1. INTRODUCTION
Individuals with more education command higher earnings in the labor market, and the return to education is higher in developing countries [Psacharopoulos (Reference Psacharopoulos1994); Schultz (Reference Schultz2002)]. Education has a positive impact on other outcomes as well, such as the health status of individuals [Chou, Liu, Grossman and Joyce (Reference Chou, Liu, Grossman and Joyce2010); Grossman (Reference Grossman1972a, Reference Grossman1972b); Lleras-Muney (Reference Lleras-Muney2005)]. Although low-income countries struggle with low levels of human capital as a barrier to development, many of these countries exhibit substantial discrepancies in education acquired by men vs. women. For example, school life expectancy is 0.5 years higher for men compared to women in countries where per capita income is less than $8,000.Footnote 1 The difference in expected schooling is about 1 year in favor of men among countries where per capita income is less than $4,000, and it goes up to 1.4 years in countries where per capita income is less than $2,000.
Gender difference in education is not an isolated phenomenon. Similar differences between men and women exist in developing countries in many other dimensions ranging from life expectancy to participation in politics, to the propensity for being exposed to violence.Footnote 2 Eradicating these inequalities is important for a number of reasons, ranging from human rights to economic development. For example, following Galor and Weil (Reference Galor and Weil1996), Lagerlöf (Reference Lagerlöf2003) shows that long-run economic development in Europe is related to the long-term trend in gender equality. Knowles, Lorgelly and Owen (Reference Knowles, Lorgelly and Owen2002) find that female schooling increases labor productivity across countries.
In this paper, we use data from Sierra Leone to investigate whether education alters women's preferences in the direction of empowerment and self-determination. Empowerment is usually understood as an exogenous change in the social or institutional structure to provide opportunities that were not previously available. Examples include changing the inheritance laws, providing property rights and voting rights, enacting laws, and implementing quotas to increase participation in the labor market and in politics [see Duflo (Reference Duflo2012) for a detailed discussion].
We define empowerment as having the knowledge along with the power and the strength to make the right decisions regarding one's own well-being. Because of social norms or lack of economic power, women may feel obligated to submit to the wishes of others (e.g. their husband or father) even though doing so would clearly decrease their well-being in some circumstances. An example is the belief that the husband is justified in beating the wife if she refuses to have sex. We argue that women are empowered internally when they can take positions to protect their own well-being. In the example above, if a woman changes her belief and declares that violence against women in the form of wife beating is not justified, we call this an indication of empowerment.Footnote 3 Another example is the attitudes toward the violent practice of female genital mutilation (FGM), which is physically and psychologically harmful to the woman affected. FGM is considered an extreme form of discrimination against women as well as a human rights violation [World Health Organization (2008)].Footnote 4 If a woman, who lives in an FGM-practicing society, alters her position on the appropriateness of the practice of FGM as a social norm, and supports its abolition, we consider this an indication of empowerment.Footnote 5
We investigate the impact of education on seven outcomes, four of which are attitudes regarding violence toward women. The issue is important, especially because a recent comprehensive report of the World Health Organization documented that 35% of women worldwide have been victims of physical and/or sexual intimate partner violence or non-partner sexual violence since age 15 years [World Health Organization (2013)]. We also analyze the extent to which education alters women's behaviors such as the use of modern contraceptives, whether she is tested for AIDS, as well as her desired number of children.
We exploit variations in schooling that were generated by an exogenous education policy reform in Sierra Leone, implemented in 2001. The program increased access to primary education, and its intensity varied between different districts of the country. We employ the Sierra Leone Demographic and Health Survey (SLDHS), administered in 2008, to investigate the extent to which exposure to enhanced educational resources had an impact on acquired schooling. Using instrumental variables, we then estimate the impact of schooling on women's preferences regarding their own health and well-being.Footnote 6 The identification strategy employed in the paper is similar to that used by Duflo (Reference Duflo2001) and Osili and Long (Reference Osili and Long2008), where differences in program intensity between districts, and differences in the exposure to the program across age cohorts are exploited. Specifically, individuals who were of the primary school age (6 to 11 years) when the policy began in 2001 were exposed to the free and compulsory primary education, while older individuals could not benefit from the program.
An increase in education can modify individuals’ behavior through a number of different mechanisms. Literacy could be one such mechanism. For example, when a woman becomes literate, she can read and learn about the health risks of unprotected sex. Alternatively, education can alter preferences by influencing people's time discounting [Becker and Mulligan (Reference Becker and Mulligan1997)], which can in turn impact behavior. Education can also improve the bargaining power of women in the household by increasing potential labor market earnings. The improved bargaining position may help women to take positions which do not conflict with their well-being. For example, in the model of Iyigun and Walsh (Reference Iyigun and Walsh2007), husband and wife bargain over household decisions, and the marital bargaining power depends on the relative income of the spouses. An increase in women's education generates higher earnings potential and more marital power, which leads to lower fertility. Basu (Reference Basu2006) presents a similar household model in which husband and wife can have different preferences. Household decisions, including the supply of child labor, are determined by the power balance between the spouses and that the power balance can be impacted by the decisions made. Chiappori, Iyigun and Weiss (Reference Chiappori, Iyigun and Weiss2009) formulate a model where investment in schooling allows individuals to earn higher wages, and it also allows them to extract a larger share of the marital surplus.
We are unable to test all possible mechanisms. But, we show that literacy is not the mechanism that changes women's preferences in Sierra Leone. Similarly, we find access to health facilities is not the reason for the change in women's attitudes. We find that the change in attitudes that is empowering for women is not likely to be driven by migration or exposure to war violence either. Although we cannot determine the particular channel through which education alters attitudes, we demonstrate that there is a strong net effect of education.
