Advertisement

Journal of Economics, Race, and Policy

, Volume 2, Issue 4, pp 225–239 | Cite as

Education with a Social Focus on Gender Attitudes: Experimental Evidence from Secondary Education in El Salvador

  • Rosangela BandoEmail author
  • Nidia Hidalgo
  • Austin Land
Original Article
  • 328 Downloads

Abstract

Gender-based violence is one of the most common social problems worldwide. It adversely affects development and the well-being of individuals. The most common prevention policy associated with gender violence is changes in attitudes through education. This paper examines the effects of an education program targeting young people between 12 and 18 years old in El Salvador. Using an experimental design, the analysis finds that the program led to changes in gender attitudes only among young women but not among young men. Both young women and men improved their knowledge about gender-related issues, but only young women carried out more conversations on the topic with peers. The analysis also identified changes in self-reported sexual behavior for both young men and women. This study provides evidence to help in the identification of effective approaches aimed at fostering a generation of young people free from gender-based violence.

Keywords

Community education Teen dating violence Gender violence Gender attitudes 

JEL Classification

I14 I24 I25 J12 J16 J18 O12 

Introduction

Gender-based violence (GBV) is a widespread social problem that affects the lives of people worldwide.1 In fact, one in every three women across the globe has been the victim of physical or sexual violence by an intimate partner (WHO 2014). GBV has adverse effects on individuals and their communities, is associated with health and psychological problems (WHO 2014),2 decreases investment and productivity, and increases public expenditure (Klugman et al. 2014).3 Klugman et al. (2014) estimate that costs associated with GBV total as much as 3.7% of gross domestic product (GDP) in some countries. Finally, GBV has adverse consequences for individual victims and their households and for the societies where those victims live. Thus, the importance of identifying effective policies to prevent GBV is clear.

Social norms and gender attitudes are one of many determinants of GBV (WHO 2014; Akerlof and Kranton 2005).4 Social norms are informal understandings or attitudes that govern the behavior of members of a group (Scott and Marshall 2009). Traditional gender roles that reinforce male control over women may make women vulnerable to physical, emotional, and sexual violence perpetrated by men (WHO 2014). Behaviors associated with gender violence begin in adolescence and manifest fully while young people are of school age (Levitt and Lochner 2001; Sampson and Laub 2003). Recent work has focused on understanding the role of education in shifting individual attitudes about gender roles to make them more equitable (Pacifici et al. 2001; Fay and Medway 2006; Wolfe et al. 2009).

This paper studies the effects of an innovative program with a social approach to changing student attitudes about gender roles. Program H and M is aimed at addressing GBV through education and the provision of life skills and has been implemented in more than 22 countries (Promundo 2013).5 The paper examines an adaptation of Program H and M in El Salvador that targeted students enrolled in 7th and 8th grades and whose ages ranged between 12 and 19. The program consisted of two components. The first consisted of educational groups where the students reflected critically and discussed gender equality. The same-sex educational groups met in weekly sessions of 1.5 hours for 3 months. The second component consisted of support for students to create and execute a social marketing campaign to promote gender equality for up to 1 month within the school. The activities associated with the campaign included workshops, puppet shows, skits, and art projects, as well as the dissemination of key messages through promotional materials such as t-shirts and posters.

The case of El Salvador is of special interest because the country is suffering from extremely high rates of violence. The homicide rate in El Salvador in 2014 was 64.2 per 100,000 populations, compared with a rate of 16.3 in the Americas and 6.2 globally in 2013 (UNODC 2014).6 In 2015, El Salvador had an estimated per capita GDP of US$4776 and a population of 6.4 million (IMF 2017). The country ranked 116th in the world on the Human Development Index and was categorized as medium-income in terms of development status (UNDP 2015).7 This context is an important aspect of the present study because the social context influences the effectiveness of programs that aim to address social norms (Krug et al. 2002; Donnelly and Ward 2015; De La Rue et al. 2016; Pacifici et al. 2001).

This study undertakes a rigorous evaluation of the H and M Programs by exploiting experimental data from El Salvador. The government implemented an adaptation of the H and M program in 17 schools randomly selected from a pool of 33 schools in the departments of La Libertad and Sonsonate between August 2014 and October 2015.8 To assess the program’s success, this paper investigated its effects on three factors that together make up a simple model of the relationship between knowledge, beliefs, social resources, behavior, and attitudes. The logic in the design of the program led to the research approach described below.

First, the paper sought to estimate the program’s impact on students’ knowledge and the frequency with which they had conversations about gender norms and GBV with their peers. The rationale is that if teachers discuss gender bias with students in the classroom, then students may talk more among themselves about gender bias out of the classroom. If they talk more about gender bias out of the classroom, then they may change their perceptions about what the group norm is. Second, the paper assessed the program’s impact on coping skills and attitudes toward gender norms. Students who have discussed gender with peers may shift attitudes and norms toward more equitable ones to fit with the group. Third, the paper examined the program’s impact on self-reported incidence of physical or emotional GBV. A change in students’ attitudes and skills to deal with conflict may lead to a reduction in GBV. Students should be endowed with appropriate techniques to better deal with conflict and stress. Whether this logic holds cannot be tested and, moreover, the relationships among the dimensions studied are likely to be closely interrelated. However, the approach allows for testing causal effects on dimensions key to the design mechanism.9

The analysis found that the H and M Program successfully changed attitudes among young women but not among men. More specifically, while young men increased their gender knowledge from 25 to 29% of correct answers on a test, they did not change their number of conversations about GBV or their ability to modulate emotional responses. Young women improved their gender knowledge from 44 to 49% of correct answers on the test. Women also increased their reporting of at least one conversation about GBV in the previous week from 49 to 63%. The paper measured attitudes using the Gender-Equitable Men (GEM) Scale. This index takes values that range from 0 to 1. Larger numbers indicate more egalitarian attitudes. The analysis found that women changed their attitudes toward more equitable ones regarding domestic chores (from 0.63 to 0.69). Changes were also observed in women’s ability to modulate emotional responses to stress, which was measured using the Coping Strategy Scale (CSS), which ranks students on a scale from 0 to 1. Higher values are associated with better control of emotions. According to this scale, women improved from 0.53 to 0.58. No program effects were found on physical or emotional violence, but there were changes in self-reported sexual activity: 14% of young women in treatment reported recent sexual activity compared with 8% of young women in the control group, and 15% of sexually active young women in treatment reported using condoms compared with 8% of young women in the control groups. No effects were found on teen dating.

This study contributes to three areas of research. First, it contributes to knowledge on the effectiveness of educational programs to address GBV in a developing country. Education is one of the most common GBV prevention strategies. As of 2013, over half of all nations in the world had implemented some version of this educational program at least once (WHO 2014). The evidence for school-based interventions in the USA and Canada shows these programs can influence knowledge and attitudes, but struggle to affect violence perpetration or victimization (De La Rue et al. 2016). Case studies from India, Balkan nations, Vietnam, and Chile indicate possible positive effects on knowledge, attitudes, and self-reported behavior (Achyut et al. 2011; Promundo 2012; Promundo 2013). Thus, this study contributes to this literature by providing evidence from a developing country.

Second, this study contributes to the understanding of research by making an explicit assessment of program effects on social interactions, a key pathway to change social norms (Donnelly and Ward 2015). In addition, many studies struggle to establish social interactions because they require identifying a reference group (Manski 2000). This paper partially addresses this constraint by focusing on effects on the school as the reference group. It was found that young women increased the frequency of GBV-related conversations, while there were no changes among men in this regard.

Third, this study provides an unbiased estimate of program effects. Many evaluations are compromised by confounding factors implied by differences between the treated and comparison groups.10 This problem is difficult to address and is often overcome by an experimental approach. Indeed, De La Rue et al. (2016) found that of 1608 studies on violence prevention, only 23 involved a valid comparison group constructed in a manner intended to address selection bias. Of these 23 studies, only 11 featured an experimental design. The experimental design used in this study allows for minimizing selection bias.

Description of Intervention

The government of El Salvador, with the support of international agencies and nongovernment organizations, created the Community Education for the Prevention of Gender Violence Program (Educación Comunitaria para la Prevención de la Violencia de Género—ECPVG) in 2013 to change attitudes, behaviors, social norms, and stereotypes linked to gender inequality. The resulting community model is aimed at promoting respect for the right to a life free of violence for women as well as nondiscrimination on the grounds of gender. The program had two components: group sessions and a social campaign.

