Maternal Mortality and Female Life Expectancy: The Importance of Gender Inequality

∗∗

Sonia Bhalotra

Damian Clarke

Joseph Gomes

University of Essex∗

University of Oxford†

University of Essex‡

Atheendar Venkataramani Massachusetts General Hospital§

June 15, 2015

Abstract We test the hypothesis that maternal mortality decline is slowed by gender prejudice. Using plausibly exogenous variation in gender prejudice across cohorts within countries, we find evidence consistent with the hypothesis. For instance, moving from a low son preference country to a country where the desired ratio of sons to daughters is a standard deviation higher reduces the standard deviation of the female advantage in life expectancy over men by 11% and leads to 89 additional maternal deaths per 100,000 live births. It also reduces the survival advantage of girls over boys by 62%. A one s.d. increase in a measure of women’s political rights leads to a decrease of about 5.4% of the s.d. of maternal deaths in the country. A one s.d. increase in a composite gender intensity of language measure leads to a 41% reduction in the s.d. of female life expectancy advantage and a 14% increase of the s.d. of maternal deaths. These measures of gender prejudice have no effects on TB infection rates. TB is a gender neutral illness and hence serves as a placebo. Finally, using historical data from the USA, we show that this phenomenon can be partially explained by the fact that areas with higher levels of gender equality are less willing (or able) to apply available technologies to improve female-specific health outcomes. ∗

ISER & Department of Economics, University of Essex, Wivenhoe Park, Colchster CO4 3SQ, United Kingdom, e-mail: [email protected] † ‡

Department of Economics, University of Oxford, UK e-mail:

ISER, University of Essex, [email protected] § ∗∗

Wivenhoe Park,

Masachusstes General Hospital, USA e-mail:

[email protected]

Colchster CO4 3SQ, United Kingdom,

e-mail:

[email protected]

We are grateful to all seminar participants in the FRG seminar at ISER, Essex, NEUDC 2014 conference in Boston University, Growth Conference in ISI Delhi Dec 2014, and RES 2015 in Manchester, for their comments and feedback.

1

Introduction ‘‘In some regions of the world inequality between women and men directly involves matters of life and death, and takes the brutal form of unusually high mortality rates for women...” Sen (2001)

Maternal mortality rates (MMR) have fallen sharply in the last decade but still remain unnecessarily high, at around 800 deaths a day. In 2013 alone, around 289,000 women died due to pregnancy or child birth related complications (WHO, 2014). Almost all these deaths (99%) are concentrated entirely in developing countries. The average MMR in low income countries in 2010 was estimated to be 452 deaths per 100,000 births, similar to the rates that prevailed in England and Wales in 1930 before the introduction of antibiotics.1 This is pertinent given that 40-50% of maternal deaths are on account of post-partum puerperal sepsis which is treatable with antibiotics. Infant mortality, which is also largely determined by infectious disease, claimed policy attention much earlier and, apparently with more commitment. We investigate whether the relative neglect of maternal mortality stems from gender prejudice: MMR may be low on a list of global health priorities because it is woman-specific. Although there appears to be no systematic test of this hypothesis, it is consistent with some anecdotal evidence. For instance, in the year 2000, MMR in India was 390 and women’s life expectancy advantage was 0.59 years. In stunning contrast to this, MMR in Brazil was 84 and womens LE advantage was 6.1 years. Brazil adopted the Right to Health and implemented Universal Health Coverage and an emphasis on women’s health ahead of other poor countries. We show that female and male life expectancy each exhibit a snug linear fit to income in country*year data, but that the ratio of female to male life expectancy exhibits a weaker relationship with income. In the OECD the average female advantage in life expectancy during 1960-2011 is 6 years. In sub-Saharan-Africa it is 2-3 years and in South Asia it is close to zero. Overall, there is massive variation in MMR (and the gender gap in life expectancy) across countries and years conditional upon income. Our hypothesis is that this variation is a function of gender prejudice. The challenge in testing this is to find changes in gender inequality that occur exogenously 1

452 per 100,000 is the average MMR for the 35 low income countries (World Bank classification). The MMR for England and Wales was 440, for Denmark it was 380 and for the US it was 673 in the year 1930 (Loudon, 1992).

2

i.e. independently of factors that may directly influence changes in MMR. In this paper we use two approaches. First, we use cohort-country variation in (a) stated son preference from crosscountry fertility surveys and (b) institutionalized political, economic and social rights of women, relying upon the variation within countries over time to avoid the problem that differences in gender prejudice across countries may be correlated with all sorts of other cultural and economic factors that predict MMR but not necessarily through prejudice. Second, we rest upon the premise that language structure embeds gender differentiation (Givati and Troiano, 2012), and use variation in language at birth to proxy deep-set (centuries old) gender attitudes. We find, for each of the three measures of gender inequality that we use, that MMR is increasing in gender inequality.2 Reducing maternal mortality (improving maternal health) is a Millennium Development goal (MDG. 5).3 If we compare this MDG to the other equally important MDG of reducing child mortality (MDG. 4) we find a striking contrast.4 While infant mortality rates (IMR) have been falling steadily in the last few decades, there is widespread perception in the literature that progress with MMR has been slow (and non-existent in some countries) till about the 1990s.5 In fact, international policy initiatives to reduce maternal mortality began as late as 1987 with the Safe Motherhood Initiative and international commitment towards maternal health was further strengthened by the 1994 International Conference on Population and Development in 1994 (Hogan et al., 2010). Maternal mortality was further accorded more importance by being chosen as one of the eight MDGs. Some estimates suggest that MMR rates were declining very slowly even in the 1990s and only started falling sharply post 2003 i.e. three years after becoming an MDG (Kassebaum et al., 2014). However, even after this there is a widespread perception that little progress was being made in reducing MMR which led to the launch of the Every Woman Every Child in 2010 initiative of the UN Secretary General and the creation of the Commission on Information and Accountability for Womens and Childrens Health (Kassebaum et al., 2014). 2

In previous work Bhalotra and Clarke (2013) show that exogenous increases in women’s education created by education programs are associated with large declines in MMR. 3

Target 5.A. Reduce by three quarters, between 1990 and 2015, the maternal mortality ratio; Target 5.B. Achieve, by 2015, universal access to reproductive health. See http://www.who.int/topics/millennium_ development_goals/maternal_health/en 4

Target 4.A: Reduce by two-thirds, between 1990 and 2015, the under-five mortality rate. See http://www. who.int/topics/millennium_development_goals/child_mortality/en/ 5

See Hogan et al. (2010) and references therein.

3

In this paper we argue that as MMR is a woman specific condition, public policy attention directed at MMR, and, accordingly, differences in life expectancy between women and men across countries are a reflection of differences in gender inequality across countries and time. Using data on health and socio-economic variables from myriad sources, and different measures of gender attitudes, we show that gender inequalities in health can be traced to differences in gender attitudes and women’s empowerment in society, conditional upon GDP. Furthermore we argue that these MMR differences across countries translate into excess female mortality in the reproductive ages and hence reduces life expectancy rates of women compared to men. While typically women tend to enjoy a life expectancy advantage over men in the developed countries, studying trends in life expectancy differences across regions throw up some striking contrasts. While in OECD countries women live longer than men by around six years, in sub-Saharan Africa women enjoy a life expectancy advantage over men by only two to three years and even less than that in South Asia. In fact, even in the 1970s women actually had a life expectancy disadvantage over men in South Asia. We focus on three specific measures of gender bias in society: First, using individual level micro data from the Demographic and Health Surveys (DHS), we construct both individual level and time varying country specific measures of stated son preference among mothers. Then we make use of a previously under-exploited dataset on women’s political, economic and social rights (as well as actual women’s parliamentary representation from the WDI), to measure women’s status in society. Finally, we exploit insights from some recent papers which argue that grammatical gender can reflect gender attitudes in society and are correlated with different gender outcomes across countries (Givati and Troiano, 2012; Gay et al., 2013), and use the gender intensity in language grammar as an exogenous measure of gender bias in society. We use two primary measures of gender inequality in health outcomes viz. the life expectancy advantage of women relative to men and the maternal mortality ratio. We also use age and gender specific mortality data from the mortality tables of the UNDP. Finally, we use data on cross country tuberculosis (TB) infection rates as a placebo to compare with MMR rates. We use cross country panel data identification relying on country and time fixed effects to purge relevant unobservables. We then extend the country fixed effects framework allowing for time varying patterns of unobserved heterogeneity that are common within groups of countries following Bonhomme and Manresa (2012). We have several interesting findings. First, we show that while there is a systematic 4

positive relation between female (and male) life expectancy with GDP (which is well documented in the literature), female life expectancy advantage has a much weaker relationship with GDP. This implies that there is more to female-male differences in life expectancy across countries than just income differences. Next, we show that MMR can explain part of the excess female mortality in reproductive ages especially in low and middle income countries. We find that a one standard deviation increase in MMR increases the excess female mortality in the reproductive ages by 35.78% standard deviations for low income countries. Then we go on to show that different measures of gender prejudice always significantly increase MMR and reduce the female life expectancy advantage. For instance, let us consider the stated son preference variable. This variable measures the mother’s desired sex ratio/composition of her children. A desired sex ratio of one implies gender neutral child preferences, while a desired sex ratio of greater than one implies son preference and the magnitude of the measure gives us the degree of son preference. As expected, South Asian countries like India have a very high degree of son preference. Pakistan (1.59), Nepal (1.42) and India (1.33) occupy 3 out of the top 5 spots in terms of stated son preference. Not surprisingly they also have very high rates of MMR and a lot fewer years of female advantage in life expectancy compared to other countries. We find that a one s.d. increase in the stated son preference of the mothers, which is roughly the difference in the stated son preference variable between Ethiopia/Bangladesh and India, knocks off around 0.38 standard deviations of the relative female advantage in life expectancy and leads to 92 additional maternal deaths. Moreover, in our overall sample, female infants have a 1.4% lower probability of dying than their male counterparts in case the mother had gender neutral child preferences. A one s.d. increase in the son preference of mothers knocks off 62% of the girl child’s survival advantage over a boy child. There are several ways in which son preference can lead to higher higher maternal mortality and morbidity. For instance, (Milazzo, 2014) shows how in order to achieve the desired sex composition of children, families in India indulge in sex selective abortion and/or son preferring fertility behaviour (repeated and closely spaced pregnancies), both of which have negative health consequences for mothers. She shows that women who change their fertility behaviour in response to son preference after having a first born girl child are more likely to develop anaemia more than a year after birth. Moreover, son preference also leads to increased domestic violence against women close to the time of birth which also leads to higher mortality. 5

Anderson and Ray (2010) undertake an accounting exercise and provide a decomposition of missing women by age and cause of death in India, China and sub-Saharan Africa. One of their main findings is that the vast majority of missing women in India and sub-Saharan Africa are of adult age. In particular they show that maternal deaths account for more than 130,000 excess female deaths in India, and around 226,000 excess female deaths in sub-Saharan Africa.6 They also find that in India the excess female deaths is also very high from a sinister category called “Injuries”, which leads to more than 225,000 combining all age categories and around 118,000 of these excess female deaths are concentrated in the reproductive age category of 15-44. This might be indicative of violence against women, and son preference might be a contributing factor (This is related to the findings of (Milazzo, 2014) for example). Apart from the stated son preference variable, we also find that our other measures of gender bias in society including, female political, economic and social rights, female representation in the parliament and the gender intensity of language grammar systematically increase maternal mortality rates and reduce the female advantage in life expectancy. On the other hand, these measures of gender prejudice have no effects on cross-country TB infection rates.7 Since TB is a gender neutral infectious disease, it serves as our placebo. The women’s political rights variable for instance, takes into account women’s rights to vote, to run for political office, to hold elected and appointed government positions, to join political parties, and to petition government officials. This variable takes discrete values between 0 and 3, with 3 representing high rights and 0 representing very low rights. In the year 2000, of the 154 countries for which we had data, only 7 countries were in the 0 rights group. This included countries like Afghanistan, Saudi Arabia, Kuwait and UAE. 18 countries were in the category 1, including Pakistan and a host of other Middle Eastern countries but not only (Russia and Bhutan also fall in this category). The majority of the countries (119), were in the category 2. And finally, 10 countries were in category 3 including the Scandinavian countries and countries like Canada, Germany and New Zealand. South Africa is an interesting case since it moved up from value 1 in 1990 to taking value 3 in 2000. We find that a one standard deviation increase in women’s political rights leads to a 0.05 s.d. fall in maternal deaths in the sample. We have several different measures of gender intensity of language grammar, which take 6

These maternal deaths are concentrated entirely in the reproductive age category of 15-44 years

7

Except for Women’s parliamentary representation which significantly reduces TB infection rates as well. Following, a well established previous literature we argue that women in parliament have a positive impact on healthcare in general. See Bhalotra and Clots-Figueras (2014) for example.

6

into account factors like number of genders, number of gendered pronouns, whether the gender system is sex based and the gender assignment system. For instance, the gender differentiated personal pronoun measure makes a distinction between languages on the basis of how many gendered pronouns they have. English for example has only one gender distinction in pronouns - “He” and “She”, whereas Spanish has four - “El” and “Ella” (“he” and “she”), “Ellos” and “Ellas” (“they” masculine and feminine), “Nosotros” and “Nosotras” (“We” masculine and feminine), “Vosotros” and “Vosotras” (“You” plural, masculine and feminine). Hence according to this measure Spanish is more gender intensive than English. We find that a one standard deviation increase in our composite gender intensity of language measure leads to a 0.41 s.d. decrease in the female life expectancy advantage and around 0.14 s.d. increase in maternal deaths in the sample. Finally, using a case study from the 1930s US,we throw light on a likely mechanism of how gender prejudice might affect maternal mortality rates. In particular we test and show that the US states which were early to grant suffrage to women experienced higher drops in MMR once the first antibiotics arrived in the year 1937 compared to other states. However, no such difference exists in pneumonia rates across early and late suffrage states. Using a difference in difference approach following Jayachandran et al. (2010), we econometrically show that while early suffrage states had significantly larger trend and level breaks for maternal mortality, for pneumonia mortality (for which there is a trend break only), we do not see any difference post the arrival of the antibiotics. Using this case study we argue that states which were early to legislate suffrage were more likely to adopt medical technologies which could be employed to directly reduce maternal mortality rates. We contribute to the literature that investigates the phenomenon of “missing women” which was first pointed out by Sen (1990) in his now classic article. While it is extremely hard to put precise figures to the exact numbers of women who are missing, the estimates range from around 60 million (Coale, 1991) to more than 100 million (Sen, 1990). Recent estimates from the World Development Report, show that around 6 million women are missing in the world every year (Wong, 2012), of which 21 percent are in their reproductive years. In fact, “other than pre-birth and in early childhood, women are most likely to be missing relative to men in childbearing years” (Duflo, 2011).8 This is in line with the findings of (Anderson and Ray, 2010, 2012) which establishes that unlike previously believed, the bulk of the excess female mortality 8 23 percent are never born, 10 percent are missing in early childhood, 21 percent in the reproductive years and 38 percent above the age of 60 (Duflo, 2011)

7

is not confined at birth, infancy and early childhood, but occurs at older ages. We extend this line of thought and show that there is excess female mortality in reproductive ages and high maternal mortality rates contributes to it. Moreover, we take a step further and establish that maternal mortality is systematically correlated with variation in gender inequality conditional upon income. Initially, the estimates of missing women were meant to represent gender discrimination (Anderson and Ray, 2010). However, Anderson and Ray (2010) underscore how the reasons for missing women is complex and gender discrimination is one possible channel. They leave the separation of role of direct gender discrimination from other factors including biological, social, environmental, behavioural, or economic, in explaining excess female mortality as future research. We on the other hand, take this forward and establish that negative gender attitudes are indeed correlated with missing women in older ages through higher maternal mortality rates. Ours is the first paper to establish for a wide set of countries, how excess mortality among older women and in particular maternal mortality can be related to gender prejudice. In a related paper, Milazzo (2014), shows how high son preference leads to higher maternal mortality and morbidity in India. We, extend this to a wide set of countries and establish that son preference indeed plays a role in explaining differences in MMR rates across these countries. We not only use micro data based estimates of the actual stated son preference of mothers, but extend the study to incorporate different measures of gender attitudes/discrimination and show how these are consistently correlated to high MMR rates and differences in life expectancy advantage between women and men across countries. Moreover, in our specifications analyzing the role of stated son preference on MMR, we often explicitly control for desired fertility and still find son preference to matter for MMR. This shows that while there is a possible fertility channel at play via which gender prejudice leads to higher MMR, there is also possibly a policy channel as well. Further, our finding that maternal mortality rates, and female life expectancy advantage are significantly related to different measures gender prejudice in society over and above income differences across societies, shows that income by itself is insufficient in explaining cross country differences in gender unequal health outcomes. This is in line with the recent findings of an interesting analysis by Jayachandran (2014b), who argues that poor countries have cultural features that exacerbate gender prejudice and that being poor by itself cannot explain why parents have strong son preference in some countries. We take this literature forward by establishing the relation of high MMR rates and differences in life expectancy advantage across countries to 8

different measures of gender prejudice including son preference, women’s rights, and exogenous measures of gender intensity of languages always controlling for differences in GDP and often controlling for country level unobservables to give a more causal flavour to our findings. There is a broad consensus in the literature that improving population health has implications for economic growth primarily via improvements in life expectancy and human capital accumulation.9 Some recent papers have underscored the importance of female health on different economic outcomes including female literacy (human capital accumulation) and female labour force participation. Albanesi and Olivetti (2009) for instance demonstrate that medical advances in healthcare in the US that led to a huge decline in maternal mortality and increased the female-male differential in life expectancy at age 20 from 1.5 years in 1920 to 6 years in 1960, led to higher female labour force participation. Again this decline in maternal mortality “also had a very strong effect on the growth in women’s educational attainment relative to men. For every standard deviation drop in maternal mortality, the female-male differential in graduation rates rises by 0.017 for college and by 0.046 for high school for the 1933-1950 birth cohorts” (Albanesi and Olivetti, 2014). Similarly, Jayachandran and Lleras-Muney (2008) demonstrate that increases in female life expectancy resulting from decreases in maternal mortality, led to an increase in literacy rates (human capital accumulation) among girls relative to boys in Sri Lanka.10 Lagerl¨ of (2003) highlights the importance of gender equality in general for long run economic growth. Again, Amiri and Gerdtham (2013) and Kirigia et al. (2006) show how MMR might affect growth and GDP. In line with this literature, we argue that closing the gender gap in health can be beneficial for the economy as a whole.11 The rest of the paper is organized as follows. In section 2, we start by documenting the trends in life expectancy and how maternal mortality rates and female life expectancy advantage are closely related to indices of gender inequality. In section 3, we summarize the main methodology used in the paper. In section 4, we estimate the contribution of maternal mortality to excess female mortality rates in the reproductive ages. In section 5, we econometrically 9

(Weil, 2005; Ashraf et al., 2008; Bloom et al., 2004; Shastry and Weil, 2003; Lorentzen et al., 2008; Aghion et al., 2010). Acemoglu and Johnson (2006) however find that exogenous improvement is life expectancy have only modest implications for growth. 10 They point out that since maternal mortality occurs early in life, an averted maternal death translates into a large life expectancy gain for women. 11 Duflo (2011) on the other hand points out the bidirectional relationship between women’s empowerment and development, but argues that a continuous policy commitment to equality for its own sake may be needed to bring about equality between men and women and development by itself need not ensure equality.

9

establish the relation of MMR to the stated son preference variable, women’s political and other rights and measures of gender intensity of language grammar. In section 6 we subject our empirical analysis to different robustness tests including the extension of the standard parametric specification to a recently developed time varying group fixed effects framework, placebo tests with TB infection rates and use of alternative MMR data from the DHS. In section 6 we conclude.

2

Descriptive Statistics

The last five decades have witnessed large and sustained increases in the life expectancy rates at birth for both men and women throughout the world. Typically women have enjoyed a life expectancy advantage over men in almost all countries of the world. However, some interesting cross country differences exist in the data. In particular, there are two striking features that emerge from studying life expectancy advantage of women over men in the developing world. First, when the AIDS epidemic struck Africa in the 1990s, women lost more in terms of life expectancy than men and subsequently their life expectancy advantage suffered a blow. However, with the arrival of anti-retro viral treatment in the mid 2000s onwards, the female advantage in life expectancy started to go up yet again (See Figure 1).12 Second, in contrast to not only all other parts of the world but also to sub-Saharan Africa, women actually started with a life expectancy disadvantage in South Asia (See Figure 2). This is not surprising given that South Asia is know to be more gender prejudiced than other regions of the world. Only since the 1980s did women finally start to enjoy a life expectancy advantage over men and this advantage keeps going up through the years. If we looked at the trends in female life expectancy advantage in the OECD countries (See appendix Figure F.1), we notice that in these countries women live longer than men by around six years, in comparison to around two to three years in sub-Saharan Africa and even less than that in South Asia. While these trends might make us believe that the differences across countries just reflect income differences and nothing more, we argue that these differences reflect more than just income differences. When we plot female life expectancy against GDP in Figure 3 we notice that female life expectancy has a positive relation with GDP regardless of the time period we consider between 1970 and 2010. In fact, a similar picture would arise if we plotted male life expectancy against 12

See http://www.who.int/gender/hiv_aids/en/ for a discussion on gender and AIDS.

10

(a)

(b)

Figure 1: In Panel 1 we plot the Female, Male and Total Life Expectancy against time. In Panel 2, we plot the Female-Male Life Expectancy advantage over time. The Life Expectancy data comes from the World Bank WDI and spans over more than 190 countries and is available for the period of 1960 - 2011. Here we plot them for the sub-Saharan Africa region (World Bank Region Classification is used).

(a)

(b)

Figure 2: In Panel 1 we plot the Female, Male and Total Life Expectancy against time. In Panel 2, we plot the Female-Male Life Expectancy advantage over time. The Life Expectancy data comes from the World Bank WDI and spans over more than 190 countries and is available for the period of 1960 - 2011. Here we plot them for the South Asia region (World Bank Region Classification is used).

GDP.13 This is not surprising and has been well documented in the literature. On the other hand, once we plot the female life expectancy advantage i.e. the log ratio of female to male life expectancy against GDP in Figure 4, we see that relationship is not that strong. This implies that there is more to gender gaps in health outcomes than just income differences across 13

See the scatter plots in ?? for example.

11

countries.

(a)

(b)

(c)

Figure 3: Female Life Expectancy is plotted against log of GDP. The Life Expectancy (and the GDP) data comes from the World Bank WDI and spans over more than 190 countries and is available for the period of 1960 - 2011. Here we plot them for the years 1970, 1990 and 2010.

(a)

(b)

(c)

Figure 4: The log of the ratio of Female to Male Life Expectancy is plotted against log of GDP. The Life Expectancy data comes from the World Bank WDI and spans over more than 190 countries and is available for the period of 1960 - 2011. Here we plot them for the years 1970, 1990 and 2010.

Now, let us look at some of the descriptive statistics on MMR rates. We first point out that MMR rates are very hard to measure and there is a lot of uncertainty regarding how exactly MMR should be measured. While MMR rates have been sharply declining for the World as a whole, there is a fair bit of variance across the world and according to some estimates the decline has been sharp only in the post 2003 period after stagnation (or at least a much slower rate of fall) in the earlier decades. For instance Kassebaum et al. (2014) estimate that “the global annual rate of change in the MMR was -0.3% from 1990 to 2003, and -2.7% from 2003 to 2013”.14 Clearly MMR rates 14

These numbers are in contrast to the numbers from the WDI/WHO, according to which the MMR fell from 320 to 210 between 2000 and 2010 which is a fall of 34.38%, and from 400 to 320 between 1990 and 2000, which represents a fall of 20%. These numbers are for the whole world and they come from the WDI/WHO.

12

have been falling more sharply during the 2000s (See Figure 5 for instance). However, despite the sharp fall in the 2000s only 16 countries (seven of which are developing) in the world are expected to achieve the MDG of reducing MMR by 75% by the year 2015 (Kassebaum et al., 2014).

Figure 5: Global maternal deaths (A) and annualised rate of change in maternal mortality ratio (B), 19902013 Shaded areas show 95% uncertainty intervals. This figure comes from Kassebaum et al. (2014) (Figure 3). We argue that the variance in maternal mortality rates across countries can be attributed to differences in gender inequality across countries. We plot MMR and Female LE advantage against the different measures of gender inequality in the appendix. For example, in Appendix section ?? and ?? we respectively plot Female LE advantage and MMR against Desired Sex Ratio and clearly notice how Female LE advantage is falling in DSR and MMR is increasing in DSR. In the next section we will show how maternal mortality rates themselves contribute to excess female mortality in the reproductive ages and to lower female life expectancy advantage. And then in the rest of the paper we will establish how differences in gender attitudes and 13

the status of women across countries can explain the differences in both the life expectancy advantage of women and MMR rates across countries.

14

3

Methodology

3.1

Gender Preferences and Women’s Health

We test our prevailing hypothesis that gender-biased preferences drive female-specific health outcomes. To do so, we begin by estimating the following regression using the panel data described in the previous section: F emaleHealthit = α + βGenderBiasit + γi + δt + (φi × t) + θXit + εit .