It should be noted that even if increased internal empowerment could be a necessary condition to improve the well-being of women, it may not be sufficient to actually achieve a tangible change in outcomes, at least in the short-run, if external obstacles remain. We find indirect evidence on this when we analyze how education impacts men's attitudes toward women's well-being. More specifically, we show that an increase in men's education does not alter men's attitudes toward women's well-being. Furthermore, we find that outcomes such as women's age at first marriage and age at first birth are not impacted by an increase in education on this margin.
Although there is a large literature that investigates the impact of education on women's marriage and fertility behavior, there exist only a few papers on the impact of education on women's preferences as defined in this paper. Friedman, Kremer, Miguel and Thornton (Reference Friedman, Kremer, Miguel and Thornton2011) find that an increase in schooling reduces young women's acceptance of male violence against women in Kenya. Samarakoon and Parinduri (Reference Samarakoon and Parinduri2015) find no impact of education on women's decision-making authority in the household, asset ownership, or community participation in Indonesia. No paper we know of has analyzed the issue by investigating the impact of education on men and women simultaneously.
The remainder of the paper is organized as follows. Section 2 provides some background information on the educational system in Sierra Leone and the Free Primary Education (FPE) policy. Section 3 describes the data used in the analysis. Section 4 describes the empirical strategy; Section 5 presents the results. Section 6 discusses potential mechanisms. Section 7 presents the results for men and Section 8 provides extensions. Section 9 is the conclusion.
2. FREE PRIMARY EDUCATION (FPE) POLICY IN SIERRA LEONE
Sierra Leone is one of the poorest countries in the world. In 2010, per capita gross national income was $340. About 70% of females and half of males of the country are illiterate, and life expectancy at birth is 46 years. The population is about 5.5 million, and the fertility rate is five births per woman. The civil war, which broke out in March 1991 and lasted for about a decade, killed at least 50,000 individuals [World Bank (2007)].Footnote 7 Even before the civil war, children faced challenges in having access to education. For example, parents were responsible for the payment of tuition and fees, teaching and learning materials, and other expenses such as textbook purchases and exam fees. One consequence of the civil war was the destruction of infrastructure, including schools. By the end of the war, 1,270 primary schools had been destroyed, representing over 70% of all schools estimated to have been in existence before the conflict started [International Monetary Fund (2001)]. As a result, more than half of the primary schools had disappeared and this led to almost 70% of school-age children being out of school [International Monetary Fund (2001)].
In 2001, Sierra Leone started implementing a policy of providing FPE for all pupils in government-owned and government-assisted schools.Footnote 8 Financed largely by domestic public funds, donor contributions from international financial institutions, and foreign governments, the FPE was part of an overall universal primary education policy aimed at constructing and rehabilitating basic education facilities as well as providing free access to education. The specific goal was to increase basic education (comprising 6 years of primary education and 3 years of secondary education) by making it free and compulsory.
By far, this has been the largest education funding initiative to be undertaken in Sierra Leone since its independence in 1961. During the period from 2001 to 2005, the government committed almost 20% of its expenditures (equivalent to 4% of the GDP) toward education. Almost half of all education expenditure was devoted to primary education [UNDP (2007)]. Consequently, the number of schools grew rapidly and by the end of 2005, there were at least 4,300 primary schools, approximately 30% of which was newly constructed since 2001.
Sierra Leone spans about 72,000 square kilometers (about 28,000 square miles, or roughly the size of South Carolina). The country consisted of 14 districts in 2001 (see Figure 1).Footnote 9 With the implementation of the FPE program, each district received new funding for education, and the program was associated with a pronounced change in primary school enrollments. The number of students doubled from 0.6 million to 1.3 million children between 2001 and 2004, and the primary school gross enrollment rate exceeded 100% in 2004.Footnote 10 In 2004, Sierra Leone had the highest primary school enrollment rates among all Sub-Saharan countries [World Bank (2007)].
Figure 1. Sierra Leone divided by districts in 2004.
3. DATA
We use data from the 2008 SLDHS. The SLDHS is part of a broader program of demographic and health surveys conducted in other developing countries. The SLDHS is a cross-sectional nationally representative household survey and is the first of this kind in Sierra Leone.Footnote 11 Interviews for the SLDHS were conducted from April to June 2008.Footnote 12 The data contain information for over 10,000 individuals who were aged 15–49 years (females) and 15–59 years (males) in 2008. Using the age reported in 2008 and the reported birth dates, we identified individuals who were 8 to 11 years old in 2001. This particular group was exposed to the FPE policy.Footnote 13 At the start of the policy in 2001, those who were above the primary schooling age (e.g. those who were 12 or 13 years) were nevertheless allowed to enroll in school. Because these older children have returned to school, the primary school gross enrollment rate rose above 100% [World Bank (2007)]. Even in 2004, more than half of the children enrolled in the first grade of the primary school were older than 6 years. This was true of the other grade levels of the primary school as well [World Bank (2007)], which means that children who were older than 11 years of age in 2001 could have also benefitted from the program. Therefore, we chose as the control group those who were 15 to 21 when the implementation of the FPE started. These individuals were beyond the primary school age during the implementation of FPE and could not benefit from the policy. We provide more details on the identification strategy and its variations in the next section, including variations in the composition of the control group.
Because district level information on the amount of education funding is not available, we impute funding received by each of the 14 districts by multiplying the total government spending on the FPE program in 2004 (which was about 14 billion Le, or about $6 million at the prevailing exchange rate) with the share of each district in local government development grants. Information on the district-level share of local government development grants and matching grants was for fiscal year 2006 [see UNDP (2007), p.79], while total funding allotment for primary education came from the Government of Sierra Leone Budget and Statement of Economic and Financial Policies for the financial year 2004. The resource variable per district is the logarithm of funding received by the district per 100 teachers. Over the period 2004–2011, Local Government Development Grants were distributed based on a formula which was initially determined by population and the infrastructure damage index. Later, this distribution of grants was based on population and the poverty index.
