Content uploaded by Russell Smyth
Author content
All content in this area was uploaded by Russell Smyth on Jul 09, 2022
Content may be subject to copyright.
She can fight her own battles:
Experimental evidence on the effects of increasing awareness about rights
and self-defense training on female empowerment in India
*
Chitwan Lalji
†
, Debayan Pakrashi
‡
, Sarani Saha
§
, Soubhagya Sahoo
**
and Russell Smyth
††
9th July 2022
Abstract
Crimes against women have increased in India. While laws protecting women’s rights exist,
under-reporting of crimes is widespread, reflecting a lack of awareness of those laws.
Employing an experiment in Bihar, we examine whether treatments designed to increase
awareness and awareness coupled with self-defense training among adolescent girls can
improve knowledge about their rights; their ability and confidence to fend off physical and
sexual assault; their health and well-being; their hopes for the future and their intentions with
respect to their education and participating in the labor force. For a subsample of participants
in grades 10 and 12, we also examine the effect of the treatments on their examination results
and whether they are more likely to pursue studies in commerce and science. In general, we
find large treatment effects on all these outcomes at endline. In addition, we find large spillover
effects on the treated adolescent’s friends and siblings. We also find that the positive effects on
both treatment groups for most of the outcomes persist six months after endline.
Keywords: awareness, self-defense training, motivation, hopes, aspirations.
JEL classification: C54, C9, J16, J21.
*
Ethical clearance for this project was received from the Indian Institute of Technology Kanpur Ethical clearance
committee, with IEC Communication No: IITK/IEC/20l9-20/I/11. This project was funded by the International
Growth Centre Bihar. We thank Abhishek Yadav and Binay Shankar for their excellent fieldwork and research
assistance.
†
Economics Area, Indian Institute of Management Kozhikode. Email: chitwan@iimk.ac.in
‡
Department of Economic Sciences, Indian Institute of Technology Kanpur. Email: pakrashi@iitk.ac.in
§
Department of Economic Sciences, Indian Institute of Technology Kanpur. Email: sarani@iitk.ac.in
**
Department of Economic Sciences, Indian Institute of Technology Kanpur. Email: ssahoo@iitk.ac.in
††
Department of Economics, Monash University, Clayton, Australia. Email: russell.smyth@monash.edu
2
1 Introduction
There has been a sharp increase in crimes against women in many developing countries (see e.g.
Alesina et al., 2021; Hoehn-Velasco et al., 2021). This is particularly true for India. In 2019, there
was one rape recorded every 16 minutes with nearly 88 rape cases filed daily in India (NCRB,
2021). In Bihar, the state in which our study is situated, crimes against women almost doubled
between 2001 and 2020 (Bihar Police, 2020). Bihar ranked second among all Indian states in
terms of the number of cases involving kidnaping and abduction of women with most of these
involving kidnapping with an intent to marry forcefully (NCRB, 2021). The majority of these
abduction cases involved adolescent girls, under 18 years. The actual numbers are believed to be
considerably higher than those recorded, primarily due to high levels of under-reporting of
gender-based violence in Bihar (State Fact Sheet Bihar (NFHS-5), 2020).
One of the reasons for such underreporting could be the lack of awareness about what constitutes
a crime against women or sexual harassment and about the various laws that exist to protect
women against such crimes. Several laws have been enacted to protect women from abuse,
assault and harassment.
1
Steps have been taken to make it easier to file a complaint and make the
criminal justice system more navigable and empathetic towards victims.
2
Other initiatives, such
as opening women police stations in some parts of India has resulted in an increase in reports of
domestic violence with victims of domestic abuse feeling more comfortable reporting crimes in
such settings (Amaral et al., 2021). Yet, despite initiatives such as this, there continues to be a
general lack of awareness among women about laws protecting their rights.
3
Crimes against women have been found to have adverse implications for the victims of such
crime, including higher incidence of anxiety and depression, reduced labor force participation,
lower wages and fewer marriage opportunities (Finkelhor et al., 1989; Koss, 1993; Lloyd &
Taluc, 1999; Sadler et al., 2000; Paolucci et al., 2001; Black et al., 2011; Rees & Sabia, 2013;
Sabia et al., 2013). Such crimes also increase safety fears among women in general. The fear of
being a victim of crime makes women more reluctant to leave the house ( Gordon & Riger, 1991;
Garcia-Reid, 2007). In India, the actual, and perceived, threat of being attacked outside the home
1
Examples are the Protection of Children from Sexual Offenses, 2012, The Sexual Harassment of Women at
Workplace (Prevention, Prohibition and Redressal) Act, 2013, Protection of Women from Domestic Violence Act,
2005 and The Criminal Law (Amendment) Act, 2013.
2
Examples are providing victims with a Protection Officer, providing shelter homes to women suffering from
domestic violence, having policewomen present at the time of the investigation, allowing for the filing of complaints
online, relaxing restrictions that the complaint has to be filed within a day of the crime, providing independent legal
assistance if required and provision of free health check-ups and medical examination.
3
See Tackling Violence Against Women: A Study of State Intervention Measures (2017)
3
deters women from working outside the home (Sudarshan & Bhattacharya, 2009; Chakraborty et
al., 2018), reducing women’s labor force participation (Mishra et al., 2021).
In this study, we conduct a randomized controlled trial (RCT) with secondary school girls, in
which participants were randomly assigned to one of two treatment groups or the control group.
Participants in the first treatment group (referred to as the ‘awareness only’ treatment) attended
a workshop designed to increase knowledge about crimes against women and how to address
them, while participants in the second treatment group (‘awareness & training’ treatment)
received self-defense training in addition to participating in the awareness workshop. Participants
in the control group did not receive any intervention. In addition to conducting surveys to
examine the effects at the end of the intervention, we also conducted a follow-up survey six
months after the endline survey to understand whether the treatment effects persist over time.
We examine if providing these adolescent girls with (a) knowledge about crimes against women,
the relevant laws and how to protect themselves; and (b) in addition to providing this information,
self-defense training increases awareness of their rights and make them more confident to defend
against an assailant. We also examine if the interventions improve the girls’ wellbeing, increase
the girls optimism about the future and make them more willing to complete secondary school,
pursue higher education and participate in the labor force. For a subsample of participants, we
examine if the interventions effect the stream of study in which they enroll and the grades realized
in their final semester exams. We also investigate whether the treatments have spillover effects
from participants to their best friend and to siblings, residing in the same household.
We find that both the interventions significantly improve the confidence level of participants,
make them more optimistic about the future and more likely to intend to complete schooling,
graduate and work. Treated individuals were also found to be more aware of the rights of women
and laws intended to protect them, relative to those in the control group. We find strong spillover
effects with respect to awareness on the friends and siblings of the treated participants. We also
find that the effects of the intervention persisted six months after endline among participants from
both treatments. Furthermore, 6.4% of the participants in the awareness only and 13.6% of the
participants in the awareness and training treatment groups were more likely to continue
education during the follow-up survey administered in the next academic year than the control
group. Participants exposed to both the treatments were also found to be more likely to opt for
STEM or professional career paths in their pre-college education. The intervention also improved
their overall academic performance across subjects during both grades 10 and 12.
4
To the best of our knowledge, this is the first rigorous study to employ a RCT to present causal
estimates for alternative pathways to improve female safety in the context of a developing
country, with a focus on understanding spillover effects to participants’ social networks.
Our contribution is related to some studies in psychology that have found that self-defense
training for women is associated with higher self-defense self-efficacy (Hollander, 2004, 2014;
Senn et al., 2017; Jordan & Mossman, 2018), greater assertiveness and self-esteem and reduced
fear and anxiety (see e.g. Ozer & Bandura, 1990; McDaniel, 1993; McCaughey, 1997; Weitlauf
et al., 2000; Hollander, 2004; Brecklin & Ullman, 2005; Brecklin, 2008; Orchowski et al., 2008).
