Results: PPVT SCORE Table 3: Regressions for PPVT Across Countries and Age Groups Age 5 Age 8 Ethiopia India Peru Vietnam Ethiopia India Peru Vietnam Wealth 1.284*** 1.380** 2.701*** 2.310*** 5.688*** 3.083*** 3.297*** 2.263*** index (0.467) (0.701) (0.561) (0.618) (1.046) (0.780) (0.363) (0.689) Mother’s 0.535*** 1.097*** 0.172 0.370** 0.783*** 0.727*** 0.421*** 0.860*** education (0.113) (0.145) (0.125) (0.146) (0.274) (0.197) (0.085) (0.179) Father’s 0.130 0.361*** 0.276** 0.551*** 0.626** 0.394*** 0.506*** 0.544*** education (0.092) (0.121) (0.136) (0.135) (0.243) (0.148) (0.087) (0.166) Mother’s 0.058 -0.037 -0.121 0.089 0.072 -0.028 -0.035 0.051 height (0.045) (0.071) (0.069) (0.062) (0.109) (0.088) (0.047) (0.080) Mother’s 0.090 -0.306 -0.023 -1.207*** -0.093 -0.061 0.122 -0.277 bargaining (0.275) (0.454) (0.368) (0.370) (0.653) (0.618) (0.248) (0.479) power Mother’s -0.033 0.108 0.180 0.156 2.235*** 2.643*** 0.504** 0.982 noncognitive (0.274) (0.429) (0.383) (0.384) (0.682) (0.665) (0.255) (0.585) skills Mother’s 0.322 1.910*** 0.883** 0.396 -0.616 -0.590 0.369 -0.309 subjective (0.260) (0.522) (0.352) (0.425) (0.687) (0.616) (0.264) (0.552) well-being Mother’s 0.585 0.253 -1.562*** -0.037 0.201 -0.412 -2.119*** -1.480*** social capital (0.343) (0.473) (0.428) (0.377) (0.915) (0.599) (0.691) (0.551) Mother’s 0.096 0.308* 0.408** 0.149 0.846*** 0.506*** 0.957*** 0.721*** aspirations (0.104) (0.173) (0.192) (0.168) (0.293) (0.172) (0.206) (0.220) for child’s education R-squared 0.3 0.26 0.3 0.36 0.47 0.23 0.45 0.32 Observations 1861 1851 1903 1747 1857 1901 1842 1848 Age 12 Age 15 Ethiopia India Peru Vietnam Ethiopia India Peru Vietnam Wealth 2.495** 1.525 2.324*** 4.543*** 5.154*** 4.834*** 2.149*** 5.147*** index (1.133) (0.965) (0.831) (1.090) (1.111) (1.220) (0.753) (1.047) Mother’s 0.492 0.263 0.466*** 0.541** 0.314 0.924*** 0.257 0.232 education (0.253) (0.208) (0.143) (0.228) (0.288) (0.265) (0.143) (0.211) Father’s 0.274 0.259 -0.116 1.001*** 0.386 0.178 0.543*** 0.224 education (0.250) (0.166) (0.185) (0.258) (0.253) (0.211) (0.188) (0.234) Mother’s 0.098 0.105 -0.066 0.254 0.131 0.186 0.010 0.041 height (0.127) (0.117) (0.088) (0.133) (0.139) (0.145) (0.091) (0.112) Mother’s 0.953 -0.106 0.712 -1.616** -0.566 0.003 -0.487 -0.249 bargaining (0.776) (0.703) (0.502) (0.663) (0.914) (0.902) (0.471) (0.632) power Mother’s 3.149*** 0.729 0.804 -0.932 1.435 0.864 -0.045 0.063 noncognitive (0.767) (0.763) (0.529) (0.720) (0.942) (0.924) (0.490) (0.593) skills Mother’s 0.283 2.852*** -1.032 -1.191 0.104 0.379 -0.159 -0.703 subjective (0.807) (0.778) (0.540) (0.694) (0.940) (0.980) (0.543) (0.656) well-being Mother’s -1.549 0.767 -0.473 0.303 0.770 1.963** -5.016*** 0.554 social capital (0.821) (0.724) (0.549) (0.633) (1.645) (0.906) (1.507) (0.646) Mother’s 1.257*** 1.961*** 0.941** 2.422*** 1.333*** 2.945*** 2.146*** 2.347*** aspirations (0.345) (0.292) (0.388) (0.676) (0.415) (0.309) (0.416) (0.528) for child’s education R-squared 0.35 0.27 0.34 0.50 0.3 0.31 0.4 0.41 Observations 953 971 672 945 962 944 652 947
Results: PPVT Score Figure 8 The Size of the Coefficient of Mother’s Aspirations for Child’s Education in Child’s PPVT Score Regressions 3 2 Coefficient 1 0 5 8 12 15 Child's age in years Ethiopia India Peru Vietnam
Results: Noncognitive Skills Index Table 4: Regressions for Child’s Noncognitive Skills Across Countries and Age Groups Age 8 Age 12 Ethiopia India Peru Vietnam Ethiopia India Peru Vietnam Wealth 0.150*** 0.056 0.069** 0.039 0.073 0.123*** 0.166*** 0.057 index (0.030) (0.030) (0.032) (0.036) (0.040) (0.042) (0.055) (0.056) Mother’s 0.004 0.011 0.020*** 0.010 -0.016 0.012 0.030*** 0.017 education (0.008) (0.006) (0.008) (0.008) (0.010) (0.009) (0.012) (0.012) Father’s 0.007 0.005 -0.002 -0.008 0.034*** 0.018** 0.010 0.026** education (0.007) (0.005) (0.008) (0.007) (0.009) (0.008) (0.014) (0.013) Mother’s -0.001 0.002 0.002 -0.001 0.002 0.007 0.006 0.001 height (0.004) (0.003) (0.004) (0.003) (0.005) (0.005) (0.007) (0.005) Mother’s -0.036 -0.039 0.003 -0.025 0.037 -0.012 -0.035 -0.063 bargaining (0.024) (0.021) (0.023) (0.022) (0.029) (0.031) (0.036) (0.033) power Mother’s 0.145*** 0.387*** 0.081*** 0.196*** 0.381*** 0.318*** 0.129*** 0.240*** noncognitive (0.025) (0.025) (0.027) (0.028) (0.033) (0.040) (0.043) (0.033) skills Mother’s 0.007 0.015 0.031 0.029 0.044 -0.044 -0.047 0.025 subjective (0.024) (0.023) (0.023) (0.026) (0.029) (0.035) (0.041) (0.034) well-being Mother’s 0.207*** 0.074*** 0.146** 0.056** 0.074** 0.025 -0.038 -0.052 social capital (0.033) (0.021) (0.062) (0.024) (0.035) (0.033) (0.039) (0.034) Mother’s -0.011 0.021*** 0.040** 0.037*** 0.018 0.032** 0.032 0.058** aspirations (0.011) (0.007) (0.016) (0.012) (0.014) (0.012) (0.026) (0.026) for child’s education R-squared 0.23 0.30 0.11 0.16 0.31 0.21 0.24 0.14 Observations 1877 1917 1921 1949 979 994 685 990 Age 15 Ethiopia India Peru Vietnam Wealth 0.135*** 0.040 0.182*** 0.111 index (0.046) (0.044) (0.050) (0.057) Mother’s 0.009 0.012 0.049*** 0.015 education (0.010) (0.010) (0.011) (0.012) Father’s 0.014 0.021** -0.020 0.007 education (0.010) (0.008) (0.014) (0.013) Mother’s 0.003 0.008 -0.005 -0.002 height (0.005) (0.005) (0.007) (0.006) Mother’s 0.045 -0.058 -0.015 0.024 bargaining (0.034) (0.032) (0.037) (0.035) power Mother’s 0.177*** 0.000 0.143*** 0.071** noncognitive (0.039) (0.033) (0.041) (0.034) skills Mother’s 0.016 0.027 0.010 0.030 subjective (0.035) (0.033) (0.040) (0.033) well-being Mother’s 0.259*** 0.160*** 0.082 0.092** social capital (0.069) (0.034) (0.121) (0.040) Mother’s 0.010 0.069*** 0.052** 0.058*** aspirations (0.013) (0.012) (0.026) (0.022) for child’s education R-squared 0.24 0.15 0.12 0.11 Observations 973 974 672 970
Results: Noncognitive Skills Index Figure 8 The Size of the Coefficient of Mother’s Aspirations for Child’s Education in Child’s Noncognitive Skills Regressions .08 .06 .04 Coefficient .02 0 -.02 5 8 12 15 Child's age in years Ethiopia India Peru Vietnam
Summary of Results • The most important predictors for height-for-age across countries and ages include: - household wealth -mother’s height and -parental education • no systematic pattern is found on the magnitude of these associations across age groups • The most important predictors for cognitive achievement across countries and ages include : -household wealth -parental education and -mother’s aspirations for the child’s education • The only systematic pattern in the magnitude of the associations across age groups is observed for mother’s aspirations for the child’s education
Summary of Results • The most important predictors for noncognitive skills across countries and ages include: - mother’s noncognitive skills -social capital -household wealth -mother’s aspirations for the child’s education -parental education • The only systematic pattern in the magnitude of the associations across age groups is observed for mother’s aspirations for the child’s education
Conclusions • There is a lack of studies in the development literature on that consider simultaneously the association of a wide range of parental background markers with children’s human capital across countries and how these associations may change with children’s age • We address this gap by using data from the Young Lives cohort study in Ethiopia, India, Peru and Vietnam to investigate the association of parental background factors with indicators of child’s human capital at ages 5, 8, 12 and 15 years • Our key findings are that across countries and age groups: -parental income is the most important predictor of child’s nutritional status and cognitive achievement across countries and at all stages of childhood -parental education has a weak or no association with children’s human capital measures -mother’s personality traits are the most important predictors of children’s noncognitive skills across countries and at all stages of childhood -the association of mother’s aspirations for child’s education with the child’s cognitive and noncognitive skills increases with children’s age
The Making of the Middle Class in Africa Mthuli Ncube and Abebe Shimeles African Development Bank CSAE Conference 2013: Economic Development in Africa 17-19 March 2013, St Catherine’s College, Oxford CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Why study the middle class in Africa? The middle class is often associated with stability and driver of social and economic reforms (e.g Sridharan, 2004, Loyza et al, 2012 ) A large middle class ushers in possibilities for social mobility and trickling down of wealth or inclusive growth (e.g. Doepke and Zilibotti, 2007; Birdsall, 2010) A large middle class is a source of dynamic economic growth and entrepreneurship (Easterly, 2001; Desgoigts and Jaramillo, 2009); CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Objectives of this study Building on existing work of the African Development Bank (2011), this study attempts to provide answers to the following questions? * What is the size of the middle class and how has it evolved over time? who are the middle class and what are their characteristics? How path dependent is a middle class status at the household level? * What explains cross-country variation in the size of the middle class? speci…cally we focus on institutions and policy. For the latter governance, education and health are examined in some detail. CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Identifying the middle class Some focus on relative de…nition where the upper and lower bounds are a certain percentage of either the median or mean income (e.g. Birdsall, Graham and Pettinato, 2000) Others use absolute de…nition such as individuals living below 2$ and 10$ per day (Banerjee and Du‡o, 2008; Milanovic and Yitzhaki, 2002; Bhalla, 2008 and others). While each de…nition has some grounding, arbitrariness cannot be avoided. In our case we used the African median weighted by population which is 0.5-0.7. most important however is to study the whole distribution provided in the Kernel density. CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