We employ four variables as measures of women's empowerment. As discussed in the introduction, we adopt the definition of empowerment as one's ability to make correct decisions about the issues that impact one's own well-being. We consider affirmative answers to the following questions as being consistent with a woman's well-being: (i) whether a wife is justified in refusing to have sex when she is tired or not in the mood; (ii) whether a wife is justified in refusing to have sex with her husband if she knows he has a sexually transmitted disease (STD); (iii) whether FGM should be stopped as a custom. The fourth question asks the respondent “whether a husband is justified in hitting or beating his wife if she refuses to have sex with him”. If the woman's answer to this question is negative (disapproval of husband beating the wife), we consider her beliefs being consistent with her well-being.
We also gauge the preference toward fertility, measured by the number of desired children. An alternative measure would have been the number of children ever born, but because we focus on young individuals (who are 15 to 28 years old in 2008), their fertility is unlikely to be completed. Furthermore, desired fertility is conceptually different from completed fertility, where the latter may diverge from the former for a variety of reasons. Two additional variables gauge whether the woman uses modern contraceptive methods and whether she has ever been tested for AIDS.
Column (2) of Table 1 displays the means and standard deviations of the full sample of females who were either 8 to 11 years old or 15 to 21 years old in 2001. The information provided in the table pertains to 2008 when the individuals were aged 15 to 28 years. Of this group, those who were 8 to 11 years old in 2001 were exposed to the FPE. This is the “treatment” group, the descriptive statistics of which are provided in column (3). Those who were 15 to 21 years old in 2001 constitute the “control” group, listed in column (4).
Table 1. Summary statistics: 2008 SLDHS sample (individual level data) – females
Note: The descriptive statistics are calculated using sample weights from the SLDHS.
On average, women desire to have 4.7 children over their lifetime. The number of desired children for women in the treated group is 4.2, while it is 4.9 for those who are not exposed to the FPE program. The rate of modern contraceptive use is lower for those in the treatment group, but this is likely because the average age in this group is lower (16.5) in comparison to the average age in the control group (25.5). This is also the likely reason for the difference in the rate of having been tested for AIDS between the two groups.
The proportion of women who indicates that a wife is justified in refusing sex when she is tired, is higher among the control group. The same is true for the question that asks whether a wife is justified in refusing to have sex with her husband if he has a STD. It is somewhat puzzling that the proportion of affirmative answers is higher for the control group which has less education. On the other hand, the proportion of married women is larger in the control group. To the extent that married women have better information and experience on the issues raised by these two questions, it could be reasonable to have higher proportion of affirmative answers among the control group. The proportion of women who thinks that wife beating cannot be justified is higher in the treatment group. Similarly, the proportion of women who thinks that FGM should be discontinued is higher among the treatment group.
Although the individuals who are exposed to the FPE are younger than those who are not exposed, the former group has more schooling on average than the latter as expected (4.4 years vs. 2.1 years). The data set contains information on religious affiliation and ethnic background. Islam and Christianity are the two dominant religions in Sierra Leone; and Temne and Mende are the two largest ethnic groups. We also have information on whether the person resides in an urban or rural area. The wealth of the person is measured by dummy variables titled Wealth1, Wealth2, and so on, to indicate the wealth quintile to which the individual's household belongs. Total wealth is calculated using all household assets, including cars and television sets, as well as dwelling characteristics such as flooring material, type of drinking water source, and toilet facilities.Footnote 14 TV, Radio, and Fridge indicate whether the household of the individual has these consumer durables. These variables are correlated with wealth. In addition, TV and radio may have a direct impact on attitudes because they may enable the person to have access to information. For example, Jensen and Oster (Reference Jensen and Oster2009) show that the introduction of cable television into rural India has changed the reported acceptability of domestic violence toward women and the preference for sons. Chong and la Ferrara (Reference Chong and La Ferrara2009) report that access to a particular TV channel, which has monopoly over soap operas (which typically contain themes related to criticism of traditional values and female empowerment and emancipation) is associated with increased divorce rates.
The data set does not contain information on the region of birth, childhood place of residence, or the place of education. As a result, we cannot determine whether the individual has received her education in a given district and then moved and was surveyed in 2008 while living in a different district of the country. The basic assumption behind our empirical analyses is that individuals have received their education in the same district where they were interviewed in 2008. However, later in the paper, we identify those who have lived in the same residence continuously since the school years. Using this sub-sample of individuals, who have clearly not relocated since the elementary school age, we obtain similar results. We also demonstrate that the district migration rates do not differ by treatment status and the level of education funding in the district. Similarly, we show that exposure to war violence is not the driver of the results and that results do not change when we modify the age composition of the control group.
4. EMPIRICAL STRATEGY
We estimate the impact of the education policy on schooling acquired by women using the specification depicted by equation (1) below:
$$\begin{equation}
{S_{ijt}} = {\varphi _0} + {\varphi _1}\left( {{\rm{FPECohor}}{{\rm{t}}_i}*{R_j}} \right) + {{\bm{X}}_i}{\bm{\Psi }} + {\gamma _j} + {\delta _t} + {{\bm{\varepsilon }}_{ijt}},
\end{equation}$$
where Sijt stands for years of education (measured in 2008) of individual i, in district j, who was born in year t. FPE Cohort i is a dichotomous indicator which takes the value of one if the cohort of the individual was exposed to the FPE policy, and zero otherwise. As described in the introduction, because of the devastation of the infrastructure during the civil war, there was significant damage to school buildings. While the government could not increase the number of teachers at the end of the civil war in 2001, it started repairing damaged classrooms and building new schools. The intensity of this effort, which differs between districts, measured as log spending per 100 teachers in 2004, is represented by Rj .
