ARM HRV CAN HUN ARM HRV CAN AUT SRB USA AUS GBR CHE POL BGR POL LUX FRA HUN ARM HRV 0 0 CAN BGR POL BGR 0 5 10 15 20 25 0 5 10 15 20 25 GGDI Baseline (σ = 3) GGDI Baseline (θ = 1.5) Notes: The left (right) panel of the figure plots the GGDI for different values of σ (θ ). For all parameters we full re-calibrate the model. 30 different values of θ between 1.25 to 2. We note that the GGDI is similar to our baseline results as long as θ > 1.5. However, for smaller values of θ , the aggregate implications of gender-based misallocation are sensitive to the particular choice of θ . Future research should aim to estimate θ directly from the microdata and to allow it to vary across countries. 9. C ONCLUSION In this paper we proposed a methodology to estimate the macroeconomic productivity losses of gender misallocation. Building on the work of Hsieh et al. (2019) we show that using data on employment shares across different activities, earnings, and human capital one can compute the overall productivity gains that an economy could reap if distortions could be abolished. This statistic, which we refer to as the Global Gender Distortion Index (GGDI), captures the productivity losses of the observed allocation relative to a situation of gender equality. As such, the GGDI is comparable across countries and time, is grounded in economic theory, has a straightforward cardinal interpretation, and captures both distortions on the labor demand side (such as wage discrimination or flatter career ladders for women) and on the labor supply side (such as institutions that discourage women to partake in full-time, market work). While the GGDI captures the correct welfare-relevant notion of aggregate productivity, it is distinct from measured GDP. The reason is that the GGDI captures all economic activity, that is, market work, home production, and unpaid activities. By contrast, GDP is confined to activities that fall within the production boundary as defined by the system of national accounts. This distinction is conceptually important in the context of developing countries where informal, non-market transactions play an outsized role. Nevertheless, the implications of gender-based misallocation on aggregate GDP can also be readily computed. In our applications, we find that the change in GDP exceeds the GGDI, highlighting that distortions are a key reason that keep too many women in home activities. Because our methodology is easy to implement, we view the GGDI as a complementary tool to other measures of gender equality and we hope that future research will make use of it. As part of this paper, we also distribute the MATLAB codes to compute the GGDI for other datasets that contain the required information on employment shares, earnings, and human capital. 31 R EFERENCES Agte, P., O. Attanasio, G. Pinelopi, R. Aishwarya Lakshmi, R. Pande, M. Peters, C. T. Moore, and F. Zilibotti: 2024, ‘Gender Gaps and Economic Growth: Why Haven’t Women Won Globally (Yet)’. Yale Working Paper. Bandiera, O., A. Elsayed, A. Heil, and A. Smurra: 2022, ‘Presidential address 2022: Economic development and the organisation of labour: Evidence from the jobs of the world project’. Journal of the European Economic Association 20(6), 2226–2270. Bhandari, A., T. Kass, T. J. May, E. McGrattan, and E. Schulz: 2024, ‘On the Nature of Entrepreneurship’. Working Paper 32948, National Bureau of Economic Research. Bloom, N., B. Eifert, A. Mahajan, D. McKenzie, and J. Roberts: 2013, ‘Does management matter? Evidence from India’. The Quarterly journal of economics 128(1), 1–51. Bridgman, B., A. Craig, and D. Kanal: 2022, ‘Accounting for household production in the national accounts’. Survey of Current Business 102(2), 1–3. Chiplunkar, G. and P. K. Goldberg: 2021, ‘Aggregate Implications of Barriers to Female Entrepreneurship’. Working Paper 28486, National Bureau of Economic Research. Chiplunkar, G. and T. Kleineberg: 2022, ‘Gender Barriers, Structural Transformation, and Economic Development’. Working paper. Duflo, E.: 2012, ‘Women Empowerment and Economic Development’. Journal of Eco- nomic Literature 50(4), 1051–1079. Feenstra, R. C., R. Inklaar, and M. P. Timmer: 2015, ‘The Next Generation of the Penn World Table’. American Economic Review 105(10), 3150–82. andez, R.: 2013, ‘Cultural Change as Learning: The Evolution of Female Labor Fern´ Force Participation over a Century’. American Economic Review 103(1), 472–500. Fletcher, E., R. Pande, and C. T. Moore: 2019, ‘Women and Work in India: Descriptive Evidence and a Review of Potential Policies’. India Policy Forum 15(1), 149–216. Fogli, A. and L. Veldkamp: 2011, ‘Nature or Nurture? Learning and the Geography of Female Labor Force Participation’. Econometrica 79(4), 1103–1138. Goldin, C.: 1994, ‘The U-shaped female labor force function in economic development and economic history’. National Bureau of Economic Research Working Paper (4707). Gollin, D.: 2002, ‘Getting Income Shares Right’. Journal of Political Economy 110(2), 458–474. Gottlieb, C., C. Doss, D. Gollin, and M. Poschke: 2024, ‘The gender division of work across countries’. IZA Discussion Paper. Greenwood, J., A. Seshadri, and M. Yorukoglu: 2005, ‘Engines of Liberation’. Review of Economic Studies 72(1), 109–133. Hsieh, C.-T., E. Hurst, C. I. Jones, and P. J. Klenow: 2019, ‘The allocation of talent and us economic growth’. Econometrica 87(5), 1439–1474. 