1
Run the World (Girls): Experimental Evidence from Tanzania
“…We’re smart enough to make these millions Strong enough to bear the children
Then get back to business.” – Beyonce
Barriers to female empowerment: Evidence from a field experiment in Tanzania
Lars Ivar Oppedal Berge, Kjetil Bjorvatn, Tausi Kida, Linda Helgesson Sekei, Vincent Somville and Bertil Tungodden*
December, 2016
Abstract
Many young girls in developing countries experience early pregnancy and lifelong dependence upon family and partners, which may prevent them from reaching their full productive and social potential. In this paper, we consider two potential barriers to female empowerment: lack of reproductive health knowledge and lack of economic opportunities, and report from a randomized control field experiment of an empowerment program involving 3900 adolescent girls in 80 schools in rural Tanzania. One group was randomly offered a training program on reproductive health, a second group was offered a program on entrepreneurship while a third group was offered both training programs. The evidence from two rounds of follow‐up surveys shows that both the entrepreneurship program and the combined program have empowered the girls in the economic domain, while the impact of the reproductive health training is more muted. These findings suggest that entrepreneurship training is more important than health training in empowering the adolescent girls. Regarding the health domain, we do not find any evidence of a treatment effect of either training program. JEL classifications: I25, J13, J24, O12.
____________________
Berge: NHH Norwegian School of Economics, Bergen, and Chr. Michelsen Institute, Bergen e‐mail:
lars.ivar.berge@nhh.no. Bjorvatn: NHH Norwegian School of Economics, Bergen, e‐mail: kjetil.bjorvatn@nhh.no. Kida:
Economic and Social Research Foundation, Dar es Salaam, e‐mail: tausi.kida@gmail.com. Somville: Chr. Michelsen Institute, Bergen, e‐mail: vincent.somville@cmi.no. Sekei: Development Pioneer Consultants, Dar es Salaam. E‐mail:
linda@dpc‐tz.com Tungodden: NHH Norwegian School of Economics, Bergen and Chr. Michelsen Institute, Bergen, e‐mail: bertil.tungodden@nhh.no. We would like to thank FEMINA HIP for excellent cooperation throughout the
research project, in addition we would like to thank Dr. Katanta Simwanza and Dr Goodluck Charles for their invaluable assistance in developing the training material. A special thanks for excellent research assistance to Juda Lyamai. We have received financial support from Research Council of Norway, NHH Norwegian School of Economics and Chr. Michelsen Institute. The project has been administered by The Choice Lab.
3 1. Introduction
Many young girls in developing countries experience early pregnancy and lifelong dependence upon family and partners, which may prevent them from reaching their full productive and social potential. Adolescent fertility rates are, for example, more than four times higher in Sub‐Saharan Africa than in OECD (108 vs 25 births per 1000 women in the ages 15‐19, data.worldbank.org), with rates often being particularly high in rural areas. Teenage pregnancies are often unplanned and in many cases involve a relationship with an older male (Dupas, 2011b). They are associated with negative effects on a woman’s health outcomes, educational attainment (Rasul, 2008, Goldin and Katz, 2000) and future employment and economic opportunities (Bailey, 2006, Miller, 2010). Female empowerment may therefore have positive consequences both in the health domain and in the economic domain, and the two domains are clearly linked. Delayed pregnancies and improved health may enable young girls to exploit economic opportunities, while an improved economic situation may cause them to decide to postpone child‐bearing with corresponding health benefits.
Health training represents the standard approach to the problem, where the underlying idea is that teenage pregnancies and risky behavior reflect lack of relevant information and personal control. Evaluations of such programs have often found that knowledge and attitudes have changed, while biological impacts (on STIs or fertility) have been less clear.1
An alternative approach focuses on entrepreneurship training, the assumption being that the underlying problem is lack of economic opportunities. Focusing mainly on
1 Gallant and Maticka‐Tyndale (2004), in an overview article on the impact of school based HIV prevention programs, find that it may be possible to influence knowledge and attitudes, but that inducing changes in sexual behavior is much more difficult. Similarly, Ross et al. (2007) studies both behavioral and biological impacts of an multicomponent adolescent sexual health intervention in Tanzania. They find that knowledge, attitudes and self reported behavior had improved, but found much more muted biological impacts when studying the occurrence of sexual transmittable diseases. Dupas (2011) finds that an information program on the risks of intergenerational sex leads to a significant reduction in childbearing among adolescent girls, and in particular those involving an older man. In contrast, she finds no effect of the official HIV/AIDS curriculum which emphasizes abstinence.
business outcomes, impact evaluations typically find that the effects of such programs are muted. 2
In the present paper we report from an empowerment program offered to adolescent girls in rural Tanzania. The program involved two training modules, one on reproductive health and the other on entrepreneurship. We evaluate the impact of the modules separately and in combination, allowing us to investigate the relative effectiveness of the two approaches, and whether there is a complementarity between them. Moreover, we conduct both a short‐term and a long‐term follow‐up study of the intervention, enabling us to shed light on the sustainability of such interventions.
The main finding of the paper is that entrepreneurship training is indeed the more effective approach to female empowerment in the economic domain. We find that girls who have received entrepreneurship training 16‐18 months after the intervention are much more likely to be involved in business activities and are happier with their economic situation. In contrast, we find very little evidence of the reproductive health training, alone or in combination with the entrepreneurship training, having an impact in the economic domain. None of the interventions have an impact in the health domain, but this null result may partly reflect that it takes more time for health changes to be observable.3 Finally, we also show that some of the positive short‐term effects of the training on gender equality perceptions do not survive in the long term, which may reflect a tension between local social gender norms and the messages of the female empowerment program. The short‐term positive effects on locus of control, however, remain and are also present in the long‐term follow‐up.