(1)

Here F emaleHealthit is a measure of health stocks or outcomes speficic to women in country i and time t. As discussed in section X.X, female health is measured in two ways: using a femalespecific mortality measure (MMR), and using a relative measure of female health stocks to male health stocks (LE advantage). We include country-specific fixed effects γ, and year specific fixed effects δ while allowing for differential trends in female health measures by country. Standard errors are always clustered at the level of the country. Consistent estimates of β—the effect of gender-biased preferences on female health— requires that the transitory error term εit contains no factors correlated with both our measure of gender bias and female health outcomes. While in our most demanding specifications we include a number of time-varying controls Xit (including levels and growth of GDP), it is unlikely that observable measures will capture all time varying country-specific factors which affect both variables. As such, we view (1) as compelling descriptive evidence, and later employ a number of alternative specifications, methodologies and falsification tests which allow us to reach causal conclusions regarding the effect of gender bias on female health. We describe these tests in the following sections 3.1.1

Measuring Gender Bias

In our most na¨ıve regressions, we use a range of accepted measures of gender bias as our independent variable in (1). Different measures are collected at the level of the individual, or the level of the country, and in each case, vary over time and space. These initial variables are: desired sex ratio; the Cingranelli et al. political, economic and social rights of women measures; and proportional representation of women in national parliaments. We provide additional descriptive statistics and discussion of these variables in section XX. In each case we progressively control for log(GDP), and an interaction between log(GDP) and our independent variable of interest. We thus estimate the effect of gender bias on woman15

specific health outcomes conditional and unconditional on country income levels, to allay concerns that both gender bias and outcomes such as MMR are jointly driven by country income. 3.1.2

Predetermined Measures of Gender Bias

Measures of gender bias used in estimation up to this point are (either directly or indirectly) based on contemporaneous decisions by a society and/or its members. Thus, even controlling for time-varying covariates, trends and fixed effects, endogeneity concerns still exist. Given these concerns, we focus on a measure of gender attitudes which is entirely pre-determined, and arguably entirely exogenous when considering contemporaneous maternal mortality and life expectancy advantage. Following Givati and Troiano (2012) and Gay et al. (2013), we use measures of the inherent gender neutrality (or gender bias) built into the grammatical structure of different languages.15 We estimate: F emaleHealthit = β0 + β1 GIIi + β2 P ercentLangi + Xit + Xi + νit ,

(2)

where F emaleHealthit is identical to that described in model (1). Here, GII, (Gender Intensity Index), refers to the measure of the gender bias inherent in the language, as coded by Givati and Troiano, and Gay et al. Precise details regarding this variable are provided in section X.X and appendix X). Given that the language of country i is essentially fixed over time, we can no longer include country fixed effects, as these capture (among other things) the language spoken in the country. We thus now control for both fixed (Xi ) and time varying (Xit ) country-level variables, including decade dummies, continent dummies, log(GDP), the log of population, religion, income and climate factors. Finally, given that GII is defined based on the majority language in each country, we control for the percentage of inhabitants of each country which speak the particular language. 15

These authors suggest that grammatical gender can influence and reflect gender attitudes in society and are correlated with different gender outcomes including maternity leave policy differences across countries (Givati and Troiano, 2012), female labour force participation, and political participation (Gay et al., 2013). Language grammar was established centuries in the past and is one of the features of language that is stable over long periods of time. Moreover, grammatical gender is something that an individual is born in to and thus arguably a more exogenous measure of gender bias in society.

16

3.1.3

Falsification Tests: Gender Neutral Disease Burden

The analysis discussed so far in this section tests the hypothesis that gender biased preferences slow the improvement of female-specific health outcomes (namely maternal mortality, and relative life expectancy). In order to provide evidence that it is indeed female-specific health outcomes which are driven by gender preferences, we run a series of falsification tests. These falsification tests consist of estimating the previous regressions, however focusing on a genderneutral health outcome. We thus focus on rates of tuberculosis (TB) infection. TB is a gender-neutral condition16 , and was the second most common infectious cause of death world wide in 2014 (WHO, 2015). Given that it affects individuals indiscriminately by gender, it is a particularly suitable placebo for our analysis. Based on this placebo, we estimate analogues to (1) and (2). For example, for (1) we estimate: T Bit = αT B + β T B GenderBiasit + γiT B + δtT B + (φTi B × t) + θT B Xit + εTitB ,

(3)

and if βˆ but not βˆT B is significantly (and economically) significant, this lends support to the hypothesis that gender preferences halt progress on (only) woman-specific conditions.

3.2

Testing Mechanisms

After documenting the effect that gender-biased preferences have on female health outcomes, we turn to testing the mechanisms that drive these results. Using the arrival of sulfanide drugs (the first antibiotics) to the United States in 1937, we test whether states which were early to legislate suffrage were more likely to adopt medical technologies which could be employed to directly reduce rates of maternal mortality. We define early legislators of suffrage as those states which passed suffrage laws prior to the mandated national reform in 1920. Using these two groups of states, we follow Jayachandran et al. (2010) and estimate the following difference-indifferences specification around the 1937 date of arrival of sulf drugs: log(M M R)st = α + β1[P ost1937]t + γ(EarlySufs × t) + δ1 (EarlySuf × P ost1937t ) +δ2 (EarlySuf × P ost1937t × t) + φt + µs + υst .

(4)

In these regressions we always weight states (s) by their population, and cluster standard errors at the level of the state. 16

According to the WHO “In most of the world, more men than women are diagnosed with tuberculosis (TB) and die from it. TB is nevertheless a leading infectious cause of death among women.”

17

By estimating (4), we are able to split the quantitatively important (exogenous) reduction in maternal deaths which occurred with the arrival of the first antibiotics into three components. The first, β, identifies the immediate general effect of sulfanide drugs on MMR, while δ1 and δ2 test whether there are larger level and trend breaks (respectively) in states that were early adopters of women’s suffrage, and which presumably had long-standing norms evolving in favor of women’s autonomy. Sulfa drugs were also important in reducing morbidity and mortality from pneumonia. Unlike maternal mortality, pneumonia mortality affects males more than females, and infants in particular. Thus, as in section 3.1.3, we employ a similar logic, and use rates of pneumonia (for which there is also a post-1937 trend break), as a falsification test. We restimate (4), replacing maternal mortality with pneumonia mortality as our dependent variable. Our placebo test is based on the coefficients δ1 and δ2 . If states with a lower preference for female equality invest less (only) in applying available technologies to female-specific illnesses, then δ1 and δ2 should be negative in (4), but not significant in the gender-neutral pneumonia version of (4) estimated for pneumonia rather than MMR.

4

The contribution of IMR and MMR to gender differences Mortality.

In this section we quantify the contribution of maternal mortality and female to male infant mortality ratio to age specific mortality rates across genders. In previous studies, MMR has been found to contribute to female life expectancy. For example, Jayachandran and LlerasMuney (2008) show how a drop in MMR contributed to increased female life expectancy in Sri Lanka. Again, Canudas-Romo et al. (2014) estimate the contribution of maternal mortality to Reproductive-Aged Life Expectancy (RALE) i.e. life expectancy calculated between ages 15 and 49. They find that over the twentieth century, 5 years of RALE were gained in developed countries and around 10% of this gain, which is approximately half a year, can be attributed to reductions in Maternal mortality. “In sub-Saharan African countries, the possible achievable gains fluctuate between 0.24 and 1.47 years, or 6% and 44% of potential gains in RALE” (Canudas-Romo et al., 2014). In this section we take this forward and establish that higher maternal mortality ratios can explain part of the excess female mortality in reproductive ages, which in turn explains part of the gender gap in life expectancy across countries. We first consider the differences in age

18

specific mortality rates across genders and also the relative contributions of IMR and MMR (if any) to these mortality rates using data from UN mortality tables. In order to do so we consider 3 distinct age classes viz. 0-14, 15-49 and 50+. The 15-49 age group is typically considered the reproductive age class for women and hence this is the class in which most maternal mortality should be concentrated. In order to compare differences between women and men, we construct the ratios of mortality rates of women to men in these three different age categories and use the log values of these ratios as our primary dependent variable.17 Since this data exists for the entire world sample, we are able to analyze differences across countries belonging to the three different income categories of high, middle and low income countries following the World Bank’s income classification. We thus regress the log ratio of female to male mortality rates (times 100,000) by different age categories on maternal mortality and infant mortality ratio for different countries in Table 1. In the three panels of Table 1 we respectively have the low, middle and high income country samples.18 In the odd numbered columns, we control for MMR and the log of GDP. In the even numbered columns we control for MMR, the log of GDP and the IMR ratio (female to male).19 In Table 1, we first notice that for both the low and middle income category countries (i.e. in Panels 1 and 2), MMR significantly increases the excess female mortality for the 15-49 age category which is the reproductive age for women. This implies that some of the excess female to male mortality in the reproductive age is explained by MMR. In other words at least some of the 21% of the 6 million women who are missing in their reproductive age every year around the world can be attributed to maternal mortality.2021 Let us now consider the marginal effects of our MMR variable (which is measured per mortality in age category X Dependent variable = ln( female ) ∗ 100000; where X can be (0-14), (15-49), or, (50 +). male mortality in age category X This data comes from Nations (2013). See appendix for more details. 17

18

See the appendix section A.1 for a list of countries belonging to each of these categories.

19

We include country fixed effects and time dummies in all columns and always use cluster robust standard errors clustered at the country level. 20 On the other hand, MMR significantly reduces excess female mortality in the 15-49 category and increases it in the 50+ category for high income countries. However, since MMR is concentrated solely in poorer countries, we cannot read too much into these numbers for developed countries. 21

Both IMR and MMR are from the WDI and hence for the whole world sample. The IMR is available for 223 countries for the years 1990, 2000, 2010, 2012, while the MMR is available for 181 countries for the years 1990, 1995, 2000, 2005, 2010, and they have common data for 180 countries for the years 1990, 2000, and 2010. The 2 panels represent Low Income and Middle Income countries from the World Bank income group classification. There are 35 low income countries and 95 middle income countries according to this classification. Again the female to male mortality ratios are available for the years 5 yearly from 1960 to 2005 and hence all these variables together are available only for the years 1990 and 2000. Hence there is a fall in observations from Col1 to Col2.

19

Table 1: Gender differences in Mortality rates & MMR: WB Income categories.

Low Income MMR ln GDP

(1) (0-14)

(2) (0-14)

(3) (15-49)

(4) (15-49)

(5) (50+)

(6) (50+)

-4.640 (4.637) -1705.0 (1698.3)

-10.34∗ (5.286) -5545.6∗ (2750.4) -0.0723 (0.851) -5055.781 5670.332 63 0.374

16.45∗∗∗ (4.701) 1928.7 (2209.5)

17.41∗∗∗ (5.284) 3612.3 (3482.2) -1.511 (1.153) 731.178 15193.921 63 0.408

-2.515 (3.886) -325.6 (1421.0)

-6.470 (5.586) -124.4 (2152.0) 0.777 (1.188) 7396.504 9631.565 63 0.218

IMR ratio (F/M) Mean Dep. Var. s.d. Dep. Var. N r2 Middle Income MMR ln GDP

-5345.207 6133.344 126 0.146 -8.961∗ (4.751) 2871.7∗∗ (1261.9)

IMR ratio (F/M) Mean Dep. Var. s.d. Dep. Var. N r2 High Income MMR ln GDP

-5931.644 14029.156 368 0.0717 -4.352 (11.39) 372.1 (3755.8)

IMR ratio (F/M) Mean Dep. Var. s.d. Dep. Var. N r2

-16468.01 25082.707 203 0.0300

-12.31∗∗ (5.769) 1508.4 (2059.6) 0.436 (0.481) -5662.593 13886.466 183 0.0951 -13.64 (13.10) -4565.3 (5793.9) -0.659 (0.718) -17515.445 24215.493 99 0.0378

973.607 15359.475 126 0.334 17.90∗∗ (8.767) -2623.8 (1823.4)

-31405.483 31456.973 368 0.108 -33.32∗∗∗ (7.408) -8484.2∗∗∗ (2630.7)

-70145.069 24225.146 203 0.128

16.08∗∗ (7.762) -5725.8∗∗ (2339.8) 1.066 (0.991) -30533.694 31325.235 183 0.200 -41.46∗∗∗ (4.960) -14227.9∗∗∗ (2540.2) 0.784∗ (0.437) -69730.243 24331.836 99 0.375

7181.443 9544.369 126 0.114 -8.991∗ (4.971) -32.36 (678.2)

9291.782 9007.864 368 0.0537 13.67∗∗ (5.538) 412.4 (1821.9)

8109.057 5983.682 203 0.179

-7.516 (5.191) 801.2 (1003.8) -0.731 (0.490) 9332.058 9256.403 183 0.103 11.69∗∗∗ (2.461) 318.5 (968.7) -0.0154 (0.224) 8218.618 5604.673 99 0.398

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is the log ratio of Female to Male mortality rates (scaled up by 100,000) in the different age groups specified in the column headers. We have used a country fixed effects panel framework with year dummies. Standard errors in parentheses are clustered at the country level. The data on mortality rates come from the United Nations, Department of Economic and Social Affairs, Population Division (2013) and is available for every 5 years from 1960 to 2005. The MMR data comes from the World Bank - WDI (based on WHO data) available for 5 time periods -1990, 1995, 2000, 2005, 2010. The IMR data also comes from the World Bank - WDI (based on WHO data) and is available for 4 time periods 1990, 2000, 2010 and 2012. Hence these regressions are based on the years 1990, 1995, 2000 and 2005 (odd columns) and the years 1990 and 2000 (even columns) and 33 low income countries, 94 middle income countries and 50 high income countries. The mean (s.d.) of the MMR & IMR variables is 30.204 (103.201) & 82470.743 (3542.108) for high income countries, 200.775 (238.522) & 81560.562 (4525.799) for middle income countries, 644.183 (338.514) & 84320.087 (3231.888) for low income countries. All the variables are 5 yearly averages.

100,000 live births throughout the paper) on the log ratio of excess female mortality rates in the reproductive age. In order to do so we consider column 4 as our baseline specification. As far as the marginal effects for low income countries are concerned, a one s.d. (312.3) increase in MMR

20

(measured as per 100,000 live births) leads to a 5.44% (= (17.41*312.3)/100000) increase in the female to male mortality ratio in the reproductive age category. This is equivalent to 744% (0.054/0.0073) of the mean of the female to male mortality in low income countries and around 35.78% of the s.d. of the ratio of female to male mortality in this age category in these countries. For middle income countries on the other hand, a one s.d. (266.332) increase in MMR leads to a 4.29 (= (16.08*266.332)/100000) percentage points increase in the female to male mortality ratio in the reproductive age category. This is equivalent to a 14.03% (0.043/0.305) decrease of the mean ratio of female to male mortality and around 13.67% of the s.d. ratio of female to male mortality in this age category in these countries. The female to male infant mortality ratio on the other hand has no significant effect on the mortality rates in any of the three different age categories.22 In other words, if the average MMR in the low income countries in our sample which is around 451 in the year 2010, came down to the average MMR of the Middle income countries which is around 145 in the same year, then this fall of around 306 in the MMR would lead to a reduction in excess female mortality in the reproductive age by around 5%. Let us now consider some individual country examples to see how different levels of MMR lead to different rates of life expectancy differences between women and men. If we consider India for example, which is a lower middle income country according to the World Bank income classification, the average MMR is 390 and the difference between female and male life expectancy is only of 0.59 years. Again, consider Nepal which also had a very high rate of MMR of 420 with the female advantage being only of 0.86 years. Again in Bangladesh, the MMR is around 466 and the women are actually at a disadvantage of 0.65 years. But, if we consider a country like Brazil which has roughly the same income level (upper middle income country) as India, they have a relatively low MMR rate of 84 and a female advantage in LE of 6.11 years. Similarly Thailand has an MMR of 55.2 and 6.07 years of female advantage.

22

In the appendix Table E.8 we provide a specification for the whole world sample pooled together.

21

5

The Importance of Gender Inequality

5.1

Desired Sex Ratio and Gender inequalities in Health

The Demographic and Health Surveys (DHS), funded by the U.S. Agency for International Development (USAID), make available many variables on the life and health outcomes of individuals in many developing countries from across the world. The data is collected by interviewing a nationally representative sample of women of child bearing age (15-49) in these countries and the standardized components of the DHS can be used to compile micro data sets comparable across the different countries. In particular, the DHS surveys allow us to measure the degree of son preference among women of a child bearing age across many developing countries. Exploiting the DHS data, in this section we first show how the stated son preference of mothers can explain part of the excess female infant mortality in developing countries. Then, using the individual mother level son preference variable we come up with a measure of country level gender bias in society using which we go on to show how countries with gender biased attitudes also have significantly higher maternal mortality ratios. In the DHS surveys, each surveyed woman is asked about the ideal number of children, the ideal number of boys, and the ideal number of girls she would like to have. Using the stated preference of the mothers we construct a variable that gives us the DSR or “Desired sex ratio - (boys/girls)” for each mother. The exact definition of the DSR variable is given by the equation 5. Given the way we construct our variable, DSR measures son preference or in other words, it measures the mother’s bias against having a girl child, which in turn reflects negative attitudes towards women. DSR ≡ Desired Sex ratio ≡

ideal no. of boys ideal no. of girls

(5)

Since our DSR variable directly measures son preference of the mothers, our first variable of interest is infant mortality. In particular we are interested in verifying whether, female children of mothers who have a higher stated son preference also have a higher probability of dying as infants than their male counterparts.23 Biologically speaking, female children actually have a lower probability of dying as infants. However, in many developing countries the opposite is often observed and hence the phenomenon of “Missing women”. Using the DHS data, we show that gender biased attitudes of mothers in favour of sons and against daughters indeed lead to a higher probability of a female infant dying before reaching the age of one. 23

Infant mortality is defined as a child dying before reaching the age of one.

22

Jayachandran (2014a) highlights the endogeneity of desired sex ratios to fertility rates. She shows how desired sex ratio rises sharply as the fertility rate falls. The DHS data allows us to construct a measure of desired fertility and in order to take into account the endogeneity underscored by Jayachandran (2014a), we always include a specification controlling for desired fertility. In Table 2 using child level regressions on a sample of more than 4.6 million child births from around 63 DHS countries, we regress the probability of infant death on the DSR or stated son preference variable and different controls.24 Infant death is a 0-1 dummy variable which takes the value 1 if the child dies before reaching the age of one. In order to run the regressions this variable has been scaled up by 100. In columns 1 & 2 we notice how desired sex ratio of boys to girls has no significant effect on infant mortality. In columns 3 & 4 on the other hand we notice how interacting the desired sex ratio variable with the female child dummy completely changes the picture. There are two terms of interest in the columns 3 & 4, the desired sex ratio term by itself and the interaction between the DSR variable and the female dummy. We notice how the desired sex ratio variable significantly increases the probability of a girl child dying as an infant (and hence reduces the probability of a boy child dying as an infant). In columns 5 &6 we also control for the desired fertility and notice that while desired fertility significantly increases infant deaths, controlling for it does not reduce neither the significance, nor the magnitude of the effect of desired sex ratio on female infant deaths. In terms of marginal effects (considering column 6), we notice that in general female infants have a 1.4% (= −0.0279 + 0.014) lower probability of dying than their male counterparts in case the desired sex ratio was equal to one i.e. in case the mother had gender neutral child preferences. On the other hand a one s.d. (0.608) increase in the son preference of mothers reduces the probability of survival of the female infant relative to a male infant to only 0.52% (= −0.0279 + 0.014 ∗ 1.608). In other words it knocks of around 62% (= 1 − 0.52/1.4) of the girl child’s survival advantage over the boy child. Moreover, this effect is highly significant. As evident from Table 2, stated son preference which reflects gender bias in society is a significant correlate of excess female infant deaths in developing countries. As we have mentioned in the previous sections, around 10% of the six million missing women around the world are missing in early childhood. On the other hand, around 21% are missing in their reproductive ages. We have already established in the previous section that a part of this can 24 These 63 countries are the ones for which the DSR variable is available. The full list of these countries can be found in the appendix.

23

Table 2: Infant mortality and Stated Son Preference

Female Desired Sex Ratio

(1) OLS -1.153∗∗∗ (0.0928) 0.0125 (0.103)

(2) Probit -0.0801∗∗∗ (0.00614) 0.00167 (0.00668)

No 4524542 0.0216

No 4524521 0.040

(3) OLS -2.716∗∗∗ (0.163) -0.551∗∗∗ (0.0997) 1.362∗∗∗ (0.121) No 4524542 0.0218

Desired Sex Ratio*Female Desired Fertility N R2 /pseudoR2

(4) Probit -0.194∗∗∗ (0.00894) -0.0389∗∗∗ (0.00660) 0.0976∗∗∗ (0.00704) No 4524521 0.041

(5) OLS -2.729∗∗∗ (0.163) -0.571∗∗∗ (0.0971) 1.369∗∗∗ (0.121) Yes 4524542 0.0220

(6) Probit -0.195∗∗∗ (0.00900) -0.0400∗∗∗ (0.00647) 0.0981∗∗∗ (0.00707) Yes 4524521 0.041

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is infant mortality (which was initially a 0-1 binary variable, but now has been multiplied by 100). In Panels 1, 3 & 5 we run OLS regressions, while in panels 2, 4 & 6 we run Probit regressions. All specifications include country and year of birth dummies, age at birth and the square of age at birth of the mother, and years of education of the mother. Standard errors in parentheses are clustered at the country level. Marginal effects have been reported for the Probit regressions. This data and hence the regressions are based on the DHS sample (developing countries) of 63 countries over the period of 1955 - 2012. The mean (s.d.) of infant deaths is 8.05 (27.2), about 48.8% of the sample is female, and the mean and s.d. of desired sex ratio is 1.158 (0.608) and that of the interaction term is 0.538 (0.676). And the mean (s.d.) of the desired fertility variable is 4.18 (2.26).

be attributed to high maternal mortality leading to excess female mortality in reproductive ages. Next, we show that gender bias in society measured by the stated son preference can also explain the variances in maternal mortality rates and we will later also show that the stated son prference also reduces the female life expectancy advantage across countries. Our MMR and life expectancy data are all at the country level while the DSR variable is at the individual mother level. In order to arrive at country level values of DSR we average the DSR variable across all mothers in the country and birth cohort (weighted by weights provided in the DHS) to come up with country and birth cohort specific DSR which is in fact the country birth cohort level Son Preference ratio. Then in order to get the country year specific numbers on son preference in the country, we take the average desired sex ratio of all women who are within the age group of 15-25 in a particular year. Do do this under the assumption that most women are likely to become mothers between the ages of 15 and 25 years. Thus by this method we come up with a country and year specific son preference measure based on the individual level stated preferences of the mothers who become mothers in those particular years. There is a fair degree of variance in our desired sex ratio variable across countries.Let us consider some examples from the country averages which we constructed aggregating the individual level data. Below are some examples of low, medium and high son preference countries with the actual average son preference ratio provided within parentheses after the country

24

names.25 • Low Son Preference countries: Dominican Republic (0.92); Haiti, Ukraine (0.94); Nicaragua (0.96), Colombia (0.99) • Medium Son Preference countries: Zimbabwe (1.08), Ghana (1.108), Tanzania (1.07) • High Son Preference countries: India (1.33), Nepal (1.42), Pakistan (1.59) As expected, South Asian countries like India have a very high degree of son preference. Pakistan, Nepal and India occupy three out of the top five spots in terms of stated son preference. Not surprisingly they also have very high rates of MMR and a lot fewer years of female advantage in life expectancy compared to other countries. On the other hand, a country like Brazil are also developing countries in the middle income group, but have lower rates of both stated son preference and MMR as well as a higher female life expectancy advantage. Again if we look at a country like Dominican Republic which has one of the lower levels of son preference in our data-set the female advantage in life expectancy is as high as 4.6. In Table 3 we show how the DSR variable affects Maternal Mortality ratios across countries. The dependent variable is always MMR defined as deaths per 100,000 live births from the WDI and is available at a frequency of 5 years from the years 1990 onwards. While in column 1, we do not control for country and year fixed effects, in columns 2, 3, 4 and 5 we do. We control for log GDP in columns 3 and 4, in column 4 we control for the interaction term of log GDP and DSR as well, and in column 5 we also add a control for desired fertility. From the different columns of Table 3 it is clear that the stated son preference variable significantly increases maternal mortality ratios. We also find evidence of heterogeneity of the effect of the stated son preference variable by GDP. In terms of marginal effects (considering column 5), we notice that a one s.d. (0.122) increase in the Desired Sex ratio leads to 92 additional maternal deaths per 100k live births for the country with the average GDP in this sample. This implies that a one s.d. increase in the DSR in a country can explain 21.12% of the mean maternal deaths in a country (= 92.77/439.189) and 27.29% of the s.d. maternal deaths in the country. As evident, these effects are quite large. 25

We provide a full list of countries by stated son preference in appendix Table A.3

25

Table 3: MMR & Desired Sex Ratio (boys/girls) Desired Sex Ratio

(1) 1155.2∗∗ (453.3)

(2) 655.0∗∗ (267.0)

(3) 667.0∗∗ (255.2) 40.90 (43.20)

(4) 923.9∗∗∗ (225.1) 12.38 (44.36)

No 310

No 310 0.442

No 307 0.444

Yes 307 0.505

log GDP Desired Sex Ratio* lGDP Desired Fertility N r2

(5) 2627.7∗∗∗ (549.9) 318.5∗∗∗ (106.4) -285.3∗∗∗ (89.61) Yes 307 0.520

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables is MMR (deaths per 100,000 births) from WDI. Country fixed effects panel regressions with year dummies have been run (except the first column where only the DSR variable has been controlled for). Standard errors in parentheses are clustered at the country level. The DSR (desired sex ratio boys/girls) data comes from the DHS and has been constructed using the questions on ideal number of boys and girls asked to the mothers. The MMR data comes from the WDI and is based on calculations from the WHO and is available for the years 1990, 1995, 2000, 2005, and 2010. The regressions are hence based on the DHS sample (developing countries) of 63 countries for the years 1990, 1995, 2000, 2005, and 2010 (for the years for which MMR data is available from the WDI). The mean (s.d.) of MMR and DSR are 439.189 (339.91) and 1.11 (0.122). All the variables are 5 yearly averages.