32 ILO: 1982, ‘Resolution concerning statistics of the economically active population, employment, unemployment and underemployment’. Jayachandran, S.: 2015, ‘The Roots of Gender Inequality in Developing Countries’. Annual Review of Economics 7(1), 63–88. Jayachandran, S.: 2021, ‘Social Norms as a Barrier to Women’s Employment in Devel- oping Countries’. IMF Economic Review 69(3), 576–595. Klasen, S.: 2019, ‘What Explains Uneven Female Labor Force Participation Levels and Trends in Developing Countries?’. The World Bank Research Observer 34(2), 161–197. Lagakos, D.: 2016, ‘Explaining Cross-Country Productivity Differences in Retail Trade’. Journal of Political Economy 124(2), 579–620. Ngai, L. R., C. Olivetti, and B. Petrongolo: 2024, ‘Gendered Change: 150 Years of Transformation in US Hours’. Working Paper 32475, National Bureau of Economic Research. Ngai, L. R. and B. Petrongolo: 2017, ‘Gender gaps and the rise of the service economy’. American Economic Journal: Macroeconomics 9(4), 1–44. Olivetti, C., J. Pan, and B. Petrongolo: 2024, ‘The Evolution of Gender in the Labor Market’. Working Paper 33153, National Bureau of Economic Research. Pennings, S.: 2022, ‘A Gender Employment Gap Index (GEGI)’. World Bank Policy Research Working Paper 9942. Rendall, M.: 2018, ‘Female Market Work, Tax Regimes, and the Rise of the Service Sector’. Review of Economic Dynamics 28, 269–89. United Nations: 2008, ‘System of National Accounts 2008’. United Nations, New York. Available from United Nations, Department of Economic and Social Affairs, Statistics Division. World Bank: 2024, ‘Women, Business and the Law 2024’. Technical report, World Bank, Washington, DC. Accessed: 2025-03-03. Young, A.: 1995, ‘The Tyranny of Numbers: Confronting the Statistical Realities of the East Asian Growth Experience’. The Quarterly Journal of Economics 110(3), 641–680. 33 APPENDIX A: THEORETICAL RESULTS This section contains additional details on the theoretical analysis. In Section A-1 we describe how we calibrate the job-specific demand shifts A jct in (9). In Section A-2 we describe how we compute the equilibrium allocations for given parameters and distortions. A-1. C ALIBRATING A jct Let w j denote the equilibrium skill prices of activity j. The optimality condition for labor demand in job j from (9) is given by 1 σ −1 1/σ σ −1 σ −1 −1 σ −1 Y wj = ∑( A j Hj ) σ Aj σ Hj σ = Aj σ Hj . (A-1) j θ −1 Total labor supply, Hj is given in equation (8) as Hj = ∑ g L g p jgθ h jg Γθ . The equilib- rium skill prices w j can be inferred from equations (4), (6), and (5). In particular, the employment share of men in activity j is given by θ θ w j h jm z jm Γθ w j h jm Γθ θ wθ jm w j h jm z jm p jm = = = = (A-2) ∑ j wθ jm ∑ j wθ jm wage jg z jm wage jg Hence, wage jg w j = p1/ θ , (A-3) jm h jm Γθ which expresses w j in terms of observables. Using (A-1) for activity j and the market sector w, yields 1 σ σ −1 Aj wj σ −1 Hj = , (A-4) Aw ww Hj which determines A j relative to Aw . To determine Aw we use that total output is given in (9) as σ σ σ −1 σ −1 σ −1 σ −1 Aj σ Y= ∑( A j Hj ) σ = ∑ H Aw j × Aw , (A-5) j j and we chose Aw to match a given level for real GDP per capita.1 1 Strictly speaking, Y is not equal to GDP, because it also includes home production and unpaid work. However, because we only use this moment to calibrate the level of A j across countries and years, this discrepancy is not important and we equalize the model-based measure Y with the empirically measured level of GDP per capita. A-1 A-2. S OLVING FOR THE E QUILIBRIUM Given the parameters and distortions, we can solve the model in an iterative way. To do so guess total labor demand in each activity Hj . Then use (A-5) to compute Y and (A-1) to compute w j . Given w j and the relative distortions, we can compute labor supply by men and women and compute aggregate labor supply. We then iterate on this guess for Hj until labor supply is equal to labor demand. A-2 APPENDIX B: EMPIRICAL RESULTS B-1. D ETAILS FOR E MPIRICAL A NALYSIS Our empirical analysis uses the Harmonized World Labor Force Survey (HWLFS), a large scale micro dataset that harmonizes over 2’500 surveys and contains individual level data on individual characteristics, education, employment and jobs from 120 countries. From this dataset, we select specific surveys to build two datasets, a cross- country and an Indian state dataset. B-1.1 Construction of variables Cross-country dataset For our cross-country dataset, we use nationally representative surveys that contain data on hours worked, wages, and industry codes for all workers. Overall, our analysis currently draws from 51 countries and our sample cover 69 percent of the World Population in 2015. The poorest country in our sample is Niger in 2005 with an GDP per capita level of USD 624 while the richest country is Luxembourg in 2005 with an GDP per capita level of USD 97973 .1 In Table B-I we list all country surveys and their year coverage. For each survey, we focus on individuals aged 25 to 60. For these, we record information on their marital status and their educational attainment. When available, we use reported years of schooling; otherwise, we impute years of schooling based on the highest completed degree. We use the main job section of the surveys to classify workers by their job type: wage worker, self-employed (either employer or own-account worker), or unpaid worker. An unpaid worker is an individual engaged in a job without receiving a formal wage or salary, typically contributing to a family business, farm, or household enterprise. Unlike own-account workers and employers, unpaid workers have no claim to the profits of the activity. They contribute economically but are not compensated in wage nor in profits. Alongside their job type, we also record information on their sector of employment and weekly hours worked. For wage workers, we measure labor income. To do so, we use data on their most recent payment, payment frequency and hours worked, to compute their hourly wage. Individuals with no main job are classified as not working, a category we also refer to as working in the home sector. Indian States data For our cross-state exercise, we use Periodic Labour Force Survey (PLFS) of 2018-19. It contains detailed data on labor force participation of a represen- tative sample of (approx.) 