The group sessions are aimed at promoting dialog and social interactions as a means of learning. The sessions covered themes such as gender, emotional education, peaceful coexistence, sex education, economic empowerment, and community participation.11 The group sessions took place between August and October 2014.12 The sessions totaled 18 h and were provided as part of the social studies curriculum. In the single-sex groups, students learned about human rights and techniques to deal with conflict. Students also received information on reproductive health.13

The second component was a social marketing and communication campaign that took place between September 2014 and October 2015. Students designed and led a campaign called “Campaign Lifestyles” (Campaña Estilos de Vida). The students used their own words and language to promote social change in their school.14 The activities associated with the campaign included workshops, puppet shows, skits, art projects, and activities related to self-care and emotional moderation, as well as the dissemination of key messages through promotional materials such as t-shirts and posters. All students in the school were exposed to the campaign component. All the activities took place within the school because activities out of the school presented security risks. This restriction allowed for defining the school as the group reference.

The motivation to expand implementation of the H and M Programs was the documented success of the program in countries where it had been implemented. Achyut et al. (2011) evaluated implementation of the program in 30 schools randomly chosen from a pool of 45 schools in India targeting student ages 12 to 14. The authors found that the program shifted students’ attitudes toward gender equality. Consistent with this, the authors found male students reported participating more in household chores and female students reported opposing gender discrimination. Other case studies based on quasiexperimental approaches include evaluations by Promundo (2012) in the Balkan countries and Vietnam. Both analyses find a shift in students’ attitudes toward gender equality.

Identification Strategy

The construction of an appropriate counterfactual is a challenge for the evaluation of any program’s impact. The counterfactual allows evaluators to estimate outcomes in the absence of the program. To this end, the government of El Salvador selected a paired randomized experimental design as described in Imbens and Athey (2016).

Treatment allocation involved two stages. In a first stage, 15 school pairs and one trio were created from the 33 total schools by nearest neighbor matching. Matches were created based on violence level, social risk classification, urban or rural location, and size of school. The social risk classification is a government index that incorporates school and community information on violence.15

In a second stage, one school from each pair was assigned to the treatment group with probability 0.5 within each group. The other school was assigned to a control group.16 For the group with three schools, the third school was assigned randomly to treatment or control with probability 0.5. This process resulted in the allocation of 17 schools to the treatment group and 16 to the control group. The matched-pair design allows for identifying program effects. The approach relies on within-pair variation between treatment and control schools.

The identification of the program’s impact relies on two assumptions. The first is that the program only affects the treatment schools, but not the control schools. Implementation was monitored to avoid interference in control schools as much as possible. For example, the activities and social campaign components of the program were designed to take place only on school grounds.

The second assumption is that treatment assignment is independent of potential outcomes. This condition was ensured by way of construction. “Data” section presents evidence.

Four hypotheses are tested. First, the null hypothesis that the average treatment effects of the program are zero is tested. This relies on the paired randomized design for identification. Thus, we rely on sampling inference. Within-pair average outcomes between students in treatment and control schools are controlled for. Specifically, the following model is estimated:
$$ {Y}_{ig}={\rho}_g+\beta {T}_{ig}+{\varepsilon}_{ig}, $$
(1)
where the subscript i indexes students and the subscript g indexes pairs. The variable Yigdenotes the outcome of interest. The term ρg denotes pair fixed effects. The term Tig denotes a dichotomous variable with value 1 for students enrolled in treatment schools and 0 if not. The null hypothesis of no program effect on Y is H0 : β = 0.

Student characteristics within schools may be correlated. Therefore, standard errors are clustered at the school level. One drawback of this approach is that the design includes approximately two schools within each pair. A traditional clustered error estimator based on the law of large numbers would overreject the null hypothesis. Therefore, this approach is complemented by considering a high correlation between errors within clusters.17 At one extreme, the errors are perfectly correlated. Effects under this scenario were tested in hypotheses three and four, which are described later in this section.

The second null hypothesis tested has no impact on the average change for a given outcome. Changes are defined as the difference in outcome values before and after program implementation. For example, the change in test scores before and after the student participated in the program is examined. The advantage of this approach is that it allows for controlling for unobserved characteristics that are constant over time. Consider no differences at baseline. In this case, the difference in changes between the treatment and control groups is equal to the difference between the two groups ex post. The identification assumption is that potential changes in outcomes are independent of treatment assignment. The model is estimated as follows:
$$ {Y}_{it}={\mu}_i+{\nu}_t+\gamma {T}_i{P}_t+{u}_{it}, $$
(2)
where the subscript i indexes students and the subscript t indexes time. The subscript t has the value 0 for baseline observations and 1 for follow-up observations. The term μi denotes student fixed effects. This term models the variation between individuals that is constant over time. The term νt denotes time fixed effects. This term models the variation between time periods that is constant between individuals. The term Ti is a dichotomous variable with value 1 for students enrolled in treatment schools and 0 if not. The term Pt is a dichotomous variable with value 0 for baseline observations and 1 for follow-up observations. The term uit models stochastic error. The null hypothesis of no average effect of the program on the changes for outcome Y is H0 : γ = 0. To estimate this model, bootstrap standard errors conglomerated by student are estimated.

The third and fourth null hypotheses revolve around the assumption that the program had no effect on any school for outcome y. More specifically, the third null hypothesis is \( {H}_0:{\overline{y}}_s(0)={\overline{y}}_s(1) \) for each school s of the 33 schools in the study. The term \( {\overline{y}}_s(0) \) denotes the average outcome y for school s without the program, and the term \( {\overline{y}}_s(1) \) denotes the same outcome for school s with the program. The alternative hypothesis is that the program had a nonzero average effect on at least one school. The fourth null hypothesis is that the program had a nonzero average effect on changes in outcome y for at least one school, \( {H}_0:{\overline{\Delta y}}_s(0)=\overline{\Delta {y}_s}(1) \) for each school s. This approach is employed as a relatively conservative method to test for effects at the school level as suggested in Imbens and Athey (2016).

The third and fourth hypotheses are sharp because they allow for estimating the exact probabilities that each is true. These hypotheses rely on the assumption that potential outcomes are known exactly. Only the treatment variable is random. The distribution of the treatment coefficient is completely determined by T. It is estimated by computing the coefficient for all possible treatment assignments of schools to receive the program. The null hypothesis is then tested. It can be observed where the observed treatment coefficient falls in this distribution. We reject the null of no effects based on the share of coefficient values above the absolute value of the actual estimate. There are many possible treatment assignments. Thus, 1000 possible assignments are randomly selected for inference. Fisher (1937) proposed randomization inference. Rosenbaum (2002) subsequently developed it further. An example of a recent application of this method is Cohen and Dupas (2010).

The tests of these four hypotheses are complementary. The first and second tests allow for assessing effects for the average student. The third and fourth tests allow for assessing effects for the average school. The four analyses are important. Both individual and social factors determine violence. The program is aimed at having an impact on outcome indicators at both levels. Thus, it is important to carry out the analysis at both levels.18

Data

Sample

The sample for this study draws from the universe of students enrolled in 7th and 8th grades in eight municipalities. These municipalities are in the Departments of La Libertad (Jayaque, Sacacoyo, Ciudad Arce, Colón, San Juan Opico, Talnique, Tepecoyo) and Sonsonate (Armenia). There are 74 schools in these municipalities. The government deemed ineligible schools participating in a program to improve education quality.19 The sample for this study consists of the 33 remaining schools. These schools lie in the municipalities of Jayaque, Sacacoyo, Colón, Talnique, Tepecoyo (Department of La Libertad), and Armenia (Department of Sonsonate). The municipalities with schools in this study have a higher proportion of urban households (as opposed to rural) and higher literacy rates compared with other municipalities in the country. Appendix 1 provides maps and a comparison table.

Two data sources provided information for this study. The first source was school-level administrative records provided by the Ministry of Education. Administrative records include school-level information on geographic location, enrollment, repetition rates, and the proportion of overage students. The data included two indicators associated with security. The first classified schools in three groups according to the level of violence in the municipality. The second classified social risk in the municipality as low, medium, or high.