2 One exception is Bjorvatn et al. (2012), who studies the impact of an edutainment for entrepreneurship television program for youth in Tanzania, which aimed at inspiring and informing young people, in particular females, about entrepreneurship. They find that the show inspired the viewers to think about business as a career opportunity, and also caused an increase in business startups, in particular among females. In contrast, business training programs targeting adult entrepreneurs often find weak effects on females, see Karlan and Valdivia (2011), Berge et al. (2015), and Giné and Mansuri (2014).
3 We plan to collect objective health data in mid‐2015 (two years after the end of the intervention), including data on pregnancies and STIs.
5
Another notable feature of our study is that we pre‐committed our analysis by registering two detailed pre‐plans. The plans were filed at the AEA RCT register at www.socialscienceregistry.org, and the plans lay out exact hypothesis and specifications to be tested; including which outcomes to study and in which dimensions to look for heterogeneous impacts.
In the next section, we discuss related literature. In section 3 we present the randomization procedures, the participants and the interventions in detail. Section 4 presents the data and estimation methods used, and section 5 presents results. We discuss our findings in section 7, while we conclude in section 8.
2. Related literature
Our paper is related to the literature evaluating different programs addressing either young females economic and/or reproductive health challenges. Most of these studies evaluate single‐pronged programs aiming to influence either the economic situation, or the reproductive health domain. However, many of the evaluated single‐pronged programs in developing countries have, at best, yielded mixed results.
In the health domain, Gallant and Maticka‐Tyndale (2004), in an overview article studying the impact of school based HIV prevention programs, found that it may be possible to influence knowledge and attitudes, but that inducing changes in sexual behavior is much more difficult.4 One notable study is Ross et al (2007), studying both behavioral and biological impacts of an multicomponent adolescent sexual health intervention in Tanzania. They found that knowledge, attitudes and self reported behavior had improved, but found much more muted biological impacts when studying the occurrence of sexual transmittable diseases. Another notable and positive study is Dupas (2011a), who evaluated an intervention in Kenya informing teenage
4 Similar conclusions are drawn by McCoy et al (2010), who find that most behavioral interventions are not successful in reducing risky sexual behavior.
girls that they faced a higher risk of being infected by HIV if they had sexual relationships with older rather than younger men. The information program caused the rate of childbearing in the target group to decrease by 28% within a year, and the rate of childbearing with men at least five years older to decrease by 61%, suggesting that teenagers are responsive to risk information.
In the economic‐opportunity domain, there are surprisingly few studies on young females and entrepreneurship in a developing country context. Card et al (2011), studying the impact of a job training program for low income youth in the Dominican Republic, find no impacts, although previous non‐experimental evidence have been more positive. Bjorvatn et al (2012) studies the impact of the edutainment for entrepreneurship television program Ruka Juu (Jump Up) in Tanzania, which aimed at inspiring and informing young people, in particular females, about entrepreneurship. The study, which involved around 2100 secondary students in Dar es Salaam, found that Ruka Juu had an important empowering effect on the female students, where the female contestants in the show were perceived as role models showing that it was possible for young females to become successful business women in Tanzania.
However, our study evaluates a program including both two single‐pronged programs and a joint program, thereby complementing a growing body of research studying the inter‐linkages between economic opportunities and reproductive health.
The study that comes closest to ours is Bandiera et al. (2014), who evaluated an intervention conducted by the micro‐finance institution BRAC in Uganda, where adolescent girls were offered “life skills” and vocational training. In particular, they found that the intervention increased by 72% the probability that the girls were involved in income generating activities, and teen pregnancies were reduced by 26%.
Their findings lend support to the hypothesis that providing economic opportunities and information to adolescent girls can fundamentally change both their health and their economic behavior. But their study cannot offer any insights into the relative
7
importance of a reproductive health campaign compared to an expansion of the economic opportunities of young girls.
Closely related, also studying a two‐pronged intervention, is Duflo et al (2014), investigating the impact of a school based HIV‐prevention program and/or school subsidies in Kenya. While each of the single‐pronged programs, in particular the cash transfers, was found to have beneficial impacts, only the combined program reduces sexual transmittable diseases, indicating important inter‐linkages between economic opportunities and fertility decisions. Another recent study that aims to reduce prevent HIV and STIs in East Africa, is de Walque et al (2014). They study whether conditional cash transfers rewarding safe sex can change behavior of young male and females in Tanzania. They find that incentives may work and can improve health outcomes, but that they only have lasting effects beyond the program ends for boys.
Furthermore, our study also relates to the broader literature on human capital, entrepreneurship and gender. A general lesson from this literature is that it is often difficult to raise female entrepreneurial income. For instance Karlan and Valdivia (2011) find very modest impacts of entrepreneurship training on female microfinance clients. However, Berge et al (2015) find no impacts on female entrepreneurs of neither business training, a cash grant, or both, suggesting that many females face other more binding constraints than lack of human and financial capital. Such constraints can not only be “internal” psychological constraints, such as unwillingness to compete or take risks, but may also be “external” constraints, such as household obligations. Although few subjects in these studies are far less than 20 years old, a hypothesis may at least be that human capital interventions may be more effective the younger the females are, as it may be difficult to implement new ideas if childbearing and marriage have already started.
Finally, our study relates to the general literature on health and information campaigns. A well‐known case is the success of Egypt’s National Control of Diarrheal Disease Project, in which infant diarrheal deaths decreased by 82% between 1982 and
1987, partly attributed to the spread of information about oral rehydration therapy (Levine et al., 2004). In the same vein, Fitzsimons et al. (2012) have recently evaluated a randomized intervention in Malawi which provided information on infant nutrition and health to mothers, where they find that the information program caused a significant improvement in the consumption of protein‐rich foods by children.