100,000 households. We restrict the sample to working age population: age 25-60 years. This leaves us with 202,696 individual level observations. PLFS classifies the employment status of an individual into following categories: wage workers (code 31, 41, 42, 51, 71, 72), Self employed (code 11, 12, 61, 62), unpaid workers (code 21). As shown in table B-II, roughly 75% of women are not in labor force, where as this number is less than 10% for men. The majority of people are self-employed, followed by regular wage and casual labor.2 1The GDP per capita numbers Are taken from the Penn World Tables (Feenstra et al., 2015). 2 Maybe surprisingly, Table B-II shows that educational attainment of the few men doing unpaid work B-1 TABLE B-I: L IST OF SURVEYS heightCountry Survey Name Year coverage Armenia Labour Force Survey 2014-2019 Australia Household, Income and Labour Dynamics in Australia 2001-2017 Austria European Union Statistics on Income and Living Conditions 2004-2020 Belgium European Union Statistics on Income and Living Conditions 2004-2005 Benin Enquˆete Modulaire Int´ ee sur les Conditions de Vie des m´ egr´ enages 2010-2015 Bhutan Labor Force Survey 2018-2020 Bolivia Encuesta de Hogares 2005-2020 Brazil Pesquisa Nacional por Amostra de Domic´ ılios 2009-2015 Bulgaria European Union Statistics on Income and Living Conditions 2008-2020 Cambodia Cambodia Labor Force and Child Labor Survey 2012-2019 Canada Labour Force Survey 1997-2020 Chile Encuesta de Caracterizacion ´ ´ Socioeconomica Nacional 1990-2017 China Family Panel Studies 2012-2016 Colombia Gran Encuesta Integrada de Hogares 2006-2019 Costa Rica Encuesta Continua de Empleo 2010-2023 Croatia European Union Statistics on Income and Living Conditions 2010-2020 Ecuador Encuesta Nacional de Empleo, Desempleo y Subempleo 2007-2018 Egypt Harmonized Labor Force Survey 2007-2017 Ethiopia National Labour Force Survey 2005-2013 France Enquˆete emploi annuelle 2003-2019 Georgia Labour Force Survey 2017-2021 Ghana Ghana Living Standard Survey 1987-2017 Greece European Union Statistics on Income and Living Conditions 2004-2020 Hungary European Union Statistics on Income and Living Conditions 2006-2020 India Indian National Sample Survey 1987-2011 India Periodic Labor Force Survey 2017-2023 Ireland European Union Statistics on Income and Living Conditions 2004-2019 Italy European Union Statistics on Income and Living Conditions 2004-2020 Japan Employment Status Survey 1997-2017 Jordan Harmonized Labor Force Survey 2005-2016 Kenya Kenya Continuous Household Survey Programme 2019-2021 Luxembourg European Union Statistics on Income and Living Conditions 2012-2015 Mexico Encuesta Nacional de Ocupacion ´ y Empleo 2005-2019 Namibia Labor Force Survey 2012-2018 Niger National Survey on Household Living Conditions and Agriculture 2011-2014 Nigeria Living Standards Measurement Survey 2010-2018 Pakistan Labor Force Survey 2010-2018 Palestinian Territories Harmonized Labor Force Survey 2000-2016 Peru Encuesta Nacional de Hogares 2007-2019 Philippines Labor Force Survey 2005-2019 Poland European Union Statistics on Income and Living Conditions 2005-2019 Rwanda Enquˆ egrale sur les Conditions de Vie des M´ ete Int´ enages 2000-2016 Serbia European Union Statistics on Income and Living Conditions 2013-2020 South Africa Labor Market Dynamics 2010-2019 South Korea Korean Labor and Income Panel Study 1998-2018 Spain European Union Statistics on Income and Living Conditions 2004-2012 Sri Lanka Labor Force Survey 1996-2022 Switzerland European Union Statistics on Income and Living Conditions 2007-2020 Uganda Uganda National Panel Survey 2009-2019 United Kingdom European Union Statistics on Income and Living Conditions 2005-2018 United States Current Population Survey 1967-2022 Uruguay Encuesta Continua de Hogares 2006-2017 is very high. This subpopulation tends to be young and work in family firms, potentially in managerial positions (see Bloom et al. (2013)). B-2 TABLE B-II: Descriptive Statistics Men Women Total Shares 49.46% 50.54% 100.00% Employment Proportions Wage work 45.68% 12.14% 28.73% Self employed 40.14% 5.82% 22.80% Unpaid work 4.14% 17.74% 11.02% Home sector 10.03% 64.29% 37.46% Total 100.00% 100.00% 100.00% Average Monthly Earnings (in rupees) Wage work 2851 1974 2408 Self employed 2530 1458 1988 Unpaid work 2292 1212 1747 Home sector 1809 - - Education (Years of Schooling) Wage work 8.92 7.85 8.38 Self employed 7.93 5.35 6.63 Unpaid work 9.88 4.64 7.23 Home sector 9.09 6.55 7.56 Notes: Descriptive statistics are calculated from Periodic Labor Force Survey of 2018-19. B-1.2 Measurement of labor income We use data on weekly hours worked (h) and hourly wages (w) to impute weekly labor income (y = wh) for all individuals in the dataset. In particular, for each country-year survey and for each gender, we run a Mincerian wage regression ln yi = δI (i) + γ yrsi + xi′ β + ε i , (B-1) where δI is a set of sector fixed effects, yrsi denotes years of schooling for individual i, and x is a set of individual characteristics such as marital status and age. For each country-year survey, we estimate the coefficients (δ ˆ, γ ˆ ) for both men and women ˆ, β separately. We