5.2

Women’s Rights and and Gender inequalities in Health

In the previous section we have established that gender biased social attitudes, in particular the desired sex ratio or the stated son preference of the mother, can explain part of the cross country variance in gender inequalities in health outcomes. In this section we study the effects of women’s empowerment, as measured by different types of women’s rights, on such inequalities. This is also an attempt to understand whether female empowerment can reduce cross country differences in gender inequalities in health outcomes. We exploit a previously under-exploited cross country rights data from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set, which provides data on three different variables measuring Political, Economic and Social Rights of women, for the period of 1981 to 2011 for around 127 (in 1981) to 192 (in 2011) countries. The women’s political rights variable, takes into account women’s right to vote, the right to run for political office, the right to hold elected and appointed government positions, the right to join political parties, and the right to petition government officials. The women’s economic rights variable takes into account the rights to equal pay for equal work; free choice of profession or employment without the need to obtain a husband or male relative’s consent; to gainful employment without the need to obtain a husband or male

26

relative’s consent; equality in hiring and promotion practices; job security (maternity leave, unemployment benefits, no arbitrary firing or layoffs, etc...); non-discrimination by employers; to be free from sexual harassment in the workplace; to work at night; to work in occupations classified as dangerous; to work in the military and the police force. And finally, the women’s social rights include the rights to equal inheritance; to enter into marriage on a basis of equality with men; to travel abroad; to obtain a passport; to confer citizenship to children or a husband; to initiate a divorce; to own, acquire, manage, and retain property brought into marriage; to participate in social, cultural, and community activities; to an education; to choose a residence/domicile; freedom from female genital mutilation of children and of adults without their consent; and the freedom from forced sterilization. All three of these variables take 4 discrete values of 0, 1, 2, and 3, with higher values indicating more rights for women. In addition we construct 2 other composite rights variables, the first one being the the first principal component of women’s political, economic and social rights and the second one only incorporates women’s political and economic rights.26 In Table 4, we regress maternal mortality ratios on the different rights variables. In this case we notice that while the three different rights variables significantly reduce maternal mortality ratios, women’s political rights has the most robust effect on MMR. Moreover, we notice that the effect of the rights variable on MMR is non-linear in the GDP level of the country and rights do more to reduce MMR in poorer countries. This significant effect of women’s political and economic rights is robust to controlling for democracy and our rights variable is not solely picking up the effects of democracy. Democracy itself also significantly reduces MMR. In order to calculate marginal effects, we consider column 6 as our preferred specification. A one standard deviation (0.547) increase in women’s political rights leads to 17 fewer maternal deaths, which represents about 7.11% of the mean maternal deaths and 5.49% of the s.d. maternal deaths in the sample. A one standard deviation (0.618) increase in women’s economic rights leads to 14 fewer maternal deaths in the country which represents about 5.66% of the mean maternal deaths in the country and 4.37% of the s.d. maternal deaths in the country. A one standard deviation (0.618) increase in women’s social rights leads to 11 fewer maternal deaths 26

The reader is directed to the appendix for a more detailed description of how each of the variables are constructed. As can be seen in Table E.9, data on women’s political and economic rights exists for the entire period of 1981- 2011, whereas data on women’s social rights was discontinued after the year 2004. Tables E.10 and E.11 give the summary statistics and correlation matrix of the different rights variables.

27

Table 4: Maternal Mortality and Women’s Rights Political Rights log GDP

(1) -17.48 (14.61) -93.60∗∗∗ (11.00)

(2) -9.836 (15.33) 1.836 (22.53)

Democracy

(3) -2.467 (15.35) 6.242 (23.43) -10.02∗∗∗ (3.686)

Rights * lGDP

(4) -365.7∗∗∗ (55.99) -91.71∗∗∗ (24.88) 47.47∗∗∗ (6.780)

(5) -346.9∗∗∗ (68.86) -85.10∗∗∗ (26.76) -6.489∗ (3.487) 45.42∗∗∗ (8.244)

Democracy * lGDP N r2 Economic Rights log GDP

821

821 0.233

757 0.253

821 0.314

757 0.316

-0.813 (15.40) -92.78∗∗∗ (10.68)

14.25 (18.62) 4.218 (23.13)

6.624 (19.73) 7.289 (23.96) -9.718∗∗∗ (3.679)

-168.2∗ (85.58) -15.72 (27.60) 22.95∗∗ (9.059)

-164.6∗ (86.29) -11.67 (28.19) -9.448∗∗ (3.713) 21.65∗∗ (9.033)

Democracy Rights * lGDP Democracy * lGDP N r2 Social Right log GDP

819

819 0.233

755 0.252

819 0.251

755 0.268

0.966 (12.17) -101.7∗∗∗ (11.60)

12.33 (14.01) 5.772 (22.34)

11.33 (14.30) 2.031 (23.00) -5.841∗ (3.374)

-151.9∗ (85.02) -17.63 (24.45) 21.16∗∗ (9.684)

-150.8∗ (86.55) -20.93 (25.13) -5.759∗ (3.406) 20.98∗∗ (9.893)

635 0.218

591 0.224

Democracy Rights * lGDP Democracy * lGDP N r2

635

635 0.198

591 0.205

(6) -256.7∗∗∗ (71.24) -108.9∗∗∗ (29.37) -72.51∗∗∗ (14.98) 35.17∗∗∗ (8.257) 9.658∗∗∗ (2.090) 757 0.368 -103.7 (77.16) -55.91∗ (32.58) -82.79∗∗∗ (14.59) 12.72 (7.929) 10.80∗∗∗ (2.003) 755 0.335 -97.87 (79.49) -74.02∗∗ (31.68) -76.74∗∗∗ (14.89) 13.54 (8.903) 10.84∗∗∗ (2.146) 591 0.282

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is MMR (deaths per 100,000 live births) from WDI. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set. The mean (s.d.) of MMR, women’s political, economic and social rights variables respectively are 238.81 (314.192), 1.863 (0.547), 1.299 (0.618) and 1.248 (0.808). All the variables are 5 yearly averages. The regressions are based on a sample of around 160 countries for the years 1990, 1995, 2000, 2005 and 2010 (panel 3 does not have the year 2010).

in the country which represents about 4.2% of the mean maternal deaths in the country and 3.33% of the s.d. maternal deaths in the country.27 These marginal effects have been calculated for the countries with the GDP of the 25th percentile in the distribution. 27

In moving from columns 5 to 6 however, the economic rights variable becomes marginally insignificant. We test for joint significance of the economic rights variable and its interaction with GDP and find them to be jointly marginally insignificant at 10% level of significance. On the other hand the social rights variable and the its interaction with GDP remain jointly significant in column 6.

28

In Appendix Table E.14 we regress MMR on the 2 composite rights measures. In Panel 1 we include all three of the rights measures (Political, Economic and Social) in the same specification and in Panel 2 we include only Political, Economic rights which gives us more observations since data on these rights are available for a longer period of time.In Appendix Table E.15 we include the three rights variables together in the same specification. As evident from the various tables, women’s rights significantly reduces the maternal mortality ratio. Particularly, the political rights has a significant and robust impact on maternal mortality.

5.3

Women’s Political Representation and Gender inequalities in Health

In this section we directly look at effect of women’s political representation on gender differences in health outcomes. The exact variable representing women’s political representation gives us the “Proportion of seats held by women in national parliaments (%)” and is based on data from the World Bank WDI, which is available for around 160 to 188 countries from around the world for the period of 1997 - 2013. Like in the previous sections we always use country fixed effects panel data regressions with year dummies. The standard errors are always clustered at the country level. In Table 5 we regress Maternal mortality ratio from the WDI on women’s political representation. Considering column 6 as our preferred specification we see that a one standard deviation (9.848) increase in women’s parliamentary representation leads to 20 less maternal deaths in the country, which represents 9.14% of the mean deaths in the sample and 7.16% of the s.d. deaths in the sample. Like in the case of the female life expectancy advantage, these effects have been calculated for a country with the average GDP and for poorer countries these effects will be much higher. From the Tables 5 we notice that women’s political representation significantly reduces the maternal mortality ratio. This is in line with the robust effect of women’s political rights on both female life expectancy and maternal mortality found in the previous section.

5.4

Gender Intensity in Language and Gender inequalities in Health

So far we have used different measures of women’s status in society, including the stated desired sex ratio of mothers from the DHS, women’s political, economic and social rights from the Cingranelli, Richards, and Clay (Cingranelli et al.) database and the political representation of women from the WDI. While all these measures that we have used so far are good reflections 29

Table 5: MMR and Women’s Political Representation womparl lgdp

(1) -2.313∗ (1.358) -99.10∗∗∗ (12.19)

(2) -3.248∗ (1.766) -48.24 (37.66)

democ

(3) -3.379∗ (1.961) -44.62 (41.13) -4.551 (4.037)

Rights * lGDP

(4) -29.69∗∗∗ (3.971) -87.44∗∗ (37.00) 3.464∗∗∗ (0.463)

(5) -30.46∗∗∗ (4.067) -83.66∗∗ (40.35) -1.807 (3.372) 3.597∗∗∗ (0.490)

659 0.425

594 0.441

Democracy * lGDP N r2

659

659 0.278

594 0.287

(6) -26.79∗∗∗ (4.234) -96.79∗∗ (39.74) -47.50∗∗∗ (17.34) 3.145∗∗∗ (0.504) 6.414∗∗∗ (2.343) 594 0.468

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is MMR (deaths per 100,000 births) from the WDI. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. Both the Life expectancy data and the women’s political representation data comes from the WDI. These regressions are based on a sample of around 38 to 44 developing countries from the DHS sample for the period of 1997 to 2011. The mean and s.d. of the dependent variable are 223.059 and 284.725, whereas the mean and s.d. of women’s representation in parliament is around 14.371 and 9.848. All the variables are 5 yearly averages.

of gender attitudes in society, they are quite likely to be endogenous. In this section we use some arguably exogenous measures of negative gender attitudes in society, to try and establish whether such negative gender attitudes have a causal effect on the observed gender differences in health outcomes across countries. Some recent papers have argued that grammatical gender can influence and reflect gender attitudes in society and are correlated with different gender outcomes including maternity leave policy differences across countries (Givati and Troiano, 2012), female labour force and political participation (Gay et al., 2013). In this section, we exploit the findings of these papers and use the gender intensity in language grammar as a measure of gender bias in society. Language grammar was established centuries in the past and is one of the features of language that is stable over long periods of time. Moreover, grammatical gender is something that an individual is born with and thus arguably a more exogenous measure of gender bias in society. For gender salience of languages we use the data and classification of Gay et al. (2013) and Givati and Troiano (2012). In both these papers the focus is on female/male distinctions in grammar.28 Givati and Troiano (2012) use the number of cases of gender differentiated pronouns for 33 languages (mostly but not entirely European) as a measure of gender neutrality of the 28

Chen (2013) shows how the future tense in different language can have an effect on an individuals saving behaviour.

30

language. According to their classification, the 33 languages can be divided into 4 distinct groups, with each group having either 0 (6 languages), 1 (10 languages), 2 (14 languages), or 4 (3 languages) gender differentiated pronouns. Languages with a higher number of gender differentiated pronouns are supposed to be less gender neutral. Gay et al. (2013) provide an alternative classification. They also focus on female/male distinctions in grammar but do not restrict themselves to personal pronouns. They use the gender related grammatical variables coming from the World Atlas of Linguistic structures and have four binary variables related to gender neutrality of languages viz. Sex-Based Intensity Index, Number Gender Intensity Index, Gender Assignment Intensity Index, Gender Pronouns Intensity Index. Their final Gender Intensity index (GII) is a sum of all these four indices (or some combination of a subset of these indices). Since, all these four indices do not exist for all countries we sometimes choose a subset of these indices in order to maximize the number of observations. Suppose we choose to use three of these indices to construct our final Gender neutrality index, then GII ∈ 0, 1, 2, 3 and all languages can be divided into 4 distinct classes of gender neutrality. For our final analysis we choose to use all eight of the following measures, the first seven of which come from Gay et al. (2013) and the final one comes from Givati and Troiano (2012). 1. Sex-Based Intensity Index (sbii) 2. Number Gender Intensity Index (ngii) 3. Gender Assignment Intensity Index (gaii) 4. Gender Pronouns Intensity Index (gpii) 5. gii0 = ngii + sbii + gaii + gpii 6. gii1 = ngii + sbii + gaii 7. gii2 = ngii + sbii + gpii 8. gtroiano = number of cases of gender differentiated pronouns. In the rest of our analysis we will refer to the different language grammar gender intensity variables as GII measures.29 Since our measures of gender intensity of languages are time 29

Appendix Tables E.16 and E.17 respectively give the summary statistics and correlation statistics of our different GII measures.

31

invariant, unlike the previous sections, we are unable to use a country fixed effects framework. However, following the previous literature, we control for an extended set of controls in order to alleviate endogeneity concerns. Our primary specification is given by: Yit = α + GIIi + Percentagei + Xit + Xi + 

(6)

where GII is our measure of gender intensity. Since the different GII measures at the country level are based on the majority language in each country, following Gay et al. (2013) whenever we run regressions using these measures we will always control for the percentage of the population that speak the particular language, P ercentagei . Xit and Xi represent our other control variables including decade dummies, continent dummies, log of GDP, the log of population, dummies for income groups from the World Bank income groups classification, the percentages of the population that is Protestant, Catholic and Muslim respectively, and the proportion of the country that is tropical or subtropical. We always use cluster robust standard errors clustered at the country level. In Tables B.4, and 6 we respectively regress the log ratio of female to male life expectancy and the MMR from the WDI on the different measures of gender intensity of language. The different columns in these tables correspond to the different GII measures listed in the paragraphs above. The three panels on the other hand, correspond to three different sets of controls. In Panel 1 for instance, apart from the GII variable we control for the percentage of the population speaking the majority language (for which the GII has been calculated), decade dummies and continent dummies. In Panel 2 we control for the log of GDP, the log of population, dummies for the World Bank Income groups classification, the percentage of population that is Protestant, Catholic and Muslim, and the proportion of the country that is tropical or subtropical in addition to the controls from Panel 1. And finally in Panel 3, we add the interaction term of log of GDP and GII in addition to the controls from the previous panels. Similarly, in Table 6 we find that most of our gender intensity of language measures significantly increase the number of maternal deaths. Moreover, there is evidence of heterogeneity of this effect by GDP. The GII variable leads to higher maternal mortality in poorer countries than in rich countries. Considering Panel 3 as our preferred specification (given the heterogeneity by GDP) and gii0 which is an aggregate of the four GII measures from Gay et al. (2013) as our GII measure, we notice that a one standard deviation (1.634) increase in GII leads to around 35 additional maternal deaths in the country which represents around 20.25% of the mean maternal deaths in the sample and around 14.01% of the s.d. maternal deaths in the 32

sample. The results in this section are in line with the findings of the previous literature which finds gender salience in language grammar has a negative impact on different measures of gender inequality in society (Gay et al., 2013; Givati and Troiano, 2012). However, ours is the first paper to underscore how the gender salience in language grammar is correlated with the gender inequalities in health outcomes. These findings further strengthen our claim that differences in gender attitudes in society can explain the differences in gender inequality in health outcomes across countries over and above differences in levels of development. Table 6: Maternal Mortality & Gender Intensity of Language

GII N r2 GII lgdp N r2 GII GII gdp lgdp N r2

(1) ngii

(2) sbii

(3) gaii

(4) gpii

(5) gii0

(6) gii1

(7) gii2

(8) gtroiano

3.054 (37.92) 610 0.468

-25.85 (47.26) 610 0.470

40.22 (40.01) 445 0.504

-44.24 (43.32) 600 0.508

-3.998 (12.34) 420 0.543

5.564 (17.99) 445 0.501

-15.04 (16.65) 570 0.490

5.499 (9.502) 405 0.355

50.03∗∗ (21.61) -71.53∗∗∗ (19.65) 575 0.701

68.52∗∗ (33.75) -72.58∗∗∗ (19.68) 575 0.702

105.0∗∗∗ (29.02) -77.63∗∗∗ (22.86) 417 0.766

56.60∗ (32.01) -66.64∗∗∗ (19.01) 562 0.704

30.24∗∗∗ (10.24) -71.86∗∗∗ (22.32) 399 0.750

40.23∗∗∗ (12.19) -74.53∗∗∗ (22.31) 417 0.762

24.90∗ (12.57) -68.26∗∗∗ (19.06) 542 0.690

2.832 (8.682) -69.24∗∗∗ (19.97) 384 0.611

368.1∗∗ (154.7) -38.32∗∗ (17.13) -52.19∗∗ (22.84) 575 0.712

162.9 (203.8) -11.88 (23.26) -64.07∗∗ (30.76) 575 0.703

665.0∗∗∗ (109.2) -69.92∗∗∗ (12.73) -27.11 (21.57) 417 0.794

499.7∗∗∗ (176.4) -55.01∗∗∗ (20.08) -47.69∗∗ (20.28) 562 0.718

213.0∗∗∗ (42.13) -23.91∗∗∗ (5.045) -4.747 (21.37) 399 0.777

249.4∗∗∗ (57.73) -27.63∗∗∗ (6.965) -14.31 (21.37) 417 0.784

149.6∗ (83.73) -16.04 (9.952) -42.03 (28.60) 542 0.699

148.2∗ (85.26) -16.70∗ (9.214) -25.53 (19.44) 384 0.636

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables in all three panels are the maternal mortality ratio (from the WDI database) . Standard errors in parentheses are clustered at the country level. The GII data come from Gay et al. (2013) and Givati and Troiano (2012). Apart from the GII variable in Panel 1 we control for the percentage of the population speaking the majority language (for which the GII has been calculated), decade dummies and continent dummies. In Panel 2 we control for the log of GDP, the log of population, dummies for the World Bank Income groups classification, the percentage of population that is Protestant, Catholic and Muslim, and the proportion of the country that is tropical or subtropical in addition to the controls from Panel 1. In Panel 3, we add the interaction term of log of GDP and GII in addition to the controls from the previous panels.

33

6 6.1

Robustness and Extensions Female Life Expectancy advantage

In this section we extend our analysis to the Female Life expectancy advantage and show how our different measures of gender prejudice also reduces the Female LE advantage. The results from this section are provided in the Appendix section B. In Table B.1 we present the results from a cross country panel of how the DSR variable reduces the female advantage in life expectancy. However, there is no significant heterogeneity by GDP level of this effect. In column 5, the coefficient for DSR and the interaction term of DSR and log GDP look separately insignificant. However, we conduct a test for joint significance which yields a F-statistic of 8.35 and a p-value of 0.0006, which allows us to strongly reject the null of joint insignificance of the two terms. In terms of marginal effects (considering say column 5), a one s.d. (0.147) increase in the desired sex ratio, which is roughly the difference in the DSR between Bangladesh/Ethiopia and India, knocks of around 1.37% of the relative female advantage in life expectancy over men for the country with the average GDP in this sample. These marginal effects imply that a one s.d. change in DSR leads to around a 23.59% (=0.0137/0.059) fall in the average life expectancy advantage of women over men and around a 38.01% (=0.0137/0.036) fall in the s.d. of life expectancy advantage of women over men in the sample. In Table B.2, we regress the log ratio of female to male life expectancy on the three different rights measures. As in the previous sections, we use country fixed effects panel data regressions with year dummies. The standard errors are always clustered at the country level. As far as the log ratio of female to male life expectancy is concerned, women’s economic and social rights are more significant correlates than women’s political rights. Also, whenever the rights variables are significant we find evidence of heterogeneity of the effects of the rights variable by GDP. We notice that the rights do more to increase the female advantage in life expectancy for poorer countries. Our results are robust to controlling for democracy and thus the rights variables are not merely picking up the effects of democracy. In order to calculate marginal effects, we consider column 6 as our preferred specification. While women’s political rights do not seem to have a significant effect on the female life expectancy advantage variable, a one standard deviation (0.682) increase in women’s economic rights leads to an increase in female life expectancy advantage by 0.15% which represents a 2.21% of the mean and 4.3% of the s.d. of log ratio of female to male life expectancy in the

34

sample. A one standard deviation (0.828) increase in women’s social rights on the other hand, leads to an increase in female life expectancy advantage by 0.24% which represents a 3.38% of the mean and 6.72% of the s.d. of log ratio of female life expectancy advantage in the sample. These marginal effects have been calculated for the countries with the 25th percentile GDP. We consider the GDP of the 25th percentile country since almost all the low income countries and several middle income countries have and average GDP in the sample which is lower than this cutoff.30 From the Table B.3 we notice that women’s political representation significantly increases the female life expectancy advantage. In Table B.3 we regress the log ratio of female to male life expectancy on women’s political representation and notice how women’s representation in parliament increases the female advantage in life expectancy and this effect is stronger in poorer countries. Considering column 6 as our preferred specification we see that a one standard deviation (10.204) increase in women’s parliamentary representation (measured by percentage of seats held by women in national parliaments) leads to a 0.06% increase in the female life expectancy advantage which is about 0.88% of the mean female life expectancy advantage and about 1.62% of the s.d. female life expectancy advantage. These effects have been calculated for a country with the average GDP. For poorer countries these effects will be much higher. In Table B.4 we notice that the gender intensity of language significantly reduces the female life expectancy advantage and this effect is robust to a wide set of controls. There is not much evidence of heterogeneity of this effect by GDP level (see Panel 3). Considering Panel 2 as our preferred specification and gii0 which is an aggregate of the four GII measures from Gay et al. (2013) as our GII measure, we notice that a one standard deviation (1.606) increase in GII leads to a 1.47% decrease in the female life expectancy advantage ratio which represents about 21.29% of the mean female life expectancy advantage and 41.16% the s.d. female life expectancy advantage. 30

In Appendix Table E.12 we regress the log ratio of female to male life expectancy on the 2 composite rights measures. In Panel 1 we include all three of the rights measures (Political, Economic and Social) in the same specification and in Panel 2 we include only Political, Economic rights which gives us more observations since data on these rights are available for a longer period of time. In Appendix Table E.12 we include the three rights variables together in the same specification

35

6.2

Infant and Child Mortality Ratios (Female/Male)

In the previous sections we have shown how our different measures of gender prejudice increase MMR rates and reduces the female advantage in life expectancy. Also, in section 4.1 using individual level infant mortality data from the DHS we showed how the stated son preference of mothers increase the probability of female infant mortality relative to male infant mortality. In this section we extend the results to our other gender prejudice variables. In particular we show how the different rights variables, and the gender intensity of language variables significantly increase cross country female/male infant and child mortality ratios. The results are provided in Appendix Section C.

6.3

Placebo Tests with TB infection rates

In the analysis presented in the previous sections of this paper, we have established that both the female advantage in life expectancy and maternal mortality ratio variables are correlated with different measures of gender inequality including the degree of son preference in society which is a reflection of gender biased attitudes in society, women’s political, economic and social rights, women’s representation in the parliament and finally the gender intensity of language grammar. We have argued that since maternal mortality is a women specific health outcome we expect it to be negatively related to different measures of negative gender attitudes and prejudice in society. However, in case our argument is valid, we would expect to see no effects of any of our gender prejudice variables on a gender neutral illness like TB.31 In this section we establish exactly that. We replicate the regressions presented in the last few sub-sections of the paper, but using TB infection rates as our primary dependent variable. We refer to these as our placebo regressions. For the sake of brevity, we provide the regression output in the appendix. As expected we notice that the desired son preference variable (Table E.18) and the rights variables (Table E.19) have no effects on TB infection rates. On the other hand, the GII variables if anything reduce TB infection rates (Table E.21). The only aberration is that women’s parliamentary representation significantly reduces TB infection rates (Table E.20). However, this does not necessarily falsify all the claims that we have made in this paper so far and is very much in line with previous 31

If anything incidence and death rates are higher among men than women, but there is evidence that men are more likely to be diagnosed of TB. See http://www.who.int/tb/challenges/gender/en/ and Thorson and Diwan (2001).

36

literature which finds women’s political representation to positively affect health outcomes in general (See Bhalotra and Clots-Figueras (2014) for example).

6.4

Time varying Group Fixed Effects Framework

In this section we use the “time varying group fixed effects framework” (GFE) recently developed by Bonhomme and Manresa (2012). This framework allows us to control for grouped patterns of unobserved heterogeneity. It is a flexible framework which allows us to use a group fixed effects model as an alternative to the country fixed effects framework. This framework leaves group membership unrestricted and allows the determination of group membership endogenously from within the data. Bonhomme and Manresa (2012) propose “ ... a framework that allows for time patterns of unobserved heterogeneity that are common within groups of individuals. Both the groupspecific time patterns and individual group membership are left unrestricted, and are estimated from the data. In particular, our time-varying specification shares with FE the fact that it leaves the relationship between observables and unobservables unrestricted, thus allowing for general forms of covariates endogeneity. The main assumption is that the number of distinct individual time patterns of unobserved heterogeneity is relatively small.” We use codes provided by Bonhomme and Manresa (2012) to determine group membership. We take a flexible approach and try specifications including different numbers of groups varying from 2 to 6. Since our LE data is more extensive and exists annually for most countries in the world, we choose to use this framework on the LE advantage regrssions. We provide our results in appendix section D where we show how our results are robust to this new methodology.

6.5

Life Expectancy Pre and Post 1990

Our MMR variable comes from the WDI and is available for the year 1990 onwards. On the other hand, the Life Expectancy variable is available from 1960 onwards. In this section we regenerate our Life Expectancy advantage regressions for the periods of pre 1990 and post 1990 and we restrict the sample to the set of countries for which we have MMR data. We notice from the below referenced tables that there are no substantial differences in our results. DSR: First, in the appendix Tables E.22 and E.23 we show the results for the pre and post 1990 sample and, in Tables E.24 and E.25, we show the same for the MMR sample. GII: First, in the appendix Tables E.26 and E.27 we show the results for the pre and post 1990 sample and, in Tables E.28 and E.29, we show the same for the MMR sample. 37

Women’s rights: First, in the appendix Tables E.30 and E.31 we show the results for the pre and post 1990 sample and, in Tables E.32 and E.33, we show the same for the MMR sample.