Another relevant example is the study by Madajewicz et al. (2007), which showed that an information campaign in Bangladesh, where households were informed about the concentration of arsenic in their wells, caused a significant increase in the use of safer wells. Similarly, Chaudhuri (2009) showed that households in India decided to adopt purification technologies once they were informed that their drinking water was infected.
The effect of an information campaign may, however, crucially depend on the target group and the type of information communicated. For example, Kremer and Miguel (2007) did not find any effect of a health‐education campaign in Kenya aiming at reducing intestinal worm infections, which may be due to the children, and not the parents, being the primary target group of the intervention (Dupas, 2011b).
Our paper adds to the existing literature on female empowerment by investigating not only the separate impacts of entrepreneurship training and reproductive health training, but also the combination of these programs. This enables us to investigate important inter‐linkages and complementarities between economic opportunities and reproductive health as well as the relative importance of the two treatments; which are important both from a theoretical and policy perspective.
3. Randomization, Participants and Interventions 3.1. Randomization procedure
During the baseline in April 2013, we sampled 80 public schools with at least 20 girls in Form IV in the regions Tabora, Singida, Morogoro and Dodoma.5
5 We considered schools that were on Femina Hips lists, receiving their free magazines related to female empowerment and / or entrepreneurship, but we excluded schools that had already established Femina‐
9
Among these schools, we randomly allocated 20 to each treatment arm, and 20 schools to the control group. That is 20 schools to entrepreneurship training (“build your life”), 20 schools to health training (“protect your life”), 20 schools to receive both entrepreneurship and health training, and 20 schools to control. The randomization was blocked by school‐size (below or above 40 girls in Form IV) and by region.6
During the baseline we reached 3 483 girls. Table 1 shows that our randomization ensured similar treatment groups; they are only significantly different on one baseline characteristic. Although at later survey rounds we also included girls that had been absent during the baseline survey, we define our main sample (and attrition from it) to be girls that were present and sampled during the baseline study; since the selection of girls into the follow up surveys may be different in the control and treatment.7
The sample size and number of schools was powered to detect changes in pregnancy rate, which is the most demanding variable to measure and therefore serves as a conservative estimate for the other variables of interest.8
3.2. The participants
Together with our partner Femina HIP, we decided to focus on girls in semi‐urban areas who were in their last year of secondary school (Form IV in Tanzania). We
clubs, which dealt with similar topics as the training programs. In addition, we did not consider private schools.
6 We followed David McKenzie and Miriam Bruhn’s recommendations in dealing with the uneven numbers in some strata and when doing the randomization we the STATA code they shared on the World Bank’s “Development Impact” blog on the 11th of June 2011. We did not make any re‐draws in order to ensure optimal balance.
7 In appendix 11‐13, we report all tables with ”full” samples, including girls not present in the baseline.
8 Taking into account the effect of clustering and the fact that we have three different treatment groups in addition to a control group, we have with the planned sample a power of 80% (with a 5% confidence level) to detect a decrease in pregnancy rate from 25% to 20% (using the approach of Hayes and Moulton, 2009).
believe that this target group has a lot to benefit from the interventions. Few girls in rural/semi‐rural secondary schools are normally able to continue schooling. They will have to consider other opportunities, including opening a small scale business, doing small scale farming, or getting a family and /or getting pregnant.
During baseline, most of them were in the age interval 16‐18, which is where we observe a sharp increase in fertility. In addition to the detailed baseline survey, we also surveyed the headmaster of each school, who provided us with detailed information about school characteristics.
From Table 1, we also note that most girls come from households that are not woman‐
headed and do not own their own businesses, reflecting that farming is the most common activity in these districts. We see that only 14% of the girls with one semester left at form IV (and most likely school) had any plans of starting a business in the near future.
Furthermore, from the control group in the short term follow up, in table 2 (column 2), we see that health knowledge seems to better than entrepreneurship knowledge.
However, from G1 we note that 28% of the girls accepts that a male beats his wife, and that as many as 31% of the girls in the control group admits to have had unprotected sex. From table 3, we note in particular that 19% of the girls in the control are involved in economic activities around one year after they quit school, and that 6% reports to be have been starting childbearing. Moreover, we note that the average response of the control girls to the question about general happiness is 3.8, on a scale 1‐5, indicating they are fairly happy with their lives. Finally, girls in the control group report a weekly income of 11 069 TZS or 6.20 USD
3.3. The interventions
After the baseline survey, in June 2013, one or two teachers per (treated) school attended a one‐week instructor session organized by Femina HIP (two weeks for the teachers involved in the combined treatment). After these instructor sessions, in July
11
– September 2013, the teachers implemented the training sessions (treatments) with all the Form IV girls of their school. During the baseline survey, the girls were asked to name two teachers they trusted and could talk with, and based on these recommendations, the headmaster at each school selected the teachers. Using internal teachers instead of external consultants is important for the scalability of the interventions.
Both the entrepreneurship and the health treatments had 8 weekly training sessions of 1.5 to 2 hours, 1 session per week, while those who got the both treatments received 16 training sessions per week. Both treatments were offered in a classroom setting, and most schools finished the treatments in 8‐10 weeks.
On average, the girls followed 6.88 sessions in the health treatment and 6.65 sessions in the entrepreneurship treatment. In the combined treatment, they attended on average 7.08 health sessions and 7.04 entrepreneurship sessions. The girls receiving training also evaluated the course, and where in general very happy with both courses,. For instance, in both trainings more than 98% of the students either agreed or strongly agreed that (i) the training was very useful for them, that (ii) the training provided them with new information and that (iii) the training was very well organized.9
Femina HIP, a leading NGO on reproductive health in Tanzania, together with the research team, designed both the entrepreneurship program and the reproductive health training. Both treatments aimed at empowering the girls, but in different domains.