Some of these studies have employed an experimental design to draw causal inference, but these
studies typically have small sample sizes. Other studies only observe the women who have
participated in the training, leading to self-selection bias. Most of this literature has focused on
self-defense training in developed countries. We differ from this literature in other ways as well.
First, while these studies focus just on self-defense training, we seek to identify the most effective
intervention between two different treatments: (1) awareness of the various laws and how to
respond to various threatening situations and (2) awareness and self-defense training. Second, in
addition to examining the effects of the interventions on self-confidence, we consider the effects
on the long-term hopes and aspirations of the participants and intentions with respect to important
socio-economic outcomes that fear of crime is known to influence. Third, we measure the effect
of the interventions on objective outcomes, such as career choices and academic performance,
measured by performance in senior secondary school exams. Fourth, we seek to understand the
spillover effects of the interventions on awareness among participants’ social networks.
The rest of the paper is organized as follows. Section 2 presents the experimental design and data,
while Section 3 outlines the timeline of the experiment. Section 4 discusses the outcome variables
used in this study and the empirical methodology that we use in this paper. Section 5 then
discusses the results, and, finally, we conclude the paper in Section 6.
2 Experimental design
We conduct a RCT with 690 girls enrolled in grades 7 to 12 in 2020-2021 selected from 60
different schools across 20 different regions in the Patna district of Bihar, India.
4
From each
4
For the purpose of our project, we restricted our sample to adolescent girls with an android phone (with facilities
for online classes). We used this eligibility criterion to select students from all villages/regions and then randomly
assigned students and villages to the treatment and control groups. Overall, 95% of the girls in the villages that we
selected had access to android phones.
5
region, adolescent girls were randomly selected from the census/region roster. The 60 schools
from the 20 regions were then randomly assigned to one of two treatment groups or the control.
Figure A1 in Appendix A shows the location of the 20 regions in the district of Patna in Bihar.
Participants in the awareness only treatment participated in a workshop designed to increase
awareness about crimes against women, the laws designed to protect women and awareness about
how to defend oneself. In addition to participating in the awareness workshop, participants in the
awareness & training treatment also received self-defense training. The control group did not
receive any intervention. There were 243 girls from seven regions in the awareness-only
treatment group, 245 girls from seven regions in the awareness and training treatment group and
202 girls from the remaining six regions in the control group.
5
We administered a census and
baseline survey before the interventions. We administered an endline survey and a survey to
capture spillovers to participants’ social networks following the intervention. We also conducted
a follow-up survey six months after the endline survey to examine if the treatment effects persist
over time. Interestingly, no participants dropped out between the baseline and endline surveys.
Participants in the awareness only treatment received the awareness training using a well-
designed awareness module. The awareness module was first administered over the phone by
trained female para-counsellors, with written support material and videos subsequently shared
with the participants. During the awareness module, participants were made aware of different
types of harassment that women face, the laws designed to protect women against harassment,
the importance of reporting crimes to the police and the methods for doing so, as well as the
plausible ways to protect themselves. The awareness module included videos explaining child
sexual harassment, inspirational videos of girls fighting back against harassment and the
procedure to make pepper spray using ingredients available at home. The module also contained
the success stories of famous women. More generally, in this module participants were
encouraged to overcome their fears, aspire to be independent and pursue further studies.
In addition to the awareness module, which was provided to participants in the first treatment,
participants in the awareness & training treatment received a well-designed self-defense training
program. The self-defense training classes included exercises to improve strength, build stamina
and provided participants with training in extensive self-defense techniques designed to combat
severe physical and sexual threats. Initially, the plan was to provide the self-defense training
5
While randomly assigning the 20 regions and 60 schools to either of the three groups (the two treatment groups
and a control group), we ensured that no student from one group was enrolled in a school located in a region assigned
to another group. This helped eliminate spillover effects across the three groups.
6
module in person. Due to COVID-19, and subsequent nationwide movement restrictions in India
resulting in schools being closed, the training class was conducted online over the phone.
6
3 Timeline and surveys
The study was conducted in five stages over a period of one year between August 2020 and July
2021. The baseline survey was conducted in August and September 2020. Both interventions–
involving the awareness and the self-defense training modules – were administered from the third
week of October 2020 to the third week of November 2020. One month post the interventions,
the endline survey was administered in December 2020. The spillover survey ran one week after
the endline survey and was completed in January 2021. The follow-up survey was then
administered in July 2021, approximately six months after the endline survey (see Figure 1).
The main purpose of the baseline survey was to collect participants' responses to questions about
their perception of safety and security, their confidence to fight back in situations in which they
were physically or sexually assaulted, their hopes and aspirations for the future, their intentions
with respect to finishing school, college, and joining the workforce. The endline survey used
eleven multiple-choice questions designed to assess knowledge gained during the interventions.
The data collected during the endline survey also allowed us to examine the effect of the
interventions on several outcomes, such as knowledge levels, confidence levels (to fight back),
intention to complete schooling, graduate and participate in the labor force.
In the spillover survey, we surveyed 197 siblings (siblings of participants residing with them in
the same household) and 504 best friends of participants (those with active mobile numbers) to
examine the extent of the spillover effects with respect to awareness on others close to the
participant that were not directly targeted via either of the interventions. The participants were
not informed beforehand that their friends or siblings would also be surveyed.
A follow-up survey was administered about six months after the endline survey to a randomly
selected sub-sample of participants (348 out of 690 participants) during which information on
outcomes similar to those captured during the endline survey were collected. The data collected
during the follow-up survey allowed us to study the persistence of the treatment effects over time.
6
The self-defense training was conducted via real time online video classes by trained female instructors. Note that
participants were very comfortable using their android phones for the self defense training since all of their classes
had been conducted online since the COVID-19 school closure. The self-defense module took approximately two
hours per day spaced over five days and was divided into three parts: (a) warm-up and revision, (b) main training,
and (c) cooling-down exercises. The content of both interventions is provided in Appendix D.
7
4 Data and Empirical Methodology
We use the following regression specification to estimate the treatment effects:
(1)
where is a range of outcomes for participant residing in region during the post-treatment
period. is an indicator for the awareness only treatment and is an indicator for the awareness
& training treatment. Our reference group is the control group.
The control variables () include the age of the participant, their grade level, their location, the
nature of their family structure, their caste category, father’s education, mother’s education,
father’s employment, mother’s employment, birth order (among siblings) and income level of
their family (relatively rich) compared to others in the village.
We estimate the intent to treat (ITT) effect, which is an average of the causal effects of receiving
the treatment. If the interventions are effective, then and will both be positive and
significant. When available, we also control for the baseline measure of a particular outcome,
. We also include region level fixed effects (
Finally, the term indicates the error term.
We cluster the standard errors at the unit of our randomization, which is by region.
We study the effect of the interventions on 19 separate outcome variables grouped under several
specific categories. The specific survey questions used to construct the outcome variables and the
procedure to construct each variable are presented in Table A1 and discussed in detail in
Appendix B. The first outcome focuses on whether the interventions improve awareness about
the rights of women. The second set of outcome variables is concerned with confidence to defend
against an assailant in the event that the participant was attacked. A third set of outcomes focus
on the participant’s hopes for the future and intention to finish school, pursue higher studies and
participate in the labor market. A fourth set of outcomes focus on the health and wellbeing of
participants, measured by self-reported mental health and happiness.