6.6

MMR from DHS data

So far in this paper we have used the MMR data from the WDI and have shown how our different measures of gender prejudice in society consistently lead to higher Maternal mortality ratios. In this section we show that our results are robust using alternative data exploiting a novel panel data set on maternal mortality constructed by Bhalotra and Clarke (2014) using sibling linked files from the Demographic and Health Survey (DHS). We undertake three specific types of exercises using the DHS data. First, in the appendix Tables E.34, E.35, E.36 and E.37 we show how our results are qualitatively similar regardless of whether we use the DHS and WDI data. In these tables we also restrict the same set of country and years in some of the specifications. Then using the DHS data we investigate the coefficients from the pre and post 1990 period to understand if there are any major changes to the coefficients across these two time periods. See Tables E.38, E.39, E.40 and E.41. Finally in Tables E.42 and E.43 we provide coefficients from the pre 1990 period. We notice that the coefficients are substantial but not significant.

6.7

US Case Study: Early Suffrage, Adoption of Medical Technology, and Maternal Mortality

This supplement explores the interplay between early suffrage and the post-1937 fall in maternal mortality driven by the arrival of the first antibiotics in the United States. Women’s suffrage was institutionalized at the national level in 1920 with the 19th amendment to the U.S. Constitution. However, prior to this, individual states had the authority to legislate women’s suffrage with respect to state and local elections. About 40% of U.S. passed suffrage laws prior to 1920. The evolution of women’s suffrage is chronicled in Miller (2008). He contends, based on work by other scholars, that early adoption of women’s suffrage was driven by slowly changing social norms with respect to role of women.32 32

E.g. “The most obvious pattern is geographic all else equal, women in western states could vote before women elsewhere in America. Some historians suggest that frontier conditions were amenable to womens suffrage because women supported restrictions on common western vices (drunkenness, gambling, and prostitution) or because the harsh realities of frontier life made it impossible to maintain traditional gender roles (Brown 1958; Grimes 1967).7 Many others argue that idiosyncratic circumstances in each state resulted in the vote for women (Larson 1971; Beeton 1986), citing rich historical evidence in support of this view.8 Quantitative studies yield

38

The role of sulfa drugs in inducing a historic decline in maternal mortality is chronicled in Jayachandran et al. (2010). Basically, log(MMR) was constant for a long period of time until 1937, when sulfa drugs were rapidly adopted in hospital and outpatient settings. Sulfa drugs were also important in reducing morbidity and mortality from pneumonia, as well. Following the main paper, it is worthwhile to ask if states adopting early women’s suffrage, which presumably had long standing norms evolving in favor of women’s autonomy gained more from the arrival of antibiotics than their later suffrage counterparts. We will look at both maternal mortality, the condition of interest, and pneumonia mortality, which affected males more than females, and infants in particular. The graphs below show that the gap between early (anytime before 1920) and late (laws passed in 1920) suffrage adopters for MMR widened after the arrival for sulfa drugs, but not so for pneumonia mortality:

(a)

(b)

Figure 6: MMR and Pneumonia in the US. We can verify this econometrically using state level mortality and suffrage data from the above papers. Basically, for the 1925-1943 time period, we can regress logged MMR against a linear year trend, a dummy = 1 for year = 1937 or later, and their interaction. These RHS variables can additionally be interacted with an indicator for early suffrage = 1, so as to assess whether the level and trend breaks are larger for these states. This specification follows that used in the aforementioned Jayachandran et al. (2010) study. As a falsification, we will do the same for logged pneumonia mortality. All models include state fixed effects, cluster S.E. at the strikingly inconclusive results (Cornwall, Dahlin, King, and Schiffman 2004). The single robust correlate of suffrage law enactment emerging from these studies is the share of women working in non-agricultural occupations (King, Cornwall, and Dahlin 2005). Although this presumably reflects changing social norms about the role of women, it evolved very gradually over time (Smith and Ward 1985; Goldin 1990) and can be distinguished econometrically from abrupt year-to-year legislative changes governing womens right to vote.”

39

state level, and weight observations (state-level) by their population: Table 7: MMR, Pneumonia Regressions

Post Post*Year Year Early Suffrage*Post Early Suffrage*Post*Year Early Suffrage*Year Constant

Observations R-squared

(1) Ln(MMR)

(2) Ln(Pneum)

-0.0917*** -0.0298 -0.0891*** -0.0049 -0.0230*** -0.00246 -0.0849** -0.0365 -0.0146** -0.00642 0.001 -0.00335 1.689*** -0.012

0.0087 -0.0215 -0.0611*** -0.0108 -0.0293*** -0.00647 -0.0459 -0.0279 -0.00674 -0.0128 0.0047 -0.0076 -0.0461*** -0.0148

868 0.951

868 0.78

* p < 0.10, ** p < 0.05, *** p < 0.01 Robust standard errors in parentheses. All models include state FE. Models weighted using state population. Early suffrage refers to the states which had suffrage pre1920 i.e. prior to when women’s suffrage was institutionalized at the national level in 1920 with the 19th amendment to the U.S. Constitution. The estimated coefficients on the suffrage interactions are almost identical when we use the absolute rates rather than the logarithms of the rates as dependent variables.

Early suffrage states had significantly larger trend and level breaks for maternal mortality. However, for pneumonia mortality (for which there is a trend break only), we do not see any difference in the “sulfa effect.” The results suggest that preferences correlated with female suffrage may have influenced the specific adoption of medical technology for a “women specific issue.” In order to examine the parallel trend assumptions underlying our double-difference specification more completely, we plot event studies figures for both causes of death. These event studies follow the specification in table 7, however now fully interact early suffrage states with each year dummy. This allows us to plot the difference between early- and late-suffrage states by year, both preceding, and after the sulfa reforms. If the parallel trend assumptions hold, so that the only difference between the two types of states arises after the date of the reform, we should see that all estimates plotted prior to the reform are not statistically distinguishable from zero. What’s more, if early suffrage states only make additional inroads in female specific

40

causes, we should see that all coefficients are insignificant for infant pneumonia, but post-reform coefficients are significant for MMR. This is precisely what we observe in figures 7 and 8. In figure 7 we see that all confidence intervals include 0 prior to the reform, however in the year of the arrival of sulfa, and in the post-reform years, early suffrage states make significant improvements in MMR indicators when compared with their late-suffrage counterparts. This suggests that prior to the reform, in the absence of sulfanid drugs, rates of maternal mortality evolved on a similar trajectory in both states, while in post-reform years, early suffrage states employ the available technology more effectively to address female-specific mortality causes. When compared to figure 8, the contrast is stark. With the exception of 1 in 18 years, there is no significant difference between the early- and the late-suffrage states in their trends in pneumonia death rates. Fundamentally, as suggested in figure 6b, both states are equally able to leverage the arrival of sulfa drugs to reduce the gender-neutral (or indeed slightly male-biased) disease burden. This reinforces the idea discussed above that there are important differences in the way that more gender-biased states are able to employ health care technology to all, and to female-speficic, health outcomes. Figure 7: Event Study - MMR in the US

−.3

−.2

MMR −.1

0

.1

MMR and Early Suffrage: Event Study

−10

−5

0

5

time Point Estimate Year −1 is omitted as the base case.

41

95% CI

Figure 8: Event Study - Pneumonia in the US

−.3

−.2

−.1

IPR 0

.1

.2

IPR and Early Suffrage: Event Study

−10

−5

0

5

time Point Estimate Year −1 is omitted as the base case.

42

95% CI

7

Conclusion

Preventable maternal mortality is still very high in many developing countries, even after falling by almost 50% since 1990 to the present day. Moreover, while IMR has been falling steadily in the last few decades, MMR rates started to fall only in the 1990s after initial stagnation. In this paper we show that MMR is a woman specific condition and differences in MMR and female life expectancy advantage across countries are a reflection of differences in gender attitudes across countries. We find that regardless of the measures of gender bias or woman’s status in society we use including the stated son preference of mothers, women’s political rights or the gender intensity of language grammar, we consistently find that that cross country differences in gender inequalities in health outcomes are a reflection of differences in gender bias across societies and go beyond differences in income. The main policy implication is that specific interventions to reduce maternal mortality and improve female health outcomes might be required even in high growth poor countries with high gender prejudice.

References Acemoglu, D. and S. Johnson (2006). Disease and development: the effect of life expectancy on economic growth. Technical report, National Bureau of Economic Research. Aghion, P., P. Howitt, and F. Murtin (2010). The relationship between health and growth: when lucas meets nelson-phelps. Technical report, National Bureau of Economic Research. Albanesi, S. and C. Olivetti (2009). Gender roles and medical progress. Technical report, National Bureau of Economic Research. Albanesi, S. and C. Olivetti (2014). Maternal health and the baby boom. Quantitative Economics 5 (2), 225–269. Amiri, A. and U.-G. Gerdtham (2013). Impact of maternal and child health on economic growth: New evidence-based granger causality and dea analysis. Anderson, S. and D. Ray (2010). Missing women: age and disease. The Review of Economic Studies 77 (4), 1262–1300.

43

Anderson, S. and D. Ray (2012). The age distribution of missing women in india. Economic and Political Weekly 47 (47-48), 87–95. Ashraf, Q. H., A. Lester, and D. N. Weil (2008). When does improving health raise gdp? Technical report, National Bureau of Economic Research. Bhalotra, S. and D. Clarke (2013). Maternal education and maternal mortality: Evidence from a large panel and various natural experiments. Mimeo. Bhalotra, S. and D. Clarke (2014). Trends in maternal mortality and gender bias. Work in Progress. Bhalotra, S. and I. Clots-Figueras (2014). Health and the political agency of women. American Economic Journal: Economic Policy 6 (2), 164–197. Bloom, D. E., D. Canning, and J. Sevilla (2004). The effect of health on economic growth: a production function approach. World development 32 (1), 1–13. Bonhomme, S. and E. Manresa (2012). Grouped patterns of heterogeneity in panel data. Technical report, Citeseer. Canudas-Romo, V., L. Liu, L. Zimmerman, S. Ahmed, and A. Tsui (2014). Potential gains in reproductive-aged life expectancy by eliminating maternal mortality: A demographic bonus of achieving mdg 5. PloS one 9 (2), e86694. Chen, M. K. (2013). The effect of language on economic behavior: Evidence from savings rates, health behaviors, and retirement assets. The American Economic Review 103 (2), 690–731. Cingranelli, D. L., D. L. Richards, and K. C. Clay.

The ciri human rights dataset.

http://www.humanrightsdata.org Version 2013.12.05. Coale, A. J. (1991). Excess female mortality and the balance of the sexes in the population: an estimate of the number of” missing females”. The Population and Development Review , 517–523. Duflo, E. (2011). Womens empowerment and economic development. Technical report, National Bureau of Economic Research. Gay, V., E. Santacreu-Vasut, and A. Shoham (2013). The grammatical origins of gender roles. Mimeo. 44

Givati, Y. and U. Troiano (2012). Law, economics, and culture: Theory of mandated benefits and evidence from maternity leave policies. Journal of Law and Economics 55 (2), 339–364. Hogan, M. C., K. J. Foreman, M. Naghavi, S. Y. Ahn, M. Wang, S. M. Makela, A. D. Lopez, R. Lozano, and C. J. Murray (2010). Maternal mortality for 181 countries, 1980–2008: a systematic analysis of progress towards millennium development goal 5. The Lancet 375 (9726), 1609–1623. Jayachandran, S. (2014a). Fertility decline and missing women. Technical report, National Bureau of Economic Research. Jayachandran, S. (2014b). The roots of gender inequality in developing countries. Technical report, National Bureau of Economic Research. Jayachandran, S. and A. Lleras-Muney (2008). Life expectancy and human capital investments: Evidence from maternal mortality declines. Technical report, National Bureau of Economic Research. Jayachandran, S., A. Lleras-Muney, and K. V. Smith (2010). Modern medicine and the twentieth century decline in mortality: Evidence on the impact of sulfa drugs. American Economic Journal: Applied Economics 2 (2), 118–46. Kassebaum, N. J., A. Bertozzi-Villa, M. S. Coggeshall, K. A. Shackelford, C. Steiner, K. R. Heuton, D. Gonzalez-Medina, R. Barber, C. Huynh, D. Dicker, et al. (2014). Global, regional, and national levels and causes of maternal mortality during 1990–2013: a systematic analysis for the global burden of disease study 2013. The Lancet. Kirigia, J. M., D. Oluwole, G. M. Mwabu, D. Gatwiri, and L. H. Kainyu (2006). Effects of maternal mortality on gross domestic product (gdp) in the who african region. African journal of health sciences 12 (3), 55–64. Lagerl¨of, N.-P. (2003). Gender equality and long-run growth. Journal of Economic Growth 8 (4), 403–426. Lorentzen, P., J. McMillan, and R. Wacziarg (2008). Death and development. Journal of Economic Growth 13 (2), 81–124. Loudon, I. (1992). Death in childbirth: an international study of maternal care and maternal mortality 1800-1950. 45

Milazzo, A. (2014). Why are adult women missing? son preference and maternal survival in india. Policy Research Working Paper (World Bank) WPS6802. Miller, G. (2008). Womens suffrage, political responsiveness, and child survival in american history. The Quarterly Journal of Economics 123 (3), 1287. Nations, U. (2013). World population prospects: The 2012 revision, dvd edition. Sen, A. (1990). More than 100 million women are missing. The New York Review of Books. Sen, A. (2001). The many faces of gender inequality. New republic, 35–39. Shastry, G. K. and D. N. Weil (2003). How much of cross-country income variation is explained by health? Journal of the European Economic Association 1 (2-3), 387–396. Thorson, A. and V. K. Diwan (2001). Gender inequalities in tuberculosis: aspects of infection, notification rates, and compliance. Current opinion in pulmonary medicine 7 (3), 165–169. Weil, D. N. (2005). Accounting for the effect of health on economic growth. Technical report, National Bureau of Economic Research. WHO (2014). Maternal mortality, fact sheet n348, updated may 2014, accessed - july 2014. Wong, Y. N. (2012). World development report 2012: Gender equality and development. In Forum for Development Studies, Volume 39, pp. 435–444. Taylor & Francis.

46

A

Data appendix

A.1

List of countries according to World Bank Income group classification.

Here we list the High, Middle and Low income countries according to the World Bank classification: High Income: Aruba, Andorra, United Arab Emirates, Antigua and Barbuda, Australia, Austria, Belgium, Bahrain, Bahamas, The, Bermuda, Barbados, Brunei Darussalam, Canada, Switzerland, Channel Islands, Chile, Curacao, Cayman Islands, Cyprus, Czech Republic, Germany, Denmark, Spain, Estonia, Finland, France, Faeroe Islands, United Kingdom, Equatorial Guinea, Greece, Greenland, Guam, Hong Kong SAR, China, Croatia, Isle of Man, Ireland, Iceland, Israel, Italy, Japan, St. Kitts and Nevis, Korea, Rep., Kuwait, Liechtenstein, Lithuania, Luxembourg, Latvia, Macao SAR, China, St. Martin (French part), Monaco, Malta, Northern Mariana Islands, New Caledonia, Netherlands, Norway, New Zealand, Oman, Poland, Puerto Rico, Portugal, French Polynesia, Qatar, Russian Federation, Saudi Arabia, Singapore, San Marino, Slovak Republic, Slovenia, Sweden, Sint Maarten (Dutch part), Turks and Caicos Islands, Trinidad and Tobago, Uruguay, United States, Virgin Islands (U.S.) Middle Income: Angola, Albania, Argentina, Armenia, American Samoa, Azerbaijan, Bulgaria, Bosnia and Herzegovina, Belarus, Belize, Bolivia, Brazil, Bhutan, Botswana, China, Cote d’Ivoire, Cameroon, Congo, Rep., Colombia, Cape Verde, Costa Rica, Cuba, Djibouti, Dominica, Dominican Republic, Algeria, Ecuador, Egypt, Arab Rep., Fiji, Micronesia, Fed. Sts., Gabon, Georgia, Ghana, Grenada, Guatemala, Guyana, Honduras, Hungary, Indonesia, India, Iran, Islamic Rep., Iraq, Jamaica, Jordan, Kazakhstan, Kiribati, Kosovo, Lao PDR, Lebanon, Libya, St. Lucia, Sri Lanka, Lesotho, Morocco, Moldova, Maldives, Mexico, Marshall Islands, Macedonia, FYR, Montenegro, Mongolia, Mauritania, Mauritius, Malaysia, Namibia, Nigeria, Nicaragua, Pakistan, Panama, Peru, Philippines, Palau, Papua New Guinea, Paraguay, West Bank and Gaza, Romania, Sudan, Senegal, Solomon Islands, El Salvador, Serbia, Sao Tome and Principe, Suriname, Swaziland, Seychelles, Syrian Arab Republic, Thailand, Turkmenistan, Timor-Leste, Tonga, Tunisia, Turkey, Tuvalu, Ukraine, Uzbekistan, St. Vincent and the Grenadines, Venezuela, RB, Vietnam, Vanuatu, Samoa, Yemen, Rep., South Africa, Zambia. Low Income: Afghanistan, Burundi, Benin, Burkina Faso, Bangladesh, Central African Republic, Congo, Dem. Rep., Comoros, Eritrea, Ethiopia, Guinea, Gambia, The, Guinea-Bissau, Haiti, Kenya, Kyrgyz Republic, Cambodia, Liberia, Madagascar, Mali, Myanmar, Mozambique, Malawi, Niger, Nepal, Korea, Dem. Rep., Rwanda, Sierra Leone, Somalia, South Sudan, Chad,

47

Togo, Tajikistan, Tanzania, Uganda, Zimbabwe.

A.2

Life Expectancy

The data on Life Expectancy comes from the World Bank WDI. These data are available annually for around 200 countries from around the world for the years 1960-2011. The World Bank WDI data is based on the following sources:(1) United Nations Population Division. World Population Prospects, (2) United Nations Statistical Division. Population and Vital Statistics Report (various years), (3) Census reports and other statistical publications from national statistical offices, (4) Eurostat: Demographic Statistics, (5) Secretariat of the Pacific Community: Statistics and Demography Programme, and (6) U.S. Census Bureau: International Database. Life Expectancy according to the World Bank WDI is defined as “... the average number of years a newborn is expected to live if mortality patterns at the time of its birth remain constant in the future. It reflects the overall mortality level of a population, and summarizes the mortality pattern that prevails across all age groups in a given year. It is calculated in a period life table which reflects a snapshot of a mortality pattern of a population at a given time. It therefore does not reflect actual mortality patterns that a person actually goes through during his/her life, which can be calculated in a cohort life table.”33 (World Development Indicators) Since the World Bank WDI provides us with data for female and male life expectancy separately, we are able to construct our female life expectancy advantage variable as follows:

Female LE advantage = ln{

Female Life Expectancy } Male Life Expectancy

(7)

33 High mortality in young age groups significantly lowers the life expectancy at birth. But if a person survives his/her childhood of high mortality, he/she may live much longer. For example, in a population with a life expectancy at birth of 50, there may be few people dying at age 50. The life expectancy at birth may be low due to the high childhood mortality so that once a person survives his/her childhood, he/she may live much longer than 50 years.“Complete vital registration systems are not common in developing countries. Therefore estimates of life expectancy must be derived from sample surveys or by applying indirect estimation techniques to registration, census, or survey data. Survey data are subject to recall error, and surveys estimating infant/child deaths require large samples because households in which a birth has occurred during a given year cannot ordinarily be preselected for sampling. Indirect estimates rely on model life tables that may be inappropriate for the population concerned. Because life expectancy at birth is estimated using infant/child mortality data and model life tables for many developing countries, similar reliability issues arise for this indicator. Extrapolations based on outdated surveys may not be reliable for monitoring changes in health status or for comparative analytical work. Annual data series from the United Nations are interpolated based on five-year estimates and thus may not reflect actual events.”

48

A.3

MMR

The data on MMR come from two different sources. First, we use the data on MMR from the WDI that is available for 181 countries but only for the 5 time periods of 1990, 1995, 2000, 2005, 2010. The relevant MMR variable from the WDI is the Maternal mortality ratio (modeled estimate, per 100,000 live births), defined as “the number of women who die from pregnancyrelated causes while pregnant or within 42 days of pregnancy termination per 100,000 live births. The data are estimated with a regression model using information on the proportion of maternal deaths among non-AIDS deaths in women ages 15-49, fertility, birth attendants, and GDP.” In other words, MMR is defined as:

MMR =

Number of Maternal Deaths × 100, 000 Number of Live Births

(8)

“The modeled estimates are based on an exercise by the Maternal Mortality Estimation Inter-Agency Group (MMEIG) which consists of World Health Organization (WHO), United Nations Children’s Fund (UNICEF), United Nations Population Fund (UNFPA), and World Bank, and include country-level time series data. For countries without complete registration data but with other types of data and for countries with no data, maternal mortality is estimated with a multilevel regression model using available national maternal mortality data and socioeconomic information, including fertility, birth attendants, and GDP.” In the robustness and extension section we also exploit a novel panel dataset on MMR calculated from the sibling linked files in the DHS, recently constructed by Bhalotra and Clarke (2014). The reader is directed to Bhalotra and Clarke (2014) for more details. The availability of this data varies a lot by years which can be seen on Table A.2

49

Table A.1: MMR from the WDI country World World World World World High income High income High income High income High income Middle income Middle income Middle income Middle income Middle income Low income Low income Low income Low income Low income

50

year 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010

MMR 400 360 320 260 210 25 20 18 16 16 370 320 290 230 190 810 740 630 520 410

Table A.2: MMR from the DHS- Availability by year year 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

No. of countries 41 42 43 43 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 45 44 43 42 41 38 38 37 37 37 35 33 31 24 20 17 14 6 2

51

A.4

UN Mortality Data

This data comes from United Nations (2013) and the main variable of interest is constructed as follows: mortality in age category X Dependent variable = ln( female male mortality in age category X ) ∗ 100000; where X can be (0-14),

(15-49), or, (50 +). Where, male mortality in age category X: Percentage of male deaths by broad age group (per 100 male total population); female mortality in age category X : Percentage of female deaths by broad age group (per 100 female total population). This data is available for 193 countries at a 5 yearly intervals, from the years “1955-1960” to the years “2005-2010”. In order to match with our other data we use the following matching: The period “1955-1960” is matched to the year 1955, the period “1960-1965” is matched to the year 1960, ..., the period “2005-2010” is matched to the year 2005.

A.5

Desired Sex Ratio

For the desired sex ratio variable we use 3 questions viz. ideal no. of children, ideal no. of daughters and ideal no. of daughters from the DHS. These questions are available for 63 of the DHS countries. We first start by dropping all the observations for which either the ideal no. of daughters or the ideal no. of daughters variables are missing. We then have 1,947,322 responses for the first question and 1,947,334 responses for the next two, out of which some are non-usable (like non-numeric responses or unusable values coded at 96) or take very high values like more than 10 ideal no. of children etc. We drop these from the sample. In fact we drop all the values which are greater than 10 for any of these questions (This includes the non-numeric responses coded at 96 as well). We then have around 1,797,720 mothers for whom we have responses to each of the above questions. We then generate our desired sex ratio variable as follows:

desired sex ratio = ideal nboys/ideal ngirls

(9)

However, given that the above equation is not defined for the cases in which the ideal no. of girls variable is 0, the above variable gets generated only for 1,458,302 mothers. In 288,867 of the missing cases, we have both ideal no. of boys and ideal no. of girls equal to zero. In these cases we just assign a value of 1 to our DSR variable. Next, we consider the cases for which ideal no. of girls = 0, while ideal no. of boys not = 0. In these cases we assign the country specific values of the 99th percentile of the values of the DSR variable. There were 50,551 such cases. Finally, after these imputations we have the DSR variable available for 1,797,720 mothers and

52

its mean and s.d. are respectively 1.12 and 0.56. List of (the 63) DHS countries for which DSR is available: Albania, Armenia, Azerbaijan, Bangladesh, Benin, Bolivia, Brazil, Burkina-Faso, Burundi, Cambodia, Cameroon, CentralAfrican-Republic, Chad, Colombia, Congo-Brazzaville, Congo-Democratic-Republic, Cote-dIvoire, Dominican-Republic, Egypt, Ethiopia, Gabon, Ghana, Guatemala, Guinea, Guyana, Haiti, Honduras, India, Indonesia, Jordan, Kazakhstan, Kenya, Kyrgyz-Republic, Lesotho, Liberia, Madagascar, Malawi, Maldives, Mali, Morocco, Mozambique, Namibia, Nepal, Nicaragua, Niger, Nigeria, Pakistan, Peru, Philippines, Rwanda, Sao-Tome-and-Principe, Senegal, SierraLeone, Swaziland, Tanzania, Togo, Turkey, Uganda, Ukraine, Uzbekistan, Vietnam, Zambia, Zimbabwe.