The second data source was two survey rounds carried out by the General Directorate of Statistics and Censuses (Dirección General de Estadística y Censos—DIGESTYC). The baseline survey was carried out in June and July of 2014, before the program began. The follow-up survey took place in October 2015.

The surveys were comprised of nine modules. The first five modules included questions on student characteristics such as family structure and home and school environments. The following three modules collected information on students’ knowledge, attitudes, and relationships. These modules included questions on violence and other risky behaviors. The last module focused on violence and insecurity in the community.

This study explores four dimensions: knowledge, skills for everyday life, attitudes, and partner relationships. All scores were standardized to scales between 0 and 1. Values closer to 1 indicate knowledge, skills, attitudes, or behaviors favoring gender equality.

The questions included in these surveys facilitated the use of four key outcome measures. Students responding to the survey read questions and statements and then chose among several potential answers from a Likert scale. For each group of related questions or statements, indices were constructed in two steps. First, each item was given one point if the student answered in a manner favoring gender equity. Second, all points were summed and divided by the total number of items. The resulting score lies in a 0 to 1 scale. Larger values indicate skills, attitudes, or behaviors that favor gender equity.

The first outcome used was the Gender-Equitable Men (GEM) Scale. Population Council/Horizons and Promundo developed the GEM scale. It is aimed at examining attitudes toward gender norms (Barker et al. 2004) by measuring attitudes about standards in intimate relationships. It also assesses differences in social expectations between men and women (Nanda 2011). This indicator is associated with the use of contraceptive methods and having multiple sexual partners and is predictive of future intimate partner violence (Nanda 2011).

The second outcome measure used was the Gender Relations Scale (GRS). This scale is aimed at measuring equality and power dynamics in intimate relationships.20 It has been associated with the future use of modern methods of family planning (Nanda 2011). The GRS and the GEM scales have been applied in at least seven countries, and experts have validated them as culturally sensitive tools (Pulerwitz and Barker 2008; Nanda 2011).

The third outcome measure that informs this study is the Coping Strategies Scale (CSS) adapted for Latin America by Londoño et al. (2006). This scale examines cognitive and behavioral resources by measuring an individual’s response to stressful situations (Chorot and Sandín 1993; Londoño et al. 2006). Questions state a hypothetical situation, and a student then identifies among options how he or she would confront that problem or challenge.

Responses to a 12-question multiple-choice test are also examined. The test covers key concepts discussed in the educational sessions. The percentage of correct responses is calculated, and this share serves as an approximation of the students’ knowledge.

Four indicators were constructed to measure student demographic characteristics: age, an assets index, both parents at home, and violence before the age of 12. The asset index was built using principal components following the approach of Fernald et al. (2008).21 The resulting indicator is associated with household consumption. Appendix 2 lists the items in the questionnaires used for the definition of the indicators used in this study.

The baseline survey collected information on 4716 students in grades 7 and 8. The follow-up survey collected information on 3654 of the 4716 baseline students. Of those 3654 students, the focus is on 2191 students (1257 young men and 934 young women) who have been in an intimate partnership (girlfriend or boyfriend). This approach is taken because the other students have not had experiences that allow them to assess the role of social norms in a relationship. Seventy percent of young men and 51% of young women reported they are currently in an intimate relationship or have been in one in the past. The survey found that having had an intimate relationship was independent of treatment at baseline (the p value for the probability of an association with treatment is 0.759 for young men and 0.189 for young women).

Students who have been in an intimate relationship differ from those who have not. For example, in relation to young men who have never had a partner, students who have had a partner are older (14.6 years compared with 14.2 years) and live in households with better socioeconomic status (0.12 against − 0.14 standard deviations in relation to the sample distribution of assets). A share of 44% of young men who have never had a partner have experienced violence before the age of 12, compared with 50% of the students who have had an intimate partner. The differences between young women who have not had an intimate relationship and those who have had one are similar to the differences among young men. A share of 42% of young woman who have never had an intimate relationship had experienced violence before the age of 12 years in contrast to 57% of those who have had an intimate relationship.

Descriptive Statistics

Table 1 displays school-level descriptive statistics for 33 schools analyzed in this study. Column 1 shows the school-level average of each indicator for the 16 schools assigned to the control group. On average, schools had 711 students enrolled, of which 52% were male. The failure rate in 2013 was 1.5% and the overage rate was 9.7%. Column 2 shows the average differences between treatment schools relative to control schools.
Table 1

Averages and balance at baseline for school characteristics

 

Control group average

Average difference (treatment-control)

p value

(1)

(2)

(3)

Enrollment

710.50

−45.31

0.703

Proportion of male students

0.52

0.00

0.816

Students who have had an intimate relationship

0.58

0.02

0.506

Students repeating the grade

0.01

0.00

0.840

Overage students

0.10

0.00

0.866

Students who trust the people in their neighborhood

0.23

0.01

0.779

Students who feel safe in their house

0.67

0.01

0.889

Estimates are based on student averages of 33 schools, of which 17 were assigned to the treatment group. Column 3 refers to the hypothesis that the program had an impact in at least one school

The main identification assumption is that school characteristics are independent of treatment status. Column 3 shows the exact p values for the probability that there is no association. All values are above 0.15. It is concluded that the randomization created two groups of schools that do not differ systematically. A threat to the identification strategy is student attrition. However, it was found that the student attrition is not related to the treatment assignment (the p value is 0.710).

Tables 2 and 3 show characteristics between the treatment and control groups at the student level at baseline for young men and young women, respectively. Column 1 shows the average among students in the control group. On average, male students are 15 and female students are 14 years of age. A share of 65% of young men and 58% of young women reported that both parents live at home. In addition, 50% of young men and 52% of young women reported experiencing violence before the age of 12. The rest of the indicators assess skills, attitudes, and behavior of students. Column 3 shows p values for school-level tests of average effects in at least one school. Column 4 shows the p values for tests of mean differences at the student level. Column 5 shows the p values for tests of differences at the student level adjusted for multiple testing. The analysis follows Anderson (2008) to make the adjustment. The adjusted p values account for the probability of false rejections for the family of results listed in each table. The results of these tests indicate that there are no differences before treatment. It is concluded that the treatment and control groups had no systematic differences at baseline.
Table 2

Averages at baseline for young men

 

Control group average

Average difference (treatment-control)

p value for school-level test

p value for average differences

Adjusted p values for multiple testing

(1)

(2)

(3)

(4)

(5)

Age in years

14.59

0.00

0.719

0.975

1.000

Asset indexa

0.13

− 0.02

0.649

0.911

1.000

Both parents live at homeb

0.65

− 0.01

0.668

0.697

1.000

Experienced violence before age 12b

0.50

− 0.01

0.402

0.641

1.000

Score on knowledge test (proportion of correct answers)

0.28

0.00

0.899

0.958

1.000

Had conversations about gender-based violence in the last weekb

0.44

0.01

0.986

0.748

1.000

Attitudes toward gender norms in reproductive healthc

0.41

0.00

0.754

0.907

1.000

Attitudes toward gender norms in domestic choresc

0.46

− 0.02

0.574

0.578

1.000

Equality and power in intimate relationshipsd

0.51

0.00

0.920

0.974

1.000

Resources to modulate an emotional response to stresse

0.44

0.01

0.979

0.699

1.000

Currently in an intimate relationshipb

0.49

− 0.01

0.760

0.811

1.000

Had sex in the last 3 monthsb

0.15

− 0.03

0.189

0.314

1.000

Perpetrated physical violence against partner in the last 6 monthsb

0.19

0.00

0.330

0.865

1.000

Perpetrated emotional violence against partner in the last 6 monthsb

0.20

− 0.02

0.351

0.612

1.000

Source: authors’ calculations

Estimates are based on 1257 young men. Column 3 refers to the hypothesis that the program had an impact in at least one school. Column 4 refers to the hypothesis that averages in treatment, and the control groups are not different. Column 5 shows adjusted p values for multiple testing. The analysis follows Anderson (2008) for the group of results listed in the table

aStandardized with a mean of 0 and a standard deviation of 1

bTakes on a value 1 if yes, 0 if no

cGender-Equitable Men Scale between 0 and 1, with values closer to 1 indicating egalitarian gender attitudes

dGender Relations Scale between 0 and 1, with values closer to 1 indicating greater equality and power in decision-making

eCoping Strategies Scale between 0 and 1, with values closer to 1 indicating better emotional control