A final set of outcomes examine seven objective educational and career choice outcomes for
participants who were enrolled in grades 10 and 12 during the intervention. These outcomes were
examination scores in Hindi, English, Mathematics, Science and an average of all examination
scores and, for participants in Grade 10, whether they elected to pursue the STEM or professional
streams. Information on awareness, current enrolment status and educational outcomes were not
collected during the baseline survey. Data for each outcome variable, except for the current
8
enrolment status and educational outcomes, were collected during the endline survey. The
educational outcome data, which were collected in the follow-up survey, allowed us to study the
effect of the treatments on academic performance and future career choices.
7
Of the 19 outcome variables that we considered, 8 were constructed using response scales to
single questions from the survey. Three outcome variables: (a) the knowledge scale measuring
awareness; (b) ability and confidence to defend against physical assault; and (c) ability and
confidence to fight back and neutralize an assailant are indices constructed by aggregating
responses to several individual questions (see Appendix B for a detailed discussion of the variable
construction). Of the seven educational outcomes, five are based on grades received in the final
semester exam and two denote career choices based on subjects selected following grade 10 i.e.,
transition from grade 10 to 11, collected during the follow-up survey.
Finally, current enrolment status is measured using a binary response and is only collected during
the follow-up survey. All outcome variables, except current enrolment status, STEM and
Professional, have been control group-standardized following (Kling et al., 2007) so that each
variable has a mean 0 and standard deviation 1 for the control group.
5 Results
5.1 Summary statistics and balance check
We begin the results section by reporting the summary statistics of the outcome variables by
treatment type at endline and follow-up in Table A2 for the full sample. Table A3 gives the
summary statistics of the educational outcomes of participants who were enrolled in 10th or 12th
grade during the intervention (and had just completed their 10th or 12th grades at the time of the
follow-up survey). Tables A4 and A5 present summary statistics of the individual characteristics
and outcome variables measured at baseline, respectively, by treatment type. As can be seen, our
7
Students enrolled in grades 10 and 12 complete board examinations at the end of the academic year and grades are
awarded by the relevant board (State board, CBSE, or ICSE). Following their grade 10 board exams students then
opt to enroll in arts, commerce, or science pre-college streams (for grades 11 and 12). This determines the subjects
in which they enroll in their final two years of secondary school. For example, students who enroll in the science
stream study physics, chemistry, mathematics, and biology with compulsory language subjects in English and Hindi.
The pre-college stream in which students enroll is crucial for determining the higher education options open to
students. For professional courses in higher education, like engineering (Bachelor of Technology) or health care
(Bachelor of Medicine or Nursing), students must enroll in the science stream for their pre-college education. During
the follow-up survey, we collected data for 131 participants in grade 10 and 39 participants in grade 12 on the marks
received in the respective board examinations. For participants who were enrolled in grade 10, we also collected
information on the stream that they selected for pre-college education during the follow-up survey.
9
treatment and control groups are well balanced in terms of individual characteristics (Table A4)
and the various baseline measures of outcomes (Table A5).
In both Table A4 and Table A5, we report the means and standard deviations of characteristics
and baseline outcomes of our sample. To derive p-values on tests of equality of means across
treatment and control groups, we regress the variable of interest on the treatment variable (equal
to 2 if in the awareness and training treatment, 1 if in the awareness only treatment and 0
otherwise) with region fixed effects and standard errors clustered at the region. We also regress
the variables of interest (at baseline) on the treatment variable (equal to 1 if in the awareness and
training treatment and zero if in the awareness only treatment) to verify the balance between the
two treatment groups. All the balance tests are done using the control group as a reference in
columns 4-5 and the awareness only group as a reference in column 6 of Tables A4 and A5.
For our sample, the balance test results for individual and household characteristics like age,
class, caste, birth order, family type (joint or nuclear), education and employment status of
parents, household size, and family income are not statistically different from the distributions
for the control group (p > 0.10) (Table A4). In addition, the balance test results are not statistically
different between the two treatment group distributions (p > 0.10).
In terms of outcomes at baseline, our samples are well balanced across all 10 outcomes (shown
in Table A5). The treatment group distributions are not statistically different from the control
group (p > 0.10). The balance test results are not statistically different between the two treatment
group distributions (p > 0.10), except for intention to complete schooling (p-value = 0.03) and
intention to graduate (p-value=0.09). Thus, of 30 tests, only two yield a p-value less than 0.10.
5.2 Effect on primary outcomes at endline survey
Table 1 reports the effects of the intervention on awareness of the rights of participants, their
ability and confidence to defend against physical assault, sexual assault and to fight back and
neutralize an assailant, the participant’s hopes for the future, as well as their intentions with
respect to completing schooling, graduating college, the highest level of education that they
intend to pursue and participation in the labor market. Column 1 reports the ITT estimates for the
awareness only treatment group and column 2 reports estimates for the awareness and training
treatment group. We check if the coefficients for the awareness only and awareness and training
10
treatments in column 1 and 2, are significantly different and the results for a Wald test
8
are
reported in Column 3. We also report the RI p-values
9
and Multiple hypothesis testing p-values
10
in Columns 4 through 7. Figure 2 shows the effect size of the intervention at the endline survey.
11
Awareness. Panel A of Table 1 reports the impact of the intervention on participants' knowledge
of different types of harassment faced by women, existing laws for protecting women against
harassment, the importance of reporting crimes, the avenues for so doing and measures that could
be employed to reduce the likelihood of being a victim of crime.
We find that participants' awareness improved significantly due to the intervention. Specifically,
participants in the awareness only treatment experienced a 3.73 standard deviation (SD)
improvement in the knowledge index (p < 0.01), relative to participants in the control group,
while for participants in the awareness and training treatment, there was a 3.97 SD improvement
in the knowledge index (p < 0.01), relative to participants in the control. We find our results to
be consistent using the randomization-based inference (RI) and Westfall-Young (WY)
adjustments (FWER-adjusted p-value). Although the estimates for awareness and training are
higher than awareness only, the Wald test statistics suggests that the impact of the intervention
on both the treatment groups is not significantly different (Table 1, Column 3). This finding
reflects that both the treatment groups received the same awareness program.
Ability & confidence to defend oneself. Table 1 (Panel B) reports the impact of the intervention
on participants’ ability and confidence to defend against physical assault, sexual assault and to
fight back and neutralize an assailant. Participants perform better on all three measures due to
the intervention. Participants in the awareness only treatment experienced a 2.82, 2.59, and 2.53
SD improvement in ability and confidence to defend against physical assault, sexual assault and
to fight back and neutralize an assailant, respectively in comparison to the control group (p <
0.01). Similarly, participants in the awareness and training treatment group experienced a 3.57,
8
We use the Wald test to test for the equality of the two estimated treatment effects, where the null is equal treatment
effect of both the interventions, for each of the variables of interest.
9
To capture the uncertainty in our estimates that comes naturally from the random allocation of participants into the
treatment groups, we also report p-values using randomization-based inference (RI) following Young (2019) that
test the null that the placebo coefficients are similar to the actual coefficients. These are constructed by randomly
shuffling the treatment groups and re-estimating our model using this placebo assignment 1,000 times. Results
reported in the following section are largely robust to using this method.
10
We have many outcomes of interest and two treatments which makes it essential to correct p-values for each
outcome that we test. We performed a robustness check of our results using the Westfall-Young (WY) adjustments
to account for joint correlation across different hypothesis tests (Westfall & Young, 1993). We report the adjusted
p-values (with 1,000 replications) of the Wald joint and Westfall-Young multiple hypothesis test of statistical
significance for equations with multiple treatment effects. Our results are consistent using this method.
11
The results are robust to the use of school fixed effects and the results were found to be similar.
11
3.59, and 3.94 SD improvement in ability and confidence to defend against physical assault,
sexual assault and to fight back and neutralize an assailant, respectively, in comparison to
participants in the control group (p < 0.01) (see Figure A2). The impact of the intervention on the
three outcomes is significantly different between treatments, with the estimates for the awareness
and training treatment significantly higher than for the awareness only treatment.