The aim of the entrepreneurship training was to economically empower the girls, inspiring and enabling them to “build your life”, providing the girls with knowledge on how to establish and run their own business. Topics included customer care,
9 The other possible answers being strongly disagree, disagree and neither agree nor disagree.
marketing, record keeping, pricing of products, personal finance, and sessions aiming at improving entrepreneurial mindset and self‐confidence.
The aim of the reproductive health‐training was to enable the girls to take control of their own body and health, or to “protect your life”. The training provided practical and objective information about reproductive health and gender empowerment, including information and guidance about contraception and the consequences of risky sexual behavior, as well as making the girls aware of basic gender equality rights.
In both treatments all the girls received their own copy of either a “build your life” or a “protect your life” booklet/magazine, and in many cases this was one of very few school related books the girls had. Those who received both the training programs, received both magazines.10
10 For a complete list of topics in the training program, see appendix 1.
13 4. Data and Estimation Methods
4.1. Short‐ and long‐term data
The immediate impact of the treatments was evaluated in a short term follow up survey conducted in October 2013, a few weeks after the training programs ended.
The purpose of the short term follow‐up was to better understand the mechanisms that may influence long‐term outcomes, and we measured changes in knowledge, behavior, views on gender equality and empowerment. In all dimensions, we measured changes in both the entrepreneurial and the reproductive health domains, as stated in our short‐term pre‐analysis plan.11
In both the baseline and the short‐term survey, we visited all 80 schools, and held common survey sessions in‐class. Knowledge was measured by the girls answering incentivized multiple choice questions about reproductive health and entrepreneurship, while behavior was captured in the health domain by a binary variable equal to one if the girl reported to not having sex or using a condom when she has sex. In the entrepreneurial domain, changes in behavioral was captured by stated plans of opening a business. Furthermore, gender equality was captured by questions revealing the girls acceptance of gender based violence and acceptance of a wife earning more than her husband.
Finally, empowerment was measured in two ways; in their willingness to compete against boys, and by answering a set of locus of control questions. Before deciding to compete or not, all girls solved a set of incentivized (piece rate) math questions. After the first set of math questions, they had to state whether they thought they were equally good, worse or better than the boys in their class (incentivized), before they were asked whether to compete (and get a higher rate if they performed better than
11 The pre‐analysis plans are registered with the AEA RCT Register at www.socialscienceregistry.org.
The short term plan has ID number AEARCTR‐0000150 and the long term plan has ID number AEARCTR‐0000511. Appendix 5 specifies which analyses that where not pre‐specified.
the average of the boys) or to work another round with a the same piece rate as in round 1.
Long term data was collected in September – October 2014. Since the large majority of all the girls were expected to have quit school, we interviewed the girls by telephone.
The aim of this survey was in particular to capture behavioral and welfare changes. In addition, we again measured changes in the gender equality and empowerment dimension, in both the entrepreneurial and reproductive health domain.
In particular, in the behavioral dimension we asked about current/former pregnancies, and whether they had started a business, and if so, the sales from this business. In addition, we also measured their patience, as the girls had to choose whether or not to get the telephone voucher (as a compensation for the interview) today, and receive 2000 TZS, or to wait one month and receive 5000 TZS. In the gender‐equality dimension, we asked the same questions about acceptance of gender based violence and a wife’s higher earnings, while we in the empowerment dimension asked whether they felt in control of their life and if they felt being useful.
Finally, and furthermost, we measure changes in the girls welfare by asking about their general, economic and health happiness, as well as if they were healthy (able to work the whole last week), and what their income (from all sources) was.
Attrition rates was low in both survey rounds. In the short term survey we reached 83% percent of the girls. In the long term study, we in the ordinary interview‐phase 82%; of the remaining 18%, we randomly draw 30% (or 188 girls) that we intensively tracked. Of the intensively tracked girls, we managed to reach 77%, giving us an effective tracking rate of 96%.
15
In table 4, we regress survey response on treatment status. While we see that girls in the health treatment were slightly less likely to be reached in the short term survey, attrition is not predicted by treatment status in the long term survey.12
4.2. Empirical Strategy
We estimate the intention to treat estimators (ITT) for each individual outcome Yi by including a dummy for each treatment group;
(1)
Where is a dummy for receiving entrepreneurship training (only), a dummy for receiving reproductive health training (only), and a dummy for receiving both treatments. is a vector of covariates from the baseline. In appendix we also report all regressions without covariates.13 For long term outcomes, we use weighted regression, were each observation from the intensive‐tracking phase is weighted equal to the inverse probability of being included in the “intensive‐tracking” sample.
We cluster standard errors at the school‐level. When testing whether the impact of the cross treatment is equal to the two separate treatments, we do not have a clear a priori hypothesis about the sign, and thus we will use a two‐sided test to test whether
.
Finally, in the appendix, we will also report heterogeneity in treatment effects. We will then introduce interaction terms, where the three treatment arms will be interacted with the relevant variable. When checking for heterogeneous effects, the equation becomes:
12 Given the slight imbalance in short‐term attrition, we present lower lee‐bounds and confidence intervals for all short term estimates in appendix 14.
13 All covariates are defined in appendix 4
1
2
3
(2) ∗ ∗
∗
Where W stands for the variable defining the heterogeneous effects of interest. We measure heterogeneous effects along four dimensions; to what extent the school is remotely located, the wealth of the family of the girl, cognitive abilities of the girls, and the girls’ age.
5. Results
In this section, we study the treatment effects of the interventions on knowledge, behavior, gender equality, empowerment and welfare, using data from the two follow up survey.