53

Table A.3: Son Preferences across countries Rank 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63

Country Pakistan Senegal Nepal India Jordan Mali Niger Nigeria Burkina-Faso Guinea Chad Bangladesh Ethiopia Burundi Congo-Democratic-Republic Egypt Guatemala Rwanda Azerbaijan Lesotho Kenya Benin Guyana Armenia Uzbekistan Swaziland Ghana Albania Vietnam Bolivia Madagascar Maldives Liberia Zimbabwe Tanzania Cameroon Sierra-Leone Mozambique Cote-d-Ivoire Indonesia Kazakhstan Kyrgyz-Republic Togo Congo-Brazzaville Turkey Honduras Zambia Morocco Sao-Tome-and-Principe Namibia Philippines Peru Malawi Brazil Gabon Cambodia Central-African-Republic Uganda Colombia Nicaragua Ukraine Haiti Dominican-Republic

Desired Sex Ratio 1.591361 1.442077 1.420883 1.333632 1.28127 1.263558 1.255963 1.243862 1.234437 1.234308 1.197923 1.190113 1.178427 1.16554 1.151482 1.145844 1.134763 1.133816 1.129156 1.127053 1.125672 1.124392 1.120784 1.118306 1.112029 1.111806 1.108479 1.108138 1.103347 1.093611 1.092916 1.089177 1.08451 1.082583 1.077299 1.074412 1.071624 1.067048 1.06387 1.059097 1.05177 1.050084 1.049242 1.049235 1.046912 1.046143 1.041258 1.038927 1.033501 1.033349 1.032981 1.02985 1.029814 1.023659 1.022349 1.018316 1.013725 1.00685 0.9871559 0.9652104 0.9464852 0.9416622 540.9223508

Desired Fertility 4.04797 5.20824 2.50707 2.4698 4.02775 6.05204 7.43121 5.76644 5.49502 5.54555 7.4775 2.39822 4.41782 4.14199 5.96472 2.86839 3.51829 4.1176 2.47213 2.88506 3.32215 4.92522 2.82163 2.58183 3.61026 2.5052 4.2259 2.55117 2.40708 2.45037 4.71422 3.11338 4.89519 3.91093 5.00151 5.42036 4.87725 5.22022 5.18669 2.80611 2.84618 3.66682 4.43386 4.97025 2.42031 2.85255 4.82581 2.90608 3.4502 3.12124 2.9735 2.41027 4.05177 2.30147 4.58298 3.35046 6.07393 4.86307 2.27366 2.72309 1.973 2.95353 2.97296

Son Preference Very High Very High Very High Very High Very High Very High Very High Very High Very High Very High Very High Very High High High High High High High High High High High High High High Medium Medium Medium Medium Medium Medium Medium Medium Medium Medium Medium Medium Low Low Low Low Low Low Low Low Low Low Low Low Low Lowest Lowest Lowest Lowest Lowest Lowest Lowest Lowest Lowest Lowest Lowest Lowest Lowest

A.6

Women’s Political Rights

“Women’s political rights include: • The right to vote • The right to run for political office • The right to hold elected and appointed government positions • The right to join political parties • The right to petition government officials The coding scheme: In measuring womens political rights we are primarily interested in two things: one, the extensiveness of laws pertaining to womens political rights; and two, government practices towards women or how effectively the government enforces the laws. Regarding the political equality of women: (0) None of womens political rights are guaranteed by law. There are laws that completely restrict the participation of women in the political process. (1) Political equality is guaranteed by law. However, there are significant limitations in practice. Women hold less than five percent of seats in the national legislature and in other high-ranking government positions. (2) Political equality is guaranteed by law. Women hold more than five percent but less than thirty percent of seats in the national legislature and/or in other highranking government positions. (3) Political equality is guaranteed by law and in practice. Women hold more than thirty percent of seats in the national legislature and/or in other high-ranking government positions.” Cingranelli, Richards, and Clay (Cingranelli et al.)

A.7

Women’s Economic Rights

Women’s economic rights include: • Equal pay for equal work • Free choice of profession or employment without the need to obtain a husband or male relative’s consent

55

• The right to gainful employment without the need to obtain a husband or male relative’s consent • Equality in hiring and promotion practices • Job security (maternity leave, unemployment benefits, no arbitrary firing or layoffs, etc...) • Non-discrimination by employers • The right to be free from sexual harassment in the workplace • The right to work at night • The right to work in occupations classified as dangerous • The right to work in the military and the police force In measuring women’s economic rights we are primarily interested in two things: one, the extensiveness of laws pertaining to womens economic rights; and two, government practices towards women or how effectively the government enforces the laws. Regarding the economic equality of women: (0) There are no economic rights for women under law and systematic discrimination based on sex may be built into the law. The government tolerates a high level of discrimination against women. (1) There are some economic rights for women under law. However, in practice, the government DOES NOT enforce the laws effectively or enforcement of laws is weak. The government tolerates a moderate level of discrimination against women. (2) There are some economic rights for women under law. In practice, the government DOES enforce these laws effectively. However, the government still tolerates a low level of discrimination against women. (3) All or nearly all of women’s economic rights are guaranteed by law. In practice, the government fully and vigorously enforces these laws. The government tolerates none or almost no discrimination against women.

A.8

Women’s Social Rights

Women’s social rights include a number of internationally recognized rights. These rights include: • The right to equal inheritance 56

• The right to enter into marriage on a basis of equality with men • The right to travel abroad • The right to obtain a passport • The right to confer citizenship to children or a husband • The right to initiate a divorce • The right to own, acquire, manage, and retain property brought into marriage • The right to participate in social, cultural, and community activities • The right to an education • The freedom to choose a residence/domicile • Freedom from female genital mutilation of children and of adults without their consent • Freedom from forced sterilization A score of 0 indicates that there were no social rights for women in law and that systematic discrimination based on sex may have been built into law. A score of 1 indicates that women had some social rights under law, but these rights were not effectively enforced. A score of 2 indicates that women had some social rights under law, and the government effectively enforced these rights in practice while still allowing a low level of discrimination against women in social matters. Finally, a score of 3 indicates that all or nearly all of womens social rights were guaranteed by law and the government fully and vigorously enforced these laws in practice. [This Variable was retired as of 2005]

57

B

Female Life Expectancy advantage Table B.1: Female Life Expectancy advantage & Desired Sex Ratio (boys/girls) Desired Sex Ratio

(1) -0.0586∗∗∗ (0.0132)

(2) -0.0701∗∗∗ (0.0141)

(3) -0.0749∗∗∗ (0.0144) 0.0101∗∗ (0.00410)

(4) -0.0692∗∗∗ (0.0149) 0.00888∗∗ (0.00408)

No 2944

No 2944 0.131

No 2564 0.235

Yes 2564 0.251

lGDP Desired Sex Ratio* lGDP Desired Fertility N r2 ∗

(5) 0.0102 (0.0688) 0.0266 (0.0167) -0.0167 (0.0153) Yes 2564 0.258

p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 The dependent variables is the log of the ratio of female to male life expectancy. Country fixed effects panel regressions with year dummies have been run (except the first column where only the DSR variable has been controlled for). Standard errors in parentheses are clustered at the country level. The data on Life Expectancy comes from The World Bank WDI. The desired sex ratio (DSR) data comes from the DHS and has been constructed using the questions on ideal number of boys and girls asked to the mothers. These regressions are hence based on the DHS sample (developing countries) of 63 countries for the years over the period of 1961 - 2011. The mean (s.d.) of the log ratio of female to male life expectancy and DSR are 0.058 (0.036), 1.121 (0.147).

58

Table B.2: Female Life Expectancy Advantage and Women’s Rights Political Rights log GDP

(1) -37.10 (159.9) 584.9∗∗ (274.9)

(2) -59.98 (160.1) 607.7∗ (342.2)

Democracy

(3) -28.56 (170.6) 635.0∗ (347.0) 2.116 (35.13)

Rights * lGDP

(4) 880.4 (549.6) 882.5∗ (448.3)

-126.8 (76.85)

(5) 921.2 (619.7) 910.9∗ (468.9) -6.386 (35.16) -128.6 (86.17)

Democracy * lGDP N r2 Economic Rights log GDP

4588

4588 0.178

4207 0.184

4588 0.182

4207 0.188

40.75 (95.72) 583.4∗∗ (280.6)

33.49 (97.44) 609.9∗ (343.3)

51.61 (103.3) 636.9∗ (347.1) -0.405 (34.38)

958.7∗∗ (452.8) 755.6∗∗ (354.7) -117.3∗∗ (51.94)

1020.9∗∗ (475.4) 791.8∗∗ (361.3) -3.515 (34.23) -123.6∗∗ (54.55)

Democracy Rights * lGDP Democracy * lGDP N r2 Social Rights log GDP

4549

4549 0.177

4168 0.184

4549 0.181

4168 0.188

99.62 (123.4) 770.3∗∗ (349.1)

63.42 (126.3) 903.4∗ (460.8)

112.6 (131.1) 951.7∗∗ (479.0) 1.193 (35.08)

1133.5∗∗ (513.7) 1097.7∗∗ (512.8) -141.5∗∗ (61.80)

1204.2∗∗ (541.0) 1151.5∗∗ (534.1) -4.390 (34.63) -145.5∗∗ (65.55)

3251 0.166

3045 0.177

Democracy Rights * lGDP Democracy * lGDP N r2

3251

3251 0.160

3045 0.172

(6) 777.7 (545.1) 991.8∗ (528.2) 121.1 (170.6) -110.4 (76.40) -18.55 (24.45) 4207 0.189 920.9∗ (466.4) 906.7∗∗ (449.1) 135.8 (184.1) -109.5∗∗ (53.86) -20.40 (26.03) 4168 0.189 1075.6∗∗ (450.5) 1321.8∗ (730.8) 197.3 (298.4) -127.9∗∗ (53.60) -30.17 (43.90) 3045 0.179

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is the log of the ratio of female to male life expectancy times 100,000. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set. These regressions are based on a sample of around 150 countries from around the world and approximately for the years 1981-2011 (Panels 1 & 2) and 1981-2004 (Panel 3). The mean (s.d.) of the dependent variable is 7058.501 (3623.966) and that of women’s political, economic and social rights variables respectively are 1.806 (0.623), 1.308 (0.682) and 1.235 (0.828).

59

Table B.3: Female LE advantage and Women’s Political Representation Log (Female LE/ Male LE) * 100000 womparl log GDP

(1)

(2)

(3)

(4)

(5)

(6)

6.084 (9.683) 204.1 (159.9)

5.253 (10.06) 136.2 (194.3)

9.931 (10.47) 209.3 (174.5) 41.95 (37.70)

85.72∗∗∗ (23.53) 260.3 (197.9) -10.47∗∗∗ (3.195)

83.67∗∗∗ (23.67) 316.1∗ (176.6) 34.86 (36.53) -9.718∗∗∗ (3.493)

2465 0.0948

2136 0.125

71.88∗∗∗ (23.25) 377.2∗∗ (189.2) 208.2 (139.8) -8.285∗∗ (3.546) -24.72 (20.11) 2136 0.129

Democracy womparl*lgdp democ*lgdp N r2

2465

2465 0.0818

2136 0.113

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is log of the ratio of female to male life expectancy times 100,000 respectively. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. Both the Life expectancy data and the women’s political representation data comes from the WDI. These regressions are based on a sample of around 150 to 179 countries from around the world for the period of 1997 to 2011. The mean and s.d. of the log ratio of female/ male LE (times 100,000) are 6844.572 and 3709.42, whereas the mean and s.d. of women’s representation in parliament is around 15.193 and 10.204.

Table B.4: Female Life Expectancy Advantage & Gender Intensity of Language

GII N r2 GII lgdp N r2 GII GII gdp lgdp N r2

(1) ngii -1676.3∗∗∗ (505.1) 5079 0.277

(2) sbii -1949.5∗∗∗ (646.9) 5079 0.279

(3) gaii -1951.8∗∗∗ (601.0) 3925 0.295

(4) gpii -1545.5∗∗∗ (511.8) 5007 0.272

(5) gii0 -654.7∗∗∗ (159.0) 3725 0.328

(6) gii1 -902.1∗∗∗ (208.4) 3925 0.321

(7) gii2 -803.0∗∗∗ (222.1) 4736 0.291

(8) gtroiano -396.5∗ (210.3) 3596 0.291

-2111.7*** (565.7) -457.9* (272.4) 4865 0.378

-3156.1*** (753.5) -413.9 (259.4) 4865 0.403

-2012.8** (790.0) 128.7 (306.4) 3733 0.384

-2188.1*** (631.7) -373.7 (273.2) 4743 0.371

-916.1*** (255.6) 123.2 (303.6) 3575 0.408

-1170.0*** (314.3) 94.42 (305.4) 3733 0.414

-1317.5*** (268.5) -382.8 (271.8) 4564 0.397

-558.7* (300.3) -345.9 (284.8) 3428 0.367

-5751.6*** (2173.4) 463.1* (237.6) -671.8** (297.8) 4865 0.387

-3330.1 (2382.4) 22.95 (278.2) -429.9 (347.0) 4865 0.403

-5354.0* (2809.2) 446.2 (344.2) -189.4 (448.8) 3733 0.390

-1730.7 (2060.1) -60.96 (270.3) -351.9 (300.3) 4743 0.371

-1821.3** (797.9) 125.3 (103.7) -218.4 (479.8) 3575 0.412

-2106.4* (1145.7) 130.6 (144.1) -184.4 (503.3) 3733 0.417

-1409.1* (803.6) 12.41 (98.70) -402.6 (326.5) 4564 0.397

-169.1 (1085.7) -48.74 (122.4) -233.9 (391.8) 3428 0.368

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables in all three panels are the log of the ratio of female to male life expectancy times 100,000 (from the World Bank WDI database). Standard errors in parentheses are clustered at the country level. The GII data come from Gay et al. (2013) and Givati and Troiano (2012). Apart from the GII variable in Panel 1 we control for the percentage of the population speaking the majority language (for which the GII has been calculated), decade dummies and continent dummies. In Panel 2 we control for the log of GDP, the log of population, dummies for the World Bank Income groups classification, the percentage of population that is Protestant, Catholic and Muslim, and the proportion of the country that is tropical or subtropical in addition to the controls from Panel 1. In Panel 3, we add the interaction term of log of GDP and GII in addition to the controls from the previous panels.

60

C

Infant and Child Mortality Ratios and Gender Inequality Table C.1: Infant Mortality Ratio (F/M) and Women’s Rights Political Rights Rights log GDP

(1)

(2)

(3)

(4)

(5)

(6)

-0.00425∗∗ (0.00215) -0.00235∗ (0.00131)

-0.00199 (0.00226) 0.00213 (0.00258)

-0.00263 (0.00273) 0.00141 (0.00285) -0.00144∗∗ (0.000640)

-0.0422∗∗∗ (0.00945) -0.00791∗∗ (0.00360) 0.00520∗∗∗ (0.00117)

-0.0531∗∗∗ (0.0105) -0.0117∗∗∗ (0.00385) -0.000793 (0.000658) 0.00652∗∗∗ (0.00135)

-0.0370∗∗∗ (0.00998) -0.0133∗∗∗ (0.00365) -0.00909∗∗∗ (0.00242) 0.00474∗∗∗ (0.00126) 0.00120∗∗∗ (0.000365) 454 0.151

Democracy Rights * lGDP Democracy * lGDP N r2 Economic Rights Rights log GDP

518

518 0.0149

454 0.0357

518 0.0738

454 0.103

-0.000164 (0.00236) -0.00237 (0.00152)

0.00287 (0.00258) 0.00222 (0.00259)

0.00218 (0.00295) 0.00154 (0.00286) -0.00139∗∗ (0.000620)

-0.0200∗ (0.0102) -0.000130 (0.00291) 0.00282∗∗ (0.00133)

-0.0201∗ (0.0109) -0.000721 (0.00319) -0.00131∗∗ (0.000620) 0.00277∗ (0.00146)

Democracy Rights * lGDP Democracy * lGDP N r2 Social Rights Rights log GDP

516

516 0.0180

452 0.0353

516 0.0348

452 0.0506

-0.00696∗∗∗ (0.00250) -0.00344∗∗ (0.00171)

0.000789 (0.00330) 0.00212 (0.00329)

0.000637 (0.00335) 0.00148 (0.00342) -0.000511 (0.000563)

-0.0436∗∗∗ (0.0146) -0.00459 (0.00407) 0.00594∗∗∗ (0.00210)

-0.0464∗∗∗ (0.0150) -0.00561 (0.00422) -0.000297 (0.000576) 0.00634∗∗∗ (0.00217)

327 0.0616

293 0.0712

Democracy Rights * lGDP Democracy * lGDP N r2

327

327 0.00281

293 0.00628

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

61

-0.0103 (0.0104) -0.00597∗∗ (0.00286) -0.0110∗∗∗ (0.00236) 0.00135 (0.00143) 0.00141∗∗∗ (0.000355) 452 0.122 -0.0422∗∗∗ (0.0147) -0.0111∗∗ (0.00461) -0.00727∗∗ (0.00364) 0.00585∗∗∗ (0.00214) 0.00109∗ (0.000596) 293 0.0944

Table C.2: Infant Mortality Ratio (F/M) and Composite Women’s Rights Political, Economic, Social Rights log GDP

(1)

(2)

(3)

(4)

(5)

(6)

-0.00545∗∗∗ (0.00165) -0.00304∗ (0.00176)

-0.00109 (0.00181) 0.00218 (0.00333)

-0.00152 (0.00188) 0.00152 (0.00347) -0.000505 (0.000578)

-0.0334∗∗∗ (0.00924) 0.00109 (0.00347) 0.00447∗∗∗ (0.00132)

-0.0385∗∗∗ (0.00969) -0.0000517 (0.00359) -0.00000224 (0.000610) 0.00512∗∗∗ (0.00138)

-0.0344∗∗∗ (0.00981) -0.00477 (0.00423) -0.00573 (0.00389) 0.00462∗∗∗ (0.00140) 0.000886 (0.000630) 292 0.0956

Democracy Rights * lGDP Democracy * lGDP N r2 Political & Economic Rights log GDP

326

326 0.00433

292 0.00970

326 0.0642

292 0.0807

-0.00173 (0.00156) -0.00207 (0.00142)

0.000592 (0.00170) 0.00237 (0.00258)

0.0000591 (0.00202) 0.00171 (0.00284) -0.00144∗∗ (0.000628)

-0.0253∗∗∗ (0.00626) 0.00337 (0.00259) 0.00324∗∗∗ (0.000819)

-0.0290∗∗∗ (0.00734) 0.00263 (0.00279) -0.000924 (0.000640) 0.00370∗∗∗ (0.00100)

516 0.0741

452 0.0962

Democracy Rights * lGDP Democracy * lGDP N r2

516

516 0.0131

452 0.0326

-0.0191∗∗∗ (0.00717) -0.00315 (0.00238) -0.00913∗∗∗ (0.00236) 0.00247∗∗ (0.000994) 0.00118∗∗∗ (0.000356) 452 0.144

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table C.3: Infant Mortality Ratio (F/M) and Women’s Parliamentary Representation Rights log GDP

(1) -0.000372∗∗∗ (0.000144) -0.000909 (0.00135)

(2) -0.000367∗∗ (0.000184) 0.000503 (0.00275)

Democracy

(3) -0.000439∗ (0.000225) -0.000510 (0.00311) -0.00123∗ (0.000716)

Rights * lGDP

(4) -0.00305∗∗∗ (0.000533) -0.00265 (0.00275) 0.000339∗∗∗ (0.0000650)

(5) -0.00328∗∗∗ (0.000583) -0.00346 (0.00301) -0.000465 (0.000692) 0.000378∗∗∗ (0.0000741)

360 0.197

304 0.220

Democracy * lGDP N r2

360

360 0.0599

304 0.0787

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

62

(6) -0.00295∗∗∗ (0.000577) -0.00432 (0.00291) -0.00456∗ (0.00260) 0.000341∗∗∗ (0.0000748) 0.000575∗ (0.000328) 304 0.238

Table C.4: Child Mortality Ratio (F/M) and Women’s Rights Political Rights log GDP

(1)

(2)

(3)

(4)

(5)

(6)

-0.00286 (0.00256) -0.0114∗∗∗ (0.00153)

0.000395 (0.00283) -0.00223 (0.00310)

0.000149 (0.00344) -0.00264 (0.00349) -0.00126∗ (0.000754)

-0.0501∗∗∗ (0.0115) -0.0148∗∗∗ (0.00438) 0.00652∗∗∗ (0.00137)

-0.0629∗∗∗ (0.0132) -0.0189∗∗∗ (0.00493) -0.000450 (0.000779) 0.00813∗∗∗ (0.00162)

-0.0461∗∗∗ (0.0130) -0.0206∗∗∗ (0.00474) -0.00905∗∗∗ (0.00266) 0.00629∗∗∗ (0.00156) 0.00124∗∗∗ (0.000389) 454 0.174

Democracy Rights * lGDP Democracy * lGDP N r2 Economic Rights log GDP

518

518 0.0659

454 0.0720

518 0.126

454 0.140

-0.000555 (0.00252) -0.0115∗∗∗ (0.00170)

0.00360 (0.00274) -0.00263 (0.00312)

0.00279 (0.00316) -0.00316 (0.00349) -0.00120 (0.000756)

-0.0329∗∗∗ (0.0117) -0.00637∗ (0.00344) 0.00450∗∗∗ (0.00143)

-0.0350∗∗∗ (0.0126) -0.00701∗ (0.00379) -0.00107 (0.000757) 0.00471∗∗∗ (0.00156)

Democracy Rights * lGDP Democracy * lGDP N r2 Rights log GDP

516

516 0.0709

452 0.0746

516 0.0985

452 0.103

-0.00725∗∗ (0.00297) -0.0132∗∗∗ (0.00199)

0.00393 (0.00413) -0.00140 (0.00446)

0.00332 (0.00420) -0.00189 (0.00464) -0.000395 (0.000742)

-0.0559∗∗∗ (0.0159) -0.0104∗∗ (0.00519) 0.00800∗∗∗ (0.00214)

-0.0559∗∗∗ (0.0167) -0.0108∗∗ (0.00542) -0.000126 (0.000728) 0.00798∗∗∗ (0.00227)

327 0.0864

293 0.0829

Democracy Rights * lGDP Democracy * lGDP N r2

327

327 0.0150

293 0.0142

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

63

-0.0255∗∗ (0.0117) -0.0121∗∗∗ (0.00372) -0.0105∗∗∗ (0.00268) 0.00332∗∗ (0.00149) 0.00137∗∗∗ (0.000393) 452 0.148 -0.0510∗∗∗ (0.0164) -0.0172∗∗ (0.00676) -0.00826∗ (0.00475) 0.00741∗∗∗ (0.00222) 0.00127 (0.000775) 293 0.104

Table C.5: Infant Mortality Ratio (F/M) and Composite Women’s Rights Political, Economic, Social Rights log GDP

(1)

(2)

(3)

(4)

(5)

(6)

-0.00551∗∗∗ (0.00207) -0.0129∗∗∗ (0.00205)

0.000934 (0.00251) -0.00109 (0.00453)

0.000297 (0.00263) -0.00157 (0.00473) -0.000453 (0.000741)

-0.0478∗∗∗ (0.0106) -0.00274 (0.00468) 0.00675∗∗∗ (0.00146)

-0.0513∗∗∗ (0.0121) -0.00376 (0.00491) 0.000249 (0.000780) 0.00715∗∗∗ (0.00166)

-0.0468∗∗∗ (0.0119) -0.00902 (0.00647) -0.00615 (0.00505) 0.00659∗∗∗ (0.00164) 0.000989 (0.000817) 292 0.114

Democracy Rights * lGDP Democracy * lGDP N r2 Political & Economic Rights log GDP

326

326 0.00938

292 0.00969

326 0.101

292 0.102

-0.00134 (0.00170) -0.0113∗∗∗ (0.00161)

0.00189 (0.00188) -0.00231 (0.00310)

0.00145 (0.00225) -0.00281 (0.00345) -0.00123 (0.000752)

-0.0329∗∗∗ (0.00743) -0.000972 (0.00311) 0.00436∗∗∗ (0.000901)

-0.0388∗∗∗ (0.00833) -0.00155 (0.00340) -0.000512 (0.000766) 0.00512∗∗∗ (0.00105)

516 0.140

452 0.153

Democracy Rights * lGDP Democracy * lGDP N r2

516

516 0.0686

452 0.0733

-0.0298∗∗∗ (0.00811) -0.00682∗∗ (0.00331) -0.00800∗∗∗ (0.00263) 0.00401∗∗∗ (0.00104) 0.00108∗∗∗ (0.000390) 452 0.179

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table C.6: Child Mortality Ratio (F/M) and Women’s Parliamentary Representation Rights log GDP

(1) -0.000433∗∗ (0.000172) -0.00881∗∗∗ (0.00160)

(2) -0.000441∗∗ (0.000221) -0.00142 (0.00325)

Democracy

(3) -0.000414 (0.000280) -0.00163 (0.00375) -0.000831 (0.000930)

Rights * lGDP

(4) -0.00393∗∗∗ (0.000718) -0.00550∗ (0.00324) 0.000439∗∗∗ (0.0000927)

(5) -0.00454∗∗∗ (0.000692) -0.00593∗ (0.00352) 0.000275 (0.000913) 0.000549∗∗∗ (0.0000870)

360 0.289

304 0.331

Democracy * lGDP N r2

360

360 0.152

304 0.150

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

64

(6) -0.00427∗∗∗ (0.000717) -0.00663∗∗ (0.00335) -0.00311 (0.00321) 0.000518∗∗∗ (0.0000900) 0.000476 (0.000391) 304 0.338

Table C.7: Infant Mortality Ratio (F/M) & GII GII N r2 GII log GDP N r2 GII GII gdp log GDP N r2