thus allow for the return to human capital, the marriage premium, and sectoral premia to differ by gender. We consider three sectors—agriculture, industry, and services—as defined by the International Standard Industrial Classification (ISIC): agriculture corresponds to ISIC Section A; industry includes Sections B to F (e.g., mining, manufacturing, and construction); and services comprise Sections G to U, including trade, transport, finance, education, health, and other service activities. We then use these coefficients to predict weekly labor income for all individuals in the dataset. Specifically, for self-employed and unpaid individuals we compute (log) earnings by ln yi = δI (i) + γ yrsi + xi′ β, (B-2) that is we use both the individual characteristics, i.e. marital status, age, and education as well as the sector of employment to predict overall earnings. For individuals that are not working we impute earnings assuming that they work in the agricultural sector, B-3 exploiting the idea that agricultural work might be the closest comparison for the appropriate opportunity cost of home production. To get a sense of the magnitude of the returns to experience, marital, and sectoral premia, we estimate regression B-1 pooling all the data and incorporating survey-year- country fixed effects. In table B-III, we reported the estimated coefficients. On average, women earn 36.3% less than men, and married individual earn 9.2% more than singles. The estimated returns to schooling amount to 8% which is in line with estimates in the literature, and our assumed return of human capital for our structural analysis (see equation B-3). We estimate sectoral income gaps of 50% and 42.6% for workers in industry and services relative to agricultural wage workers. Weekly Income Women -0.364 (0.021) Married 0.092 (0.015) Years of Education 0.080 (0.004) Age 0.008 (0.001) Industry 0.491 (0.076) Services 0.417 (0.078) Adj. R-squared 0.908 Within R-squared 0.211 Observations 22’782’947 TABLE B-III: Mincer Regression. Notes. This table shows the estimated coefficients of regression B-1. The left hand side variable is weekly income of wage workers, i.e. hourly wage times weekly hours worked. We pool all country year surveys and incorporate survey, county and years fixed effects. Standard errors, clustered at the country, year and survey level, are reported in parentheses. B-1.3 Cross-country data: Income groups In Tables 1 and 2 we reported gender differences in educational attainment and the type of work for countries of different income. Table B-IV reports the individual countries by income group. B-1.4 Gender Earnings Gaps over Time In Figure B-1 below we plot the gender income gap, that is average earnings in the wage sector of women relative to men, against GDP per capita in 2014.3 . For the vast majority of countries the gender income gap is below one, that is women earn less than men. The cross-country pattern is suggestive of a U-shape relationship between income per capita and the gender income gap. 3 In case a country does not report data in 2014, we take the closest year B-4 TABLE B-IV: I NCOME G ROUPS Income Group Countries High income Australia, Austria, Belgium, Canada, Chile, Croatia, France, Greece, Hungary, Ireland, Italy, Japan, Korea, Rep., Luxembourg, Poland, Spain, Switzerland, United Kingdom, United States, Uruguay Middle income Armenia, Benin, Bhutan, Bolivia, Brazil, Bulgaria, Cambodia,China, Colombia, Costa Rica, Ecuador, Egypt, Arab Rep., Georgia, Ghana, India, Jordan, Kenya, Namibia, Nigeria, Palestine, Pakistan,Peru, Philippines, Serbia, South Africa, Sri Lanka Low income Ethiopia,Niger, Rwanda, Uganda F IGURE B-1: R AW G ENDER INCOME GAP ACROSS COUNTRIES 1.25 HUN PHL 1 JOR NAM ECU EGY CRI LKA SRB HRV BOL BTNCOL GRC POL BGR ESPFRA KHM KEN PER BRA ITA Gender Income Gap ZAF CHL LUX URY BEL CAN USA IRL .75 ETH IND GEO CHN AUS RWA GHA AUT PAK ARM KORGBR CHE JPN .5 UGA NGA .25 0 1'000 3'000 10'000 30'000 100'000 GDP per capita (PPP, real) Notes: This figure plots the gender income gap for all countries in our cross-country dataset against the log of GDP per capita as provided by Feenstra et al. (2015). Gender income gap is the ratio of weekly income of women relative to men. Weekly income is the product of hourly wage and weekly hours worked. B-5 F IGURE B-2: P REDICTED G ENDER I NCOME G AP ACROSS C OUNTRIES B Y A CTIVITY 1.25 1.25 1 PHL JOREGY NAM 1 ECU Gender Income Gap (Predicted) Gender Income Gap (Predicted) SRB HUN SRB BTNCOL HRV PHL HUN POL BOL CRI POL GRC GRC KHM ZAF BGR LKA KHM COL BGR HRV PER BRA ITA ESPFRA RWA BRA FRA ITA ETH UGA GHA LUX ECU CRI ESP IRL CHL URY CAN USA BEL ETH ZAF LKA BEL CAN USA LUX .75 KEN GEO IRL .75 BOL ARM URY CHN AUS KEN PER CHL NGA IND ARM AUT NGA EGY GEO GBR AUS RWA KORGBR IND CHN PAK CHE UGA BTN NAM KOR AUT CHE GHA .5 .5 NER PAK JPN JPN .25 .25 0 0 1'000 3'000 10'000 30'000 100'000 1'000 3'000 10'000 30'000 100'000 GDP per capita (PPP, real) GDP per capita (PPP, real) ( A ) Wage work ( B ) Self employment 1.25 1.25 NAM ZAF LUX 1 1 CRI HUN POL GRC PHL ESP Gender Income Gap (Predicted) Gender Income Gap (Predicted) SRB HRV ECU BRA BGR USA IRL KHM KOR BEL KHM LKA EGY PHL ECU EGY CHL ITA ARMGEO .75 KEN NGA .75 ETH KEN INDBOL URY FRA ETH PAK PER GEO UGA JOR COL AUS BOL CAN PAK IND PER CHN JPN NER CHE BTN RWA AUT GBR .5 .5 BTN NER RWA GHA GHA NGA .25 .25 0 0 1'000 3'000 10'000 30'000 100'000 1'000 3'000 10'000 30'000 100'000 GDP per capita (PPP, real) GDP per capita (PPP, real) ( C ) Unpaid work ( D ) Home work Notes: Each panel plots the predicted gender income gap against log GDP per capita (PPP, real) for a