The primary school age in Sierra Leone is 6 to 11 years. Those who were 6 years old in 2001 when the FPE policy started were 13 years old in 2008 when they answered the survey questions. However, the minimum age of those who were part of the survey in 2008 is 15 and these individuals were 8 years old in 2001. Therefore, we consider those individuals who were 8 to 11 years old in 2001 (who were 15 to 18 in 2008) as part of the treatment group; and FPE Cohort takes the value of one for these individuals. As mentioned in the data section, when the program started in 2001, those who were above the primary schooling age (older than 11 years) were nevertheless allowed to enroll in school, which means that children who were older than 11 years of age in 2001 could have also benefitted from the program. Therefore, the control group consists of those who were 15 to 21 in 2001. We also entertain alternative classifications of the control group which do not alter the results appreciably.
The vector Xi includes attributes of the individuals such as religion, ethnic background, marital status, employment status, residence in an urban area, the proportion of population in the age group of the person in the district, and indicators of wealth. The equation includes district fixed effects γj , year of birth fixed effects δt , and a standard white noise error term ε ijt .Footnote 15
To investigate the extent to which schooling had an impact on women's preferences, we estimate equation (2).
$$\begin{equation}
{Y_{ijt}} = {\alpha _0} + {\alpha _1}{S_{ijt}} + {{\bm{X}}_i}{\bm{\Phi }} + {\mu _j} + {\lambda _t} + {\tau _{ijt}},
\end{equation}$$
where Yijt represents the response of individual i who was born in year t, and who resides in district j to questions on health, sexual activity and husband-wife relations, shown in Table 1. S ijt stands for the level of schooling of the individual; μj and λt represent district fixed effects and year of birth fixed effects, respectively. Because unobservable characteristics of the person, represented by the error term τijt , have an impact on her preferences (Y) and because these unobservables could be correlated with her level of schooling (S), Equation (2) is estimated using instrumental variables, where schooling is instrumented by exposure to the FPE and intensity of the program in the district (FPECohorti *Rj ). The factors that may be correlated with schooling as well as with the outcomes analyzed in the paper (Y) may include unobservable attributes of the families and the communities.
A central assumption is that any district attributes that may impact the attitudes of people who reside in those districts are unrelated to the distribution of the funds to districts. Lack of correlations between funds received by districts and some district characteristics in 2004 supports this claim. For example, the biggest ethnic minorities in Sierra Leone are the Temne and the Mende. Correlations between district funding and the proportion by Temne and Mende in the district are 0.085 and −0.145, respectively, while the correlation between funding and population per square kilometer is 0.90. Similarly, the allocation of funds was not influenced by the leadership style of district leaders (e.g. whether or not district leaders have a more progressive social attitude).Footnote 16 If variation in access to schooling between districts due to the FPE program is related to funding of other local public goods, such as health facilities and clinics, then our instrument may impact attitudes through the influence of the provision of such public goods. However, we show in the paper that access to clinics and health services are not related to education.
5. RESULTS
Column (1) of Table 2 displays the result obtained from estimation of equation (1), which is the first-stage regression of the impact of the FPE program on schooling. Column (2) reports the same specification after dropping all covariates other than (FPECohort * R). The estimated impact of the program remains the same.Footnote 17 These specifications imply that doubling the education spending per teacher in a district increases young women's education by an average of 0.7 years.
Table 2. The impact of the FPE program on schooling – females
Notes: Standard errors in parentheses are clustered at the treatment-district level. Statistical levels of significance: * indicates p < 0.1, ** indicates p < 0.05, *** indicates p < 0.01. The sample comprises the FPE cohort (those born during the period 1990 to 1993) as the treatment and those females born during the period 1980 to 1985 as the control group. FPE Cohort*R is the product of the FPE Cohort dummy and the logarithm of district-level funding allocated for primary school education for every 100 teachers in 2004. Regressions are weighted by sample weights from the SLDHS.
In Table 3, we provide the results of the instrumental variables regressions where women's behavior regarding using modern contraceptives and testing for AIDS is analyzed, along with the analysis of their preferences regarding physical violence against them, and the issues that impact their health. The top panel of Table 3 reports the results of the regressions that do not include any control variables. The bottom panel displays the results obtained from the models that include all control variables. In the interest of space, the estimated coefficients of the control variables are not reported in the paper, but the full results of the bottom panel of Table 3 are provided in Appendix Table B.1. Regressions employ sampling weights, and standard errors are clustered in two ways. First, we classify the treatment cohort into two groups as younger and older (those who were 8–9 years old in 2001 and those who were 10 or 11 in 2001). The same was done for the control group (those in the control group who are younger than 18 years in 2001, and those who are older than 18 years). We clustered the standard errors by old/young in treatment/control and district. This specification provides 56 clusters (14 districts; and two age-groups each in treatment and control). The p-values of the estimated coefficients are provided in parentheses under the coefficients. Alternatively, we created clusters by treatment status and district. Because this formulation provides only 28 clusters, we also calculated the bootstrapped p-values for this method of clustering. These are reported in two rows of [brackets] under the coefficients.
Table 3. The impact of schooling on preferences – instrumental variables regressions – females
Notes: p-values of the estimated coefficients with 56 clusters are in (parentheses), p-values with 28 clusters and bootstrapped p-values with 28 clusters [in brackets] are shown in rows 2 and 3 below the coefficients. Regressions are weighted by sample weight from the SLDHS. Regressions include controls for religion, ethnicity, urban residence, marital status, whether respondent's household has a radio, whether respondent's household has a television, and a set of wealth status indicators and the percentage share of the respondent's age group.