(1) 0.0230∗∗ (0.00936) 488 0.125

(2) 0.0244∗∗ (0.0111) 488 0.121

(3) 0.0313∗∗ (0.0125) 356 0.164

(4) 0.0172∗ (0.0103) 484 0.0959

(5) 0.00881∗∗ (0.00360) 336 0.158

(6) 0.0119∗∗ (0.00477) 356 0.169

(7) 0.00976∗∗ (0.00421) 456 0.129

(8) -0.00419 (0.00453) 320 0.338

0.0239∗ (0.0125) -0.00855 (0.00544) 341 0.200

0.0304∗ (0.0175) -0.00888 (0.00562) 341 0.205

0.0314∗∗ (0.0155) -0.0123∗ (0.00632) 247 0.291

0.0221∗ (0.0119) -0.00808 (0.00548) 333 0.186

0.0106∗ (0.00587) -0.0110∗ (0.00650) 236 0.268

0.0136∗ (0.00758) -0.0110∗ (0.00637) 247 0.293

0.0126∗ (0.00653) -0.00809 (0.00553) 321 0.204

-0.00328 (0.00367) -0.0147∗∗ (0.00644) 227 0.455

0.0947∗ (0.0535) -0.00851 (0.00539) -0.00410 (0.00586) 341 0.216

0.107∗∗ (0.0539) -0.00958∗ (0.00538) -0.00195 (0.00679) 341 0.222

0.133∗∗ (0.0597) -0.0126∗∗ (0.00633) -0.00328 (0.00847) 247 0.317

0.0595 (0.0445) -0.00462 (0.00512) -0.00647 (0.00592) 333 0.189

0.0353∗ (0.0189) -0.00321 (0.00208) -0.00191 (0.00920) 236 0.282

0.0365 (0.0255) -0.00300 (0.00282) -0.00441 (0.00950) 247 0.301

0.0390∗∗ (0.0191) -0.00338 (0.00208) -0.00248 (0.00629) 321 0.214

-0.0295 (0.0209) 0.00300 (0.00227) -0.0224∗∗ (0.00863) 227 0.464

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table C.8: Child Mortality Ratio (F/M) & GII

GII N r2 GII log GDP N r2 GII GII gdp log GDP N r2

(1) ngii 0.0254∗∗ (0.0103) 488 0.248

(2) sbii 0.0246∗∗ (0.0121) 488 0.237

(3) gaii 0.0319∗∗ (0.0132) 356 0.278

(4) gpii 0.0108 (0.00992) 484 0.220

(5) gii0 0.00797∗∗ (0.00367) 336 0.269

(6) gii1 0.0122∗∗ (0.00513) 356 0.283

(7) gii2 0.00889∗∗ (0.00432) 456 0.238

(8) gtroiano -0.00614 (0.00564) 320 0.359

0.0295∗∗ (0.0139) -0.0146∗∗∗ (0.00522) 341 0.367

0.0397∗∗ (0.0195) -0.0150∗∗∗ (0.00542) 341 0.377

0.0396∗∗ (0.0178) -0.0182∗∗∗ (0.00627) 247 0.442

0.0238∗ (0.0123) -0.0145∗∗∗ (0.00548) 333 0.364

0.0127∗ (0.00663) -0.0165∗∗ (0.00639) 236 0.417

0.0178∗∗ (0.00870) -0.0164∗∗ (0.00627) 247 0.449

0.0150∗∗ (0.00703) -0.0143∗∗∗ (0.00539) 321 0.368

-0.00464 (0.00379) -0.0212∗∗∗ (0.00696) 227 0.527

0.139∗∗ (0.0607) -0.0131∗∗ (0.00604) -0.00777 (0.00548) 341 0.393

0.135∗∗ (0.0581) -0.0119∗∗ (0.00562) -0.00645 (0.00657) 341 0.395

0.156∗∗ (0.0678) -0.0145∗∗ (0.00714) -0.00777 (0.00871) 247 0.465

0.0791 (0.0519) -0.00683 (0.00599) -0.0122∗ (0.00615) 333 0.368

0.0530∗∗ (0.0208) -0.00523∗∗ (0.00225) -0.00165 (0.00934) 236 0.441

0.0575∗∗ (0.0284) -0.00521∗ (0.00307) -0.00495 (0.00984) 247 0.464

0.0520∗∗ (0.0204) -0.00473∗∗ (0.00215) -0.00648 (0.00607) 321 0.382

-0.0351 (0.0286) 0.00350 (0.00308) -0.0302∗∗ (0.0116) 227 0.536

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

65

D

Group Fixed Effects Framework

D.1

Desired Sex ratio (boys/girls) and the Time Varying Group Fixed Effects estimator

Table D.1: Female LE advantage and DSR (boys/girls): Time varying GFE framework

Desired Sex Ratio log GDP N r2

(1) group0 -0.0762∗∗∗ (0.0140) 0.0102∗∗∗ (0.00387) 2564

(2) group2 -0.0812∗∗∗ (0.0140) 0.0105∗∗∗ (0.00364) 2564

(3) group3 -0.0817∗∗∗ (0.0131) 0.0111∗∗∗ (0.00348) 2564

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

66

(4) group4 -0.0596∗∗∗ (0.0137) 0.00921∗∗∗ (0.00250) 2564

(5) group5 -0.0555∗∗∗ (0.0143) 0.00935∗∗∗ (0.00245) 2564

(6) group6 -0.0505∗∗∗ (0.0158) 0.00887∗∗∗ (0.00221) 2564

D.2

Women’s Rights: Groups Fixed effects framework Table D.2: Female LE advantage and Women’s Rights: Time varying GFE framework

Political Rights Rights log GDP Democracy Rights * lGDP Democracy * lGDP N r2 Economic Rights Rights log GDP Democracy Rights * lGDP Democracy * lGDP N r2 Social Rights Rights log GDP Democracy Rights * lGDP Democracy * lGDP N r2

(1) group0

(2) group2

(3) group3

(4) group4

(5) group5

(6) group6

739.8 (535.2) 929.6∗∗ (467.3) 103.0 (163.9) -102.3 (75.39) -15.02 (23.40) 4207

740.8 (539.6) 900.2∗∗ (370.5) 25.20 (148.0) -100.8 (76.77) -0.558 (20.47) 4207

649.6 (539.7) 1049.7∗∗∗ (361.6) 67.60 (146.1) -81.85 (76.37) -10.33 (20.33) 4207

393.2 (526.3) 848.4∗∗∗ (284.9) -76.04 (130.9) -49.39 (74.82) 18.68 (17.88) 4207

150.0 (479.6) 576.4∗∗∗ (195.9) -127.0 (129.0) -17.21 (64.33) 22.65 (18.04) 4207

396.9 (435.5) 731.6∗∗∗ (154.7) -97.20 (131.7) -57.30 (58.98) 12.66 (18.12) 4207

998.2∗∗ (465.6) 870.3∗∗ (391.7) 112.9 (177.1) -118.4∗∗ (53.68) -16.24 (25.05) 4168

1056.9∗∗ (483.1) 843.7∗∗∗ (311.2) 20.07 (154.6) -127.8∗∗ (57.26) 0.960 (21.22) 4168

989.6∗∗ (484.2) 844.0∗∗∗ (274.1) 21.32 (153.1) -122.2∗∗ (56.97) -1.603 (21.00) 4168

787.0 (504.1) 829.5∗∗∗ (230.3) -85.84 (141.1) -91.51 (59.30) 19.62 (19.11) 4168

1044.6∗∗ (456.4) 532.2∗∗∗ (143.6) -64.76 (141.9) -135.8∗∗∗ (52.22) 9.036 (19.25) 4168

515.0 (462.3) 564.3∗∗∗ (128.1) -124.9 (135.5) -60.75 (55.31) 17.10 (18.53) 4168

1151.2∗∗∗ (443.9) 1152.8∗ (618.0) 142.8 (278.3) -133.6∗∗ (52.67) -21.30 (41.11) 3045

1208.8∗∗ (491.5) 1063.0∗∗ (471.4) 28.76 (226.6) -141.2∗∗ (60.94) 0.402 (31.47) 3045

949.4∗∗ (395.4) 663.6∗∗∗ (218.4) -160.7 (169.9) -120.1∗∗ (48.22) 25.08 (24.07) 3045

805.8∗ (414.9) 691.7∗∗∗ (170.5) -277.7∗ (158.8) -90.59∗ (51.46) 46.87∗∗ (22.50) 3045

702.1∗ (414.8) 540.8∗∗∗ (165.9) -317.3∗∗ (158.0) -81.86 (51.09) 52.13∗∗ (22.22) 3045

384.8 (374.5) 442.5∗∗∗ (147.3) -297.7∗∗ (148.7) -51.68 (45.45) 42.24∗∗ (20.87) 3045

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

67

D.3

Women’s Parliamentary representation: Group FE approach

Table D.3: Female LE advantage and Women’s Parliamentary Representation: Time varying GFE framework

Women’s Parliamentary Representation Rights log GDP Democracy Rights * lGDP Democracy * lGDP N r2

(1) group0

(2) group2

(3) group3

(4) group4

(5) group5

(6) group6

72.63∗∗∗ (23.22) 438.6∗∗ (174.9) 184.2 (135.7) -8.206∗∗ (3.523) -19.81 (19.25) 2101

76.66∗∗∗ (23.58) 595.5∗∗∗ (161.4) 91.98 (130.8) -8.576∗∗ (3.635) -7.033 (18.24) 2101

66.65∗∗∗ (22.52) 655.5∗∗∗ (155.9) 137.6 (131.5) -6.984∗∗ (3.376) -11.86 (17.98) 2101

53.88∗∗ (25.67) 732.2∗∗∗ (140.7) 9.828 (120.5) -5.457 (3.633) 11.42 (16.20) 2101

56.64∗∗ (23.66) 740.4∗∗∗ (107.8) 229.1∗ (117.4) -6.765∗∗ (3.204) -21.13 (15.05) 2101

72.57∗∗∗ (20.58) 805.1∗∗∗ (99.75) 210.4∗∗ (106.7) -8.426∗∗∗ (2.791) -17.59 (13.88) 2101

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

68

D.4

GII in the GFE framework Table D.4: Female LE advantage and gii0: Time varying GFE framework

gii0 log GDP N r2

(1) group0 -572.1∗∗ (267.4) 221.6 (228.9) 3575 0.282

(2) group2 46.71 (191.0) -6.787 (156.0) 3575 0.612

(3) group3 -321.3 (225.9) 463.7∗∗∗ (123.8) 3575 0.732

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

69

(4) group4 131.2 (99.45) 231.2∗∗ (116.0) 3575 0.810

(5) group5 73.63 (96.46) 64.26 (77.82) 3575 0.844

(6) group6 -142.2∗ (77.81) 199.4∗∗ (88.04) 3575 0.872

E

Other Tables Table E.1: Life Expectancy by decade and region

Middle Middle Middle Middle Middle

Latin Latin Latin Latin Latin

East East East East East

& & & & &

North North North North North

Africa Africa Africa Africa Africa

sub-Saharan Africa sub-Saharan Africa sub-Saharan Africa sub-Saharan Africa sub-Saharan Africa America & Caribbean America & Caribbean America & Caribbean America & Caribbean America & Caribbean

Least Least Least Least Least

developed developed developed developed developed

Region World World World World World (all income levels) (all income levels) (all income levels) (all income levels) (all income levels) OECD members OECD members OECD members OECD members OECD members South Asia South Asia South Asia South Asia South Asia (all income levels) (all income levels) (all income levels) (all income levels) (all income levels) (all income levels) (all income levels) (all income levels) (all income levels) (all income levels)

decade 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000

LE Male 53.7481 59.4446 62.2737 64.3076 66.8484 48.8723 54.6846 58.9167 66.2945 68.9153 65.4151 67.6281 70.1225 72.5199 75.3114 46.2999 52.2994 56.9288 59.7967 63.0256 40.7349 44.6767 47.7784 48.6621 51.2612 56.1503 59.8519 63.269 66.6794 69.6879

LE Female 57.4589 63.4192 66.6568 68.7142 71.0582 50.4199 56.978 64.5732 69.5498 72.472 71.4261 74.1939 76.9595 79.0504 81.0886 44.8177 51.7196 57.2863 61.3009 65.6846 43.5057 47.6637 50.8705 51.3173 53.0988 60.2377 64.7885 69.48 73.1856 76.0673

LE (F - M) 3.710785 3.974562 4.383033 4.406686 4.209783 1.547623 2.293405 5.656537 3.255312 3.556709 6.011083 6.56582 6.837035 6.530491 5.777154 -1.48221 -0.579765 0.3575199 1.504152 2.659069 2.770884 2.987085 3.092166 2.65522 1.837569 4.087476 4.936647 6.210935 6.50623 6.379373

High income High income High income High income High income Middle income Middle income Middle income Middle income Middle income Low income Low income Low income Low income Low income UN classification UN classification UN classification UN classification UN classification

1960 1970 1980 1990 2000 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000

66.064 67.9009 70.0511 71.6693 74.0566 50.0512 58.0974 61.3373 63.8108 66.5391 42.8778 46.182 50.7575 53.1788 56.7958 41.8294 44.9399 49.463 52.4964 56.6317

72.5002 75.0255 77.2724 78.8137 80.5404 52.2621 60.8193 64.8435 67.5757 70.3576 44.9947 48.4645 53.0638 55.4021 58.8922 43.5694 46.8438 51.4468 54.4204 58.6034

6.436127 7.124637 7.221369 7.144483 6.483738 2.210822 2.721917 3.50621 3.764846 3.818524 2.11688 2.282499 2.306221 2.223319 2.096442 1.739977 1.903901 1.983829 1.923992 1.971654

countries: countries: countries: countries: countries:

70

Table E.2: MMR from the WDI (our sample of countries) year 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010

Obs 181 181 181 181 181 51 51 51 51 51 95 95 95 95 95 35 35 35 35 35

Mean 303.41 277.16 239.40 203.14 169.17 47.29 39.33 25.84 19.90 18.65 258.85 225.28 200.96 172.97 145.81 797.54 764.51 654.91 552.03 451.91

Std. Dev. 378.68 361.32 307.33 265.35 224.05 166.67 138.63 62.38 37.74 33.81 304.66 262.87 226.20 197.04 164.70 326.25 367.43 320.29 291.73 266.97

Min 6 2 4 2 2 6 2 4 2 2 8 10 10 9 4 73 98 82 77 65

Max 1600 1900 1300 1100 1100 1200 1000 450 270 240 1600 1200 970 820 730 1300 1900 1300 1100 1100

High High High High High Middle Middle Middle Middle Middle Low Low Low Low Low

Region World World World World World Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income

Table E.3: MMR from the WDI (our sample of countries) year 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010 1990 1995 2000 2005 2010

Obs 181 181 181 181 181 51 51 51 51 51 95 95 95 95 95 35 35 35 35 35

Mean 303.41 277.16 239.40 203.14 169.17 47.29 39.33 25.84 19.90 18.65 258.85 225.28 200.96 172.97 145.81 797.54 764.51 654.91 552.03 451.91

Std. Dev. 378.68 361.32 307.33 265.35 224.05 166.67 138.63 62.38 37.74 33.81 304.66 262.87 226.20 197.04 164.70 326.25 367.43 320.29 291.73 266.97

71

Min 6 2 4 2 2 6 2 4 2 2 8 10 10 9 4 73 98 82 77 65

Max 1600 1900 1300 1100 1100 1200 1000 450 270 240 1600 1200 970 820 730 1300 1900 1300 1100 1100

High High High High High Middle Middle Middle Middle Middle Low Low Low Low Low

Region World World World World World Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income

Table E.4: Desired Sex ratio year 1970 1975 1980 1985 1990 1995 2000 2005 2010 1970 1975 1980 1985 1990 1995 2000 2005 2010 1970 1975 1980 1985 1990 1995 2000 2005 2010

Obs 24 26 26 26 26 26 26 26 24 33 37 37 37 37 37 37 37 34 57 63 63 63 63 63 63 63 58

Mean 1.1404 1.1325 1.1345 1.1351 1.1339 1.1322 1.1280 1.1306 1.0950 1.1170 1.1093 1.1103 1.1099 1.1104 1.1076 1.0994 1.0892 1.0811 1.1269 1.1189 1.1203 1.1203 1.1201 1.1178 1.1112 1.1063 1.0868

Std. Dev. 0.1447 0.1302 0.1220 0.1161 0.1108 0.1016 0.0928 0.1295 0.0854 0.2099 0.1510 0.1390 0.1337 0.1321 0.1309 0.1276 0.1284 0.1405 0.1841 0.1421 0.1318 0.1264 0.1233 0.1195 0.1146 0.1295 0.1201

Min 0.9726 0.9711 0.9732 0.9669 0.9558 0.9492 0.9413 0.9280 0.9138 0.6818 0.9165 0.9204 0.9221 0.9266 0.9315 0.9344 0.9280 0.9149 0.6818 0.9165 0.9204 0.9221 0.9266 0.9315 0.9344 0.9280 0.9138

Max 1.6385 1.5894 1.5572 1.5197 1.4802 1.4265 1.3438 1.6110 1.2340 1.8416 1.6534 1.6346 1.6227 1.6182 1.5913 1.5575 1.5739 1.5929 1.8416 1.6534 1.6346 1.6227 1.6182 1.5913 1.5575 1.6110 1.5929

Low Low Low Low Low Low Low Low Low Middle Middle Middle Middle Middle Middle Middle Middle Middle

Region Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income World World World World World World World World World

Table E.5: Women’s Political Rights Year 1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010

Obs 29 30 33 32 34 34 35 64 65 75 76 98 100 101 36 37 43 45 56 56 56 129 132 151 153 188 190 192

Mean 1.310345 1.348333 1.490909 1.520312 1.735294 1.945098 2.028571 1.484375 1.639231 1.589333 1.740132 1.787755 1.865333 1.975248 1.513889 1.556757 1.72093 1.8 1.864286 2.082143 2.16369 1.453488 1.55 1.605298 1.711765 1.801064 1.943509 2.039931

s.d. 0.67366 0.50863 0.530397 0.628857 0.547658 0.4914 0.487986 0.654464 0.544324 0.506064 0.378948 0.482106 0.452352 0.432298 0.769843 0.824803 0.817825 0.795442 0.774228 0.635641 0.547484 0.691584 0.634233 0.616824 0.587444 0.593187 0.525215 0.482983

72

Min 0 0.6 0 0 0 0 0.25 0 0 0 0.8 0 0.6 0.75 0 0 0 0 0 0 0.75 0 0 0 0 0 0 0.25

Max 2 2 2 2 2.2 3 3 3 3 2.2 2.25 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3

Low Low Low Low Low Low Low Middle Middle Middle Middle Middle Middle Middle High High High High High High High

Region Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income World World World World World World World

Table E.6: Women’s Economic Rights Year 1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010

Obs 28 30 32 32 34 34 35 63 65 74 76 98 100 101 35 37 43 45 56 56 56 126 132 149 153 188 190 192

Mean 1.107143 1.016667 1.032292 0.960938 1.035294 0.852941 0.678571 1.103175 1.223077 1.177703 1.160088 1.279592 1.280667 1.153465 1.6 1.643243 1.774419 1.69037 1.8 1.792857 2 1.242063 1.293939 1.31868 1.274401 1.390426 1.355088 1.313802

s.d. 0.61399 0.486248 0.407236 0.417893 0.363387 0.400935 0.456052 0.501278 0.479589 0.429171 0.414622 0.546962 0.52004 0.637348 0.705024 0.585254 0.625773 0.578077 0.561815 0.643812 0.743151 0.625249 0.55982 0.56985 0.543711 0.592388 0.628825 0.796445

Min 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.2 0 0 0 0 0 0 0 0 0

Max 2 2 2 2.4 2 1.8 1.75 2.5 2.4 2.2 2.4 3 2.8 3 2.5 2.6 3 2.6 3 3 3 2.5 2.6 3 2.6 3 3 3

Low Low Low Low Low Low Low Middle Middle Middle Middle Middle Middle Middle High High High High High High High

Region Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income World World World World World World World

Table E.7: Women’s Social Rights Year 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005

Obs 28 30 32 32 34 33 62 65 74 76 97 97 35 37 42 45 56 53 125 132 148 153 187 183

Mean 0.892857 0.89 0.864063 0.870833 0.902941 0.848485 1.016129 1.082308 1.010811 1.021053 1.160825 1.273196 1.714286 1.608108 1.821429 1.855556 1.767857 1.877358 1.184 1.185985 1.209122 1.235076 1.295722 1.371585

s.d. 0.58305 0.529867 0.490923 0.454034 0.47 0.744615 0.59346 0.576319 0.555068 0.519055 0.709835 0.832356 0.964714 0.911159 0.938482 0.957163 1.004555 0.975201 0.784261 0.726839 0.776089 0.777816 0.838126 0.929469

73

Min 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Max 2 2 2 2 2 3 2 2.5 2.5 2.6 3 3 3 3 3 3 3 3 3 3 3 3 3 3

Low Low Low Low Low Low Middle Middle Middle Middle Middle Middle High High High High High High

Region Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income Income World World World World World World

E.1

Mortality and MMR Table E.8: Gender differences in Mortality rates & MMR: All countries

MMR lgdp 5

(1) (0-14) -3.799 (2.605) 2213.6∗∗ (950.6)

IMR ratio (F/M) Mean Dep Var. s.d. Dep Var. N r2

-8894.328 17798.454 697 0.0310

(2) (0-14) -7.450∗∗ (3.437) 706.1 (1475.2) 0.0737 (0.320) -8953.037 17450.25 345 0.0455

(3) (15-49) 7.708 (5.075) -2912.0∗ (1486.3)

-36834.996 36503.055 697 0.0493

(4) (15-49) 4.479 (5.933) -5942.8∗∗∗ (1761.3) 0.866∗∗ (0.391) -36072.162 36329.557 345 0.131

(5) (50+) -2.328 (2.989) -194.6 (781.9)

8565.819 8379.994 697 0.0322

(6) (50+) -2.443 (3.762) 857.5 (880.3) -0.284 (0.183) 8659.101 8460.996 345 0.0610

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is the log ratio of Female to Male mortality rates (scaled up by 100,000) in the different age groups specified in the column headers. We have used a country fixed effects panel framework with year dummies. Standard errors in parentheses are clustered at the country level. The data on Mortality rates come from the United Nations, Department of Economic and Social Affairs, Population Division (2013)and is available for every 5 years from 1960 to 2005. The MMR data comes from the World Bank - WDI (based on WHO data) available for 5 time periods -1990, 1995, 2000, 2005, 2010. The IMR data also comes from the World Bank - WDI (based on WHO data) and is available for 4 time periods 1990, 2000, 2010 and 2012 (and the IMR variable has been scaled up by 100,000 as well). Hence these regressions are based on 178 countries and the years 1990, 1995, 2000 and 2005 (odd columns); and 177 countries and the years 1990 and 2000 (even columns) from across the world. All the variables are 5 yearly averages.

74

E.2

Desired Sex Ratio

75

E.3

Women’s Rights Table E.9: Rights data - No. of countries by year year 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

Political 127 129 130 131 131 132 132 133 133 130 127 144 149 150 149 153 152 150 152 152 188 156 185 186 188 190 190 188 190 191 192

Economic 119 118 124 130 131 130 128 131 128 130 127 142 150 148 147 151 150 147 151 151 187 156 185 186 188 190 190 190 191 191 192

Social 118 117 124 129 128 131 129 130 129 128 125 137 149 144 147 151 151 149 150 150 186 154 168 171 0 0 0 0 0 0 0

Composite1 118 116 122 127 128 129 127 130 126 127 117 132 148 142 144 149 149 146 149 149 185 154 167 171 0 0 0 0 0 0 0

Composite2 119 118 124 129 131 130 128 131 128 129 123 140 149 148 147 151 150 147 151 151 187 156 184 186 188 190 190 188 190 191 192

Table E.10: Summary statistics Variable Political Economic Social Composite1 Composite2

Mean 1.786 1.323 1.235 0 0

Std. Dev. 0.647 0.697 0.84 1.436 1.16

76

Min. 0 0 0 -3.435 -3.286

Max. 3 3 3 3.882 3.022

N 4830 4779 3395 3352 4766

Table E.11: Correlation of different Rights measures Variables Political Economic Rights Social Rights Composite1 Composite2

Political 1.000 0.347 0.449 0.707 0.821

Economic

Social

Composite1

Composite2

1.000 0.723 0.874 0.821

1.000 0.894 0.693

1.000 0.939

1.000

Table E.12: Log Life Expectancy Ratio and Women’s Rights

Political, Economic, Social right lgdp

(1) indep1

(2) indep2

(3) indep3

(4) indep4

(5) indep5

22.53 (108.2) 907.3∗ (468.4)

61.47 (110.6) 961.9∗∗ (485.7) -0.469 (34.79)

1020.1∗∗ (459.9) 979.7∗∗ (487.2) -135.8∗∗ (62.70)

1087.7∗∗ (497.4) 1035.9∗∗ (505.3) -13.31 (34.68) -140.7∗∗ (68.45)

1004.1∗∗ (410.2) 1175.8∗ (704.3) 137.9 (278.6) -129.1∗∗ (56.92) -22.55 (41.17) 3009 0.181

Democracy Rights * lGDP Democracy * lGDP N r2 Political & Economic right lgdp

3211 0.157

3009 0.168

3211 0.168

3009 0.180

-7.644 (95.06) 609.6∗ (344.0)

13.75 (100.7) 638.7∗ (347.9) -0.123 (34.33)

732.0∗∗ (328.9) 646.2∗ (349.1) -96.51∗∗ (40.56)

788.1∗∗ (363.0) 676.1∗ (353.7) -10.55 (34.29) -101.8∗∗ (45.12)

4536 0.182

4157 0.189

Democracy Rights * lGDP Democracy * lGDP N r2

4536 0.176

4157 0.182

712.3∗∗ (336.2) 768.0∗ (442.6) 89.97 (174.0) -91.77∗∗ (41.46) -14.62 (24.76) 4157 0.190

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is the log of the ratio of female to male life expectancy times 100,000. country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set.