specific work activity as outlined in B-1.2. The gender income gap is the ratio of predicted weekly income for women relative to men. GDP data is from Feenstra et al. (2015). B-6 B-2. M EASUREMENT FOR S TRUCTURAL A NALYSIS To calibrate our model, we require three objects from the HLWFS microdata: (i) the job-specific employment shares for each group g, p jg ; (ii) average earnings of group g in activity j, wage jg ; (iii) group g’s average human capital for activity j, h jg . The employment shares p jg are directly observed in the HLWFS data. To measure the stocks of human capita, h jg , we rely on Mincerian wage regressions to project observed schooling into unobserved human capital. Specifically, we assume an annual return of human capital of 8% and hence compute h jg according to h jg = exp 0.08 × yrs of schooling jg , (B-3) where yrs of schooling jg denotes average years of schooling of individuals of group g in activity j. Finally, we compute wage jg directly from observed earnings. Note that wage jg is the geometric average of earnings. Empirically, we observe the arithmetic average of earnings, e jg . Given the Fr` echet distribution, we can convert the arithmetic average of earnings e jg to the geometric average wage jg using the relationship Γ˜ wage jg = e jg Γ (1 − 1 θ) γem where Γ ˜ =e θ . γem is Euler–Mascheroni constant (≈ 0.5772). B-3. D EMAND AND S UPPLY D ISTORTIONS A CROSS THE W ORLD In Figure 4 in the main text, we reported the GGDI in the cross-section of countries. The GGDI measures the productivity increase induced by eliminating both supply and demand distortions. In Figure B-3 below we report the results separately: In the left panel we report the productivity consequences of eliminating demand distortions (keeping supply distortions in place), in the right panel we eliminate supply distortions but keep demand distortions untouched. Figure B-3 shows that, on average, demand distortions are economically more costly for aggregate productivity and that the gains from dismantling either distortions are higher in poor countries. B-7 F IGURE B-3: D EMAND V S . S UPPLY D ISTORTIONS ( A ) Demand Distortions ( B ) Supply Distortions 20 20 % change in Y from removal of demand distortions % change in Y from removal of supply distortions EGY PAK 15 15 EGY PSE IND JOR 10 PAK 10 LKA LKA ETH JOR RWA NER IND ECU BOL 5 PSE 5 UGA ECU PER MEX BOL PHL BTN BTN NER MEX PHL CRI PER BEN GHA KOR BEN COL CRI KEN BRA ETH CHL GRC NGA BRA NGA CHL ITA GHA COL IRL URY KHM NAM GEO JPN RWA KOR ITA URY KHM NAM GRC ZAF ESP BEL UGA KEN 0 CHN AUS AUT USA SRB HRV HUN BEL USA LUX SRB GBR CHE FRA CAN ARM FRA LUX ZAF BGR POL ESP AUT BGR HRV HUN POL CAN JPNGBR AUSIRL 0 ARM CHN GEO CHE 7 8 9 10 11 12 7 8 9 10 11 12 GDP per capita (in log) GDP per capita (in log) Notes: Panel A plots the increase in GDP from removal of labor demand distortions and Panel B plots the same for the case of removal of labor supply distortions. The sample consists of all country observations near the year 2014. F IGURE B-4: D ISTORTIONS AND F EMALE L ABOR F ORCE PARTICIPATION ( A ) Demand Distortions ( B ) Supply Distortions 1 1 UGA BEN ECU JPN ETH RWA UGA CAN UGA BOL NER GBR NER UGA ETH KEN GHA PER URY PER RWA BTN USA CHE RWA URY FRA AUS AUT CAN CAN LUX KEN .8 ETH NGA BOL GEO BGR KOR .8 ETH RWA NGA BTN COL KEN ECU COL BRA MEX CHL BEL KHM KEN PHL LKA PER CHN CRI JPN IRL PER JPN KHM NAM HUN GBR CHL GBR USA LUX GHA BGR URY ITAFRA USA GHA BGR URY FRA JPN FRA USA PAK PHL POL ESP LUX BRA MEX LUX NER BTN NER IND BTN GBR IND NGA LKA CHE NGA CRI BGR HUN CHE KHM ARM HUN AUT AUS KHM GHA EGY HUN AUT AUS BEN BOL BEN BOL KOR AUT Female LFP Female LFP ECU NAM COL ZAF HRV POL PAK NAM NAMECU COL POL ITA BEN GEO GRC GEO POL EGY BEL IRL BEL BEL IRL AUS CHE .6 PSE BRA SRB CHL KOR .6 BRA CHL HRV KOR PHL ITA PHL ITA HRV ESP GEO HRV ESP IRL ARMCHN ARM CHN ESP CRI CRI SRB GRC JOR MEX SRB MEX ZAF SRB ARM ZAF GRC GRC LKA LKA .4 IND .4 IND CHN JOR PSE PAK PAK EGY Baseline EGY Baseline .2 .2 PSE No Demand Distortions (τ) PSE No Supply Distortions (z) JOR JOR 1000 3000 10000 30000 100000 1000 3000 10000 30000 100000 GDP per capita GDP per capital Notes: Panel A (Panel B) shows female labor force participation (FLFP) in the baseline calibration and in an economy with no demand (no supply) distortions for all countries, plotted against their GDP per capita. The size of points are proportional to the country’s population. The sample for the plot is all country observations near the year 2014. B-8 F IGURE B-5: T HE GGDI AND THE CHANGE IN GDP ( A ) Demand Distortions ( B ) Supply Distortions EGY 20 20 Change in total output (GGDI) Change in total output (GGDI) PAK PSE Change in measured GDP EGY Change in measured GDP % change in measured GDP and total output % change in measured GDP and total output EGY PAK JOR 15 EGY 15 PSE IND PAK JOR 10 PAK LKA IND JOR JOR LKA 10 LKA IND ECU PSE IND ETH LKA ECU NER ETH RWA 5 NER PSE PHL BOL BTN MEX NER RWA BEN BOL PHL BTN NER PER COL CRI MEX MEX PER CHL BOL ETH BEN COL CRI BRA NGA CHL CRI ETH KHM NGA URY BOL BTN PER ECU GHA BRA URY GRC ITA KOR 5 UGA UGA GHA PHL ECU PER BRA MEX GRC KOR RWA RWA KHM KEN KEN GHA NAM NAM SRB GRC HRV HUN KOR ITA LUX CHN UGA SRB BEL CAN USA LUX KEN PHL BTN GEO CRI CHL ITA IRL 0 ZAF BGR HUN HRV POL POL FRA ESP JPN ESP AUT GBR AUS IRL CHE BEN GHA ZAF KOR ESP CHN GEO BGR GBR KHM KEN NGA NAM COL BRA ARM BEL JPN AUS ARM GEO AUS IRL CHE NGA SRB URY CHL GRC ITA GBR AUT CHE ARM KHM NAM GEOCOL JPN IRL USA CHN URY HRV FRA ZAF POL HUN ESP BEL AUS AUTUSA CHE LUX CHN SRB BGR GBRCAN ARM HUN HRV POL FRA CAN LUX BGR 0 -5 1000 3000 10000 30000 100000 1000 3000 10000 30000 100000 