Neither the point estimates nor their statistical significance change appreciably between the two panels of Table 3, although the bootstrapped standard errors are larger. Columns (1) and (2) of the bottom panel of Table 3 show that an additional year of education increases women's propensity to use modern contraception by about 11 percentage points and the propensity for being tested for AIDS by 10 percentage points.Footnote 18 These are substantial impacts as the proportion of women who use modern contraception in the sample is 17%, and 13% of the women in the sample has been tested for AIDS. Column (3) demonstrates that an additional year of schooling generates a decrease in the desired number of children by 0.30. The fertility rate in Sierra Leone is five children per woman and the desired number of children in our sample is 4.7. Thus, according to our estimate, the desired number of children of a woman who completes primary schooling and receives 6 years of education will be 1.8 fewer children than the sample average (or a total of three children). Whether this decrease in desired fertility would translate into an actual decline in the number of children born depends on the interplay between the woman's and her husband's preferences along with other social expectations and pressures.Footnote 19 As we demonstrate below, an increase in schooling for men does not translate into a significant change in men's preference for their desired number of children, and education does not impact two behavioral outcomes: women's age at first marriage and age at first birth.
Columns (4) to (7) present the results where we investigate the impact of education on women's attitudes regarding their own health and well-being. Columns (4) and (5) display the results of the models where the dependent variable measures whether the woman believes it is justified to refuse sex with the husband if the wife is tired (column 4), or if the husband has an STD (column 5). In both cases, an increase in schooling makes women more likely to answer in the affirmative, although the impact is far from being statistically significant in case of the “refusing sex when tired” question. An increase in education by 1 year increases a woman's propensity by about 12 percentage points to declare that a wife is justified to refuse sex when the husband has an STD. At the sample mean, this translates into an increase in the attitude by 25%.
Columns (6) and (7) present the results where the attitudes toward physical violence against women are analyzed. Column (6) shows that one additional year of schooling increases women's propensity to disapprove wife beating by about 8 percentage points. About 77% of all women in the sample believe that wife beating is not justified (see Table 1). Thus, the result reported in column (6) represents a sizable impact on the attitudes regarding acceptability of violence against women. Table 1 shows that about only 30% of women in the sample believe that FGM should be discontinued. An increase in education by 1 year, triggered by the FPE program, increases women's propensity to declare that FGM should be stopped by about 9 percentage points (column 7 of Table 3).Footnote 20
These results indicate sizable gains in attitudes that reflect women's empowerment. Similarly, large impacts are reported by Jensen and Oster (Reference Jensen and Oster2009) in a different domain. Their results showed that in India, access to cable TV reduces the number of situations in which women consider it acceptable for a man to beat his wife by 10%, and women's reported preference to have the next child be a boy by 22%.
Estimating these models (Equation 2) using OLS instead of instrumental variables generated results that are qualitatively similar to those reported in Table 3. More specifically, these OLS specifications produced coefficients for education that had the same signs as those reported in Table 3, and that were always significant at the 1% level or better, but the OLS coefficients were smaller in absolute value in comparison to the instrumental variables coefficients. These OLS results are reported in Appendix Table B.2.
In summary, the results demonstrate that an increase in schooling alters women's preferences in favor of protecting their own well-being. An increase in education makes women more likely to declare that a woman is justified in refusing sex when there is a health risk or when she is tired.Footnote 21 An increase in education also makes women more likely to disapprove acts of physical violence against them. Footnote 22 These results indicate that education changes women's preferences toward self-determination.
6. POTENTIAL MECHANISMS
It is possible that districts which received greater resources for education spending also received more funding for other public goods such as local health clinics. In this case, the use of modern contraception and getting tested for HIV can be impacted by increased access to health care facilities and health services. In other words, if education captures a general increase in local public good spending, then education would also impact access to local health facilities.Footnote 23 In the data set, we identified two questions that help gauge access to health services. The first question asks women if they gave birth in a government hospital, government health care center, government health post, or other public health facility (the main alternative is to give birth at home). Running the IV regression as in Table 3 showed that education had no impact on the propensity to give birth at a government facility since the end of the civil war (coef. of education = −0.096, se = 0.086, N = 983; mean of the dependent variable = 0.25 for the treatment group, = 0.27 for the control group). Second, we created a dichotomous indicator that takes the value of one if the last source of contraceptive was a government hospital, government health center, family planning clinic, mobile clinic, field worker, or other public source. Education had no impact on the propensity to use one of these government-provided sources for contraception (coef. of education = 0.001, se = 0.033, N = 253). These results indicate that education has no impact on using government health services, which implies that the impact of education on attitudes is unlikely to be driven by access to government provided health services.
6.1. Potential Migration
The data set does not allow us to identify the district in which the individuals received their education. There is, however, one question in the survey that asks individuals how long (in years) they have been living in their current residence. For each individual, we calculate the number of years since they completed their schooling. If the number of years in the current residence is greater or equal to the years since the highest grade completed, the individual has not moved to another district since completing schooling. We call such people “non-movers”. Note that the question about residence pertains to the location of the physical dwelling and not to the city or the district of residence. This means, for example, that a person who has completed her schooling 10 years ago but lived in the same residence continuously for only 9 years is considered a “mover”. In reality, this particular individual may have moved to another residence in the same district – thus, her place of current residence and her place of schooling may be the same. However, there is no way to distinguish between the cases in which the move was to a new residence and where the move was to a new district. Put differently, this crude measure forces us to classify anybody as a “mover” who has not lived in the current place of residence longer than the number of years since schooling is completed. As a result, we end up with those who definitely went to school in the same district as where they were surveyed in 2008, but we omit those who have moved to a different residence in the same district. Consequently, we lose more than 1,000 observations, some of which are non-movers who had to be classified as movers. Nevertheless, the regression results obtained from the sample of imputed “non-movers” revealed that the point estimates of schooling were smaller in comparison to those reported in Table 3, but the main inference did not change.Footnote 24
Our imputed district-level migration rates are higher in the control group (48%) in comparison to the treatment group (20%).Footnote 25 To investigate whether the district-level migration rates differ by treatment cohort and by the amount of education funding received by district, we classified each district into above-median resource or below-median resource district based on the education funding (R in equation 2) received by the district. We calculated the migration rates (the proportion of movers) in each district among those who are treated by the reform and those who are in the control group. Running a dif-in-dif regression on this sample revealed that the migration rates were not different between the treatment and control groups by the level of education funding.Footnote 26