77

Table E.13: Log Life Expectancy Ratio and Women’s Rights

wopol wosoc wecon lgdp

(1) indep1 -129.5 (192.6) 49.97 (110.4) 79.27 (103.7) 891.8∗ (462.6)

Democracy

(2) indep2 -112.2 (195.2) 84.19 (114.9) 119.4 (106.0) 946.0∗ (480.8) 2.095 (34.61)

wopol gdp wosoc gdp wecon gdp

(3) indep3 1077.7 (900.6) 726.8∗ (436.4) 863.3 (528.9) 1527.7∗∗ (759.6)

-171.0 (137.6) -89.81∗ (49.14) -106.2∗ (62.97)

(4) indep4 1203.1 (963.1) 796.3∗ (454.7) 854.2 (547.6) 1612.1∗∗ (799.3) -12.31 (34.41) -186.5 (147.2) -95.80∗ (51.75) -100.3 (65.04)

Democracy * lGDP N r2 wopol wecon lgdp

3211 0.158

3009 0.170

3211 0.171

3009 0.184

-63.47 (156.0) 36.65 (93.28) 600.8∗ (340.8)

-33.27 (166.3) 52.98 (98.56) 630.9∗ (345.4) 0.612 (34.39)

776.8 (539.0) 852.2∗∗ (427.3) 975.4∗∗ (450.0)

-115.2 (75.20) -102.0∗∗ (49.41)

823.6 (597.8) 913.5∗∗ (444.6) 1019.1∗∗ (471.3) -9.828 (33.90) -118.2 (83.18) -108.8∗∗ (51.32)

4536 0.183

4157 0.190

Democracy wopol gdp wecon gdp Democracy * lGDP N r2

4536 0.176

4157 0.183

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is the log of the ratio of female to male life expectancy times 100,000. country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set.

78

(5) indep5 1100.7 (840.8) 734.7∗ (418.5) 826.9 (539.3) 1690.8∗ (911.3) 119.0 (268.9) -171.9 (130.3) -87.17∗ (47.48) -96.71 (63.90) -19.57 (39.73) 3009 0.185 715.5 (531.7) 853.2∗ (449.1) 1080.4∗∗ (523.6) 98.55 (173.3) -104.5 (74.69) -100.1∗ (52.49) -15.74 (24.72) 4157 0.191

Table E.14: Maternal Mortality from WDI (per birth) and Women’s Rights Political, Economic, Social right lgdp

(1)

(2)

(3)

(4)

(5)

1.106 (8.573) 6.353 (19.99)

1.120 (9.188) 5.307 (19.50) -4.201 (2.843)

-90.48* (52.86) 5.125 (20.83)

12.35* (6.283)

-94.44* (54.93) 4.382 (20.51) -3.585 (2.849) 12.98** (6.541)

-89.16* (52.54) -58.13** (27.33) -71.59*** (15.69) 12.33* (6.254) 10.53*** (2.350) 394 0.280

Democracy Rights * lGDP Democracy * lGDP N r2 Political & Economic right lgdp

409 0.167

394 0.188

409 0.192

394 0.215

5.499 (7.924) 17.44 (23.48)

3.491 (8.406) 23.92 (21.31) -7.176** (3.277)

-156.7*** (34.00) 21.90 (23.02)

20.90*** (3.850)

-153.6*** (34.92) 28.37 (20.88) -5.120* (3.092) 20.37*** (3.943)

763 0.311

704 0.322

Democracy Rights * lGDP Democracy * lGDP N r2

763 0.247

704 0.261

-123.9*** (32.90) -11.87 (25.27) -58.17*** (12.79) 16.38*** (3.694) 7.878*** (1.831) 704 0.361

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is MMR (deaths per 100,000 births) from WDI constructed by Bhalotra et al. country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set.

79

Table E.15: Maternal Mortality from WDI (per birth) and Women’s Rights wopol wosoc wecon lgdp

(1) 22.93** (10.41) 3.431 (13.54) -14.10 (11.52) 10.11 (19.11)

Democracy

(2) 18.81* (10.54) 5.878 (13.88) -14.22 (11.98) 7.838 (18.20) -4.014 (2.734)

wopol gdp wosoc gdp wecon gdp

(3) -61.01 (54.80) -70.08 (80.55) -108.1* (59.35) -39.20 (29.53)

11.96* (7.014) 9.408 (8.985) 12.81* (6.909)

(4) -83.63 (57.87) -62.64 (80.87) -109.8* (61.40) -46.68 (29.38) -3.338 (2.738) 14.68* (7.556) 8.935 (9.054) 12.99* (7.167)

Democracy * lGDP N r2 wopol wecon lgdp

409 0.185

394 0.201

409 0.214

394 0.233

-4.711 (13.96) 12.95 (9.991) 15.80 (22.97)

-6.552 (14.44) 10.95 (10.62) 22.22 (20.68) -7.217** (3.293)

-230.9*** (60.92) -112.4** (48.74) -57.39** (26.25)

30.76*** (6.976) 15.18*** (5.337)

-229.1*** (62.32) -109.7** (51.61) -49.38** (24.36) -4.976 (3.107) 30.56*** (7.208) 14.62*** (5.604)

763 0.316

704 0.327

Democracy wopol gdp wecon gdp Democracy * lGDP N r2

763 0.249

704 0.263

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variable is MMR (deaths per 100,000 birth) from the WDI. country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set.

80

(5) -85.15 (56.48) -50.23 (72.25) -106.1* (58.95) -106.4*** (35.51) -69.44*** (14.56) 14.47* (7.396) 7.456 (8.103) 12.74* (6.942) 10.23*** (2.178) 394 0.295 -196.3*** (60.37) -85.28* (47.63) -78.07*** (28.24) -59.29*** (13.39) 26.89*** (6.947) 10.77** (5.067) 8.076*** (1.936) 704 0.367

E.4

Gender Intensity of Language Table E.16: Summary statistics Variable ngii sbii gaii gpii gii0 gii1 gii2

Mean 0.445 0.680 0.677 0.339 2.453 1.935 1.513

Std. Dev. 0.499 0.468 0.47 0.475 1.671 1.249 1.236

Min. 0 0 0 0 0 0 0

Max. 1 1 1 1 4 3 3

N 128 128 93 124 86 93 117

Table E.17: Cross-correlation table Variables SBII GAII GPII gii0 gii1 gii2

NGII 0.548 0.698 0.793 0.934 0.906 0.910

SBII 1.000 0.595 0.480 0.813 0.855 0.773

GAII

GPII

gii0

gii1

gii2

1.000 0.646 0.848 0.871 0.732

1.000 0.889 0.789 0.888

1.000 0.983 0.982

1.000 0.949

1.000

81

E.5

Placebo Tests with TB infection rates - Regressions Table E.18: TB and Desired Sex ratio: DSR Desired Sex Ratio

(1) -48.12 (194.5)

(2) -98.36 (263.4)

ln GDP

(3) -140.7 (266.4) -54.46∗ (31.75)

Desired Fertility

(4) -131.9 (294.7) -55.22 (33.66) 8.259 (90.02)

Desired Sex Ratio * lGDP N r2

1407

1407 0.0438

1393 0.0625

1393 0.0626

(5) -804.1 (519.0) -174.5∗∗ (86.96)

110.6 (75.55) 1393 0.0661

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables TB is the Incidence of tuberculosis (per 100,000 people). Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level.The data on TB comes from the World Bank WDI and is available for the years 1990- 2012 yearly for almost all countries in the world. The desired sex ratio data comes from the DHS and has been constructed using the questions on ideal number of boys and girls asked to the mothers. Since the answers are available only for the survey years, to arrive at the yearly data either 20, 25 or 30 is added to the mother age. This data and hence the regressions are based on the DHS sample (developing countries) of 63 countries over the period of 1969 - 2012. The mean (s.d.) of dsr 20 is 1.125 (0.11), dsr 25 is 1.13 (0.11), dsr 30 is 1.134 (0.113). The mean (s.d.) of TB 20 is 250.252 (226.683), TB 25 is 248.002 (225.705), TB 30 is 246.618 (224.869)

82

(6) -794.3∗ (467.5) -173.9∗∗ (79.76) 4.040 (89.41) 109.7 (65.91) 1393 0.0662

Table E.19: TB infection rates from WDI and Women’s Rights Political Rights lgdp

(1) 2.323 (8.302) -46.09∗∗∗ (10.67)

(2) 2.693 (8.396) -37.97∗∗ (16.69)

Democracy

(3) 0.604 (8.878) -39.99∗∗ (17.79) 3.066 (2.970)

Rights * lGDP

(4) -24.02 (42.99) -44.60∗∗ (17.98)

3.494 (5.130)

(5) -29.22 (49.47) -47.48∗∗ (18.95) 3.253 (2.971) 3.932 (5.837)

Democracy * lGDP N r2 Economic Rights lgdp

3555

3555 0.0357

3163 0.0396

3555 0.0367

3163 0.0406

-4.285 (8.180) -47.51∗∗∗ (9.992)

-3.404 (8.612) -39.12∗∗ (17.32)

-5.438 (9.936) -41.14∗∗ (18.30) 3.362 (3.022)

-6.028 (30.40) -39.43∗∗ (17.63)

0.324 (3.165)

-11.02 (33.00) -41.82∗∗ (18.65) 3.374 (3.022) 0.692 (3.435)

Democracy Rights * lGDP Democracy * lGDP N r2 Social Rights lgdp

3543

3543 0.0359

3152 0.0414

3543 0.0359

3152 0.0414

2.059 (14.39) -56.39∗∗∗ (8.058)

2.814 (15.25) -50.22∗∗∗ (14.33)

3.464 (16.51) -55.12∗∗∗ (15.63) 2.233 (2.132)

-29.26 (39.49) -54.66∗∗∗ (13.95)

4.131 (3.456)

-27.30 (41.98) -59.40∗∗∗ (15.01) 2.208 (2.122) 3.972 (3.671)

2207 0.0386

2035 0.0411

Democracy Rights * lGDP Democracy * lGDP N r2

2207

2207 0.0372

2035 0.0399

(6) -29.48 (48.40) -47.30∗∗ (22.45) 3.611 (12.42) 3.961 (5.662) -0.0528 (1.703) 3163 0.0406 -10.61 (31.99) -42.30∗ (22.39) 2.587 (12.57) 0.636 (3.282) 0.116 (1.700) 3152 0.0415 -29.50 (41.63) -50.98∗∗∗ (17.48) 13.34 (9.441) 4.303 (3.627) -1.713 (1.372) 2035 0.0425

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables TB is the Incidence of tuberculosis (per 100,000 people). Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level.The data on TB comes from the World Bank WDI and is available for the years 1990- 2012 yearly for almost all countries in the world. The Mean (s.d.) of women’s political, economic and social rights variables respectively are 1.806 (0.625), 1.311 (0.695), and 1.264 (0.842).

83

Table E.20: TB infection rates from WDI and Women’s Parliamentary Representation (1) 0.252 (0.796) -38.03∗∗∗ (13.16)

womparl lgdp

(2) 0.231 (0.806) -27.98 (18.88)

(3) 0.259 (0.966) -26.73 (21.23) 1.937 (4.587)

Democracy Rights * lGDP

(4) -7.136∗∗ (3.136) -38.96∗∗ (18.65) 0.941∗∗∗ (0.360)

(5) -8.844∗∗ (3.413) -39.48∗ (20.80) 2.874 (4.564) 1.195∗∗∗ (0.414)

2727 0.0767

2276 0.0960

Democracy * lGDP N r2

2727

2727 0.0583

2276 0.0689

(6) -8.544∗∗ (3.574) -41.00∗ (24.05) -1.686 (15.58) 1.159∗∗∗ (0.434) 0.649 (1.949) 2276 0.0964

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables TB is the Incidence of tuberculosis (per 100,000 people). Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level.The data on TB comes from the World Bank WDI and is available for the years 1990 - 2012 yearly for almost all countries in the world. The mean (s.d.) of TB are 154.508 (213.754), whereas the mean and s.d. of women’s representation in parliament is around 15.525 (10.313).

Table E.21: TB infection rates from WDI & Gender Intensity of Language

GII N r2 GII lgdp N r2 GII GII gdp lgdp N r2

(1) ngii -65.65∗∗∗ (21.27) 2636 0.340

(2) sbii -97.12∗∗∗ (27.22) 2636 0.373

(3) gaii -42.80 (27.44) 1919 0.305

(4) gpii -132.9∗∗∗ (32.46) 2623 0.325

(5) gii0 -27.42∗∗∗ (8.619) 1820 0.380

(6) gii1 -30.68∗∗∗ (11.08) 1919 0.349

(7) gii2 -42.16∗∗∗ (9.867) 2471 0.405

(8) gtroiano -1.502 (4.661) 1771 0.187

-35.19∗ (18.16) -40.50∗∗∗ (12.75) 2513 0.558

-38.72 (26.41) -39.88∗∗∗ (12.58) 2513 0.556

18.86 (28.92) -21.57∗∗ (10.48) 1818 0.582

-70.99∗∗ (29.27) -49.50∗∗∗ (14.85) 2458 0.520

-2.300 (7.580) -22.18∗∗ (10.06) 1741 0.579

0.718 (9.335) -22.01∗∗ (10.58) 1818 0.580

-23.23∗∗ (10.45) -38.70∗∗∗ (12.23) 2370 0.563

-0.253 (4.404) -27.63∗∗∗ (6.095) 1674 0.546

-115.2 (81.19) 9.611 (8.381) -45.16∗∗∗ (14.58) 2513 0.560

-108.2 (89.88) 8.742 (9.346) -46.03∗∗∗ (16.76) 2513 0.558

52.91 (151.6) -4.241 (16.72) -18.58 (16.22) 1818 0.582

-216.3∗ (111.0) 17.99 (12.30) -55.44∗∗∗ (16.24) 2458 0.524

-5.287 (34.80) 0.391 (3.908) -23.25 (15.03) 1741 0.579

2.361 (44.29) -0.217 (5.139) -21.55 (15.92) 1818 0.580

-63.16∗ (36.44) 5.131 (3.911) -46.89∗∗∗ (15.59) 2370 0.566

14.97 (22.44) -1.739 (2.345) -23.16∗∗∗ (8.154) 1674 0.548

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables in all three panels are the TB infection rates (from the WDI database) . Standard errors in parentheses are clustered at the country level. The GII data come from Gay et al. (2013) and Givati and Troiano (2012). Apart from the GII variable in Panel 1 we control for the percentage of the population speaking the majority language (for which the GII has been calculated), decade dummies and continent dummies. In Panel 2 we control for the log of GDP , the log of population, dummies for the World Bank Income groups classification, the percentage of population that is Protestant, Catholic and Muslim, and the proportion of the country that is tropical or subtropical in addition to the controls from Panel 1. In Panel 3, we add the interaction term of log of GDP and GII in addition to the controls from the previous panels.

84

E.6

Life Expectancy Advantage Pre and Post 1990

Table E.22: Female Life Expectancy advantage & Desired Sex Ratio (boys/girls) before 1990

DSR

(1) indep1 -0.0446∗∗∗ (0.00877)

(2) indep2 -0.0362∗∗∗ (0.00875)

lgdp

(3) indep3 -0.0390∗∗∗ (0.0100) -0.00173 (0.00367)

ideal 15 25

(4) indep3b -0.0388∗∗∗ (0.00962) -0.00152 (0.00357) -0.00367 (0.00330)

DSR gdp N

1417

1417

1224

1224

(5) indep4 0.0383 (0.0562) 0.0164 (0.0147)

-0.0170 (0.0131) 1224

(6) indep5 0.0353 (0.0556) 0.0159 (0.0145) -0.00330 (0.00330) -0.0163 (0.0130) 1224

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables is the log of the ratio of female to male life expectancy. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level.The data on Life Expectancy comes from The World Bank WDI. The desired sex ratio (DSR) data comes from the DHS and has been constructed using the questions on ideal number of boys and girls asked to the mothers. Since the answers are available only for the survey years, to arrive at the yearly data either 20, 25 or 30 is added to the mother age. This data and hence the regressions are based on the DHS sample (developing countries) of 63 countries over the period of 1969 - 1990. The mean (s.d.) of dsr 20 is 1.145 (0.157), dsr 25 is 1.145 (0.165), dsr 30 is 1.144 (0.178).

85

Table E.23: Female Life Expectancy advantage & Desired Sex Ratio (boys/girls) after 1990

DSR

(1) indep1 -0.0352∗∗∗ (0.0124)

(2) indep2 -0.0676∗ (0.0345)

lgdp

(3) indep3 -0.0626∗ (0.0341) 0.00654 (0.00399)

ideal 15 25

(4) indep3b -0.0524 (0.0368) 0.00582 (0.00428) 0.00870 (0.00736)

DSR gdp N r2

1211

1211 0.240

1208 0.257

1208 0.268

(5) indep4 0.00246 (0.102) 0.0184 (0.0181)

-0.0110 (0.0165) 1208 0.260

(6) indep5 0.0236 (0.0948) 0.0196 (0.0170) 0.00911 (0.00719) -0.0127 (0.0151) 1208 0.272

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables is the log of the ratio of female to male life expectancy. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level.The data on Life Expectancy comes from The World Bank WDI. The desired sex ratio (DSR) data comes from the DHS and has been constructed using the questions on ideal number of boys and girls asked to the mothers. Since the answers are available only for the survey years, to arrive at the yearly data either 20, 25 or 30 is added to the mother age. This data and hence the regressions are based on the DHS sample (developing countries) of 63 countries over the period of 1990- 2012. The mean (s.d.) of dsr 20 is 1.131 (0.116), dsr 25 is 1.137 (0.118), dsr 30 is 1.141 (0.122).

Table E.24: Female Life Expectancy advantage & Desired Sex Ratio (boys/girls) before 1990; MMR country sample

DSR

(1) indep1 -0.0446∗∗∗ (0.00877)

(2) indep2 -0.0362∗∗∗ (0.00875)

lgdp

(3) indep3 -0.0390∗∗∗ (0.0100) -0.00173 (0.00367)

ideal 15 25

(4) indep3b -0.0388∗∗∗ (0.00962) -0.00152 (0.00357) -0.00367 (0.00330)

DSR gdp N r2

1417

1417 0.0776

1224 0.180

1224 0.184

(5) indep4 0.0383 (0.0562) 0.0164 (0.0147)

-0.0170 (0.0131) 1224 0.190

(6) indep5 0.0353 (0.0556) 0.0159 (0.0145) -0.00330 (0.00330) -0.0163 (0.0130) 1224 0.194

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables is the log of the ratio of female to male life expectancy for the MMR sample. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level.The data on Life Expectancy comes from The World Bank WDI. The desired sex ratio (DSR) data comes from the DHS and has been constructed using the questions on ideal number of boys and girls asked to the mothers. Since the answers are available only for the survey years, to arrive at the yearly data either 20, 25 or 30 is added to the mother age. This data and hence the regressions are based on the DHS sample (developing countries) of 63 countries over the period of 1969 - 1990. The mean (s.d.) of dsr 20 is 1.145 (0.157), dsr 25 is 1.145 (0.165), dsr 30 is 1.144 (0.178).

86

Table E.25: Female Life Expectancy advantage & Desired Sex Ratio (boys/girls) after 1990; MMR country sample

DSR

(1) indep1 -0.0352∗∗∗ (0.0124)

(2) indep2 -0.0676∗ (0.0345)

lgdp

(3) indep3 -0.0626∗ (0.0341) 0.00654 (0.00399)

ideal 15 25

(4) indep3b -0.0524 (0.0368) 0.00582 (0.00428) 0.00870 (0.00736)

DSR gdp N r2

1211

1211 0.240

1208 0.257

1208 0.268

(5) indep4 0.00246 (0.102) 0.0184 (0.0181)

-0.0110 (0.0165) 1208 0.260

(6) indep5 0.0236 (0.0948) 0.0196 (0.0170) 0.00911 (0.00719) -0.0127 (0.0151) 1208 0.272

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables is the log of the ratio of female to male life expectancy for the MMR sample. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level.The data on Life Expectancy comes from The World Bank WDI. The desired sex ratio (DSR) data comes from the DHS and has been constructed using the questions on ideal number of boys and girls asked to the mothers. Since the answers are available only for the survey years, to arrive at the yearly data either 20, 25 or 30 is added to the mother age. This data and hence the regressions are based on the DHS sample (developing countries) of 63 countries over the period of 1990- 2012. The mean (s.d.) of dsr 20 is 1.131 (0.116), dsr 25 is 1.137 (0.118), dsr 30 is 1.141 (0.122).

87

Table E.26: Female Life Expectancy advantage & Gender Intensity of Language before 1990

GII N r2 GII lgdp N r2 GII GII gdp lgdp N r2

(1) ngii -2000.0∗∗∗ (591.0) 3089 0.224

(2) sbii -2116.4∗∗∗ (734.0) 3089 0.213

(3) gaii -2090.3∗∗∗ (781.8) 2459 0.234

(4) gpii -1696.2∗∗ (646.7) 3029 0.204

(5) gii0 -788.0∗∗∗ (211.2) 2309 0.285

(6) gii1 -1088.9∗∗∗ (282.5) 2459 0.281

(7) gii2 -884.5∗∗∗ (269.3) 2849 0.226

(8) gtroiano -389.0 (274.8) 2189 0.239

-1809.4∗∗ (715.4) 163.6 (277.5) 2352 0.386

-2770.8∗∗∗ (929.9) 178.4 (268.7) 2352 0.405

-2010.4∗∗ (901.8) 626.0∗∗ (310.1) 1915 0.351

-1560.3∗∗ (736.8) 279.6 (286.6) 2285 0.385

-1015.3∗∗∗ (303.5) 636.9∗∗ (317.2) 1834 0.381

-1339.3∗∗∗ (365.3) 593.1∗ (311.6) 1915 0.396

-1144.9∗∗∗ (388.0) 267.7 (289.9) 2194 0.402

-770.5∗∗ (300.3) 140.2 (225.8) 1754 0.373

-4166.5 (2923.8) 319.8 (342.1) -6.697 (341.6) 2352 0.390

-4210.5 (3240.4) 201.4 (418.2) 36.99 (441.8) 2352 0.407

-3252.4 (3825.9) 178.1 (508.8) 488.9 (539.2) 1915 0.352

695.1 (2630.5) -325.8 (392.0) 423.5 (376.6) 2285 0.387

-726.9 (1275.6) -42.55 (186.3) 762.0 (690.6) 1834 0.382

-1050.4 (1649.4) -42.84 (232.6) 691.1 (662.5) 1915 0.396

-867.3 (1225.0) -40.30 (174.4) 339.8 (480.2) 2194 0.402

895.8 (1394.9) -228.5 (191.1) 728.8 (560.7) 1754 0.383

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables in all three panels are the log of the ratio of female to male life expectancy times 100,000 before 1990 (from the World Bank WDI database). Standard errors in parentheses are clustered at the country level. The GII data come from Gay et al. (2013) and Givati and Troiano (2012). Apart from the GII variable in Panel 1 we control for the percentage of the population speaking the majority language (for which the GII has been calculated), decade dummies and continent dummies. In Panel 2 we control for the log of GDP, the log of population, dummies for the World Bank Income groups classification, the percentage of population that is Protestant, Catholic and Muslim, and the proportion of the country that is tropical or subtropical in addition to the controls from Panel 1. In Panel 3, we add the interaction term of log of GDP and GII in addition to the controls from the previous panels.

88

Table E.27: Female Life Expectancy advantage & Gender Intensity of Language after 1990

GII N r2 GII lgdp N r2 GII GII gdp lgdp N r2

(1) ngii -1031.9∗ (579.3) 2275 0.318

(2) sbii -1013.2 (682.7) 2275 0.313

(3) gaii -1106.4 (747.9) 1813 0.345

(4) gpii -1244.2∗∗ (562.9) 2231 0.337

(5) gii0 -339.2∗ (202.2) 1703 0.379

(6) gii1 -479.7∗ (266.5) 1813 0.350

(7) gii2 -478.7∗ (243.7) 2099 0.331

(8) gtroiano -283.3 (222.1) 1615 0.322

-1583.5∗∗ (649.3) -687.7∗ (390.5) 2150 0.543

-1990.5∗∗ (869.8) -658.3∗ (393.0) 2150 0.544

-1482.2∗ (764.2) -365.3 (393.8) 1688 0.601

-2118.8∗∗∗ (740.3) -585.6 (385.7) 2095 0.570

-601.4∗∗ (283.2) -423.1 (392.4) 1611 0.612

-736.5∗∗ (343.8) -455.0 (394.8) 1688 0.606

-1053.9∗∗∗ (339.2) -609.9 (383.7) 2007 0.563

-503.8 (327.5) -525.9 (457.1) 1526 0.533

-3011.8 (2815.4) 168.8 (286.2) -774.9∗ (437.2) 2150 0.545

389.5 (3067.0) -291.8 (323.0) -444.9 (516.1) 2150 0.547

-4604.7 (2872.7) 381.4 (347.5) -635.6 (510.1) 1688 0.605

-1766.8 (2955.1) -43.31 (361.9) -571.2 (411.2) 2095 0.570

-1346.7 (998.4) 94.65 (124.4) -687.1 (580.1) 1611 0.614

-978.6 (1301.5) 31.07 (157.5) -522.8 (583.6) 1688 0.606

23.60 (1247.9) -136.3 (144.1) -389.0 (478.9) 2007 0.566

111.1 (1669.7) -70.13 (178.5) -353.4 (584.9) 1526 0.534

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables in all three panels are the log of the ratio of female to male life expectancy times 100,000 after 1990 (from the World Bank WDI database). Standard errors in parentheses are clustered at the country level. The GII data come from Gay et al. (2013) and Givati and Troiano (2012). Apart from the GII variable in Panel 1 we control for the percentage of the population speaking the majority language (for which the GII has been calculated), decade dummies and continent dummies. In Panel 2 we control for the log of GDP, the log of population, dummies for the World Bank Income groups classification, the percentage of population that is Protestant, Catholic and Muslim, and the proportion of the country that is tropical or subtropical in addition to the controls from Panel 1. In Panel 3, we add the interaction term of log of GDP and GII in addition to the controls from the previous panels.