GDP per capita GDP per capita Notes: The figure plots GGDI and the change in measured GDP for all country observations near the year 2014 against their GDP per capita. GGDI measure the increase in total economic activity (including home production) due to the removal of labor demand (panel A) and labor supply (panel B) gender distortions. Measured GDP is the increase in GDP following the removal of labor demand (panel A) and labor supply (panel B) gender distortions. APPENDIX C: REPLICATION PACKAGE C-1. R EPLICATION PACKAGE The GGDI replication code is written in MATLAB. It consists of two main functions distortions() and equilibrium(). For a given data set, distortions() provide the magnitudes of the distortions of labor demand and labor supply implied by the model. For a given set of distortion parameters, equilibrium() provides the level of output and sectoral labor allocations implied by the model. This can be used to estimate the counterfactual output and labor allocations. We describe both functions, along with the format of input and output vectors in detail below. C-1.1 distortions() [tau,z,parameters] = distortions(data,theta,sigma) The function distortions() takes 3 inputs. data is a ‘table’ object with observations at the level of country x year. It should have the following variables: Country name, year, sectoral participation rates ( p jg ), sectoral earnings (wage jg ), sectoral human capital (h jg ), and gender proportions in the economy (q g ) where j ∈ {h, u, s, w} (home, unpaid, self-emplyed, wage work) and g ∈ {w, m} (women, men). These variables should have the following column titles in the table: country,year,p_hm,p_um,p_sm,p_wm,p_hw,p_uw,p_sw,p_ww,wage_hm,wage_um, wage_sm,wage_wm,wage_uw,wage_sw,wage_ww,h_hm,h_um,h_sm,h_wm,h_hw,h_uw, h_sw,h_ww,q_m,q_w A recommended approach is to have a dataset with the above variable names in .csv file and import it in ‘table’ format in MATLAB using readtable() function. As mentioned in the paper, we use theta=1.5 and sigma = 3 throughout our analysis. The function distortions() gives 3 outputs. tau and z are arrays of dimension s x 2 x 4 which represent country-year x group x C-1 occupation observations in the order country-year x {w, m} x {h, u, s, w}. For example, distortions related to women in the white collar sector are given by \tau (:,1,4) and z(:,1,4) parameters() is a ‘structure’ object which contains the values of other parameters estimated by the model. These include A j , w j , Y (Baseline), p, h, q, θ , γ, σ, country, year. These objects are defined in the paper. p and h are the arrays representing p jg and h jg with the same dimension of s x 2 x 4 representing country-year x group x job-type observations. q represents q g and is of dimension s x 2. C-1.2 equilibrium() output = equilibrium(tau,z,parameters) The function equilibrium takes 3 inputs. tau and z are of same dimensions as above: s x 4 x 2. Including the same arrays as the one given by the output of the function distortion() gives the implied model values for the baseline economy. Alternatively, entering different values of tau and z will provide counterfactual estimates of the output and sectoral labor allocations implied by the model. parameters is the one which is given as output of the distortions() function. output is a ‘structure’ object whose elements are Y_model, w_model and p_model. p_model represents pog whose dimensions are s x 4 x 2 as described above. C-1.3 Example The following code highlights how to calculate counterfactual GDP and labor alloca- tions for the case when labor demand distortions are set to 0. L ISTING 1: MATLAB example clear clc close a l l % Set Parameters theta = 1 . 5 ; sigma = 3 ; % Read Data data = r e a d t a b l e ( ” data . csv ” ) ; % Estimating Distortions [ tau , z , parameters ] = d i s t o r t i o n s ( data , t h e t a , sigma ) ; % S e t t i n g c o u n t e r f a c t u a l l a b o r demand d i s t o r t i o n s t a u c f = tau ; tau cf ( : , 1 , : ) = 0; % Estimating c o u n t e r f a c t u a l output output = e q u i l i b r i u m ( t a u c f , z , parameters ) ; C-2 APPENDIX D D-1. GGDI E STIMATES Tables D-I to D-VIII contain GGDI estimates and implied GDP change for all country- year observations. D-1 TABLE D-I: GGDI Estimates – Part 1 of 9 Armenia Year 2014 2015 2016 2017 2018 2019 GGDI 0.64% 0.64% 0.66% 0.61% 0.71% 1.14% GDP change 1.82% 1.99% 1.74% 1.93% 2.97% 3.10% Australia Year 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 GGDI 1.09% 1.04% 1.12% 1.00% 1.09% 1.10% 1.01% 0.83% 1.04% 1.04% 1.36% 1.08% 1.23% 1.12% 1.05% 0.87% 0.79% GDP change 2.15% 2.08% 2.32% 2.19% 2.04% 1.88% 1.82% 1.59% 1.89% 1.92% 2.37% 1.92% 2.21% 2.15% 1.98% 1.73% 1.72% Austria Year 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 GGDI 1.93% 2.05% 1.91% 1.79% 1.96% 1.24% 1.78% 1.55% 1.72% 1.60% 1.22% 1.08% 1.35% 1.04% 1.07% 1.06% 0.98% GDP change 3.48% 3.64% 3.46% 3.35% 3.23% 2.53% 2.78% 2.69% 2.80% 2.64% 2.41% 2.52% 2.53% 2.16% 1.96% 1.90% 1.71% D-2 Belgium Year 2004 2005 GGDI 1.30% 1.74% GDP change 3.40% 3.28% Benin Year 2010 2015 GGDI 4.79% 6.76% GDP change 5.47% 8.33% Bhutan Year 2018 2019 2020 GGDI 6.20% 8.33% 7.00% GDP change 8.19% 10.53% 8.88% Bolivia Year 2005 2006 2007 2008 2009 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 GGDI 9.61% 8.38% 9.63% 7.51% 7.66% 7.07% 6.80% 6.45% 7.08% 6.80% 5.75% 6.09% 5.99% 5.23% 5.79% GDP change 11.77% 10.90% 11.86% 9.41% 9.42% 8.96% 8.57% 8.36% 8.75% 8.86% 7.82% 8.03% 8.15% 7.14% 8.19% Notes: GGDI shows the % change in total output from removal of gendered labor market distortions. GDP changes shows the associated change in measured