6.2. Is Literacy a Mechanism?
In the SLDHS, literacy was evaluated based on the ability of the respondents to read a sentence placed in front of them. We created a dummy variable, titled Literate, which takes the value of one if the individual was able to read the whole sentence. We defined a second variable, Literate2, which identifies those who could read the entire sentence as well those who could read only parts of the sentence placed in front of them. This second variable determines those who are literate or semi-literate. As shown in Table 1, 26% of the women in our sample are literate, and 32% are literate or semi-literate. This is consistent with the 30% literacy rate for females who are 15 years of age and older in 2009, reported by the World Bank [UNESCO Institute for Statistics (2011)].Footnote 27
We investigated the relationship between reported schooling and literacy among those who are part of the control group (those who are not exposed to the FPE program) as well as for those who are treated by the program. Among those with no schooling, the literacy rate is essentially zero in both groups. Among the group of women who have 1 to 3 years of schooling, the literacy rate is zero if they are part of the control group, and 7.5% of this group is semi-literate or literate. Of women who have 1 to 3 years of education and who are part of the treatment cohort, the literacy rate is about 9%; and about 46% of this group is semi-literate or literate. The same pattern emerges among those with 4 to 6 years of schooling: the literacy (semi-literacy) rate is 29 (72)% in the treatment group and 6 (27)% in the control group. The fact that the literacy rate is extremely low among the control-group women who have some primary education (1 to 6 years) may be attributable to the fact that some of these women went to school during the civil war years and therefore it is likely that they did not receive high-quality instruction during those years. On the other hand, the FPE program started post-civil war. Therefore women, who are part of the treatment cohort and who went to school under the FPE program, are not impacted by the interruptions of the civil war.Footnote 28 It is remarkable that less than one-third (28.9%) of those who received 4 to 6 years of schooling under the FPE program can read a complete sentence.Footnote 29
To investigate more formally the relationship between education and literacy, we estimated regressions where literacy is explained by education and all control variables, and education is instrumented as before. The results have shown that an increase in education on this margin has no statistically significant impact on literacy in Sierra Leone.Footnote 30 This implies that literacy is not the mechanism through which education impacts the attitudes of women.
6.3. Exposure to War Violence
The civil war in Sierra Leone started in 1991 and ended in 2001. The schooling reform started at the end of the war in 2001. This means that people in the control group, who were 15 to 21 years old in 2001, were 5–11 years old when the war started in 1991. On the other hand, people in the treatment group were not yet born, or they were at most 1 year old when the war started. This regularity suggests that the results reported so far may not be due to the impact of education, but they may be because of exposure to war. In other words, the control group is exposed to war while the treatment group is not; and if exposure to violence during the war has “hardened” the older cohorts, we might be capturing the thoughts and attitudes of these hardened individuals in comparison to those who did not experience war violence.
The hypothesis that those who are exposed to violence during the civil war in Sierra Leone are “hardened” is not supported by data. Bellows and Miguel (Reference Bellows and Miguel2009) find that individuals who experience violence during the civil war in Sierra Leone display higher levels of political mobilization and engagement and they make more local public goods contributions in comparison to non-victims. If this finding carries over to attitudes toward violence, then women who were exposed to the civil war would be less tolerant of violence, rather than more tolerant.Footnote 31 Nevertheless, to address this concern, we used detailed data on war atrocities compiled by the Sierra Leone Truth and Reconciliation Commission (TRC). The TRC was initiated in 2000 by the Parliament of Sierra Leone under Section 6(1) of the TRC Act which states that the objective of the Commission, among other things is “to create an impartial record of violations and abuses of human rights and international humanitarian law related to the armed conflict in Sierra Leone, from the beginning of the conflict in 1991. . .” [Truth (2004a) pp. 3–4]. Creation of this commission is provided for under Article XXVI of the Lomé Peace Agreement of July 1999.Footnote 32
The general function of the Commission is to “investigate, report on the nature and extent of the human rights violations and abuses, and on the context in which these violations and abuses occurred” [Truth (2004c)]. In carrying out this function, the Commission provided an opportunity for both victims and perpetrators to share their experiences. Although not every possible victim or perpetrator was interviewed, information was collected from almost all of the chiefdoms,Footnote 33 the purpose of which was to record the experiences of the population, paying attention to specific groups such as women, children, and amputees. It is important to note that victims were not compensated for providing an account of their experiences to the Commission.
A total of 7,706 detailed statements made by Sierra Leoneans (living at home as well as refugees in Gambia, Guineas, and Nigeria) were collected by the Commission. Approximately, one-third of the statements were given by women while two-thirds were given by men. Of those women providing statements, roughly 79% were victims of violations while about 84% of the men were victims. Another 16% of the females indicated they were witnesses to violence against family members while 10% of the men witnessed violence against one or more family members. This detailed account of these individuals was compiled, processed, and presented in a statistical appendix to the Commission's report [see Truth (2004b)].