Intentions & hope. We report the effect of the intervention on the future hopes of participants
and their intentions with respect to completing schooling, graduating college, highest level of
education pursued and participating in the labor market in Panel C of Table 1. The intervention
has a significant positive effect on each of these measures (see Figure A3).
With respect to the future hopes of participants, we find an improvement of 2.36 and 4.01 SD for
the awareness only and awareness and training treatments (p < 0.01), respectively, relative to
participants in the control group. The impact of the intervention on both the treatment groups is
significantly different (p < 0.01) with the magnitude of the effect for participants in the awareness
and training group considerably larger than that in the awareness only group.
Participants in the awareness only treatment experienced a 1.48 and 2.09 SD improvement in
intention to complete school and graduate from college (p < 0.01), respectively, in comparison
to participants in the control group. Participants that received the awareness and training
treatment experienced a 3.10 and 3.53 SD improvement in intention to complete school and
graduate college (p < 0.01), respectively, relative to those in the control group. The impact of the
intervention on both the treatment groups is significantly different (p < 0.01), with the estimates
for participants in the awareness and training treatment considerably higher (> 1 SD) than those
in the awareness only treatment. With regard to the highest level of education one wishes to
pursue, we find significant improvement for both the treatment groups. The highest level of
education one intended to pursue improved 0.96 SD for participants in the awareness only (p <
0.01) and 1.44 SD for participants in the awareness and training (p < 0.01) treatment groups,
relative to participants in the control group, and are significantly different (p < 0.05).
Finally, we consider the effect of the intervention on intention to participate in the labor market.
We find that participants that received the awareness only treatment experienced a 1.35 SD
improvement in intention to participate in the labor force (p < 0.01), while participants in the
awareness and training treatment group experienced a 2.49 SD improvement (p < 0.01) in this
measure, in comparison to participants in the control group. The impact of the intervention on
both treatment groups is significantly different (p < 0.01).
12
Health and wellbeing. We report the results for mental health and happiness in Panel D of Table
1. We find significant improvement in the mental health and happiness of participants in both
treatment groups. Specifically, the mental health of participants in both treatments improves ~2
SD compared to the control group. The impact of the intervention on the happiness of participants
in the awareness and training treatment group is double that of participants in the awareness only
treatment group and significantly different (p < 0.01) (see Figure A4).
5.3 Effect on primary outcomes at follow-up
For the follow-up survey, we randomly selected a representative sample of 348 participants. We
report the mean and standard deviation of the outcome variables at baseline (and endline) of the
follow-up sample and non-follow-up sample in Table A6. Moreover, we also report the mean and
standard deviation of the individual characteristics at baseline of the follow-up sample and non-
follow-up sample in Panel B of Table A6. We test the equality between means (t-test p-values)
of the outcome variables at baseline (and endline) and individual characteristics of follow-up and
non-follow-up samples and report the results in columns 7 and 8 of Table A6. The follow-up
sample is found to be balanced with the non-follow-up sample.
We also report the balance test of the baseline individual characteristics in the follow-up sample
over three treatment groups. Table A7 reports the balanced summary statistics for individual
characteristics and Table A8 reports the baseline outcome variables, respectively. We use the
same method used in Table A4 and Table A5 for the balance check between different treatment
groups. We find that the treatment group distributions are not statistically different from the
control group (p > 0.10), except for mental health, which yields a p-value of 0.07 for the
awareness only treatment group. The balance test results are not statistically different between
the two treatment group distributions (p > 0.10), except for defending against sexual assault,
which yields a p-value of 0.08, and the happiness scale, which yields a p-value of 0.01.
We next discuss results from the follow-up survey, conducted six months after the endline survey.
We report the results in Table 2 and Figure A7. We continue to find a large and significant effect
for awareness for both the treatment groups in the follow-up survey, although the effect sizes
decline compared to endline. Participants that received the awareness only treatment are found
to have a 1.89 SD improvement in the knowledge index (p < 0.01), while participants that receive
the awareness and training treatment have a 1.89 SD improvement (p < 0.01), relative to
participants in the control group. Consistent with the findings at endline, the impact of the
intervention on both the treatment groups is not significantly different (p > 0.10).
13
Of the three self-defense measures, only ability and confidence to defend oneself against sexual
assault was surveyed at the follow-up. We find a significant improvement of 1.13 SD and 1.40
SD on ability and confidence to defend oneself against sexual assault for the awareness only and
awareness and training treatment groups, respectively, in the follow-up survey, relative to
participants in the control group. Hence, the effect sizes are smaller than the endline survey and
the impact of the intervention on both treatment groups is not significantly different (p > 0.10).
The effect of the intervention on hope for the future and intention with respect to completing
schooling, graduating college, highest education pursued and participating in labor market are all
found to be persistent after six months. We find a significant improvement of 0.77 SD and 1.28
SD for hope for the future, 1.14 SD and 1.26 SD for intention to complete schooling, 1.10 SD
and 1.96 SD for intention to graduate from college, 1.15 SD and 1.52 SD for highest level of
education one wishes to pursue, and 1.32 SD and 1.46 SD for intention to participate in the labor
market for the awareness only and awareness and training treatment groups, respectively, in the
follow-up survey, relative to participants in the control group. While the effect of the two
treatments on schooling and labor market intentions persist over time, there is an overall decline
in the effect size on these outcomes compared to the endline survey.
There is a significant improvement for both self-reported mental health and happiness for
participants in both the treatment groups in the follow-up survey. We find a significant
improvement of 0.35 SD and 1.11 SD for mental health, and 1.20 SD and 1.29 SD for the
happiness index for the awareness only and awareness and training treatment groups,
respectively, in the follow-up survey, relative to participants in the control group. While the effect
size declines, the effects continue to be significant at a p-value less than 0.01.
Finally, we collected information on the effect of the intervention on whether participants were
more likely to continue their education the following academic year.
12
We find that 6.4% and
13.6% of the participants in the awareness only and awareness and training treatment groups,
respectively, were significantly more likely to enroll in school or college than the control group
(p < 0.01), reinforcing the findings for intention to complete schooling and graduate.
12
The intervention was implemented in October-November 2020, while the follow-up survey was conducted during
the next academic year (July 2021).
14
5.4 Effect on educational outcomes for participants in grades 10 and 12 at follow-up
In this section, we discuss the impact of the intervention on the educational outcomes and future
career choices of participants in grades 10 and 12, six months after endline. The results are in
Table 3. We also report mediation results, in which hope for the future is a channel. Figure 3 and
Figure A5 graphically show the effect size of both the treatments on educational outcomes.
In Panel A of Table 3, we examine the impact of the intervention on participants’ scores in Hindi
and English, as well as the average marks across all subjects in the grades 10 and 12 exams.
Participants in both treatments performed better than those in the control across the board. For
Hindi, participants in the awareness only (awareness and training) treatment performed 1.48 SD
(p < 0.01) (2.72 SD (p < 0.01)) better than participants in the control group. For English,
participants in the awareness only (awareness and training) treatment performed 1.98 SD (p <
0.01) (2.72 SD (p < 0.01)) better than participants in the control group and for overall
performance, participants in the awareness only (awareness and training) treatment performed
2.81SD (p < 0.01) (2.94 SD (p < 0.01)) better than participants in the control group
In Panel B of Table 3, we report the impact of the intervention on the results of participants in
grade 10 in Mathematics and Science and whether they subsequently enrol in the STEM or
professional pathways in grade 11. For both subjects participants in the treatment groups perform
better than those in the control group. While the effect size for the awareness and training
treatment is larger than awareness only, the difference is insignificant.