GDP, which exclude home sector. TABLE D-II: GGDI Estimates – Part 2 of 9 Brazil Year 2009 2011 2012 2013 2014 2015 GGDI 4.14% 4.25% 4.04% 3.97% 3.74% 3.60% GDP change 6.22% 6.37% 6.04% 6.01% 5.76% 5.74% Bulgaria Year 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 GGDI 0.91% 0.80% 0.72% 0.50% 0.33% 0.34% 0.35% 0.54% 0.61% 0.48% 0.55% 0.55% 0.52% GDP change 1.77% 1.52% 1.39% 1.04% 0.83% 0.71% 0.59% 1.08% 1.22% 0.92% 1.10% 1.08% 1.22% Cambodia Year 2012 2019 GGDI 2.51% 2.55% GDP change 4.28% 3.80% D-3 Canada Year 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 GGDI 1.13% 1.09% 0.98% 0.99% 0.96% 0.88% 0.76% 0.78% 0.84% 0.82% 0.72% 0.75% 0.68% 0.65% 0.63% 0.55% 0.57% GDP change 2.24% 2.14% 1.99% 1.97% 1.91% 1.73% 1.58% 1.57% 1.64% 1.57% 1.40% 1.39% 1.30% 1.26% 1.28% 1.19% 1.17% Canada Year 2014 2015 2017 2018 2019 2020 GGDI 0.59% 0.57% 0.54% 0.43% 0.56% 0.51% GDP change 1.25% 1.30% 1.18% 1.07% 1.18% 1.10% Chile Year 1990 1992 1994 1998 2000 2003 2006 2009 2011 2013 2015 2017 GGDI 8.33% 9.35% 9.38% 6.80% 6.30% 6.11% 5.47% 5.01% 4.40% 3.98% 3.04% 2.87% GDP change 12.12% 12.69% 12.25% 9.98% 9.36% 9.01% 7.99% 7.78% 6.80% 6.22% 5.28% 4.92% China Year 2012 GGDI 1.97% GDP change 3.51% Notes: GGDI shows the % change in total output from removal of gendered labor market distortions. GDP changes shows the associated change in measured GDP, which exclude home sector. TABLE D-III: GGDI Estimates – Part 3 of 9 Colombia Year 2006 2007 2008 2009 2010 2012 2013 2014 2015 2016 2017 2018 2019 GGDI 5.19% 4.45% 5.11% 4.54% 3.62% 3.43% 3.19% 3.08% 3.08% 2.82% 2.90% 2.76% 2.80% GDP change 7.51% 7.07% 7.68% 6.81% 6.24% 5.83% 5.63% 5.33% 5.33% 5.17% 5.35% 5.22% 5.35% Costa Rica Year 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 GGDI 5.95% 5.36% 3.44% 4.04% 4.42% 4.26% 4.44% 4.30% 3.88% 4.34% 3.55% 3.75% 4.14% 3.62% GDP change 8.52% 8.00% 5.94% 6.69% 7.22% 7.17% 7.54% 7.04% 6.67% 6.90% 7.32% 7.02% 6.89% 6.70% Croatia Year 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 GGDI 0.56% 0.69% 0.77% 0.61% 0.63% 0.47% 0.58% 0.74% 0.78% 0.74% 0.83% GDP change 1.98% 1.97% 2.26% 2.03% 2.20% 1.74% 1.83% 2.06% 1.98% 1.80% 2.00% D-4 Ecuador Year 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 GGDI 7.14% 7.19% 6.63% 7.35% 7.53% 7.41% 7.73% 7.72% 7.15% 6.33% 5.78% 6.23% GDP change 8.23% 8.40% 7.89% 8.81% 9.02% 8.90% 9.26% 9.01% 8.34% 7.42% 6.86% 7.40% Egypt, Arab Rep. Year 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 GGDI 23.06% 24.04% 23.79% 22.08% 22.38% 19.53% 22.79% 22.95% 21.36% 19.06% 18.37% GDP change 28.06% 29.32% 29.19% 26.94% 27.54% 26.37% 28.48% 28.90% 27.41% 25.42% 25.29% Ethiopia Year 2005 2013 GGDI 11.09% 8.85% GDP change 12.49% 10.31% France Year 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 GGDI 1.29% 1.24% 1.22% 1.07% 0.96% 1.02% 0.96% 1.04% 1.03% 1.00% 0.88% 0.78% 0.78% 0.74% 0.78% 0.83% 0.80% GDP change 2.51% 2.39% 2.42% 2.20% 2.09% 2.05% 1.96% 1.97% 1.94% 1.91% 1.73% 1.68% 1.63% 1.66% 1.73% 1.73% 1.64% Notes: GGDI shows the % change in total output from removal of gendered labor market distortions. GDP changes shows the associated change in measured GDP, which exclude home sector. TABLE D-IV: GGDI Estimates – Part 4 of 9 Georgia Year 2017 2018 2019 2020 2021 GGDI 1.91% 1.56% 1.62% 2.36% 2.14% GDP change 3.22% 3.45% 3.52% 4.47% 4.52% Ghana Year 1987 1988 1991 1998 2005 2008 2017 GGDI 4.15% 6.70% 8.21% 6.46% 7.34% 5.00% 4.16% GDP change 6.01% 8.97% 8.82% 7.09% 8.47% 6.06% 5.78% Greece Year 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 GGDI 5.51% 5.00% 4.71% 4.30% 3.93% 3.36% 2.53% 2.55% 2.72% 2.29% 2.96% 2.33% 2.47% 2.36% 2.72% 2.67% 2.80% GDP change 7.49% 7.13% 7.20% 6.69% 6.06% 5.25% 4.21% 4.90% 4.99% 4.71% 5.51% 4.64% 4.78% 4.73% 5.08% 4.90% 5.14% D-5 Hungary Year 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 GGDI 1.44% 1.17% 1.11% 1.15% 0.91% 0.90% 0.83% 0.99% 0.75% 0.82% 0.66% 0.84% 0.40% 0.51% 0.75% GDP change 2.85% 2.39% 2.42% 2.45% 2.07% 1.99% 2.03% 2.14% 1.62% 1.71% 1.38% 1.56% 0.83% 1.26% 1.44% India Year 1987 1999 2004 2005 2006 2007 2011 2017 2018 2019 2020 2021 2022 GGDI 10.52% 10.28% 11.68% 10.55% 11.37% 12.39% 12.77% 14.06% 14.24% 13.39% 11.90% 12.15% 12.35% GDP change 12.31% 11.77% 13.46% 11.96% 13.09% 14.18% 14.73% 17.24% 17.61% 16.40% 15.40% 15.33% 15.04% Ireland Year 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 GGDI 3.31% 3.37% 3.85% 3.24% 3.23% 3.16% 2.67% 1.92% 1.64% 1.73% 1.85% 2.09% 2.15% 1.95% 1.76% 1.55% GDP change 5.22% 4.94% 5.13% 4.36% 4.46% 4.47% 3.79% 3.40% 2.70% 2.71% 2.96% 3.35% 3.49% 3.23% 2.85% 2.73% Italy Year 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 GGDI 3.61% 3.56% 3.69% 3.82% 3.62% 3.29% 3.18% 3.04% 3.08% 2.67% 2.85% 2.61% 2.75% 2.69% 3.17% 2.91% 2.43% GDP change 6.16% 6.01% 6.05% 6.12% 5.86% 5.74% 5.57% 5.21% 5.10% 4.76% 4.76% 4.76% 4.83% 4.81% 5.07% 4.66% 4.42% Notes: GGDI shows the % change in total output from removal of gendered labor market distortions. GDP changes shows the associated change in measured GDP, which exclude home sector. TABLE D-V: GGDI Estimates – Part 5 of 9 Japan Year 1997 2002 2007 2017 GGDI 4.15% 3.30% 2.98% 1.92% GDP change 5.28% 4.61% 4.00% 2.66% Jordan Year 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2016 GGDI 18.65% 18.74% 15.78% 16.16% 16.38% 15.46% 14.88% 14.51% 14.79% 14.59% 9.55% GDP change 27.93% 27.80% 23.68% 23.92% 23.47% 22.58% 21.76% 21.49% 22.48% 23.13% 17.26% Kenya Year 2019 2020 2021 GGDI 3.32% 4.54% 4.10% GDP change 4.19% 5.69% 5.10% D-6 Korea, Rep. Year 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2011 2012 2013 2014 2015 GGDI 3.92% 4.47% 3.94% 4.73% 3.96% 4.92% 4.69% 4.35% 4.14% 3.99% 3.91% 