On Data and Method The main source of data for this paper comes from Demographic Health Surveys (DHS) for 42 countries covering the 1990s and 2000s. A pseudo-panel constructed on the basis of age-sex cohort was also used to look into mobility across classes and also role of education and health as important pathways. In addition we report results from rich panel data set from Ethiopia that covers 10 years in …ve waves to analyze the dynamics of the middle class. CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Note on data and methods A composite asset index was constructed for each household using the Multiple Correspondence Analysis method (MCA). This is a method close to a Principal Components analysis and is appropriate for the type of response in the data (mostly categorical) The MCA helps establish weight for the assets based on optimal variance attributed to each of the categories. CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
MCA in brief k W j = ∑ i = 1 a i c ij i represents the k assets that individual j possesses at a point in time to achieve a welfare level W j , which could be cardinal or unit free (ordinal) depending on how the components enter the welfare measure. CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
MCA in brief 11 m i ∑ ∑ asset index = w ij Z ij i = 1 j = 1 w ij are the computed weights A question Q i with m i answer choices is transformed into a set of binary question Z ij , j = 1 ... m i in such a way that choosing modality j of question Q i is equivalent to Z ik = 0 for k 6 = j and Z ij = 1 . CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Results The average size of the middle class that we estimated from the DHS data is strikingly closer to the …gure reported in AfDB(2011): * in the late 2000 (2006-2009) the size of the middle class in Africa on the average was around 14%. * AfDB(2011) reported 13.4% for 2010 based on consumption expenditure lying between $4-$20 per day per person. Our report from DHS data showed clear sign of increasing middle class from 5% in the 1990s to 15%. AfDB(2011) did not show much improvement. CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Trends in the size of middle class Figure 1: Trend in the size of middle class and Gini coe¢cient in asset index CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Results Asset-based estimates of middle class are highly correlated with consumption based measures CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Results Except for a few, most countries recorded growth of the middle class since the 1990s Change in the middle class size(%population) CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Size of middle class, poverty and inequality It is possible for the size of the middle class to decrease following a dramatic decline in poverty and inequality (see case of Egypt: based on DHS Figure 4) Figure 4: Kernel density of Asset Index for Egypt CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Size of middle class, poverty and inequality Or for size of middle class to increase followed also by signi…cant decline in poverty and inequality (case of Ghana : based on DHS-Figure 5) Figure 5: Kernel density for Ghana CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Size of middle class, poverty and inequality Or size of middle class to decline following an increase in poverty and inequality (case of Madagascar) Figure 6: Kernel density for Madagascar CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Mobility into and out of middle class status Table 1: Transition Matrix CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
(cont’d) Table 1: Transition Matrix CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