Table 3

Average and balance at baseline for young women

 

Control group average

Average difference (treatment-control)

p value for school-level test

p value for average differences

Adjusted p values for multiple testing

(1)

(2)

(3)

(4)

(5)

Age in years

14.41

0.07

0.760

0.381

1.000

Asset indexa

0.14

− 0.08

0.786

0.541

1.000

Both parents live at homeb

0.58

− 0.01

0.727

0.787

1.000

Experienced violence before age 12b

0.52

0.10

0.020

0.039

0.429

Score on knowledge test (proportion of correct answers)

0.38

0.00

0.965

0.961

1.000

Had conversations about gender-based violence in the last weekb

0.49

0.03

0.661

0.486

1.000

Attitudes toward gender norms in reproductive healthc

0.45

0.00

0.685

0.982

1.000

Attitudes toward gender norms in domestic choresc

0.55

0.02

0.475

0.499

1.000

Equality and power in intimate relationshipsd

0.57

0.01

0.363

0.253

1.000

Resources to modulate emotional response to stresse

0.45

− 0.01

0.413

0.312

1.000

Currently in an intimate relationshipb

0.46

0.06

0.087

0.022

1.000

Had sex in the last 3 monthsb

0.05

0.00

0.693

0.893

1.000

Suffered physical violence by partner in the last 6 monthsb

0.13

0.00

0.788

0.909

1.000

Suffered emotional violence by partner in the last 6 monthsb

0.27

− 0.03

0.326

0.118

1.000

Source: authors’ calculations

Estimates are based on 934 young women. Column 3 refers to the hypothesis that the program had an impact in at least one school. Column 4 refers to the hypothesis that averages in treatment, and the control groups are not different. Column 5 shows adjusted p values for multiple testing. The analysis follows Anderson (2008) for the group of results listed in the table

aStandardized with a mean of 0 and a standard deviation of 1

bTakes on a value 1 if yes, 0 if no

cGender-Equitable Men Scale between 0 and 1, with values closer to 1 indicating egalitarian gender attitudes

dGender Relations Scale between 0 and 1, with values closer to 1 indicating greater equality and power in decision-making

eCoping Strategies Scale between 0 and 1, with values closer to 1 indicating better emotional control

Results

This section describes the impact analysis of the ECPVG program. Tables 4 and 5 list impacts in three categories of outcomes for young men and women, respectively. Panel A groups the first category of outcomes. It includes immediate outcomes related to students’ knowledge and gender-related socialization. Panel B groups outcomes related to intermediate program outcomes: life skills and attitudes. Panel C lists outcomes related to final program outcomes: behaviors associated with intimate relationships. The logic of the program is that if students learn and discuss gender roles, they will improve life skills and change their attitudes toward gender discrimination. As a result, students will change the dynamics in their intimate relationships.
Table 4

Impacts of the education program H and M on learning, social interactions, life skills, attitudes, and relationships of young men

 

Control group average

Average difference (treatment-control)

Average difference with student fixed effects

p value school average difference

p value school fixed effects

Adjusted p values for multiple testing

(1)

(2)

(3)

(4)

(5)

(6)

Panel A. Learning and social interactions

  Score on knowledge test (proportion of correct answers)

0.25

0.05 (0.01)***

0.04 (0.01)***

0.027

0.013

0.150

  Had conversations about gender-based in the last weeka

0.46

0.06 (0.03)*

0.04 (0.04)

0.161

0.340

0.440

Panel B. Life skills and attitudes

  Attitudes toward gender norms in reproductive healthb

0.40

0.04 (0.03)

0.03 (0.03)

0.343

0.346

0.565

  Attitudes toward gender norms in domestic choresb

0.67

0.02 (0.05)

0.03 (0.03)

0.941

0.657

1.000

  Resources to modulate an emotional response to stressc

0.54

0.00 (0.02)

− 0.01 (0.02)

0.605

0.550

1.000

Panel C. Intimate relationships

  Currently in an intimate relationshipa

0.53

− 0.01 (0.04)

0.01 (0.03)

0.567

0.917

1.000

  Equality and power in intimate relationshipsd

0.59

0.02 (0.03)

0.01 (0.02)

0.900

0.847

1.000

  Had sex in the last 3 monthsa

0.24

0.03 (0.04)

0.07 (0.03)***

0.474

0.036

0.194

  Perpetrated physical violence against partner in the last 6 monthsa

0.09

0.03 (0.03)

0.05 (0.03)

0.580

0.268

1.000

  Perpetrated emotional violence against partner in the last 6 monthsa

0.14

0.00 (0.03)

0.03 (0.03)

0.824

0.449

1.000

Source: authors’ calculations

Estimates are based on 1257 young men, of whom 604 were assigned to the treatment group. Column 3 corresponds to the hypothesis that the program had an average effect on outcomes in changes at the student level. Column 4 shows the p value for the test of the hypothesis that the program had average effects on at least one school, in levels. Column 5 shows the p value for the hypothesis that the program had average effects on at least one school, in changes. Column 6 shows the p values adjusted for multiple testing for the group of results listed in the table. The symbols *, **, and *** indicate that the coefficient estimates are statistically significantly different than 0 at the 0.10, 0.05, and 0.01 levels, respectively

aTakes on a value of 1 if yes, 0 if no

bGender-Equitable Men Scale between 0 and 1, with values closer to 1 indicating egalitarian gender attitudes

cCoping Strategies Scale between 0 and 1, with values closer to 1 indicating better emotional control

dGender Relations Scale between 0 and 1, with values closer to 1 indicating greater equality and power in decision-making

Table 5

Impacts of the education program H and M on learning, social interactions, life skills, attitudes, and relationships of young women

 

Control group average

Average difference (treatment-control)

Average difference with student fixed effects

p value school average difference

p value school fixed effects

Adjusted p values for multiple testing

(1)

(2)

(3)

(4)

(5)

(6)

Panel A. Learning and social interactions

  Score on knowledge test (proportion of correct answers)

0.44

0.05 (0.02)**

0.05 (0.01)***

0.022

0.004

0.042

  Had conversations about gender-based violence in the last weeka

0.49

0.18 (0.03)***

0.14 (0.04)***

0.014

0.025

0.082

Panel B. Life skills and attitudes

  Attitudes toward gender norms in reproductive healthb

0.45

0.04 (0.03)

p = 0.164

0.05 (0.02)**

0.199

0.167

0.201

  Attitudes toward gender norms in domestic choresb

0.63

0.06 (0.03)**

0.04 (0.03)

p = 0.109

0.009

0.085

0.136

  Resources to modulate an emotional response to stressc

0.53

0.02 (0.02)

p = 0.211

0.05 (0.02)**

0.140

0.066

0.131

Panel C. Intimate relationships

  Currently in an intimate relationshipa

0.52

0.07 (0.05)

− 0.02 (0.04)

0.475

0.732

0.415

  Equality and power in intimate relationshipsd

0.60

0.02 (0.01)**

0.01 (0.01)

p = 0.319

0.030

0.284

0.304

  Had sex in the last 3 monthsa

0.08

0.07 (0.03)***

0.06 (0.02)***

0.034

0.063

0.082

  Suffered physical violence by partner in the last 6 monthsa

0.07

0.03 (0.02)

0.04 (0.02)

0.269

0.416

0.304

  Suffered emotional violence by partner in the last 6 monthsa

0.18

0.05 (0.04)

p = 0.216

0.08 (0.05)*

0.237

0.365

0.304

Source: authors’ calculations

Estimates are based on 934 young women, of whom 448 were assigned to the treatment group. Column 3 corresponds to the hypothesis that the program had an average effect on outcomes in changes at the student level. Column 4 shows the p value for the test of the hypothesis that the program had average effects on at least one school, in levels. Column 5 shows the p value for the hypothesis that the program had average effects on at least one school, in changes. Column 6 shows the p values adjusted for multiple testing for the group of results listed in the table. The symbols *, **, and *** indicate that the coefficient estimates are statistically significantly different than 0 at the 0.10, 0.05, and 0.01 levels, respectively

aTakes on a value of 1 if yes, 0 if no

bGender-Equitable Men Scale between 0 and 1, with values closer to 1 indicating egalitarian gender attitudes

cCoping Strategies Scale between 0 and 1, with values closer to 1 indicating better emotional control

dGender Relations Scale between 0 and 1, with values closer to 1 indicating greater equality and power in decision-making

Knowledge and Social Interactions

Panel A in Tables 4 and 5 shows results from the knowledge test. Young men and women in the control group responded correctly to 25 and 44% of the questions, respectively. The first row in column 2 indicates that the program increased test scores by 5 percentage points for young men and women. This difference is statistically significant. The impact is equivalent in magnitude to one additional correct response to 16 test questions. The next row displays statistics on conversations about themes related to gender in the last week. Column 2 indicates the program led to an increase in conversations. The share of respondents reporting conversations increased from 0.46 to 0.52 (or 15%) among young men and from 0.49 to 0.67 (or 35%) among young women. The average differences are statistically significant for both groups. Column 3 shows this difference is not significant for young men when considering baseline differences (p value of 0.272). Columns 4 and 5 show school-level test results. The results are consistent with the student-level analysis.