The commission identified a variety of atrocities during the war ranging from torture to rape, from amputation to assault/beating, from sexual slavery to forced labor. A total of 40,242 violent acts took place during the war. These incidents are identified by gender of victim, year, and district in which they took place. Using these data, we calculated the number of atrocities against females per 1,000 females in each district. Using this variable as “treatment”, we ran the basic models as in Table 3, instrumenting education by this treatment. In these regressions, the coefficient of education was never statistically different from zero.Footnote 34
We also investigated whether accounting for the possible direct impact of violence changes the results obtained from the basic model, reported in Table 3. This is done by adding to the models the variable that measures district level per capita war atrocities, described above, as an additional control variable. The estimated impact of education remained the same. As an alternative measure of exposure to violence, we used each person's age and location at the end of the war to determine exposure to years of violence since she was 2 years of old.Footnote 35 This measure of exposure to war violence did not alter the estimated coefficients of schooling in any meaningful way either.
Taken together, the results indicate that an increase in schooling changes the attitudes of women regarding the issues that matter for their well-being but that this effect is not due to women's improved reading ability due to increased schooling. The results are not due to access to health services or because of the impact of war violence either. Thus, the mechanism through which education impacts attitudes is unclear. Education's impact could be due to the exposure to teachers/lectures even though such exposure does not significantly enhance literacy. It could also be due to some other channel such as socialization experiences while in school; it could also be the result of multiple influences.
7. THE IMPACT ON MEN
In this section, we investigate the extent to which the increase in educational attainment due to the FPE program has impacted men's attitudes regarding women's well-being. Table 4 displays the descriptive statistics for men. A comparison between men and women reveals an interesting and surprising picture. For example, while 56% of women indicated that a wife would be justified in refusing sex when she is tired (see Table 1), 68% of men believe that such a refusal is justified. Similarly, the proportion who thinks that a wife is justified in refusing sex if the husband has an STD is higher among men than women. Along the same lines, the proportion of men who thinks that wife beating is justified is lower than the proportion of women who thinks the same (0.23 vs. 0.36), and the proportion of individuals who thinks that FGM should be discontinued is higher among men than women (0.38 vs. 0.30). These are surprising findings because one would expect that the rate of support for the statements in favor of women's well-being would be higher among women. This outcome, however, is not an artifact of the data we employ. It is consistent with a health report on Sierra Leone [Statistics Sierra Leone and ICF Macro (2009b)] and could be a reflection of a culture that fosters particular behavioral patterns for women and promotes a subservient attitude to men. Similarly, El Feki writes that in Egypt, where FGM is a custom, mothers, rather than fathers, decide on whether to circumcise their daughters and that men are traditionally not involved in this decision-making process (El Feki Reference El Feki2013).
Table 4. Summary statistics: 2008 SLDHS sample – males
Note: The descriptive statistics are calculated using sample weights from the SLDHS.
Table 5 presents the instrumental variables regressions for men. This table is the counterpart to the bottom panel of Table 3, which presented the same information for women. Table 5 shows that, as was the case for women, an additional year of schooling increases the likelihood to use modern contraceptive and to have been tested for AIDS among men. The point estimates are smaller than those estimated for women, but the baselines are lower for men. Thus, the impact of education on AIDS testing is similar between men and women, but the impact on modern contraceptive use for men is half of the impact on women (65% vs. 30%). On the other hand, an increase in education on this margin (the mean years of schooling is 5 years among men) has no impact on men's attitudes toward women's well-being.
Table 5. The impact of schooling on preferences – instrumental variables regressions – males
Notes: p-values of the estimated coefficients with 56 clusters are in (parentheses); p-values with 28 clusters and bootstrapped p-values with 28 clusters [in brackets] are shown in rows 2 and 3 below the coefficients. Regressions are weighted by sample weight from the SLDHS. Regressions include controls for religion, ethnicity, urban residence, marital status, whether respondent's household has a radio, whether respondent's household has a television and a set of wealth status indicators and the percentage share of the respondent's age group.
That education does not impact men's attitudes could be because average education of men was higher before the initiation of the program (average education of the control cohort of men is 4.7 years, while it is 2.1 years for women), and therefore the program might not be as impactful on men. On the other hand, there is other evidence (from a middle-income country) that an increase in education, due to an education reform, does not generate a change in religious beliefs and religiosity of men at a higher education margin (from 5 to 8 years), even though the same reform does change women's beliefs [Cesur and Mocan (Reference Cesur and Mocan2013)].
Although increased education changes women's attitudes in matters that impact women's well-being, the fact that an increase in education (on this margin) has no influence on men's preferences may suggest that the impact of education on changes in behavior may be unclear.
It should also be noted that in the sample the average age difference between husbands and wives is 11 years. Thus, husbands’ attitudes, who are for the most part, members of an older generation, are less likely to be impacted by the reform. This suggests that the change in women's attitudes, due to an increase in women's education, may not register a change in behavior such as age at marriage and age at child birth.
To investigate this point, we ran the main regressions by using three other outcome variables for women: whether they are married, age at first marriage and age when they gave birth to their first child.Footnote 36 The results show that an increase in women's education, due to the FPE policy, has no impact on the propensity of marriage (coef. of education = 0.04, se = 0.06). Similarly, education does not influence women's age at first marriage (coef. of education = 0.41, se = 0.29), or their age at first birth (coef. of education = 0.03, se = 0.48). These results support the hypothesis that the change in attitudes (on this margin) may not translate into actual behavioral changes.