To examine the channel through which the intervention affects these educational outcomes, we
control for future hope at the endline as a mediator and report the results in columns 3-5 in Table
3. We find that an improvement in hope about the future due to the intervention positively affects
performance in Hindi, English, Mathematics and overall performance (p < 0.10).
With respect to career choice, participants in the awareness only (awareness and training)
treatment are 62.20% (97.70%) more likely (p < 0.01) to end up pursuing a career in STEM.
Similarly, participants in the awareness only (awareness and training) treatment are 67.60%
(92.60%) more likely (p < 0.01) to end up pursuing a professional career, than those in the control.
5.5 Social desirability bias
Social desirability can bias the treatment effects on self-reported outcomes, such as confidence
and future aspirations. For instance, larger social desirability bias in the treatment groups, due to
15
experimenter demand effects, may lead to an upward bias in effect sizes for self-reported
outcomes. To examine this issue, we included the social desirability scale, following Crowne &
Marlowe (1960); Reynolds (1982) and Dhar et al. (2018), in the baseline survey. The social
desirability scale measures a person's tendency to give socially desirable answers.
13
We also
construct a low socially desirable score dummy variable using the median as the cut-off. To test
whether the treatment effects are biased, we report the heterogeneity analysis based on the low
social desirability score dummy in Table 4. For the treatment effects in the follow-up survey, we
report the results in Table A10 in the Appendix.
We find that positive treatment effects for self-reported confidence and motivations are similar
for participants with a low and high tendency for social desirability bias, except for one instance
in the endline survey. We find that participants in the treatment groups with low social desirability
scores have a positive and significant effect (p < 0.10) on ability and confidence to defend against
sexual assault. Thus, there is some upward bias in the treatment effect for this outcome variable
in the awareness and training treatment group, compared to participants in the control group. We
find similar results for treatment effects in the follow-up survey except for intention with respect
to highest level of education and intention to participate in labor force, which have negative and
positive significant effects (p < 0.10), respectively.
We also checked for social desirability bias in treatment effects on ability and confidence to
defend oneself, hopes for the future and intentions with respecxt to education and work by
trimming the extreme responses in the treatment arm only as well as in both treatment and control
arms. The description of the trimming method and the results are reported in Appendix C.2. We
find consistent and significant treatment effects for all the primary outcomes.
5.6 Spillover effects
To estimate the spillover effects of the intervention with respect to awareness levels on siblings
and best friends, we estimate the regression specification as in equation (1). The results are
presented in Table 5. We find that both treatments have a positive and significant effect on
awareness levels of the participants’ best friends and siblings. For siblings, the awareness and
training intervention has stronger effects than the awareness only intervention, while for friends,
both treatments have similar effects. The awareness and training intervention had much stronger
effects on knowledge dissemination among siblings than best friends. The relatively lower level
13
For details on the construction of the social desirability scale see Appendix C.1
16
of dissemination among friends could be due to the Indian government's restrictions on
movement to restrict the spread of COVID-19, as the data were collected during that period.
5.7 Heterogenous spillover effects
We separate out the estimated effects in Table 5 for different sub-samples and present the results
in Figure A6 and Table A11. We seek to isolate the channels through which the treatments may
affect the awareness of siblings and friends of participants who had taken part in the experiment.
Siblings. We find that treatments lead to a substantial and statistically significant increase in
awareness among both brothers and sisters of participants (Panel A of Table A11). While the
magnitude of the effect size is higher for sisters than brothers, the difference between the
respective coefficients between brothers and sisters is statistically not significant.
Similarly, both treatments lead to a substantial and statistically significant increase in awareness
among both younger and older sibling sub-samples. The magnitude of the effect sizes is greater
for younger siblings than older siblings, although the difference between the coefficients of the
older and younger siblings is only statistically significant for the awareness only intervention.
Friends. We find that both the treatments lead to a substantial and statistically significant increase
in awareness among both – caste in-group and out-group friend – sub-samples (Panel B of Table
A11). Further, the impact on awareness among friends who belong to the same caste as the
participant is greater than the impact on friends of a different caste. This may be due to a greater
affinity of participants with friends in their own caste. The coefficients on treatments in both sub-
samples are significantly different for the awareness and training intervention.
5.8 Heterogeneous treatment effects
We estimate whether treatment effects vary by mother’s education,
14
whether the participant felt
unsafe at baseline,
15
and whether the participant lacked freedom at home at baseline.
16
To do so
we estimate the following regression specification:
14
Mother non-graduate dummy variable takes the value 1 if the education level of the participant’s mother is below
college degree, and 0 otherwise.
15
To construct the feeling unsafe dummy variable, we use the participant’s perception of safety at baseline (on an
11-point Likert scale) and use the median cut-off.
16
Degree of freedom is the sum of whether the participant was allowed to go to four different kinds of places (the
market, outside the house after dark, a friend’s place, and school/college) alone, with someone accompanying them
or not allowed at all. It takes a value between three and 12 with a higher score indicating less freedom. Low freedom
is equal to 1 if the degree of freedom at home is higher than the median at baseline, and 0 otherwise. Information for
each of mother’s education, feeling unsafe, no intention to participate in the labor force and having low freedom at
home was collected in the baseline survey.
17
(2)
where
is the heterogeneity dummy (m) set equal to 1 if education level of the participant’s
mother is less than college, the participant feels unsafe or lacks freedom at home at baseline.
The results are in Table 6. We find that for participants whose mother’s education level is less
than a college degree there is significant improvement in intention to complete schooling,
intention to graduate college, intention with respect to the highest level of education attained and
intention to participate in the labor force for both treatments, relative to the control. Participants
with less educated mothers may be less aware about women’s rights at baseline and, thus, the
marginal returns to the interventions are greater for this group on the relevant outcome variables.
For participants who felt unsafe at baseline, we find that both the treatments lead to a statistically
significant improvement in the intention to complete schooling and intention to graduate college,
but only a statistically significant improvement in the intention to participate in the labor force
for the awareness and training treatment group. The marginal benefits of the awareness and
training treatment are expected to be much higher for those who felt unsafe at baseline as the self-
defense training would make these individuals feel more confident in dealing with an assailant
or when placed in risky situations outside the home.
Finally, we examine heterogeneity by the degree of freedom at home of the participants. We find
that both the treatments lead to statistically significant improvement in the intention to graduate
college and intention to participate in the labor force compared to the control group for
participants with a low degree of freedom at home at baseline. This result is expected as this
subgroup of participants would reap considerable benefits stemming from greater awareness.
5.9 Mediation analysis
Finally, we examine the channels through which the interventions affect the intention to complete
schooling, intention to graduate college, intention to participate in the labor force and happiness.
To do so, we estimate the following specification.
(3)
18
where;
is the mediator n. We consider three mediators: awareness, measured by the
knowledge index, ability and confidence to defend against sexual assualt and hope for the future.
We estimate the above specification sequentially, adding one mediator at a time to study the role
of each channel. Since we have three mediators and four outcomes of interest, we estimate model
(3) twelve times. For example, for the outcome variable intention to complete schooling, we first
employ the knowledge index as a mediator, then employ the knowledge index and ability and
confidence to defend against physical assault as mediators and finally employ all three mediators
to get the direct effects after controlling for the mediation effects.
17
We repeat this for the other
outcome variables: intention to graduate, intention to participate in the labour force and the
happiness index. In all specifications, the error terms are clustered at the region level.