Life Skills and Attitudes

Panel B in Tables 4 and 5 shows the estimated impact of the program on life skills and attitudes. Considering the GEM scales on sexual and reproductive health and on domestic chores, the respective values are 0.40 and 0.67 for young men (Table 4, panel B, column 1). The respective values for young women are 0.45 and 0.63 (Table 5, panel B, column 1). Program effects on life skills and attitudes were found to differ by gender. Among young men, there is no significant difference between the treatment and control groups in any outcome (Table 4, panel B, column 2). The student-level fixed-effects model (column 3) and the school-level model analysis (columns 4 and 5) lead to consistent conclusions. As for young women, a 0.06 point increase in the GEM scale for domestic chores is estimated for students in the treatment group (Table 5, panel B, column 2). Column 3 indicates no effect when student fixed effects are included. Column 4 shows the results of a more conservative test. It indicates a significant difference in changes at the school level, consistent with estimates shown in column 2. When considering differences at baseline, statistically significant differences are found. Consider two GEM indexes: effects are found on attitudes toward gender norms in reproductive health, and differences are also found in the CSS index of resources to modulate emotional response to stress. Both indicators had an additional change of 0.05 points. Columns 4 and 5 indicate that the null of no effects at the school level at a 15% significance level is rejected (p = 0.140 for levels and p = 0.066 for changes). No effects are found at the school level for the GEM scale for attitudes associated with reproductive health (p = 0.199 for levels and p = 0.167 for changes). The conclusion is that the program had no average effect on life skills or attitudes among male students. However, the program had mean effects on attitudes toward gender norms in domestic chores, and it also had effects on resources to modulate an emotional response among young women.

Intimate Relationships

Tables 4 and 5 show outcomes related to relationships. Respective shares of 53% and 52% of young men and women reported they are currently having an intimate relationship. Differences between treatment and control groups were not significant for either sex (column 2). No evidence was found of effects in the GRS scale on equality and power among young men. Among young women, a statistically significant difference of 0.02 points between treatment and control was found (Table 5, column 2). However, this difference is not significant once preexisting differences are controlled for (Table 5, column 3, p = 0.319). Shares of 24% of young men and 8% of young women reported having had sex in the last 3 months. No statistically significant differences among young men in outcome changes were found (Table 4, column 3). Among young women, there was an increase from 8 to 15% in the share reporting recent sexual activity (Table 5, columns 1 and 2). This difference is statistically significant and consistent across models. Among young men, 9 and 14% reported having committed physical and emotional violence in the last 6 months, respectively. Among young women, 7% reported having suffered physical violence, and 18% reported that they suffered emotional violence from their partner in the past 6 months. Program effects on emotional violence among young women were found (column 3), but this difference is not consistent across models.

Robustness Checks

The results show that the program improved students’ knowledge of gender-based violence. The results suggest that the program promoted discussions on the subject among women, and that women changed attitudes and skills to modulate emotional responses to stress. In addition, the program increased the reporting of sexual relations among young men and women.

The findings are consistent under randomization inference and comparisons of means. It is concluded that the method by which standard errors were calculated does not lead to overrejecting hypotheses in individual tests. In addition, the estimates between pair fixed effects models and student fixed effects were found to be consistent. Differences in means controlling for baseline outcomes and controlling for student characteristics are estimated. Findings are consistent and not statistically different to those presented in Tables 4 and 5. Appendix 3 shows the estimates. Thus, it is concluded that randomization was successful, and it is unlikely that the findings of the analysis are explained by unobserved differences.

A threat to the analysis is that the number of statistical tests implies false rejections. Tables 4 and 5 test 10 outcomes by sex, for a total of 20 statistical tests. Column 6 in each table reports p values to account for false rejection rates by the family of outcomes listed in each table following Anderson (2008). Few differences are found. Considering knowledge effects among men, and the GEM scale on household chores and emotional responses among women, no program effects at the 10% significance level are rejected. However, no program effects cannot be rejected at the 15% significance level (p = 0.150, p = 0.136, and p = 0.131, respectively). In contrast, the absence of effects at the 15% level of significance for sex among men (p = 0.194) cannot be rejected. It is concluded that the program only led to improvements in students’ knowledge about gender.

Discussion

This section has three objectives: (1) compare the results of this study with those from similar quantitative studies, (2) compare the results with qualitative evidence, and (3) list the scope and limitations of the evidence presented in this study.

In terms of the first objective, this study is aimed at identifying similar studies in order to compare results. De La Rue et al. (2016) estimated the average effects of a similar program on student knowledge and attitudes. He found effects of 0.22 and 0.14 standard deviations, respectively. Consistent with the present study, De La Rue et al. did not find effects on either perpetration of or victimization from violence. They reach this conclusion based on a meta-analysis of 23 quantitative studies. Of those studies, the three closest to the present study are Fay and Medway (2006); Pacifici et al. (2001); and Wolfe et al. (2009). The programs for the three studies were conducted in secondary schools where teachers provided the instruction. Evaluation in all three cases relied on experimental designs.

Fay and Medway (2006) studied a program for the prevention of rape for 154 students in the first year of secondary school. The study focused on a school in a rural community in South Carolina. The evaluation sample consisted of six in-school classes. Evaluators randomly assigned half of the classes to treatment. The evaluation of the program found that it improved students’ knowledge on the subject, but attitudes toward dating violence did not change. The evaluation did not seek to determine effects on GBV, and the authors did not mention the cost of the program.

Pacifici et al. (2001) studied the effect of an 18-h program that is aimed at changing the knowledge of young people about sexual harassment and dating violence. The evaluation sample consisted of 547 upper-secondary students in two suburban schools in the USA. The evaluation relied on an experimental design. The authors found changes in the students’ knowledge and attitudes. The cost of the programs was not mentioned.

Wolfe et al. (2009) measured the effects of a program that taught 21 lessons to 14- to 15-year-old students in southwestern Ontario. The study took place in 20 public schools, of which 10 were randomly assigned to treatment. The authors found that the lessons decreased the rate of intimate partner physical violence (from 9.8 to 7.4%). The evaluation did not measure effects on knowledge or attitudes. The program had a cost of US$16 per student.

The conclusion is that the findings in the present study on effects on learning and attitudes are consistent with the other empirical studies. The effects are similar to those of programs implemented in different contexts. However, the evidence from El Salvador differs in terms of the increase in sexual activity reported by young women.

In terms of the second objective, the results of the present study are compared with those of three qualitative studies. The first two, Rosekrans (2015) and Landa (2015), focused on the same program and population as that considered by the present study. Rosekrans (2015) conducted a study based on 10 focus groups with 50 students in five of the same 17 treatment schools examined in the present study. The author found that students perceived that the program had resulted in changes in beliefs, attitudes, and practices. Young women reported these perceptions more often. In the second study, Landa (2015) reported perceptions of changes in learning and attitudes. Teachers faced discipline challenges in some sessions with young men, but the author also reported classroom participation among men was better than that of women.