8. EXTENSIONS
As mentioned earlier, those who were 8 to 11 years old in 2001 were exposed to the FPE policy. Individuals who were 15 to 21 years in 2001 constitute the “control group” because at the start of the implementation of the policy, those who were older than the primary schooling age (e.g. children who were 12 or 13 years) were still allowed to enroll in school. Nevertheless, we also estimated models by using those who were 12 to 16 years in 2001 as the control group. Because some individuals in this group have benefitted from the FPE, the estimated impact of education is expected to be weaker. The results show that this is in fact the case. The point estimates are smaller, and the impact of schooling is statistically significant only for contraceptive use, having been tested for AIDS, and reporting that a wife is justified in refusing sex if the husband has an STD. Finally, we experimented with yet another specification. Here, instead of dropping the individuals who were 12 to14 years old in 2001, and therefore who might or might not have benefitted from the FPE policy, we kept them in the sample but assigned them a value of 0.5. Put differently, in this specification those who are in the treatment group (ages 8 to 11 years in 2001) are assigned the value of one, those who are in the control group (15 to 21 years) are assigned a value of zero and the questionable middle group (ages 12 to 14 years) is assigned a value of 0.5. The point estimates were smaller than those reported in Table 3 as expected; but they were not substantially different.
The models include the employment status of the individual, but employment is a choice variable which could be impacted by education. Similarly, wealth could be endogenous as well. To investigate the sensitivity of the results, we re-estimated the models without employment and wealth. The results remained the same. Similarly, the results were insensitive to the exclusion of year-of-birth fixed effects.Footnote 37
9. CONCLUSION
The impact of schooling on earnings is well-documented and an increase in education improves health outcomes. These direct impacts of education on human capital and labor productivity are important ingredients for economic development. Increased education can produce other benefits for the individuals and the society by changing attitudes and preferences.Footnote 38 In this paper, we investigate whether an increase in schooling changes women's preferences regarding the issues that are directly related to women's well-being.
We use data from Sierra Leone, where a substantial education program was started in 2001. The program provided increased access to education for children who were primary-school age, but did not benefit children who were older. We exploit the variation in access to the program generated by date of birth and the variation in resources between various districts of the country. We analyze whether an increase in schooling, triggered by the program, had an impact on women's attitudes toward matters that impact women's health and on attitudes regarding violence against women.
The household survey includes health-related questions on the acceptability of attitudes such as whether a wife is justified in refusing to have sex if the husband has an STD, or whether a wife is justified in refusing sex when she is tired. It also includes questions gauging attitudes toward violence against women such as whether beating a wife is justified if she refused to have sex and whether the custom of FGM should be eradicated. Attitudes for violence against women are particularly important as the World Health Organization recently declared that more than one-third of all women around the world have been victims of physical or sexual violence [World Health Organization (2013)]. We also employ questions that involve health behaviors such as the use of modern contraceptives and whether the person has been tested for AIDS, as well as the desired number of children.
We find that an increase in education changes women's preferences. Specifically, an additional year of schooling makes women more likely to declare that a wife is justified in refusing sex when she is tired or when the husband has an STD. The same increase in schooling makes women more likely to disagree with the statement that wife beating is justified and more likely to declare that the practice of FGM should be stopped. These results indicate that education empowers women because an increase in education makes women more intolerant of practices that conflict with their well-being. An increase in education also reduces the number of desired children and increases the propensity to use modern contraception and to be tested for AIDS.
We can determine a respondent's literacy based on their ability to read a sentence as asked in the survey. We find that a switch from being illiterate to being literate is not the source of the change in attitudes. This is because an increase in education has no strong impact on women's literacy in Sierra Leone. The results are not driven by differential exposure to violence during the civil war of those who did and did not benefit from the education program. The results are not driven by access to health services either.
It is unclear through which particular mechanism education alters women's preferences. The impact could be due to a variety of factors ranging from the influence of teachers and peers as role models to socialization experiences in school. Nevertheless, an increase in education has a causal effect on preferences, much like the causal effect of education on health, although the exact mechanism remains also unclear in case of health outcomes [Mocan and Altindag (Reference Mocan and Altindag2014); Chou, Liu, Grossman and Joyce (Reference Chou, Liu, Grossman and Joyce2010); Cutler and Lleras-Muney (Reference Cutler and Lleras-Muney2010); Grossman (Reference Grossman2008)].
Even though education empowers women so that they can voice preferences that are aligned with their own well-being, whether or not this would translate into actual behavioral change is unclear. This is because we also find that an increase in men's education, triggered by the program, does not alter men's attitudes regarding women's well-being. If men resist change, it is unclear if women's well-being would be improved in the short-run.Footnote 39 In fact, we show that an increase in women's education on this margin has no impact on women's propensity to get married, their age at first marriage and age at first birth. At the same time, it is also possible that a change in women's preferences, detected in this paper, could facilitate a change in men's preferences over a period of time. Furthermore, much like the transmission of human capital from mothers to daughters reported by previous research, it is possible that the change in women's preferences could have an impact on their daughter's preferences. Regardless, these findings are potentially important for women's well-being to the extent that exogenous changes in the legal and political structure may not be effective in improving women's well-being quickly if women themselves don't believe that such changes are necessary.
APPENDIX A. PREFERENCE QUESTIONS AS APPEARING IN THE SLDHS
-
1. If you could go back to the time you did not have children and could choose exactly the number of children to have in your whole life, how many would that be? Or for those with no living children, the following question was asked: If you could choose exactly the number of children to have in your whole life, how many would that be?
-
2. Is a wife justified in refusing to have sex with her husband when she is tired or not in the mood?
-
3. Have you ever used any (any of the contraceptive methods)?
-
4. I don't want to know the results, but have you ever been tested to see if you have AIDS?
-
5. Husband and wives do not always agree in everything. If a wife knows her husband has a disease that she can get during sexual intercourse, is she justified in refusing to have sex with him?
-
6. Sometimes a husband is annoyed or angered by the things that his wife does. In your opinion, is a husband justified in hitting or beating his wife in the following situations: if she refuses to have sex with him? (Other situations were given to the respondent as well)
-
7. Do you think that this practice (female genital cutting) should be continued, or should it be stopped?
APPENDIX B.
Table B.1. The impact of schooling on preferences – instrumental variables regressions – females
Table B.2. The impact of schooling on preferences – ordinary least squares regressions – females