In Panel A of Table 7 we report the mediation analysis where the outcome variable is intention
to complete schooling. The treatment effects decline as we control for additional mediators. The
average effect of the treatment on intention to complete schooling operates through ability and
confidence to defend against physical assault and hope for the future and both variables are
statistically significant at the 1% level. Similarly, for intention to graduate college (Panel B) and
intention to participate in the labor force (Panel C), the average effects of the treatment on
participants operates through ability and confidence to defend against physical assault and hope
for the future and both coefficients are statistically significant at the 1% level. However,
awareness is found to have no significant indirect effect.
18
Thus, overall, we find that participants'
ability and confidence to defend against physical assault and hopes for the future
19
are important
channels, while awareness, proxied by the the knowledge index, was not an effective channel.
6 Conclusion
Crimes against women are a major problem in many developing countries, including India.
Several studies suggest that fear of crime among girls and women has negative effects on female
empowerment, including loss of human capital and reduced willingness to work outside the home
(Chakraborty et al., 2018; Mishra et al., 2021; Sudarshan & Bhattacharya, 2009). For a sample
17
The results are robust to the order in which the mediators are added with only slight variation in the magnitudes
of the coefficients and no change in sign or statistical significance.
18
A Wald test of the difference between the treatment effects without any mediators and with the knowledge index
as the only mediator finds that the knowledge index is insignificant for all the outcome variables except intention to
complete schooling.
19
In Table A12 we report the OLS and 2SLS estimates for hope for the future at the endline survey, where we
instrument ‘Hope for the future’ (measured at endline survey) using the random assignment to the treatment. The
outcomes - intention to complete schooling, intention to graduate college, intention to participate in the labor force,
and happiness - in columns (1)-(4) are z-scores (with mean 0 and SD 1). We find hope for the future to be a highly
effective (p < 0.01) channel in improving each of these outcomes.
19
of adolescent females, we examine the effect of an awareness treatment and awareness and
training treatment, not only on knowledge of what constitutes a crime and how to report it, ability
and confidence to fight back against an assailant and on their health and wellbeing, but also on
aspirational outcomes - their hopes for the future and intentions with respect to schooling and
participation in the labor force. For a subsample of participants that we resurvey at follow-up, we
find that the treatment effects continue to be persistent six months after endline.
We find evidence of spillover effects with respect to awareness from treated individuals to their
social networks. We also find that the treatments have positive effects on objective outcomes for
a subsample of participants enrolled in grades 10 and 12 at the time of the intervention.
Specifically, the interventions have a positive effect on examination scores and increase the
likelihood that participants follow commerce and/or science pathways.
Many studies point to the importance of developing strategies to reduce the fear of crime. Most
of these strategies seek to ensure that adult women feel more able and confident to defend
themselves if placed in a risky situation. Our study is unique in being targeted at adolescent girls
and not only examining the effect of the interventions on their ability and confidence to fight off
an assailant, but also their career goals and hopes for the future. There is increasing interest among
economists in the returns to investing in programs in childhood and adolescence (see e.g. Conti
et al., 2016; Gertler et al., 2014; Heckman et al., 2013). Our findings suggest that including an
awareness module and self-defense training in the school curriculum may help create a pool of
confident and independent girls, who can make better informed decisions about their career goals,
such as the decision to pursue higher education, graduate college and participate in the labor
force. Our findings for the positive effects of the intervention on objective outcomes in the form
of examination results and study pathways reinforce the effects of the treatments on subjective
professional goals and hope for the future. Our results also suggest that making awareness about
rights and self-defense training part of the school curriculum can also assist to disseminate
information about women's safety among participants’ close networks, such as siblings and
friends, further increasing the pool of confident and independent girls and women.
20
References
Alesina, A., Brioschi, B., & La Ferrara, E. (2021). Violence Against Women: A Cross-cultural
Analysis for Africa. Economica, 88(349), 70–104. https://doi.org/10.1111/ecca.12343
Amaral, S., Bhalotra, S., & Prakash, N. (2021). Gender, Crime and Punishment: Evidence from
Women Police Stations in India (CESifo Working Paper No. 9002). Social Science
Research Network. https://doi.org/10.2139/ssrn.3827615
Bart, P., & O’Brien, P. (1985). Stopping rape: Successful survival strategies.
Bihar Police. (2020). Crime Data and Achievements.
http://biharpolice.bih.nic.in/menuhome/CDA.htm
Black, M., Basile, K., Breiding, M., Chen, J., Merrick, M., Smith, S., Stevens, M., & Walters, M.
(2011). National intimate partner and sexual violence survey: 2010 summary report.
Centers for Disease Control and Prevention.
Brecklin, L. (2008). Evaluation outcomes of self-defense training for women: A review.
Aggression and Violent Behavior, 13(1), 60–76.
Brecklin, L., & Ullman, S. (2005). Self-defense or assertiveness training and women’s responses
to sexual attacks. Journal of Interpersonal Violence, 20(6), 738–762.
Chakraborty, T., Mukherjee, A., Rachapalli, S. R., & Saha, S. (2018). Stigma of sexual violence
and women’s decision to work. World Development, 103, 226–238.
https://doi.org/10.1016/j.worlddev.2017.10.031
Conti, G., Heckman, J. J., & Pinto, R. (2016). The Effects of Two Influential Early Childhood
Interventions on Health and Healthy Behaviour. The Economic Journal, 126(596), F28–
F65. https://doi.org/10.1111/ecoj.12420
Crowne, D., & Marlowe, D. (1960). A new scale of social desirability independent of
psychopathology. Journal of Consulting Psychology, 24(4), 394.
https://doi.org/10.1037/h0047358
Dhar, D., Jain, T., & Jayachandran, S. (2018). Reshaping adolescents’ gender attitudes: Evidence
from a school-based experiment in India (No. W25331). National Bureau of Economic
Research. https://www.nber.org/papers/w25331
Finkelhor, D., Hotaling, G. T., Lewis, I. A., & Smith, C. (1989). Sexual Abuse and Its
Relationship to Later Sexual Satisfaction, Marital Status, Religion, and Attitudes. Journal
of Interpersonal Violence, 4(4), 379–399. https://doi.org/10.1177/088626089004004001
Garcia-Reid, P. (2007). Examining social capital as a mechanism for improving school
engagement among low income Hispanic girls. Youth & Society, 39(2), 164–181.
https://doi.org/10.1177/0044118X07303263
Gertler, P., Heckman, J., Pinto, R., Zanolini, A., Vermeersch, C., Walker, S., Chang, S. M., &
Grantham-McGregor, S. (2014). Labor market returns to an early childhood stimulation
intervention in Jamaica. Science, 344(6187), 998–1001.
https://doi.org/10.1126/science.1251178
21
Gordon, M., & Riger, S. (1991). The female fear: The social cost of rape. University of Illinois
Press, Urbana and Chicago.
Heckman, J., Pinto, R., & Savelyev, P. (2013). Understanding the Mechanisms through Which
an Influential Early Childhood Program Boosted Adult Outcomes. American Economic
Review, 103(6), 2052–2086. https://doi.org/10.1257/aer.103.6.2052
Hoehn-Velasco, L., Silverio-Murillo, A., & de la Miyar, J. R. B. (2021). The great crime
recovery: Crimes against women during, and after, the COVID-19 lockdown in Mexico.
Economics & Human Biology, 41, 100991. https://doi.org/10.1016/j.ehb.2021.100991
Hollander, J. (2004). “I can take care of myself” the impact of self-defense training on women’s
lives. Violence Against Women, 10(3), 205–235.
https://doi.org/10.1177/1077801203256202
Hollander, J. (2014). Does Self-Defense Training Prevent Sexual Violence Against Women?