5.1. Short term evidence
Table 5 presents short term impacts on knowledge, behavior, gender equality and empowerment. Column 2 shows that girls receiving entrepreneurship training alone or together with the reproductive health training, perform significantly better on a set of 7 multiple choice questions on business and entrepreneurship, getting 0.16 and 0.32 standard deviations more correct answers than the control group. However, as we see from column 1, there is no similar impact on health knowledge from the reproductive health training or the cross‐treatment. In particular, we note that the performance of the control group is much better on the health questions (4,8 correct questions out of seven) than on business questions (1,9 correct questions out of five), indicating either that the health questions we asked where too easy, or that they simply know these issues very well. From the p‐values of equality, we also note that cross‐treated answer significantly better the business questions than those who only received entrepreneurship training (p=0.022), even though those receiving only health training answer very similarly to the control group, at least indicating that the absorptive capacity and the quality of training was not reduced for those receiving both training
17
programs. Even though it was not significant, we also note that the coefficient on the cross treatment in column 1 is much larger than the coefficient on health.
Column 3 and 4 show impacts on short term behavior. In column 3 the outcome is an indicator variable equal to one if the girl reports to have had sexually safe behavior;
that is if she either don’t have had sex, or if she uses a condom when she has sex. 69%
of the girls in the control group report to abstain or be using condoms. The treatments do not have a significant impact on this outcome. Column 4 show the impact on behavior in the economic domain, and we see that while only 15% in the control group have any plans of opening a business once the school year is completed, 53%
(0.15+0.38) of those receiving entrepreneurship training and 57% (0.15+0.42) of those receiving both training programs intend to do so, clearly indicating that the entrepreneurship training program have inspired them. Furthermore, we note the similar coefficients for those receiving either both training programs or the entrepreneurship program only, indicating that the additional health‐training is not a key element for making business plans.
Column 5 and 6 show the impact on two measures capturing views on gender equality. In column 5, the outcome is the percentage of time the girl answers “yes” on the question “do you agree that a husband is justified in hitting or beating his wife if..”. In the control group, the girls answer “yes” in 28% of the cases, while those receiving health training are much less inclined to do so. In column 6, the indicator of gender equality is the girls acceptance of women’s higher earnings. We ask whether the girls agree that “it is acceptable to me that a wife earns more money than her husband”, where they answer on a scale from 1 to 5, where 1 is strongly disagrees and 5 is “strongly agrees”. In the control group, the average response is 3.8, and we see that all treatments significantly influences the girls acceptance of women earning more, indicating that the health training also influenced views on gender equality in the “economic domain”. We note though, that the biggest impacts are among those receiving both treatments or entrepreneurship training, with identical estimates (0.28 SDs).
Column 7 and 8 show that the treatments had more muted impacts on our experimental and psychological measures of empowerment. In column 7 the outcome is an indicator variable of whether they choose to compete against the boys or not.
However, while 33% of the girls in the control group choose to compete against the boys, rates are not significantly different among the treated. This is also true if we also control for initial performance, or for measures of willingness to take risk – indicating that preferences for competing is hard to influence. Our second measure of empowerment in column 8 is seemingly also difficult to “move”. The outcome is an index of seven questions intending to capture different aspects of whether the girls feel in control of their lives and have positive attitudes towards themselves. Column 8 indicate that the health training caused a small positive increase.14
In sum, the findings from the short term survey indicate that the different treatments have worked more or less as intended, and findings are also mostly in line with what one could expect; the entrepreneurship training influenced outcomes mostly in the
“economic domain”, the reproductive health training influenced outcomes mostly (albeit less) in the health domain, while we see that the cross‐treatment influenced outcomes in both domains.
14 Lee‐bounds for the health‐treatment are presented in appendix 14, and we note that confidence intervals for the two gender equality outcomes always are negative (wife beating) and positive (wife earner), whlie the other outcomes are more fragile, given the little impacts found in the main specifications.
19 5.2. Long term evidence
Table 6 present long term impacts on behavior, gender equality and empowerment. In column 1, the outcome is whether the girl has started childbearing / given birth, and we see that only 5.6% of the girls in the control group report to have done so.15 We see no reduction in childbearing among the treated girls. However, when looking at behavior in the economic domain, impacts are substantial. Those who received entrepreneurship training are 12 percentage points more likely to be involved in a business activity (having own income from farming or business), while those receiving both treatments are 18 percentage points more likely to involved in such income generating activities. These results are also reflected when looking at the impact on the girls sales’ (from farming or business). The outcome is the hyperbolic sine transformation of the sales, and we can therefore approximate the coefficients to percentage change in sales.16 We see that girls receiving entrepreneurship training or the combined treatment have sales more than double the size of the girls in the control group, while we also note a positive and large, but insignificant effect on the girls receiving the health‐training. From the p‐values of equality, we note that while the impact of the health training is significantly different to the other treatments, the impact of the entrepreneurship training and the combined training is not significantly different, even though the difference in the size of the coefficient is big (1.43 vs 2.03).
The findings on business activity and sales are also in line with the short term impacts on business plans; although we note that treatment effect is slightly smaller in the long term when looking at the business‐dummy (column 2) compared with the plans in the short term (table 5, column 4), reflecting that not all business plans were put into practice.
15 ”Giving birth” is typically regarded to be a question that most respondent will remember and answer correctly, while it is likely more error in responses regarding current/pas pregnancies/abortions. We plan to collect objective health data on the girls in the summer of 2015, with aim of also getting biological markers on current pregnancies and sexually transmittable diseases.
16 The hyperbolic sine‐transformation share most features with log‐transformation, but it is also defined for zeros and negative numbers (Burbidge et al , 1988).