The third study was an evaluation by Promundo (2013) of three case studies in India, Brazil, and Chile. Of the three, the Chilean program is the most similar to that implemented in El Salvador. The programs implemented in India and Brazil focused on adults. The study in Chile focused on young men between 14 and 19 years old who attended workshops in schools, health centers, or community centers. The evaluators identified young men with similar sociodemographic characteristics. The authors compared the average before-and-after changes among young men who participated in the program with those who did not. Participants reported an improvement in knowledge about types of violence, displayed more egalitarian attitudes according to the GEM scale, and self-reported a reduction in violence. The study findings were consistent with perceived effects on attitudes in other programs implemented for adults in Brazil, India, and Rwanda. The evidence from the present study in El Salvador is consistent with the qualitative findings in Chile, and similar to those from Brazil, India, and Rwanda about knowledge and attitudes among adults.

In terms of the third objective of this section, the present study faces several major limitations. The first is that it cannot be empirically proven there was no contamination of the control group, although the government of El Salvador took steps to avoid it. Innoceti (2014) investigated the possibility of contamination. She interviewed teacher trainers and studied administrative monitoring reports. The investigation found a teacher in the control group who attended the training motivated by a personal invitation. The teacher was allowed to stay, but was asked not to discuss or promote the course content. Innoceti (2014) also noted that teachers in the treatment group reported not discussing training contents. The teachers considered the contents to be culturally sensitive outside the context of the program. Thus, they preferred not to talk about it informally.

The second limitation of the present study is the possibility that some students migrated. No systematic attrition was found in the sample, however, so migration of students is not believed to have been a threat to identification. It is also possible that students from treatment schools interacted with students in control schools. If the program had any effect on the control group, then the estimates would be biased toward zero. The study focuses on students who have been in an intimate partnership. Estimation based on the full sample of students is consistent with that of the restricted analysis. Appendix 4 shows results.

Another limitation is that the study is based on information self-reported by the students. In this case, the estimates may be affected by measurement error. This threat is perhaps more severe given the sensitive nature of the questions students were asked to answer. It was made clear to students that their participation was voluntary, and their responses were confidential. Surveyors made it clear that students could fill out the questionnaires independently and confidentially. However, people tend not to report perpetration of or victimization from violence. This may be the case even in safe environments and when the reporting process adheres to ethical guidelines (Bloom 2008). Therefore, estimates here of violence rates and other outcome indicators may be a lower bound of actual parameters.22

A special outcome that may be misreported is that related to sexual activity. Two alternative explanations are considered for the increase in reports on sexual relationships. The first is that the program indeed increased the rate of sexual activity by around 6 percentage points, as reported. However, 75% of young men who reported having sex also reported using a condom. There were no differences in reports of condom use between the treatment and control groups. Among young women who reported having had sex, 52% in the control group reported condom use compared with 65% in the treatment group. If there was an actual increase in sexual activity, the results suggest it may have been safer.

Another possible explanation is that the changes in reports of sexual activity result from changes in knowledge and attitudes. Interviews were subject to ethical guidelines and were aimed at providing a safe and private environment for respondents. However, there is always a proportion of people who do not report sexual activity for fear of social stigma (Bloom 2008; Vivo et al. 2013). Indeed, 24% of young men report having sex in the last 3 months, while among young women, the rate was only 8%. It is plausible that the effect observed is due to underreporting at baseline among women. Reporting behavior may have changed, as young women experienced a change in attitudes toward gender norms. Unfortunately, however, this study relies on self-reported activity, so changes in underreporting and changes in sexual activity cannot be distinguished from one another.

This study focuses on an exposure period of 3 months of education and 1 month of participation in a social marketing campaign. The exposure period of the study is similar to that of other studies of similar programs (De La Rue et al. 2016). For example, Fay and Medway (2006), Pacifici et al. (2001), and Wolfe et al. (2009) evaluate programs with durations of 3, 1.43, and 7 weeks, respectively. De La Rue et al. (2016) concludes that future studies should examine how social norms evolve with longer exposure to treatment. He also emphasizes the importance of analyzing the evolution of treatment effects over time. We agree with this idea, believe that other studies should examine this aspect of dynamic exposure, and see no barriers to replicating the present study in similar contexts. These relevant questions are left for future studies.

Conclusions

Gender-based violence is one of the most prevalent damaging behaviors to society. Life skills programs targeted toward young people may reduce violence (WHO 2014), but policy design demands more rigorous evidence on their effectiveness.

This paper studies the effect of an education program with a focus on changing gender attitudes and social norms among young people ages 13 to 18. The program is aimed at changing knowledge, beliefs, social resources, behavior, and attitudes among high school students in public schools. In 2013, the government of El Salvador implemented the program in 17 schools.

The program changed gender attitudes among young women but not among young men. However, the program led to an increase in knowledge among young men, as the proportion of their correct answers on a multiple-choice gender knowledge test increased from 25 to 29%. No evidence was found among young men of effects on conversations regarding GBV, on their abilities to modulate emotional responses to stress, or on their attitudes toward gender norms.

The program also led to an increase in knowledge among young women. Their proportion of correct answers on the knowledge test increased from 44 to 49%. The percentage of young women who discussed gender issues in the previous week increased from 49 to 63%. Young women also changed attitudes toward a more egalitarian view related to household chores (from 0.63 to 0.67 on the GEM scale). They also increased the score regarding their control of emotions when responding to stress from 0.53 to 0.58 on the CSS scale. Young women reported more sexual activity (from 8 to 14%). No evidence was found of program effects on intimate partner violence.

These findings inform us about the effectiveness of programs with a school-based approach to change attitudes. The evidence presented in this study shows that education-based policies can impact social norms. The use of educational tools for social change faces significant challenges. Consistent with studies in other countries, this study found that the program was more effective among young women. Indeed, women would lose less privilege in the process of social change. No strong evidence was found in favor of change in behaviors.

To our knowledge, this is the first rigorous evaluation of this type of program in Latin America, and certainly in El Salvador. Most of the evaluation literature related to similar programs pertains to developed countries. Therefore, the evidence from El Salvador helps to provide external validity. It provides information about the effectiveness of GBV prevention programs that target young people. It also shows that the effectiveness of these programs may be robust to contexts characterized by high violence.

More evidence is needed to determine whether other elements are necessary to influence behaviors. This study focuses on a sample of 17 schools. Students had sessions for 3 months and a campaign over an 18-month period. The outcome measures rely on self-reported information. Other studies can better explore the role of complementary inputs. Other studies can also study how short-term impacts compare with long-term ones. These efforts are necessary to identify effective approaches. This work will contribute to support public efforts to foster a generation of young people free from gender-based violence.

Footnotes

  1. 1.

    GBV is defined as “any act of gender-based violence that results in, or is likely to result in, physical, sexual, or psychological harm or suffering to women, including threats of such acts, coercion or arbitrary deprivation of liberty, whether occurring in public or in private life” (The citation “United Nations 1994” has been changed to “United Nations, 2014” to match the author name/date in the reference list. Please check if the change is fine in this occurrence and modify the subsequent occurrences, if necessary.United Nations 2014).

  2. 2.

    GBV is associated with increased incidence of cancer and heart disease and may trigger mental health problems such as depression and anxiety, lead to substance abuse problems such as alcoholism and drug addiction, and lead to sexual and reproductive health problems such as unwanted pregnancies and HIV (WHO 2014).

  3. 3.

    GBV is also associated with adverse outcomes for victims’ households. For example, GBV affects the investment decisions of victims. Households where women have a diminished role in decisions exhibit lower spending on children’s health and education. (Duflo 2003; Duflo and Udry 2004). In addition, children in homes affected by domestic violence are more likely to perpetrate or suffer violence as adults (WHO 2014). The harmful effects of GBV can also extend beyond the victim’s household. For example, children who suffer from GBV at home may perform worse in the classroom (Carrell and Hoekstra 2010).

  4. 4.

    Social norms are not the only determinant of GBV. The incidence of GBV is associated with multiple risk factors at the individual, community, and social levels (WHO and London School of Hygiene and Tropical Medicine 2010; Krug et al. 2002). Some economists have argued that the behavior of individuals is explained in part by their upbringing (Heckman et al. 2014; Rangel 2006). Other economists have emphasized the role of social norms and people’s need to follow them (Akerlof and Kranton 2005). Others have found that the environment in which an individual develops may determine violent behavior (Sampson 2016; Carrell and Hoekstra 2010; Kling et al. 2005; Card and Dahl 2011; Heller et al. 2017). Factors such as gender, parental education, and the community presence of criminal organizations influence the likelihood of a person suffering from intimate partner violence.