89

Table E.28: Female Life Expectancy advantage & Gender Intensity of Language before 1990; MMR sample

GII N r2 GII lgdp N r2 GII GII gdp lgdp N r2

(1) ngii -1998.0∗∗∗ (593.0) 3086 0.223

(2) sbii -2116.5∗∗∗ (734.1) 3086 0.213

(3) gaii -2087.8∗∗∗ (785.5) 2456 0.233

(4) gpii -1694.1∗∗ (649.9) 2996 0.203

(5) gii0 -787.8∗∗∗ (212.1) 2306 0.284

(6) gii1 -1088.3∗∗∗ (283.4) 2456 0.281

(7) gii2 -883.6∗∗∗ (270.0) 2846 0.225

(8) gtroiano -386.6 (276.3) 2186 0.239

-1809.4∗∗ (715.4) 163.6 (277.5) 2352 0.386

-2770.8∗∗∗ (929.9) 178.4 (268.7) 2352 0.405

-2010.4∗∗ (901.8) 626.0∗∗ (310.1) 1915 0.351

-1560.3∗∗ (736.8) 279.6 (286.6) 2285 0.385

-1015.3∗∗∗ (303.5) 636.9∗∗ (317.2) 1834 0.381

-1339.3∗∗∗ (365.3) 593.1∗ (311.6) 1915 0.396

-1144.9∗∗∗ (388.0) 267.7 (289.9) 2194 0.402

-770.5∗∗ (300.3) 140.2 (225.8) 1754 0.373

-4166.5 (2923.8) 319.8 (342.1) -6.697 (341.6) 2352 0.390

-4210.5 (3240.4) 201.4 (418.2) 36.99 (441.8) 2352 0.407

-3252.4 (3825.9) 178.1 (508.8) 488.9 (539.2) 1915 0.352

695.1 (2630.5) -325.8 (392.0) 423.5 (376.6) 2285 0.387

-726.9 (1275.6) -42.55 (186.3) 762.0 (690.6) 1834 0.382

-1050.4 (1649.4) -42.84 (232.6) 691.1 (662.5) 1915 0.396

-867.3 (1225.0) -40.30 (174.4) 339.8 (480.2) 2194 0.402

895.8 (1394.9) -228.5 (191.1) 728.8 (560.7) 1754 0.383

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables in all three panels are the log of the ratio of female to male life expectancy times 100,000 before 1990 (from the World Bank WDI database) for the MMR sample. Standard errors in parentheses are clustered at the country level. The GII data come from Gay et al. (2013) and Givati and Troiano (2012). Apart from the GII variable in Panel 1 we control for the percentage of the population speaking the majority language (for which the GII has been calculated), decade dummies and continent dummies. In Panel 2 we control for the log of GDP, the log of population, dummies for the World Bank Income groups classification, the percentage of population that is Protestant, Catholic and Muslim, and the proportion of the country that is tropical or subtropical in addition to the controls from Panel 1. In Panel 3, we add the interaction term of log of GDP and GII in addition to the controls from the previous panels.

90

Table E.29: Female Life Expectancy advantage & Gender Intensity of Language after 1990; MMR sample

GII N r2 GII lgdp N r2 GII GII gdp lgdp N r2

(1) ngii -1055.1∗ (586.3) 2266 0.319

(2) sbii -1013.0 (682.7) 2266 0.313

(3) gaii -1143.6 (761.3) 1804 0.346

(4) gpii -1283.5∗∗ (570.7) 2200 0.338

(5) gii0 -349.2∗ (204.7) 1694 0.381

(6) gii1 -489.7∗ (268.8) 1804 0.351

(7) gii2 -487.5∗ (245.7) 2090 0.332

(8) gtroiano -298.9 (227.0) 1606 0.323

-1583.5∗∗ (649.3) -687.7∗ (390.5) 2150 0.543

-1990.5∗∗ (869.8) -658.3∗ (393.0) 2150 0.544

-1482.2∗ (764.2) -365.3 (393.8) 1688 0.601

-2118.8∗∗∗ (740.3) -585.6 (385.7) 2095 0.570

-601.4∗∗ (283.2) -423.1 (392.4) 1611 0.612

-736.5∗∗ (343.8) -455.0 (394.8) 1688 0.606

-1053.9∗∗∗ (339.2) -609.9 (383.7) 2007 0.563

-503.8 (327.5) -525.9 (457.1) 1526 0.533

-3011.8 (2815.4) 168.8 (286.2) -774.9∗ (437.2) 2150 0.545

389.5 (3067.0) -291.8 (323.0) -444.9 (516.1) 2150 0.547

-4604.7 (2872.7) 381.4 (347.5) -635.6 (510.1) 1688 0.605

-1766.8 (2955.1) -43.31 (361.9) -571.2 (411.2) 2095 0.570

-1346.7 (998.4) 94.65 (124.4) -687.1 (580.1) 1611 0.614

-978.6 (1301.5) 31.07 (157.5) -522.8 (583.6) 1688 0.606

23.60 (1247.9) -136.3 (144.1) -389.0 (478.9) 2007 0.566

111.1 (1669.7) -70.13 (178.5) -353.4 (584.9) 1526 0.534

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables in all three panels are the log of the ratio of female to male life expectancy times 100,000 after 1990 (from the World Bank WDI database) for the MMR sample. Standard errors in parentheses are clustered at the country level. The GII data come from Gay et al. (2013) and Givati and Troiano (2012). Apart from the GII variable in Panel 1 we control for the percentage of the population speaking the majority language (for which the GII has been calculated), decade dummies and continent dummies. In Panel 2 we control for the log of GDP, the log of population, dummies for the World Bank Income groups classification, the percentage of population that is Protestant, Catholic and Muslim, and the proportion of the country that is tropical or subtropical in addition to the controls from Panel 1. In Panel 3, we add the interaction term of log of GDP and GII in addition to the controls from the previous panels.

91

Table E.30: Female Life Expectancy advantage & Women’s Rights before 1990 Political Rights lgdp

(1) 34.62 (91.18) 351.7 (230.6)

(2) 0.983 (95.10) 248.1 (292.4)

Democracy

(3) -25.77 (94.09) 284.5 (300.3) 13.35 (21.65)

Rights * lGDP

(4) 640.7 (622.5) 422.3 (419.7)

-95.56 (86.79)

(5) 684.8 (639.3) 480.1 (434.6) 13.79 (21.36) -106.3 (89.21)

Democracy * lGDP N r2 Economic Rights lgdp

1096

1096 0.0269

1044 0.0284

1096 0.0301

1044 0.0326

78.28 (83.05) 356.5 (236.0)

57.62 (86.25) 259.5 (301.0)

104.1 (77.24) 302.2 (309.2) 11.84 (23.39)

564.6 (435.2) 365.4 (361.0)

-68.99 (55.89)

649.8 (402.5) 415.4 (370.9) 11.77 (23.50) -74.73 (53.81)

Democracy Rights * lGDP Democracy * lGDP N r2 Social Rights lgdp

1069

1069 0.0295

1016 0.0329

1069 0.0317

1016 0.0355

-7.670 (89.62) 377.2 (241.1)

-40.94 (91.37) 271.5 (311.9)

13.03 (83.13) 318.9 (321.1) 11.41 (23.01)

53.96 (380.1) 287.7 (346.0)

-13.21 (52.75)

65.00 (372.5) 327.8 (358.0) 11.36 (23.06) -7.325 (55.22)

1063 0.0295

1010 0.0310

Democracy Rights * lGDP Democracy * lGDP N r2

1063

1063 0.0294

∗

1010 0.0310

(6) 587.8 (594.8) 559.0 (473.5) 178.6 (139.4) -91.63 (82.38) -23.56 (20.29) 1044 0.0341 577.0 (377.0) 544.6 (431.3) 237.4 (162.2) -63.11 (49.55) -32.14 (23.28) 1016 0.0383 -26.48 (335.9) 466.9 (431.1) 244.7 (172.4) 6.382 (50.29) -33.23 (24.74) 1010 0.0340

p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 The dependent variable is the log of the ratio of female to male life expectancy times 100,000 before 1990. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set. The Mean (s.d.) of women’s political, economic and social rights variables respectively are 1.548 (0.679), 1.3 (0.643) and 1.178 (0.795).

92

Table E.31: Female Life Expectancy advantage & Women’s Rights after 1990 Political Rights lgdp

(1) 32.20 (112.0) 694.0∗∗∗ (174.1)

(2) 19.16 (113.3) 729.4∗∗∗ (235.8)

Democracy

(3) 34.24 (105.5) 727.4∗∗∗ (231.7) -33.01 (28.64)

Rights * lGDP

(4) 428.9 (464.4) 832.1∗∗∗ (253.6)

-53.83 (57.40)

(5) 470.0 (518.2) 838.1∗∗∗ (247.5) -36.51 (28.31) -56.74 (63.21)

Democracy * lGDP N r2 Economic Rights lgdp

2729

2729 0.206

2597 0.207

2729 0.207

2597 0.208

-61.55 (84.64) 692.2∗∗∗ (188.7)

-64.70 (85.76) 722.2∗∗∗ (250.0)

-52.31 (88.42) 729.4∗∗∗ (246.0) -37.59 (28.82)

117.8 (429.5) 744.2∗∗∗ (253.0)

-22.51 (50.13)

125.1 (452.1) 751.7∗∗∗ (249.0) -38.12 (28.94) -21.82 (52.70)

Democracy Rights * lGDP Democracy * lGDP N r2 Social Rights lgdp

2716

2716 0.208

2585 0.211

2716 0.208

2585 0.211

-11.66 (82.13) 748.0∗∗∗ (176.3)

-27.15 (83.03) 817.4∗∗∗ (258.1)

-11.28 (81.27) 861.1∗∗∗ (279.5) -42.69 (26.50)

-166.1 (313.9) 799.0∗∗∗ (257.3)

17.78 (35.85)

-209.7 (333.2) 835.6∗∗∗ (277.8) -42.54 (26.59) 25.27 (37.90)

1799 0.235

1709 0.247

Democracy Rights * lGDP Democracy * lGDP N r2

1799

1799 0.235

1709 0.246

∗

(6) 398.4 (519.0) 891.8∗∗∗ (275.7) 58.71 (158.2) -48.80 (62.27) -14.32 (24.01) 2597 0.209 79.56 (452.1) 815.6∗∗∗ (282.2) 55.73 (159.1) -15.31 (52.60) -14.18 (24.03) 2585 0.212 -197.8 (333.4) 697.4∗∗ (338.8) -223.3 (184.9) 23.00 (37.82) 28.46 (28.92) 1709 0.249

p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 The dependent variable is the log of the ratio of female to male life expectancy times 100,000 after 1990. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set. The Mean (s.d.) of women’s political, economic and social rights variables respectively are 1.916 (0.597), 1.328 (0.723) and 1.26 (0.875).

93

Table E.32: Female Life Expectancy advantage & Women’s Rights before 1990; MMR sample Political Rights lgdp

(1) 34.62 (91.18) 351.7 (230.6)

(2) 0.983 (95.10) 248.1 (292.4)

Democracy

(3) -25.77 (94.09) 284.5 (300.3) 13.35 (21.65)

Rights * lGDP

(4) 640.7 (622.5) 422.3 (419.7)

-95.56 (86.79)

(5) 684.8 (639.3) 480.1 (434.6) 13.79 (21.36) -106.3 (89.21)

Democracy * lGDP N r2 Economic Rights lgdp

1096

1096 0.0269

1044 0.0284

1096 0.0301

1044 0.0326

78.28 (83.05) 356.5 (236.0)

57.62 (86.25) 259.5 (301.0)

104.1 (77.24) 302.2 (309.2) 11.84 (23.39)

564.6 (435.2) 365.4 (361.0)

-68.99 (55.89)

649.8 (402.5) 415.4 (370.9) 11.77 (23.50) -74.73 (53.81)

Democracy Rights * lGDP Democracy * lGDP N r2 Social Rights lgdp

1069

1069 0.0295

1016 0.0329

1069 0.0317

1016 0.0355

-7.670 (89.62) 377.2 (241.1)

-40.94 (91.37) 271.5 (311.9)

13.03 (83.13) 318.9 (321.1) 11.41 (23.01)

53.96 (380.1) 287.7 (346.0)

-13.21 (52.75)

65.00 (372.5) 327.8 (358.0) 11.36 (23.06) -7.325 (55.22)

1063 0.0295

1010 0.0310

Democracy Rights * lGDP Democracy * lGDP N r2

1063

1063 0.0294

∗

1010 0.0310

(6) 587.8 (594.8) 559.0 (473.5) 178.6 (139.4) -91.63 (82.38) -23.56 (20.29) 1044 0.0341 577.0 (377.0) 544.6 (431.3) 237.4 (162.2) -63.11 (49.55) -32.14 (23.28) 1016 0.0383 -26.48 (335.9) 466.9 (431.1) 244.7 (172.4) 6.382 (50.29) -33.23 (24.74) 1010 0.0340

p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 The dependent variable is the log of the ratio of female to male life expectancy times 100,000 before 1990 for the MMR sample. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set. The Mean (s.d.) of women’s political, economic and social rights variables respectively are 1.548 (0.679), 1.3 (0.643) and 1.178 (0.795).

94

Table E.33: Female Life Expectancy advantage & Women’s Rights after 1990; MMR sample Political Rights lgdp

(1) 32.20 (112.0) 694.0∗∗∗ (174.1)

(2) 19.16 (113.3) 729.4∗∗∗ (235.8)

Democracy

(3) 34.24 (105.5) 727.4∗∗∗ (231.7) -33.01 (28.64)

Rights * lGDP

(4) 428.9 (464.4) 832.1∗∗∗ (253.6)

-53.83 (57.40)

(5) 470.0 (518.2) 838.1∗∗∗ (247.5) -36.51 (28.31) -56.74 (63.21)

Democracy * lGDP N r2 Economic Rights lgdp

2729

2729 0.206

2597 0.207

2729 0.207

2597 0.208

-61.55 (84.64) 692.2∗∗∗ (188.7)

-64.70 (85.76) 722.2∗∗∗ (250.0)

-52.31 (88.42) 729.4∗∗∗ (246.0) -37.59 (28.82)

117.8 (429.5) 744.2∗∗∗ (253.0)

-22.51 (50.13)

125.1 (452.1) 751.7∗∗∗ (249.0) -38.12 (28.94) -21.82 (52.70)

Democracy Rights * lGDP Democracy * lGDP N r2 Social Rights lgdp

2716

2716 0.208

2585 0.211

2716 0.208

2585 0.211

-11.66 (82.13) 748.0∗∗∗ (176.3)

-27.15 (83.03) 817.4∗∗∗ (258.1)

-11.28 (81.27) 861.1∗∗∗ (279.5) -42.69 (26.50)

-166.1 (313.9) 799.0∗∗∗ (257.3)

17.78 (35.85)

-209.7 (333.2) 835.6∗∗∗ (277.8) -42.54 (26.59) 25.27 (37.90)

1799 0.235

1709 0.247

Democracy Rights * lGDP Democracy * lGDP N r2

1799

1799 0.235

1709 0.246

∗

(6) 398.4 (519.0) 891.8∗∗∗ (275.7) 58.71 (158.2) -48.80 (62.27) -14.32 (24.01) 2597 0.209 79.56 (452.1) 815.6∗∗∗ (282.2) 55.73 (159.1) -15.31 (52.60) -14.18 (24.03) 2585 0.212 -197.8 (333.4) 697.4∗∗ (338.8) -223.3 (184.9) 23.00 (37.82) 28.46 (28.92) 1709 0.249

p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 The dependent variable is the log of the ratio of female to male life expectancy times 100,000 after 1990 for the MMR sample. Country fixed effects panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The rights data comes from the Cingranelli, Richards, and Clay (Cingranelli et al.) data set. The Mean (s.d.) of women’s political, economic and social rights variables respectively are 1.916 (0.597), 1.328 (0.723) and 1.26 (0.875).

95

E.7

MMR from DHS data Table E.34: Dependent Variable: MMR from 2 sources and DSR Desired Sex ratio ln GDP N r2 Desired Sex ratio ln GDP Desired Sex ratio * ln GDP Desired Fertility N r2

MMR DHS 4297.6∗∗∗ (1545.4) -284.1 (177.1) 298 0.249

MMR WDI 958.4∗ (519.4) -34.04 (70.79) 182 0.459

MMR WDI 667.0∗∗ (255.2) 40.90 (43.20) 307 0.444

MMR DHS 383.0 (878.5) 8.197 (77.35) 182 0.0923

6259.1 (4596.6) 109.4 (1193.3) -366.5 (1036.3) 123.2 (141.2) 298 0.254

2572.5∗∗ (1225.4) 258.6 (282.0) -269.1 (247.5) 89.56 (84.24) 182 0.472

2627.7∗∗∗ (549.9) 318.5∗∗∗ (106.4) -285.3∗∗∗ (89.61) 225.6∗∗∗ (69.20) 307 0.520

-2308.7 (2225.6) -503.5 (387.5) 469.2 (362.1) -97.40 (123.8) 182 0.111

∗

p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 The dependent variable is MMR (deaths per 100,000 births) Country Fixed Effects Panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The data on DSR (desired sex ratio -boys/girls) comes from the DHS and has been constructed using the questions on ideal number of boys and girls asked to the mothers. This data and hence the regressions are based on the DHS sample (developing countries). The MMRs in column 1 and 3 come from the WDI, while the MMRs in columns 2 and 4 come from the DHS. Columns 1 and 3 give all the country year observations for which DHS data is available whereas columns 2 and 4 restrict the sample to only the country-year observations for which both the datasets are available. As is evident, in Panel 2 we control for Desired Sex ratio * ln GDP and Desired Fertility.

96

Table E.35: MMR and Women’s Political Rights

right Rights * lGDP lgdp 5 democ 5 N r2

(1) MMR DHS -462.1∗ (242.0) 71.37∗∗ (34.95) -168.7 (124.9) 3.758 (8.455) 213 0.115

(2) MMR WDI -331.1∗∗ (140.6) 53.90∗∗ (20.71) -98.20 (83.65) -5.040 (8.337) 174 0.495

(3) MMR WDI -242.2∗∗∗ (63.18) 32.11∗∗∗ (7.384) -41.37 (26.08) -6.840∗ (3.642) 725 0.316

(4) MMR DHS -373.2 (300.5) 60.76 (42.66) -128.3 (125.9) 8.854 (9.952) 174 0.111

-427.6∗ (237.4) 65.94∗ (34.32) -178.1 (125.5) -24.23 (48.68) 4.547 (7.482) 213 0.117

-276.5∗∗ (130.6) 45.30∗∗ (18.67) -117.8 (88.56) -57.83 (36.51) 8.541 (6.545) 174 0.505

-185.8∗∗∗ (59.68) 25.81∗∗∗ (6.849) -77.55∗∗ (29.88) -77.35∗∗∗ (15.08) 10.40∗∗∗ (2.162) 725 0.378

-338.2 (285.2) 55.24 (40.17) -140.9 (131.5) -24.98 (58.82) 5.474 (9.038) 174 0.115

right Rights * lGDP lgdp 5 democ 5 democ 5 gdp N r2 ∗

p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 The dependent variable is MMR (deaths per 100,000 births) Country Fixed Effects Panel regressions with year dummies have been run. Standard errors in parentheses are clustered at the country level. The MMRs in column 1 and 3 come from the WDI, while the MMRs in columns 2 and 4 come from the DHS. Columns 1 and 3 give all the country year observations for which DHS data is available whereas columns 2 and 4 restrict the sample to only the country-year observations for which both the datasets are available.

Table E.36: MMR and GII

MMR DHS GII N r2 MMR WDI GII N r2

(1) ngii

(2) sbii

(3) gaii

(4) gpii

(5) gii0

(6) gii1

(7) gii2

(8) gtroiano

89.02 (196.0) 137 0.290

89.02 (196.0) 137 0.290

338.7∗∗ (160.1) 124 0.343

-43.72 (142.5) 152 0.329

14.93 (108.8) 119 0.323

44.68 (119.5) 124 0.323

-148.7 (150.9) 124 0.314

-41.53 (253.9) 43 0.645

50.03∗∗ (21.61) 575 0.701

68.52∗∗ (33.75) 575 0.702

105.0∗∗∗ (29.02) 417 0.766

56.60∗ (32.01) 562 0.704

30.24∗∗∗ (10.24) 399 0.750

40.23∗∗∗ (12.19) 417 0.762

24.90∗ (12.57) 542 0.690

2.832 (8.682) 384 0.611

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

97

Table E.37: MMR and GII Identical sample

MMR WDI GII N r2 MMR DHS GII N r2

(1) ngii

(2) sbii

(3) gaii

(4) gpii

(5) gii0

(6) gii1

(7) gii2

(8) gtroiano

349.5∗∗∗ (97.70) 82 0.837

349.5∗∗∗ (97.70) 82 0.837

203.7∗∗ (71.27) 75 0.809

136.9∗ (74.96) 89 0.807

215.3∗∗∗ (46.68) 70 0.875

215.7∗∗∗ (48.30) 75 0.855

253.3∗∗∗ (66.44) 72 0.860

-546.6 (455.3) 22 0.950

88.23 (83.96) 82 0.599

88.23 (83.96) 82 0.599

148.0∗∗∗ (49.28) 75 0.647

2.810 (40.37) 89 0.610

65.09∗ (32.34) 70 0.666

101.3∗∗∗ (23.79) 75 0.655

-7.705 (49.56) 72 0.630

37.24 (138.8) 22 0.596

* p < 0.10, ** p < 0.05, *** p < 0.01 The dependent variables in both Panels 1 & 2 are the maternal mortality ratio (from WDI and DHS respectively) restricted to the same set of countries and years. Standard errors in parentheses are clustered at the country level. Apart from the GII variable we control for the percentage of the population speaking the majority language (for which the GII has been calculated), decade dummies and continent dummies, log of GDP, the log of population, dummies for the World Bank Income groups classification, the percentage of population that is Protestant, Catholic and Muslim, and the proportion of the country that is tropical or subtropical.

Table E.38: MMR from DHS and Stated Son Preference: Prepost 1990: 5 yearly data

DSR lgdp 5

(1) pre 1990 8171.1∗ (4505.4) -371.0 (256.3)

(2) post 1990 423.0 (890.7) 10.44 (79.44)

(3) pre 1990 18760.8 (12592.8) 1909.1 (2478.4) -2071.6 (2147.2)

(4) post 1990 -1984.0 (2111.1) -487.0 (372.5) 454.0 (346.8)

116 0.275

171 0.0976

116 0.291

171 0.108

DSR gdp Desired Fertility N r2

(5) pre 1990 19216.2 (12533.8) 2183.7 (2557.8) -2361.7 (2226.4) 390.5 (324.6) 116 0.300

(6) post 1990 -2215.8 (2207.8) -489.1 (382.7) 459.3 (358.5) -100.4 (127.1) 171 0.116

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table E.39: MMR from DHS and Stated Son Preference: Prepost 1990: Annual data

DSR lgdp

(1) pre 1990 5519.9 (3462.6) 100.7 (130.9)

(2) post 1990 905.5 (818.5) -25.68 (67.30)

(3) pre 1990 13746.1∗ (7286.5) 1911.7 (1225.4) -1682.8 (1100.2)

(4) post 1990 -858.5 (1766.9) -388.3 (358.1) 328.9 (330.0)

746 0.121

719 0.0554

746 0.130

719 0.0575

DSR gdp Desired Fertility N r2

Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01

98

(5) pre 1990 14145.1∗ (7134.2) 2041.0∗ (1203.7) -1816.7 (1081.2) 245.9 (200.5) 746 0.131

(6) post 1990 -1130.4 (1772.7) -403.3 (363.7) 339.2 (337.9) -104.2 (115.0) 719 0.0606

Table E.40: MMR from DHS and GII: Prepost 1990: 5 yearly data (1) pre 1990 13.66 (237.3) -944.5∗∗ (349.4) 49 0.586

GII lgdp 5 N r2

(2) post 1990 65.09∗ (32.34) 28.65 (49.96) 70 0.666

Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table E.41: MMR from DHS and GII: Prepost 1990: Annual data

GII N r2

(1) pre 1990 97.08 (158.5) 318 0.239

(2) post 1990 53.13 (37.91) 270 0.524

Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01

Table E.42: MMR from DHS and Stated Son Preference: Pre 1990 sample

DSR lgdp N r2

(1) DSR 20 679.0 (677.5) 127.4 (129.6) 704 0.177

(2) DSR 25 944.3 (650.6) -53.96 (81.75) 602 0.194

(3) DSR 30 715.2 (573.7) -47.16 (61.58) 426 0.241

Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01

Table E.43: MMR from DHS and GII: Pre 1990 sample

GII lgdp N r2

(1) ngii 306.8 (363.3) -251.5*** (71.50) 360 0.219

(2) sbii 306.8 (363.3) -251.5*** (71.50) 360 0.219

(3) gaii 510.6* (286.5) -410.5*** (110.2) 320 0.261

(4) gpii -70.09 (217.8) -367.9*** (73.55) 408 0.302

Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01

99

(5) gii0 97.08 (158.5) -373.3*** (101.1) 318 0.239

(6) gii1 145.3 (172.6) -367.0*** (106.2) 320 0.242

(7) gii2 -129.6 (144.3) -348.0*** (86.24) 338 0.233

(8) gtroiano -821.6 (799.5) -256.4** (85.18) 140 0.504

F F.1

Figures Life Expectancy Trends

(a)

(b)

Figure F.1: Female Life Expectancy is plotted against time. The Life Expectancy data comes from the World Bank WDI and spans over more than 190 countries and is available for the period of 1960 - 2011. Here we plot them for the High income: OECD region (World Bank Region Classification is used).

(a)

(b)

Figure F.2: Female Life Expectancy is plotted against time. The Life Expectancy data comes from the World Bank WDI and spans over more than 190 countries and is available for the period of 1960 - 2011. Here we plot them for the whole World sample for the years 1970, 1990 and 2010. The World Bank Region Classification is used.

100