Moreover, in order to measure underlying time preferences, our final measure of behavior is from an incentivized choice where the girls could either receive a telephone voucher of 2000 TZS immediately, or 5000 TZS in one month. However, as the estimates in column 4 show, there were no significant impacts on the treated girls’
choices.
Column 5 and 6 present the long term impact on gender equality, and we see that positive treatment impacts found in the short term have not survived.17
In table 6, column 7 and 8 we present treatment‐impacts on the extent the girls feel in control of their life, and whether they feel useless or not, which may be a good psychological measures of empowerment. The results in column 7 indicate that girls receiving the health training, or both training programs, to a lesser extent agree that
“I have little control about things that happen to me”. However, as we see from column 8, the treatments did not significantly influence the girls answers to the question “I certainly feel useless at times”. Both these questions were included in the empowerment‐index in the short term, and where both positively influenced by the treatments then. Seemingly, it is therefore not only in the gender equality domain that positive short term impacts on attitudes fade away in the long‐term.
Table 7 presents the impacts on the girls’ welfare. We asked the girls about their general happiness (using a scale 1‐5), as well as how happy they were with their health and economic situation. Somehow in line with the other treatment impacts in table 5 and 6, we see that the impact on “health‐happiness” are modest, perhaps also explained by a higher general happiness level in this domain, probably reflecting that the girls are young, and well as the fact that relatively few have started childbearing.
17 Note that in the “wife beating” question, we only asked about acceptance of violence in the case “she goes out without telling him”, which was the sub‐question where we found the highest impact in the short term. The “wife earner”‐ outcome were unchanged, however.
21
In the economic domain, however, we see that all treatments have positive impacts, also the reproductive health training, which is in line with the positive albeit insignificant impacts of the health training on sales in table 6 (column 3). However, when looking at general happiness in column 1, only the impact on the cross‐treated is significant, with these girls being 0.25 standard deviations more happy. Notably, this is the only long‐term outcome where the impact of the entrepreneurship training is significantly different (and smaller) from the combined treatment.
Furthermore, in column 4 the outcome is the number of days the girls were too sick to work the previous week, which is an objective measure of welfare and wellbeing. We note the very low levels of sickness among the control girls, with the average control girl only being sick 0.2 days the previous week, and we find no treatment impact in this dimension.,
Our final outcome is income (column 5), which is the sum of all income the girls gets in a normal week. As with the sales‐outcome, we use the inverse hyperbolic sine transformation of the income and we observe that the coefficient on the combined treatments are large (0.47), albeit insignificant. The simple reason for the income estimates not being influenced as much as the sales figures, is that both sales and costs (investments and inputs purchases) have increased,
To sum up the long term evidence – we see that the entrepreneurship training, either alone or together with the health training, have had significant and large impacts on business activities and sales, while impacts in the health domain are much more muted. While we observed in the short term that the combined treatment moved more dimensions than the entrepreneurship training, this picture is less clear in the long term, with there only being a significant difference between these two treatment groups in their answers on their general happiness.
5.3. Heterogeneous effects.
In tables 8 we report heterogeneous effects on selected outcomes from the short and long term surveys, while we show heterogeneous impacts for all outcomes in appendix 7. First, we note that entrepreneurship knowledge was consistently improved among most groups receiving the entrepreneurship training, while health knowledge only were improved among girls from “wealthy” household, receiving the combined treatment.
Next, we see that willingness to compete was only increased among girls in remote areas receiving the combined treatment, while there are no impacts among other groups.
From the long term, we see that economic activity and sales have improved consistently in all groups for those having received entrepreneurship training or the combined interventions, and we also see that the health training have increased sales in remote areas, among non‐wealthy girls, and among young girls, indicating that the this treatment also empowered certain girls in the economic domain.
Furthermore, we see that general happiness have consistently improved among girls receiving the combined treatment, and that it also have improved among non‐remote, wealthy, and low cognition girls receiving the health training. Finally, we see that income have increased among non‐remote and young girls.
23 7. Concluding remarks
What is the more effective strategy of female empowerment when targeting young girls in developing countries; providing reproductive health information and knowledge about the risks associated with unprotected sex and early child bearing; or providing entrepreneurship knowledge and skills that can improve their economic outlook?
Based on two rounds of follow‐up surveys, both the entrepreneurship program and the combined program have empowered the girls in the economic domain, while the impact of the health training is more muted. Thus, our findings at this point suggest that entrepreneurship is more important than health training as a catalyst of change for adolescent girls.
Young women’s low economic development is therefore explained by a lack of economic opportunities rather than a lack of information about reproductive health and gender equality. This a very important conclusion from a policy perspective. It indicates that promoting entrepreneurship and self‐employment among the younger generations is a more effective tool to fight poverty than family planning programs.
Whether the two types of training can also have larger effects on health outcomes in the longer run is still an open question that should be answered in future research.
References
Bailey, M. J. 2006. More power to the pill: the impact of contraceptive freedom on womenʹs life cycle labor supply, Quarterly Journal of Economics, 121: 289‐320.
Bandiera, O., Buehren, N., Burgess, R., Goldstein, M., Gulesci, S., Rasul, I., and Sulaiman, M. 2014. Women’s empowerment in action: Evidence from a randomized control trial in Africa, London School of Economics, Working Paper.
Berge L. I. O., Bjorvatn, K. Tungodden, B. 2015. Human and financial capital for microenterprise development: Evidence from a field and lab experiment, Management Science, forthcoming.
Bjorvatn, K., Cappelen, A. W., Helgesson, L. S., Sørensen E. Ø., and Tungodden, B. 2012. Edutainment for entreprenurship; Evidence from a field experiment in Tanzania, mimeo, NHH Norwegian School of Economics.