(cont’d) Table 1: Transition Matrix CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Ethiopian Case Transition matrix for self-reported wealth status in urban Ethiopia: 1994-2004 CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Survival function for middle class status in Ethiopia Strong state dependence CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Cross-country correlates of size of middle class: Exogenous factors and initial conditions are important Higher level of ethnic fractionalization is correlated with low incidence of middle class in Africa CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Cross country correlates Size of the middle class is strongly correlated with initial level of development CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
The Role of Institutions: Random e¤ects estimation of determinants of asset index Governance, ethnicity, education and health remain important correlates with the making of middle class CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
The Role of Social Capital: Trust is important even after controlling for the e¤ect of inequality in assets Random e¤ects estimation of determinants of asset index CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Education: The role of education may slow down with increase in its supply Regression coe¢cients of education and mean years of education CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Ethnic Division Growth in the size of middle class is much slower in countries where ethnic division is high suggesting again some structural bottlenecks Growth rate in asset index and some initial conditions CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Concluding Remarks Size of middle class is rising in most African countries: which is a good thing The probability of maintaining a middle class status is also fairly high with real possibilities to move up as well as slip back into poverty. Ethnicity, initial per capita income, level of trust among citizens shape the evolution of middle class CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Concluding Remarks From a policy perspective, it is evident that improving governance conditions and investing on education and health can take countries a long way in improving the size of the middle class. Would nurturing the size of the middle class compatible with short term and long term interests of national governments? We do not know at this point. CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Thank You CSAE Conference 2013: Economic Developme Ncube & Shimeles (AfDB) The Making of the Middle Class in Africa / 29
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Violence against women in Sub-Saharan Africa Andreas Kotsadam 1 Sara Cools 2 1 University of Oslo 2 BI Norwegian Business School 19 March 2013 Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Motivation Domestic violence is prevalent in all societies but to different degrees. It entails large costs in terms of women’s health, productivity, shame, and fear. Fear of violence affect more women than those actually beaten. Other family members suffer, in particular children. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary What we do Using microdata for over 540 000 women and almost 200 000 men we: 1 Examine the variation in acceptance and actual wife beating across time and space. 2 Explore explanations at both the individual and contextual level. 3 Explore hypotheses regarding conflict, religion, and education using spatial data, historical exposure and reforms. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Preview of Results Female employment at the individual as well as at the societal level is associated with more wife beating. So is living in a community with more wealth inequality. Individual attitudes toward beating predicts actual violence, also living in communities that accept wife beating. We find no heightened risk of exposure to wife beating during conflicts. Having more education, or a partner with more education is correlated with less risk of wife beating but we find no effects of education using educational reforms. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Outline Previous literature and testable hypotheses 1 Data 2 Results from basic regressions 3 A closer look at religion, conflicts, and education. 4 Summary 5 Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Previous literature There are many different explanations for why men abuse their partners. Evolutionary psychology and radical feminism both claim that control of women’s sexuality is central. The large variation across time and space suggests social factors are important. We focus on resources, religion, inequality, and conflict. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Resources Individual level resources (wealth, employment, education) lead to autonomy and autonomy is argued to reduce violence. On the other hand, increased resources may lead to a backlash: 1 A threat to male dominance as resources carry symbolic value. 