  5. 5.

    In 2002, Promundo, the PAPAI Institute (Brazil), Salud y Género (Mexico), and ECOS designed Program H (H stands for hombres, the Spanish word for men). The program is aimed at addressing GBV through a change in social norms through an education and life skills program (Promundo 2013). In 2006, these organizations and the World Education Organization designed Program M (for mujeres, meaning women in Spanish) for women. The government of El Salvador received support from Promundo and the Inter-American Development Bank in 2013 to adapt the program.

  6. 6.

    Murder victims provide an insightful measure of the relationship between gender norms and violence, even though they represent only a fraction of all victims of GBV. In 2012, half of worldwide murders of women were perpetrated by an intimate partner or a family member (United Nations 2015). In contrast, among men, one in 20 was murdered by an intimate partner or family member (United Nations 2015).

  7. 7.

    In 2008, El Salvador had the highest rate of maternal mortality in Latin America and the seventh highest in the world (UNODC 2016).

  8. 8.

    The program adaptation was conducted by the Ministry of Education, the Salvadoran Institute for the Advancement of Women (Instituto Salvadoreño de para el Desarrollo de la Mujer—ISDEMU), and the Secretariat of Social Inclusion (Secretaría de Inclusión Social—SIS) with the support of Promundo and the Inter-American Development Bank. The resulting “Do Your Part” Program (Educación Comunitaria para la Prevención de la Violencia de Género—ECPVG) was designed and implemented within the framework of a national strategy in El Salvador to prevent GBV.

  9. 9.

    More broadly, there are interventions that take different approaches to prevent violence among youth. For example, Heller et al. (2017) find that programs that help youth slow down and reflect on their automatic thoughts are effective at preventing violence. They find no supportive evidence for alternative mechanisms such as emotional intelligence, social skills, self-control, or mentoring effects.

  10. 10.

    For example, schools that participate in educational programs may tend to be precisely those that need them the most. These schools stand to gain the most from participation. A simple comparison of schools that implement this program with those that do not would underestimate the impact of the program. Schools that receive the program would not only have higher rates of violence, but also a climate that was more vulnerable to violence. A comparison would confound differences in school climates with program effects.

  11. 11.

    More specific content can be found in the program manuals (SIS et al. 2014).

  12. 12.

    To implement this component, the government sponsored training programs for 44 teachers (21 men and 23 women). Training took place during a 4-week period and included 8 educational days for a total of 64 h. The teachers received a manual with instructions on how to lead group sessions. After training and during implementation, teachers received about six visits by trainers. The visits are aimed at ensuring that teachers received the necessary support to implement the program in accordance with its design.

  13. 13.

    For example, the techniques included the “talking stick” exercise in which only the student who has the stick can speak while the others must listen. Other examples of techniques included removing oneself from the situation and using nonaggressive language.

  14. 14.

    The campaign was originally scheduled to take place during July and August 2015. Administrative delays in processes related to procurement of the required inputs led to implementation delays.

  15. 15.

    The measurement of school-level social risk relies on a set of indicators that include violence events within the school and violence in the neighborhood. It covers violence that was perpetrated against, or that affects students, teachers, or strangers. It also includes measurements of the perceptions of community members about security. The index includes the presence of gangs, weapon possession, and drug trading in the community.

  16. 16.

    The numbers were selected from a uniform discrete distribution [0, 1]. The vector of numbers was generated on the website www.random.org in May 2013. This resource generates numbers using electromagnetic noise in the atmosphere.

  17. 17.

    A common approach to address this problem is to estimate clustered errors at the school level using the bootstrap method with adjusted degrees of freedom described by Cameron and Miller (2011). However, this approach is not applicable in this case because there are only two schools in 15 out of 16 of the matched groups. Therefore, it is not possible to identify pair fixed effects.

  18. 18.

    A traditional approach consists of making a simple comparison of averages in outcomes between treatment and control schools. This approach is not appropriate here because there are only 33 schools (the unit of randomization) in the sample. Averages were chosen as the focus statistic in the analysis because doing so simplifies the interpretation of results. The student fixed effects model is considered in order to ensure that preexisting differences among students from both schools are incorporated.

  19. 19.

    The Integrated Systems Program (Sistemas integrados).

  20. 20.

    An intimate relationship is defined as an interpersonal relationship that involves physical or emotional intimacy.

  21. 21.

    The indicator is constructed based on the following assets: the number of bedrooms, toilet, shower, floor type, Internet, television, radio, cable service for television, washing machine, computer, sewing machine, and automobile. The indicator is standardized to a mean of 0 and a standard deviation of 1.

  22. 22.

    The dropout on violence report between baseline and follow up may result from trends and seasonal variations. The baseline was collected in the summer months, while the follow-up was collected in the fall. In the USA, it has been observed seasonal patterns of intimate partner violence. Violence tends to peak in the summer months. Based on data from 1993 to 2010, Lauritsen and White (2014) found intimate violence in the fall was lower in 9 percentage points than that in the summer.

Notes

Funding Information

Generous funding for the preparation of this paper was provided by the Inter-American Development Bank.

Compliance with Ethical Standards

Conflict of Interest

The authors declare that they have no conflict of interest.

Disclaimer

The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent.

Supplementary material

41996_2019_37_MOESM1_ESM.docx (163 kb)
ESM 1 (DOCX 162 kb)