Violence Against Women, 20(3), 252–269. https://doi.org/10.1177/1077801214526046
Hollander, J., & Cunningham, J. (2020). Empowerment self-defense training in a community
population. Psychology of Women Quarterly, 44(2), 187–202.
https://doi.org/10.1177/0361684319897937
Jordan, J., & Mossman, E. (2018). “Back off buddy, this is my body, not yours”: Empowering
girls through self-defense. Violence against Women, 24(13), 1591–1613.
https://doi.org/10.1177/1077801217741217
Kling, J. R., Liebman, J. B., & Katz, L. F. (2007). Experimental Analysis of Neighborhood
Effects. Econometrica, 75(1), 83–119. https://doi.org/10.1111/J.1468-
0262.2007.00733.X
Koss, M. P. (1993). Detecting the Scope of Rape: A Review of Prevalence Research Methods.
Journal of Interpersonal Violence, 8(2), 198–222.
https://doi.org/10.1177/088626093008002004
Lloyd, S., & Taluc, N. (1999). The effects of male violence on female employment. Violence
Against Women, 5(4), 370–392. https://doi.org/10.1177/10778019922181275
McCaughey, M. (1997). Real knockouts: The physical feminism of women’s self-defense. NYU
Press.
McDaniel, P. (1993). Self-defense training and women’s fear of crime. Women’s Studies
International Forum, 16(1), 37–45.
Mishra, A., Mishra, V., & Parasnis, J. (2021). The asymmetric role of crime in women’s and
men’s labour force participation: Evidence from India. Journal of Economic Behavior &
Organization, 188, 933–961. https://doi.org/10.1016/j.jebo.2021.06.015
National Family Health Survey (NFHS-4), 2015–16. (2017). International Institute for Population
Sciences (IIPS). http://rchiips.org/nfhs/NFHS-4Report.shtml
NCRB. (2021). Crime in India: 2020. National Crime Records Bureau, Ministry of Home Affairs,
Government of India. https://ncrb.gov.in/en/Crime-in-India-2020
22
Norris, J., Nurius, P. S., & Dimeff, L. A. (1996). Through her eyes: Factors affecting women’s
perception of and resistance to acquaintance sexual aggression threat. Psychology of
Women Quarterly, 20(1), 123–145. https://doi.org/10.1111/J.1471-
6402.1996.TB00668.X
Orchowski, L. M., Gidycz, C. A., & Raffle, H. (2008). Evaluation of a sexual assault risk
reduction and self-defense program: A prospective analysis of a revised protocol.
Psychology of Women Quarterly, 32(2), 204–218. https://doi.org/10.1111/J.1471-
6402.2008.00425.X
Ozer, E. M., & Bandura, A. (1990). Mechanisms Governing Empowerment Effects: A Self-
Efficacy Analysis. Journal of Personality and Social Psychology, 58(3), 472–486.
https://doi.org/10.1037/0022-3514.58.3.472
Paolucci, E. O., Genuis, M. L., & Violato, C. (2001). A meta-analysis of the published research
on the effects of child sexual abuse. The Journal of Psychology, 135(1), 17–36.
https://doi.org/10.1080/00223980109603677
Rees, D. I., & Sabia, J. J. (2013). Forced Intercourse, Mental Health, and Human Capital.
Southern Economic Journal, 80(2), 324–344. https://doi.org/10.4284/0038-4038-
2013.015
Reynolds, W. (1982). Development of reliable and valid short forms of the Marlowe‐Crowne
Social Desirability Scale. Journal of Clinical Psychology, 38(1), 119–125.
Sabia, J. J., Dills, A. K., & DeSimone, J. (2013). Sexual violence against women and labor market
outcomes. American Economic Review, 103(3), 274–278.
https://doi.org/10.1257/AER.103.3.274
Sadler, A., Booth, B., & Nielson, D. (2000). Health-related consequences of physical and sexual
violence: Women in the military. Obstetrics & Gynecology, 96(3), 473–480.
Senn, C., Eliasziw, M., & Hobden, K. (2017). Secondary and 2-year outcomes of a sexual assault
resistance program for university women. Psychology of Women Quarterly, 41(2), 147–
162. https://doi.org/10.1177/0361684317690119
State Fact Sheet Bihar (NFHS-5). (2020). Ministry of Health and Family Welfare, Government
of India.
Sudarshan, R., & Bhattacharya, S. (2009). Through the magnifying glass: Women’s work and
labour force participation in urban Delhi. Economic and Political Weekly, 59–66.
Tackling Violence Against Women: A Study of State Intervention Measures. (2017). Bhartiya
Stree Shakti, Ministry of Women and Child Development, Government of India.
Weitlauf, J., Smith, R., & Cervone, D. (2000). Generalization effects of coping-skills training:
Influence of self-defense training on women’s efficacy beliefs, assertiveness, and
aggression. Journal of Applied Psychology, 85(4), 625.
Westfall, P., & Young, S. (1993). Resampling-based multiple testing: Examples and methods for
p-value adjustment (Vol. 279). John Wiley & Sons.
23
Young, A. (2019). Channeling fisher: Randomization tests and the statistical insignificance of
seemingly significant experimental results. The Quarterly Journal of Economics, 134(2),
557–598.
24
Tables & Figures
Tables
Table 1: Treatment effects at endline
Endline
Wald Test
RI p-values
Multiple hypothesis
testing
A-Only
A&T
β(1)= β(2)
A-Only
A&T
Wald
Joint
Westfall-
Young
Outcome Variables
(1)
(2)
(3)
(4)
(5)
(6)
(7)
A. Awareness
Knowledge index
3.732***
3.969***
0.127
0.000
0.000
0.000
0.000
(0.120)
(0.070)
B. Ability & Confidence to Defend Oneself
Defend physical
assault
2.818***
3.566***
0.007
0.001
0.000
0.001
0.001
(0.230)
(0.076)
Defend sexual assault
2.592***
3.593***
0.000
0.000
0.000
0.001
0.001
(0.198)
(0.054)
Fight back &
neutralize assailant
2.533***
3.943***
0.000
0.000
0.001
0.001
0.001
(0.069)
(0.060)
C. Intention & Hope
Hope for the future
2.361***
4.099***
0.000
0.000
0.001
0.000
0.000
(0.295)
(0.071)
Intention to complete
schooling
1.483***
3.105***
0.000
0.000
0.001
0.001
0.001
(0.325)
(0.097)
Intention to graduate
college
2.095***
3.531***
0.000
0.000
0.000
0.001
0.001
(0.156)
(0.120)
Highest level of
education
0.965***
1.437***
0.044
0.001
0.000
0.000
0.000
(0.216)
(0.051)
Intention to
participate in the
labor force
1.355***
2.486***
0.000
0.001
0.000
0.000
0.000
(0.132)
(0.120)
D. Health & Happiness
Mental health
2.055***
1.996***
0.917
0.000
0.046
0.001
0.001
(0.542)
(0.083)
Happiness index
1.661***
3.209***
0.000
0.000
0.001
0.000
0.000
(0.242)
(0.067)
Observations
690
690
-
-
-
-
-
Note: A-Only and A&T are Awareness Only and Awareness & Training treatment groups, respectively. All outcomes
are standardized indices so that the control group has a mean zero and standard deviation one. The standardized
baseline index/value of the outcome variables is controlled for in all regressions except for the Knowledge index.
Columns (1) & (2) present the estimated treatment effects of A-Only & A&T, respectively. Refer to Table A1 for a
detailed description of all the outcome variables. All regressions also control for individual characteristics, such as
age (in years), grade (7th to 12th), rural dummy (rural versus other regions of residence), joint family versus other
types of family structure, caste category of the participant (General, OBC (other backward castes) and SC/ST
(Scheduled Caste/Scheduled Tribe)), father’s education (graduate and non-graduate), mother’s education (graduate
and non-graduate), father’s employment (employed and unemployed), mother’s employment (employed and
unemployed), birth order (among siblings), income level of family (relatively rich) compared to others in the village
(which takes a value 1 if prosperous, very comfortable and reasonably comfortable versus others, i.e., just getting
25
along, poor or very poor), and household size (number of members in the family). We also include region fixed effects.