2 Violence may be used to reinstate men’s bargaining power. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Religion Religious traditions shape attitudes, both at the individual and societal level. Whether this influence is in a conservative and patriarchal direction is unclear and contested. Nunn (2011) finds that Protestantism in Africa is associated with higher gender equality in education while Catholicism is associated with less. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Inequality Inequality among men and households is claimed to be a risk factor for wife beating (Jewkes 2002). So is inequality between men and women (True 2012). Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Conflict Increased violence against women in times of conflict may not be solely driven by military strategies. Several studies find domestic violence to increase during conflicts (see True 2012 for an overview). Mechanisms are thought to be hypermasculinity and a celebration of armed masculinity. La Mattina (2013), however, finds no evidence of increased generalized wife beating after the conflict in Rwanda. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Demographic and Health Surveys DHS data with standardized surveys has been collected in developing countries since the 1980s. Since the 1990s DHS surveys include questions on attitudes toward wife beating. Data on actual experience of domestic violence has been collected since the late 1990s in a special module. Women of fertile age (15-49) are always interviewed and recently a smaller subset of men are also included. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Attitudes toward wife beating For women there are 50 surveys with attitudes toward wife beating including 540 842 persons. 21 517 clusters, in 242 regions, in 29 countries for the years 1992-2011. 195 188 men from 22 countries between 1999-2011. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Measures (1) Respondents are asked if a husband is justified in beating his wife if she: 1 goes out without telling him, 2 neglects the children, 3 argues with him, 4 refuses to have sex with him, 5 or burns the food. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Measures (2) We create two different variables from these questions: 1 Beat=1 if the person agrees with at least one of the statements. 55 % of women. 2 Nrbeat= number of statements the respondent agrees with. 1.6 on average. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Actual wife beating Only women who have ever had a parther are asked. We have data from 21 surveys including 108 087 women. 9 426 clusters, in 154 regions, in 15 countries for the years 2003-2011. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Measures (1) A special domestic violence module is used with: Specifically trained staff. Strict protocol to ensure privacy. A modified Conflict Tactics Scale with many different questions. 12 different questions ranging from pushing, shaking and slapping to attacking with gun, knife or other weapon. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Measures (2) We create two different variables from these questions: 1 Physical violence=1 if the woman has ever experienced any type of abuse: 32 %. 2 Violence last year= If the respondent has been abused last year: 27 %. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Results from basic regressions All these regressions: Include year and region fixed effects. Include region specific time trends. Cluster the standard errors at the DHS cluster level. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Attitudes toward violence (Table 2 column 4) VARIABLES beat urban -0.024*** (0.004) age -0.009*** (0.001) age2 0.009*** (0.001) working 0.008*** (0.003) schoolyears -0.012*** (0.000) husband_schoolyears -0.004*** (0.000) Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Continued (Table 2 column 4) VARIABLES beat number_children 0.007*** (0.000) wealth_quintile -0.007*** (0.001) christian -0.007 (0.005) muslim 0.017*** (0.006) Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Actual violence (Table 4 column 4) VARIABLES physical urban 0.017*** (0.005) age 0.006*** (0.001) age2 -0.013*** (0.002) working 0.050*** (0.003) schoolyears -0.003*** (0.001) husband_schoolyears -0.003*** (0.000) Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Continued (Table 4 column 4) VARIABLES physical number_children 0.010*** (0.001) wealth_quintile 0.003 (0.002) muslim -0.082*** (0.009) christian -0.009 (0.008) Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Nonlinear relationship with education Attitudes and experience of violence by level of educational attainment .8 .6 S hare of w om en .4 .2 0 0 1 2 3 beat physical_violence Source: Own calculations based on DHS data Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Household wealth is only correlated with attitudes Attitudes and experience of violence by wealth quintile .6 S hare of w om en .4 .2 0 1 2 3 4 5 beat physical_violence Source: Own calculations based on DHS data Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Contextual level variables We aggregate our variables of main interest to the cluster and regional level using the Jacknife method. We also create a gini coefficient of wealth inequality. We include country fixed effects and country specific time trends. Standard errors are clustered at the regional level when regional variables are included. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Results of contextual variables (1) The cluster level seems to be more important than the regional level. This is also confirmed by multilevel regressions. More education for women seems unrelated to risk of wife beating. While living in areas where men are more educated is more dangerous. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Results of contextual variables (2) Having more women working in the cluster is correlated with more violence. So is living in a cluster with more Christians. But living in a context with more Muslims is correlated with less violence. Living in a more wealth unequal area is strongly correlated with more violence. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary The relationship between attitudes and actual beating At the individual level, thinking wife beating is ok is correlated with an 8 ppt higher actual experience. Contextual acceptability of wife beating is a strong predictor of being a victim, also when controlling for individual attitudes. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary A closer look at Christianity Following Nunn (2010, 2011) we use data on historical missionary influence from a map by Roome (1924). We know the position of all Catholic and Protestant missionary stations in Africa at the time. Nunn (2010) shows that these missions explain Christianity today and Nunn (2011) shows that women with ancestors exposed to Protestant missions have higher education. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Missionary stations and wife beating Reduced form results show that living close to a historical location of a Protestant (but not a Catholic) mission is correlated with less acceptance toward wife beating. Living close to where any type of mission was situated is correlated with more actual wife beating. As we believe the link is via more religiosity we instrument Christianity with historical exposure to missions and find that the second stage effects of being Christian is strongly correlated with less acceptance of albeit more actual wife beating. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary No causal interpretation Even though the first stages are strong and plausible and we are unable to reject that the instruments valid we do not give the results a causal interpretation. Missionary stations were not allocated randomly. Nunn (2010, 2011) controls for historic railway lines, explorer routes, soil quality, and water resources among other things. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Conflicts We use the UCDP GED dataset with areas exposed to deadly conflicts in Africa since 1989. The dataset includes the starting and end dates of conflicts as well as the number of deaths. We merge the conflict polygons with our DHS clusters and calculate the distance in time to exposure to conflicts. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Education Following Fenske (2012) we exploit a number of educational reforms to see the effects on wife beating (he looks at polygamy). In particular, we exploit the primary school expansion in Nigeria in 1976 (Osili and Long 2008), the expansion of secondary school in Zimbabwe in 1980 (Aguero and Bharadway 2012, Aguero and Ramachandran 2012), and the extension of primary school by one year in Kenya in 1985 (Chicoine 2012) as sources of exogenous variation. We find no effects of education on violence using these reforms, but neither can we reject quite large effects. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Summary (1) Wife beating is widely accepted in SSA and the levels of exposure are high. Female employment is associated with more violence as is living in places where more women work. Being Muslim or living in a Muslim community is correlated with less risk of wife beating. Christianity seems correlated with less acceptance, albeit more actual wife beating. Andreas Kotsadam, Violence against women Sara Cools