References

  1. Achyut P, Bhatla N, Khandekar S, Maitra S, Verma RK. Building support for gender equality among young adolescents in school: findings from Mumbai, India. New Delhi: ICRW; 2011. https://www.icrw.org/wp-content/uploads/2016/10/GEMS-Building-Support-for-Gender-Equality-Adolescents.pdf. Accessed 15 January 2019Google Scholar
  2. Akerlof GA, Kranton RE. Identity and the economics of organizations. J Econ Perspect. 2005;19(1):9–32.CrossRefGoogle Scholar
  3. Anderson ML. Multiple inference and gender differences in the effects of early intervention: a reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects. J Am Stat Assoc. 2008;103(484):1481–95.CrossRefGoogle Scholar
  4. Athey, S., Imbens, G.W. Chapter 3 - The econometrics of randomized experiments. In: Vinayak Banerjee A, Duflo E, editors. Handbook of economic field experiments. North-Holland, Volume 1; 2017. p. 73–140.Google Scholar
  5. Barker G, Nascimento M, Segundo M, Pulerwitz J. How do we know if men have changed? Promoting and measuring attitude change with young men. In: Ruxton S, editor. Gender equality and men. Oxford: Oxfam; 2004.Google Scholar
  6. Bloom SS. Violence against woman and girls: a compendium of monitoring and evaluation indicators. U.S. Agency for International Development; 2008. http://www.endvawnow.org/uploads/browser/files/M&E%20Indicators-MEASURE-2008.pdf. Accessed 30 June 2019.
  7. Cameron AC, Miller DL. Robust inference with clustered data. In: Ullah A, Giles DE, editors. Handbook of empirical economics and finance: CRC Press; 2011. p. 1–28.Google Scholar
  8. Card D, Dahl GG. Family violence and football: the effect of unexpected emotional cues on violent behavior. Q J Econ. 2011;126(1):103–43.CrossRefGoogle Scholar
  9. Carrell SE, Hoekstra ML. Externalities in the classroom: how children exposed to domestic violence affect everyone’s kids. Am Econ J Appl Econ. 2010;2(1):211–28.CrossRefGoogle Scholar
  10. Chorot P, Sandín B. Escala de estrategias de coping revisado. Madrid: UNED; 1993.Google Scholar
  11. Cohen J, Dupas P. Free distribution or cost-sharing? Evidence from a randomized malaria prevention experiment. Q J Econ. 2010;125(1):1–45.Google Scholar
  12. De La Rue L, Polanin JR, Espelage DL, Pigott TD. A meta-analysis of school-based interventions aimed to prevent or reduce violence in teen dating relationships. Rev Educ Res. 2016:1–28.Google Scholar
  13. Donnelly PD, Ward CL. Oxford textbook of violence prevention: epidemiology, evidence and policy. Oxford: Oxford University Press; 2015.Google Scholar
  14. Duflo E. Grandmothers and granddaughters: old-age pensions and intrahousehold allocation in South Africa. World Bank Econ Rev. 2003;17(1):1–25.CrossRefGoogle Scholar
  15. Duflo E, Udry C. Intrahousehold resource allocation in Côte d’Ivoire: social norms, separate accounts and consumption choices. Cambridge: National Bureau of Economic Research Working Paper 10498; 2004.Google Scholar
  16. Fay KE, Medway FJ. An acquaintance rape education program for students transitioning to high school. Sex Educ. 2006;6:223–36.CrossRefGoogle Scholar
  17. Fernald LCH, Gertler PJ, Neufeld LM. The importance of cash in conditional cash transfer programs for child health, growth and development: an analysis of Mexico’s Oportunidades. Lancet. 2008;371(9615):828–37.CrossRefGoogle Scholar
  18. Fisher RA. The design of experiments. Second ed. Edinburgh: Oliver & Boyd; 1937.Google Scholar
  19. Heckman JJ, Humphries JE, Kautz T, editors 2014. The myth of achievement tests: the GED and the role of character in American life. Chicago: The University of Chicago Press; 2014.Google Scholar
  20. Heller SB, Shah AK, Guryan J, Ludwig J, Mullainathan S, Pollack HA. Thinking, fast and slow? Some field experiments to reduce crime and dropout in Chicago. Q J Econ. 2017;132(1):1–54.CrossRefGoogle Scholar
  21. Innoceti C. Informe sobre el levantamiento línea base: retroalimentación relativa al desarrollo de la prueba piloto. Informe para el Proyecto Modelo de Promoción Comunitaria para la Prevención de la Violencia de Género. San Salvador, El Salvador. 2014; November. Unpublished.Google Scholar
  22. International Monetary Fund (IMF). World economic outlook database; 2017. http://www.imf.org/external/pubs/ft/weo/2012/01/weodata/index.aspx. Accessed 19 January 2017.
  23. Kling JR, Ludwig J, Katz LF. Neighborhood effects on crime for female and male youth: evidence from a randomized housing voucher experiment. Q J Econ. 2005;120(1):87–130.Google Scholar
  24. Klugman J, Hanmer L, Twigg S, Hasan T, McCleary-Sills J, Santa Maria J. Voice and agency: empowering women and girls for shared prosperity. Washington, DC: World Bank; 2014.CrossRefGoogle Scholar
  25. Krug EG, Dahlberg LL, Mercy JA, Zwi AB, Lozano R, editors. World report on violence and health. Geneva: World Health Organization; 2002.Google Scholar
  26. Landa E. Análisis cualitativo del proceso de seguimiento al componente educativo: en el marco del proyecto ATN/JF-13201-ES, modelo de promoción comunitaria para la prevención de la violencia de género. El Salvador; 2015. Unpublished.Google Scholar
  27. Lauritsen JL, White N. Seasonal patterns in criminal victimization trends. Washington, DC: U.S. Department of Justice; 2014. Special Report No. NCJ245959.Google Scholar
  28. Levitt SD, Lochner L. The determinants of juvenile crime. In: Gruber J, editor. Risky behavior among youths: an economic analysis. Chicago: University of Chicago Press; 2001.Google Scholar
  29. Londoño NH, Henao López GC, Puerta IC, Posada S, Arango D, Aguirre-Acevedo DC. Propiedades psicométricas y validación de la escala de estrategias de coping modificada (EEC-M) en una muestra colombiana. Univ Psychol. 2006;5(2):327–49 Pontificia Universidad Javeriana, Colombia. http://www.redalyc.org/articulo.oa?id=64750210. Accessed 18 March 2019.Google Scholar
  30. Manski CF. Economic analysis of social interactions. J Econ Perspect. 2000;14(3):115–36.CrossRefGoogle Scholar
  31. Nanda G. Compendium of gender scales. Washington, DC: FHI 360/C-Change; 2011.Google Scholar
  32. Pacifici C, Stoolmiller M, Nelson C. Evaluating a prevention program for teenagers on sexual coercion: a differential effectiveness approach. J Consult Clin Psychol. 2001;69:552–9.CrossRefGoogle Scholar
  33. Promundo. Engaging men to prevent gender-based violence: a multi-country intervention and impact evaluation study. Washington, DC: Report for the UN Trust Fund; 2012.Google Scholar
  34. Promundo IPAPAI. Salud y Género. ECOS. Program H|M|D: a toolkit for action/engaging youth to achieve gender equity. Rio de Janeiro, and Washington, DC: Promundo; 2013. http://promundoglobal.org/wp-content/uploads/2015/01/Program-HMD-Toolkit-for-Action.pdf. Accessed 9 January 2017Google Scholar
  35. Pulerwitz J, Barker G. Measuring attitudes toward gender norms among young men in Brazil: development and psychometric evaluation of the GEM scale. Men Masculinities. 2008;10(3):322–38.CrossRefGoogle Scholar
  36. Rangel MA. Alimony rights and intrahousehold allocation of resources: evidence from Brazil. Econ J. 2006;116(5130):627–58.CrossRefGoogle Scholar
  37. Rosekrans K. Trip report: evaluation of the project community education for the prevention of gender violence in El Salvador. UC Berkeley/CEGA and the Inter-American Development Bank; 2015. Unpublished.Google Scholar
  38. Rosenbaum PR. Observational studies. New York: Springer-Verlag; 2002.CrossRefGoogle Scholar
  39. Sampson RJ. The characterological imperative: on Heckman, Humphries, and Kautz’s “the myth of achievement tests: the GED and the role of character in American life.”. J Econ Lit. 2016;54(2):493–513.CrossRefGoogle Scholar
  40. Sampson RJ, Laub JH. Life-course desisters? Trajectories of crime among delinquent boys followed to age 70. Criminology. 2003;41:555–92.CrossRefGoogle Scholar
  41. Scott J, Marshall GA. Dictionary of sociology. Oxford: Oxford University Press; 2009.CrossRefGoogle Scholar
  42. Secretaría de Inclusión Social (SIS), Instituto Salvadoreño de Desarrollo de la Mujer (ISDEMU), Ministerio de Educación (MINED), Inter-American Development Bank (IDB). ¡Cabal, dicho y hecho! Buxos contra la violencia: modelo comunitario para la prevención de violencia de género. San Salvador, El Salvador; 2014.Google Scholar
  43. United Nations. Declaración sobre la eliminación de la violencia contra la mujer A/RES/48/104, 2014. http://www.un.org/es/comun/docs/?symbol=A/RES/48/104. Accessed 28 March 2017.
  44. United Nations. The world’s women 2015: trends and statistics. New York: United Nations, Department of Economic and Social Affairs, Statistics Division; 2015. www.unstats.un.org/unsd/gender/worldwomen.html. Accessed 10 November 2016Google Scholar
  45. United Nations Development Program (UNDP). Human development report 2015: work for human development. Statistical annex. UNDP; 2015. http://hdr.undp.org/sites/default/files/hdr_2015_statistical_annex.pdf. Accessed 20 January 2017.
  46. United Nations Office on Drugs and Crime (UNODC). Global study on homicide 2013 trends, contexts, data. Vienna: UNODC; 2014.Google Scholar
  47. United Nations Office on Drugs and Crime (UNODC). UNODC statistics. Vienna: UNODC; 2016. https://www.unodc.org/documents/data-and-analysis/statistics/Homicide/Homicides_by_sex.xls. Accessed 5 January 2017Google Scholar
  48. Vivo S, Saric D, Muñóz L-P, McCoy S, Bautisa Arredondo S. A toolkit for the measurement of youth risk behavior. Washington DC: IDB Social Protection and Health Division, Inter-American Development Bank; 2013.Google Scholar
  49. Wolfe DA, Crooks CV, Jaffe P, Chiodo D. A school-based program to prevent adolescent dating violence. Arch Pediatr Adolesc Med. 2009;163(8):692–9.CrossRefGoogle Scholar
  50. World Health Organization (WHO). Global status report on violence prevention 2014. Geneva: WHO; 2014. http://www.who.int/violence_injury_prevention/violence/status_report/2014/report/report/en/. Accessed 3 January 2017Google Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  1. 1.Inter-American Development BankWashington, DCUSA
  2. 2.Inter-American Development BankSan SalvadorEl Salvador
  3. 3.University of CaliforniaBerkeleyUSA

Personalised recommendations