4.50% 4.11% 3.83% 3.23% 3.66% 2.88% GDP change 7.98% 7.88% 8.17% 7.86% 7.12% 7.53% 7.52% 7.35% 6.82% 6.48% 6.32% 6.56% 5.84% 5.63% 4.97% 5.54% 4.84% Korea, Rep. Year 2016 2017 2018 GGDI 2.04% 2.54% 2.09% GDP change 4.10% 4.24% 3.72% Luxembourg Year 2012 2014 2015 GGDI 1.06% 0.82% 0.63% GDP change 2.12% 1.87% 1.61% Namibia Year 2012 2013 2014 2016 2018 GGDI 2.88% 2.94% 2.60% 2.18% 1.89% GDP change 3.98% 3.84% 3.71% 3.25% 2.78% Notes: GGDI shows the % change in total output from removal of gendered labor market distortions. GDP changes shows the associated change in measured GDP, which exclude home sector. TABLE D-VI: GGDI Estimates – Part 6 of 9 Niger Year 2011 2014 GGDI 5.62% 9.32% GDP change 6.15% 11.34% Nigeria Year 2010 2012 2015 2018 GGDI 5.43% 5.27% 3.35% 3.54% GDP change 6.89% 7.25% 4.41% 4.36% Pakistan Year 2010 2011 2013 2014 2015 2018 GGDI 23.16% 22.05% 21.71% 20.05% 19.57% 21.11% GDP change 27.28% 26.06% 25.89% 23.83% 23.20% 25.12% D-7 Palestine Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 GGDI 18.19% 16.09% 14.73% 15.03% 15.66% 17.44% 16.69% 16.94% 15.55% 16.18% 17.79% 19.16% 18.04% 18.44% 16.33% 16.72% 18.00% GDP change 26.79% 25.18% 23.97% 23.64% 23.43% 24.74% 23.21% 23.18% 22.46% 22.87% 24.16% 24.83% 24.33% 24.36% 22.81% 23.44% 24.50% Peru Year 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 GGDI 6.92% 6.22% 6.75% 5.88% 5.62% 5.64% 5.85% 5.62% 5.84% 5.37% 5.45% 4.88% 5.09% GDP change 8.15% 7.56% 7.94% 6.99% 6.81% 6.76% 7.05% 6.77% 7.02% 6.52% 6.51% 5.82% 6.03% Philippines Year 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 GGDI 6.04% 5.83% 5.78% 5.87% 5.61% 5.43% 5.32% 5.67% 5.59% 5.53% 5.38% 6.61% 6.61% 6.01% 5.58% GDP change 8.39% 8.15% 8.06% 8.22% 7.96% 7.73% 7.60% 8.02% 7.88% 7.77% 7.65% 8.87% 9.22% 8.44% 7.75% Poland Year 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 GGDI 0.85% 0.84% 0.93% 1.07% 0.81% 0.66% 0.71% 0.77% 0.64% 0.48% 0.44% 0.42% 0.44% 0.33% 0.16% GDP change 2.30% 2.37% 2.40% 2.40% 2.02% 1.79% 1.69% 1.81% 1.54% 1.17% 1.26% 1.20% 1.12% 0.83% 0.23% Notes: GGDI shows the % change in total output from removal of gendered labor market distortions. GDP changes shows the associated change in measured GDP, which exclude home sector. TABLE D-VII: GGDI Estimates – Part 7 of 9 Rwanda Year 2000 2005 2013 2016 GGDI 9.30% 11.00% 8.42% 6.87% GDP change 9.32% 11.16% 9.05% 7.30% Serbia Year 2013 2014 2015 2016 2017 2018 2019 2020 GGDI 1.36% 1.16% 1.00% 0.68% 0.83% 0.95% 1.19% 1.13% GDP change 3.37% 3.06% 2.85% 2.48% 2.49% 2.73% 2.86% 2.81% South Africa Year 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 GGDI 2.41% 2.23% 2.16% 1.72% 1.63% 1.59% 1.66% 1.38% 1.23% 1.35% GDP change 4.14% 3.91% 3.73% 3.32% 3.32% 3.31% 3.42% 3.02% 2.95% 3.03% D-8 Spain Year 2004 2005 2006 2007 2008 2009 2010 2011 2012 GGDI 2.56% 3.53% 2.92% 2.74% 1.92% 1.37% 1.29% 1.43% 1.46% GDP change 5.16% 5.89% 5.02% 4.65% 3.42% 2.80% 2.76% 3.05% 3.08% Sri Lanka Year 1996 1997 1998 1999 2001 2002 2003 2004 2011 2012 2013 2014 2015 2016 2017 2018 2019 GGDI 8.60% 9.50% 9.48% 10.03% 10.50% 11.29% 10.84% 11.09% 12.03% 11.49% 11.10% 11.58% 11.45% 11.66% 10.66% 12.32% 11.75% GDP change 12.56% 13.43% 12.89% 13.55% 14.34% 14.70% 14.34% 14.77% 15.29% 15.18% 14.30% 15.07% 14.73% 14.52% 13.43% 14.70% 14.32% Sri Lanka Year 2020 2021 2022 GGDI 12.67% 11.85% 12.32% GDP change 15.22% 14.34% 14.62% Switzerland Year 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 GGDI 1.81% 2.25% 0.93% 0.89% 1.03% 1.08% 1.11% 0.92% 0.98% 0.81% 0.52% 0.45% 0.59% 0.58% GDP change 3.23% 3.80% 1.94% 1.77% 2.02% 1.92% 2.12% 1.88% 1.94% 1.77% 1.21% 1.00% 0.92% 0.90% Notes: GGDI shows the % change in total output from removal of gendered labor market distortions. GDP changes shows the associated change in measured GDP, which exclude home sector. TABLE D-VIII: GGDI Estimates – Part 8 of 9 Uganda Year 2009 2010 2011 2013 2016 2018 2019 GGDI 7.34% 6.49% 6.32% 4.99% 4.81% 4.60% 5.49% GDP change 7.85% 7.45% 6.76% 5.34% 5.36% 5.37% 5.98% United Kingdom Year 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 GGDI 0.65% 0.65% 0.43% 0.47% 1.20% 1.43% 1.30% 0.94% 1.19% 1.00% 1.02% 1.13% 1.23% 0.73% GDP change 1.32% 1.30% 1.19% 0.77% 1.98% 2.18% 2.06% 1.85% 2.01% 1.85% 1.96% 2.02% 1.73% 1.52% United States Year 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 GGDI 7.32% 6.95% 6.68% 6.57% 6.46% 6.22% 5.58% 5.13% 4.80% 4.52% 4.19% 3.88% 3.49% 3.24% 3.16% 3.04% 2.60% GDP change 8.69% 8.35% 8.10% 7.97% 7.92% 7.72% 7.04% 6.62% 6.48% 6.14% 5.76% 5.31% 4.80% 4.56% 4.43% 4.27% 3.80% D-9 United States Year 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 GGDI 2.48% 2.44% 2.38% 2.15% 1.96% 2.08% 2.04% 1.88% 1.81% 1.43% 1.38% 1.34% 1.43% 1.40% 1.39% 1.18% 1.23% GDP change 3.63% 3.47% 3.36% 3.07% 2.81% 2.94% 2.90% 2.74% 2.67% 2.29% 2.18% 2.20% 2.25% 2.20% 2.15% 1.92% 2.03% United States Year 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 GGDI 1.23% 1.23% 1.20% 1.32% 1.42% 1.34% 1.31% 1.09% 1.05% 1.14% 1.01% 1.13% 1.11% 1.18% 1.10% 1.04% 1.10% GDP change 2.03% 2.11% 2.08% 2.25% 2.28% 2.23% 2.12% 1.96% 1.98% 2.06% 1.98% 2.07% 2.07% 2.20% 2.06% 1.97% 2.06% United States Year 2018 2019 2020 2021 2022 GGDI 1.13% 0.92% 0.91% 0.90% 0.88% GDP change 2.03% 1.71% 1.67% 1.71% 1.67% Uruguay Year 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 GGDI 3.86% 3.62% 3.22% 3.15% 2.87% 2.52% 2.38% 2.49% 2.29% 1.91% 1.95% 1.76% GDP change 5.48% 5.17% 4.61% 4.54% 4.18% 3.75% 3.63% 3.76% 3.53% 3.17% 3.17% 2.95% Notes: GGDI shows the % change in total output from removal of gendered labor market distortions. GDP changes shows the associated change in measured GDP, which exclude home sector.