Introduction Previous literature and testable hypotheses Data Results from basic regressions A closer look at religion, conflicts, and education. Summary Summary (2) Living in a community with more wealth inequality is associated with more violence against women. Attitudes are important predictors of actual wife beating, both at the individual and contextual level. We find no heightened risk of exposure to domestic violence during conflicts. Having more education, or a partner with more education is correlated with less risk of wife beating but we find no effects of education using reforms in three countries, but we can not reject effects either. Andreas Kotsadam, Violence against women Sara Cools
Decomposition of Inequality across the Poor and Population Subgroups for Multidimensional Counting Approach Sabina Alkire and Suman Seth The Centre for the Study of African Economies 2013, Oxford 19 March 2013
Introduction Recent development in multidimensional poverty measurement – Approaches for Cardinal data (Chakravarty, Mukherjee and Ranade 1998, Tsui 2002, Bourguignon and Chakravarty 2003, Massoumi and Lugo 2008, Alkire and Foster 2011) – Counting Approaches for Binary data (Bossert, Chakravarty and D’Ambrosio 2009, Jayaraj and Subramanian 2009, Alkire and Foster 2011, Rippin 2011) Consideration of Inequality in poverty measurement has been customary since Sen (1976) – Three I’s of poverty (Jenkins and Lambert 1997) 2
Consideration of Inequality in Poverty Analysis Natural for measures in cardinal approach Not straightforward for measures in counting approach However, inequality can be captured across deprivation counts, if we take c i to be cardinally meaningful – Deprivation count vector c = (c 1 , ..., c n ); 0 < c i < 1 3
Consideration of Inequality in Poverty Analysis Fine tune a poverty measure to capture inequality – Bossert, Chakravarty and D’Ambrosio 2009 • Uses symmetric or generalized mean across deprivation counts – Jayaraj and Subramanian 2009 and Rippin (2011) • Weights deprivation counts by themselves (like FGT) Primarily used for ranking but not suitable for understanding inequality within groups and between groups 4
What Type of Inequality Matters? Should the consideration for inequality be based on relative or absolute distances in deprivations? – ‘Leftist’ vs. ‘rightist’ viewpoint (Kolm 1976) Example: c 1 = (0,0, 0.1 , 0.3 ) and c 2 = (0,0, 0.4 , 1 ) Which vector is more unequal across the poor (Union)? – Relative (scaling): c 1 has more inequality (Hard to defend) – Absolute (difference): c 2 has more inequality 5
Example: Two States of India (Union) State A State B Deprivation Score in Millions Deprivation Score in Millions Not deprived 5.4 Not deprived 4.8 0-0.3 24.1 0-0.3 21.2 0.3-0.6 3.0 0.3-0.6 24.4 0.6-0.8 0.2 0.6-0.8 9.3 0.8-0.9 - 0.8-0.9 1.9 0.9-1 - 0.9-1 1.0 Total Poor 27.2 Total Poor 56.8 Total Population 32.6 Total Population 62.6 Which state has more inequality among the poor (Union)? GE(2): 0.253 Gini: 0.372 GE(2): 0.144 Gini: 0.304 A: Kerala, B: Rajasthan, Year: 2006 from Alkire and Seth (2013) 6
Solution? We argue: ‘distance’ is more appropriate than ‘scaling’ in understanding inequality in counting framework Then a. Should we create a poverty index that is sensitive to absolute inequality? b. Should we use a separate inequality measure to analyze inequality among the poor? One advantage of (b) is that it can be used to analyze inequality within groups and between groups 7
Which Inequality Measure? It depends on the additional requirements that we want the measure to satisfy – Additive Decomposability • Overall = Total within-group + between-group – Total within group = population weighted average of all within groups • population share weighted decomposability (Chakravarty 2001) – Permutation invariance – Zero inequality when everybody has same deprivation score – Increase in inequality due to regressive transfer (Dalton) 8
Which Inequality Measure? The only absolute inequality measure that satisfies these properties is variance ( its positive multiple, technically ) – (Chakravarty 2001) V(x) = αΣ i (x i – µ (x)) 2 /n where, V(x): positive multiple of variance of vector x µ (x): mean of elements in x n: population size of x α > 0 9
Revisit the Example State A State B Deprivation Score in Millions Deprivation Score in Millions Not deprived 5.4 Not deprived 4.8 0-0.3 24.1 0-0.3 21.2 0.3-0.6 3.0 0.3-0.6 24.4 0.6-0.8 0.2 0.6-0.8 9.3 0.8-0.9 - 0.8-0.9 1.9 0.9-1 - 0.9-1 1.0 Total Poor 27.2 Total Poor 56.8 Total Population 32.6 Total Population 62.6 α = 100 V: 1.30 V: 4.69 Delamonica and Minujin (2007) and later Roelan et al. (2010) and Roche (2013) use standard deviation: Not decomposable 10
Recommend
More recommend