Burbidge, J. B., Magee, L. and Robb, A. L. 1988. Alternative Transformations to Handle Extreme Values of the Dependent Variable, Journal of the American Statistical Association. 83(401): 123‐127
Card, David, Ibarraran, P., Regalia, F., Rosas, D. and Soares Y. 2011. The Labor Market Impacts of Youth Training in the Dominican Republic: Evidence from a Randomized Evaluation.” Journal of Labor Economics, 29 (2)
Chaudhuri, S. Jalan, J., J. and. Somanathan, E. 2009. Awareness and the Demand for Environmental Quality: Survey Evidence on Drinking Water in Urban India, Environment and Development Economics, 14: 665‐692,
de Walque, D. Dow, W. H., Nathan, R. 2014. Rewarding Safer Sex: Conditional Cash Transfers for HIV/STI Prevention, Policy Research Working Paper 7099, The World Bank
Duflo, E., Dupas, P., Kremer, M. 2014. Education, HIV and Early Fertility:
Experimental Evidence from Kenya”, NBER Working Paper No. 20784.
Dupas, P. 2011a. Do Teenagers respond to HIV risk information? Evidence from a field experiment in Kenya, American Economic Journal: Applied Economics, 3: 1‐34.
Dupas, P. 2011b. Health Behavior in Developing Countries, Annual Review of Economics , 3: 425‐449.
25
Fitzsimons, E., Malde, B., Mesnard, A., Vera‐Hernández M. 2012. Household responses to information on child nutrition: experimental evidence from Malawi, IFS Working Paper 12/07. London: Institute for Fiscal Studies.
Gallant, M. and Maticka‐Tyndale, E. 2004. School‐Based HIV Prevention Programs for African Youth, Social Science and Medicine 58: 1337‐51.
Giné X, Mansuri G. 2014. Money or ideas? A field experiment on constraints to entrepreneurship in rural Pakistan. Policy Research Working Paper 6959, The World Bank, Washington, DC.
Goldin, C. and Katz, L. F. 2000. The power of the pill: oral contraceptives and womenʹs career and marriage decisions, National Bureau of Economic Research.
Hayes, R. J. and Moulton, L. H. 2012. Cluster randomised trials, CRC press, Boca Raton.
Karlan, D. and Valdivia, M. 2011. Teaching entrepreneurship: Impact of business training on microfinance clients and institutions, Review of Economics and Statistics, 93:
510‐527.
Kremer, M., Miguel, E. 2007. The Illusion of Sustainability, Quarterly Journal of Economics, 122(3): 1007‐1065.
Kling, J. Liebman, J, and Katz 2007 L. Experimental Analysis of Neighborhood Effects,’ Econometrica, 75, 83–119.
Lee, D. S. 2009. Training, wages, and sample selection: Estimating sharp bounds on treatment effects. Review of Economic Studies, 76: 1071‐1102.
Levine, R. (editor). 2004. Millions Saved: Proven Successes in Global Health Center for Global Development, ISBN 0‐88132‐372‐1
Madajewicz, M., Pfaff, A., van Geen, A., Graziano. J., Hussein. I., Momotaj, H.
2007. Can information alone change behavior? Response to arsenic contamination of groundwater in Bangladesh, Journal of Development Economincs, 84: 731–754
McCoy, S.L, Kangwende, R. A. and Padian, N. C. 2010. Behavior Change Interventions to Prevent HIV Infection Among Women Living in Low and Middle Income Countries: A Systematic Review, AIDS and Behavior 14: 469‐82.
Miller, G. 2010. Contraception as development? New evidence from family planning in Colombia, Economic Journal, 120: 709‐736.
Rasul, I. 2008. Household bargaining over fertility: Theory and evidence from Malaysia, Journal of Development Economics, 86: 215‐241.
Ross, D. A. Changalucha J, Obasi, A. I, Todd J, Plummer M. L, Cleophas‐Mazige B, et al. 2007. Biological and behavioural impact of an adolescent sexual health intervention in Tanzania: a community‐randomized trial. AIDS 2007 Sep 12;21(14):1943‐55.
27 Table 1 Treatment‐Control Balance
(1) (2) (3) (4) (5) (6)
Equality of
Full
sample
Control Health Entrepre‐
neurship
Combined Means (F) p‐value
Remote school 0.44 0.43 0.46 0.47 0.41 0.98
(0.50) (0.50) (0.50) (0.50) (0.49)
Wealthy household 0.56 0.54 0.59 0.58 0.52 0.75
(0.50) (0.50) (0.49) (0.49) (0.50)
High cognition 0.62 0.62 0.56 0.66 0.65 0.069
(0.48) (0.49) (0.50) (0.47) (0.48)
Age > 17 0.49 0.48 0.50 0.48 0.52 0.76
(0.50) (0.50) (0.50) (0.50) (0.50)
N Form IV girls 59.91 55.01 59.29 66.28 58.44 0.39
(17.22) (14.97) (15.98) (22.14) (11.00)
Woman headed hh. 0.20 0.18 0.20 0.22 0.19 0.3
(0.40) (0.39) (0.40) (0.42) (0.39)
Business owner 0.25 0.28 0.24 0.24 0.23 0.78
(0.43) (0.45) (0.43) (0.43) (0.42)
Health knowledge 2.27 2.25 2.33 2.23 2.29 0.42
(0.86) (0.86) (0.82) (0.86) (0.89)
Entrepre. knowledge 1.38 1.35 1.37 1.38 1.44 0.61
(0.79) (0.77) (0.77) (0.79) (0.83)
Risk averse 0.47 0.48 0.44 0.46 0.52 0.32
(0.50) (0.50) (0.50) (0.50) (0.50)
Business plans 0.14 0.13 0.15 0.12 0.18 0.12
(0.35) (0.33) (0.36) (0.32) (0.38)
Observations 3483 869 856 938 820
Note: The table reports average values in the respective treatment arms in columns (2)‐(5), and column present the overall average values. Column (6) present the p‐value for an F‐test of the equality of means across all four groups, after regressing each variable on the 3 treatment dummies, using clustered standard errors (80 clusters/schools). The sample consists of girls surveyed in the baseline survey. All variables are defined in the appendix.
.
Table 2 Descriptive Statistics of Short‐term Outcomes
(1) (2) (3) (4) (5)
All Control Health Entrepr. Combined
mean/sd mean/sd mean/sd mean/sd mean/sd
H1. Health knowledge 4.83 4.79 4.82 4.79 4.92
(1.37) (1.42) (1.36) (1.38) (1.33)
H2. Entrepreneurship knowledge 2.01 1.90 1.88 2.04 2.23
(1.02) (1.02) (0.94) (1.00) (1.08)
B1. Safe sex 0.65 0.69 0.63 0.60 0.67
(0.48) (0.46) (0.48) (0.49) (0.47)
B2. Business plans 0.36 0.15 0.17 0.53 0.59
(0.48) (0.36) (0.38) (0.50) (0.49)
G1. Acceptance of wife beating 0.26 0.28 0.23 0.30 0.22
(0.25) (0.24) (0.25) (0.25) (0.24)
G2. Wife earner 4.00 3.82 3.94 4.11 4.12
(1.10) (1.13) (1.12) (1.08) (1.03)
E1. Compete 0.33 0.33 0.29 0.37 0.33
(0.47) (0.47) (0.46) (0.48) (0.47)
E2. Empowerment index ‐0.01 ‐0.01 0.04 ‐0.01 ‐0.05
(0.44) (0.44) (0.45) (0.45) (0.43)
Observations 2873 741 680 781 671
Note: The table reports average values (standard deviations) of short‐term outcomes in the respective treatment arms in columns (2)‐(5), and column present the overall average values. The sample consists of girls surveyed both in the baseline and short‐term survey. All outcomes are defined in the appendix.
29
Table 3 Descriptive Statistics of Long‐term Outcomes
(1) (2) (3) (4) (5)
All Control Health Entrepren, Combined
mean/sd mean/sd mean/sd mean/sd mean/sd
B1 ‐ Childbearing 0.06 0.06 0.06 0.06 0.08
(0.25) (0.23) (0.24) (0.24) (0.27)
B2 ‐ Economic Activity 0.27 0.19 0.23 0.31 0.36
(0.44) (0.39) (0.42) (0.46) (0.48)
B3 ‐ Sales (ihst) 2.56 1.45 2.06 2.99 3.80
(7.30) (5.98) (6.94) (7.62) (8.30)
Sales (TZS) 9250 2722 16916 6974 10903
(120312) (13482) (219263) (36908) (98361)
B4 ‐ Patience 0.42 0.45 0.41 0.41 0.41
(0.49) (0.50) (0.49) (0.49) (0.49)
G1 – Wife Beating 0.01 0.01 0.01 0.01 0.01
(0.11) (0.11) (0.09) (0.11) (0.11)
G2 – Wife‐earner 3.57 3.46 3.57 3.69 3.57
(1.27) (1.28) (1.27) (1.24) (1.29)
E1 ‐ Little control 3.35 3.44 3.30 3.37 3.27
(1.27) (1.22) (1.28) (1.27) (1.29)
E2 ‐ Feeling useless 2.52 2.58 2.55 2.48 2.48
(1.38) (1.36) (1.39) (1.36) (1.40)
W1 –Happiness ‐ General 3.90 3.80 3.95 3.79 4.07
(1.14) (1.14) (1.16) (1.19) (1.04)
W2 ‐ Happiness ‐ Health 4.55 4.51 4.55 4.53 4.61
(0.76) (0.80) (0.75) (0.76) (0.71)
W3 ‐ Happiness – Econ. sit. 2.72 2.55 2.75 2.79 2.82
(1.38) (1.38) (1.36) (1.38) (1.39)
W4 ‐ Sick 0.25 0.25 0.27 0.26 0.24
(0.44) (0.43) (0.44) (0.44) (0.43)
W5 ‐ Income (ihst) 15.66 15.45 15.45 15.57 16.20
(5.49) (5.44) (5.86) (5.65) (4.91)
Income (TZS) 12897 11069 14007 12923 13693
(21884) (14771) (28916) (18511) (23224)
Observations 2950 752 723 782 693
Note: The table reports average values (standard deviations) of long‐term outcomes in the respective treatment arms in columns (2)‐(5), and column present the overall average values. The sample consists of girls surveyed both in the baseline and long‐term survey. All outcomes are defined in the appendix.
Table 4 Attrition
(1) (2)
Attrition
Short Term Long Term
Health ‐0.06* ‐0.02
(0.03) (0.03)
Entrepreneurship ‐0.02 ‐0.03
(0.04) (0.04)
Combined ‐0.03 ‐0.03
(0.04) (0.03)
Constant (and mean in the control group)
0.85*** 0.83***
(0.03) (0.02)
Observations 3483 3483
P‐val. Equality of means (F)
0.33 0.8
Note: The table reports regressions on attrition. The outcome variable is an indicator variable taking the value of one if the girl was reached in the short-term (1) or the long-term follow-up survey (2). The sample is defined as all girls being surveyed in the baseline survey. The p‐value of equality of means comes from the overall F‐test of the regression. Standard errors clustered at the school in parentheses (80 clusters); * p<0.10, ** p<0.05, *** p<0.01.