Wald statistics in Column (3) indicate if the estimated coefficients associated with the A-Only and A&T treatments
differ significantly. RI-test p-values in Columns (4) & (5) are based on a two-sided randomization inference test (with
1,000 replications) for both treatments (Young, 2019). Column (6) reports p-values for the Wald joint-test of the
significance (adjusted p-value with 1,000 replications) of treatment measures in each equation (Young, 2019). Column
(7) reports Westfall-Young multiple hypothesis test of the significance (adjusted p-value with 1,000 replications) of
any treatment measure in each equation (Westfall & Young, 1993). Robust standard errors clustered at the region level
are in parentheses in columns (1) - (2). *** p<0.01, ** p<0.05, * p<0.1
Table 2: Treatment effects at the follow-up survey
Follow-up
Wald Test
RI p-values
Multiple
hypothesis testing
A-Only
A&T
β(1)= β(2)
A-Only
A&T
Wald
Joint
Westfall-
Young
Outcome Variables
(1)
(2)
(5)
(6)
(7)
(8)
(9)
A. Awareness
Knowledge index
1.891***
1.887***
0.978
0.000
0.001
0.001
0.001
(0.063)
(0.121)
B. Ability & Confidence to Defend Oneself
Defend sexual assault
1.129***
1.401***
0.137
0.001
0.004
0.001
0.001
(0.131)
(0.085)
C. Intention & Hope
Current enrolment status
0.064***
0.136***
0.004
0.195
0.425
0.318
0.318
(0.011)
(0.015)
Hope for the future
0.772***
1.277***
0.007
0.002
0.324
0.003
0.003
(0.070)
(0.132)
Intention to complete
schooling
1.140***
1.260***
0.439
0.241
0.057
0.104
0.104
(0.116)
(0.083)
Intention to graduate
college
1.105***
1.957***
0.000
0.000
0.115
0.001
0.001
(0.096)
(0.111)
Highest level of education
1.154***
1.521***
0.003
0.001
0.001
0.001
0.001
(0.086)
(0.045)
Intention to participate in
the labor force
1.322***
1.464***
0.271
0.000
0.029
0.001
0.001
(0.081)
(0.102)
D. Health & Happiness
Mental health
0.354***
1.106***
0.0037
0.018
0.466
0.035
0.035
(0.079)
(0.194)
Happiness index
1.204***
1.294***
0.563
0.000
0.015
0.001
0.001
(0.089)
(0.109)
Observations
348
348
-
-
-
-
-
Note: Refer to footnote of Table 1. All outcomes except Current enrolment status are standardized indices so that the
control group has mean zero and standard deviation one. The standardized baseline index/value of the outcome variables
is controlled for in all regression except for the Knowledge index & Current enrolment status. Current enrolment status
takes the value one if the participant is continuing education during follow-up survey, and zero otherwise. During the
follow-up survey seven students dropped out of school for various reasons, including financial difficulties and safety
concerns.
26
Table 3: Treatment effects on educational outcomes at follow-up survey
Follow-up
Mediation results
Wald Test
RI p-values
Multiple hypothesis
testing
A-Only
A&T
A-Only
A&T
Mediator
β(1) = β(2)
A-Only
A&T
Wald
Joint
Westfall-
Young
Outcome Variables
(1)
(2)
(3)
(4)
(5)
(7)
(8)
(9)
(10)
(11)
A. Students in Grade 10 & 12
Hindi score
1.477***
2.719***
1.150***
1.946**
0.185*
0.081
0.000
0.001
0.001
0.001
(0.238)
(0.649)
(0.253)
(0.831)
(0.104)
English score
1.927***
2.724***
1.638***
2.040**
0.164*
0.254
0.000
0.004
0.001
0.001
(0.260)
(0.774)
(0.266)
(0.835)
(0.092)
Overall performance
2.808***
2.940***
2.432***
2.051***
0.213*
0.805
0.000
0.001
0.001
0.001
(0.290)
(0.421)
(0.326)
(0.636)
(0.111)
Observations
170
170
170
170
170
-
-
-
-
-
B. Students in Grade 10
Mathematics score
0.730*
1.527***
1.397
1.846*
0.241*
0.179
0.000
0.003
0.000
0.000
(0.396)
(0.486)
(1.158)
(1.062)
(0.138)
Science score
1.987***
2.116***
0.487
0.490
0.087
0.78
0.006
0.136
0.012
0.012
(0.261)
(0.506)
(0.665)
(0.530)
(0.107)
STEM
0.622***
0.977***
0.371
0.655**
0.049
0.155
0.011
0.000
0.001
0.001
(0.139)
(0.266)
(0.225)
(0.261)
(0.045)
Professional
0.676***
0.926***
0.026
0.231
0.031
0.236
0.211
0.013
0.027
0.027
(0.126)
(0.202)
(0.205)
(0.194)
(0.030)
Observations
131
131
131
131
131
-
-
-
-
-
Note: Hindi, English, overall performance, mathematics and science are scores in the school board examination held after the endline survey and are standardized indices so
that the control group has mean zero and standard deviation one. STEM is an indicator variable that takes the value 1 if the participant has opted for science as a major for pre-
college education (11th grade) and 0 otherwise. Professional is an indicator variable which takes the value 1 if the participant has opted for commerce or science as major for
pre-college education (11th grade), and 0 otherwise. Columns (1) & (2): the estimated treatment effects of A-Only & A&T in the follow-up survey, respectively. Columns (3) -
(5): the estimated treatment effects of A-Only, A&T, and the Mediator (endline Hope for the future) in the follow-up survey, with all covariates, respectively. Refer to footnote
of Table 1.
27
Table 4: Robustness check for social desirability bias at endline
Variables of interest
A-Only
A&T
LSDS
A-Only ✕ LSDS
A&T ✕ LSDS
Outcome Variables
(1)
(2)
(3)
(4)
(5)
A. Awareness
Knowledge index
3.468***
3.869***
-0.141
0.213
0.158
(0.125)
(0.171)
(0.149)
(0.159)
(0.153)
B. Ability & Confidence to Defend Oneself
Defend physical assault
3.311***
3.906***
-0.106
0.018
0.026
(0.109)
(0.237)
(0.146)
(0.150)
(0.160)
Defend sexual assault
2.751***
3.734***
-0.183
0.085
0.249*
(0.120)
(0.195)
(0.128)
(0.148)
(0.134)
Fight back & neutralize
assailant
3.049***
4.466***
-0.189
0.189
0.200
(0.115)
(0.135)
(0.149)
(0.148)
(0.164)
C. Intention & Hope
Hope for the future
2.663***
4.236***
0.071
-0.075
-0.126
(0.149)
(0.302)
(0.123)
(0.173)
(0.153)
Intention to complete
schooling
2.263***
3.438***
-0.117**
0.095
0.103
(0.111)
(0.343)
(0.048)
(0.090)
(0.085)
Intention to graduate
college
2.335***
3.589***
0.111
0.013
-0.019
(0.084)
(0.184)
(0.095)
(0.095)
(0.124)
Highest level of education
0.856***
1.544***
-0.005
0.030
0.039
(0.108)
(0.241)
(0.074)
(0.116)
(0.100)
Intention to participate in
labor force
1.353***
2.420***
-0.124
0.062
0.167
(0.083)
(0.181)
(0.085)
(0.110)
(0.109)
D. Health & Happiness
Mental health
1.089***
1.181*
0.027
-0.004
-0.002
(0.173)
(0.606)
(0.133)
(0.183)
(0.188)
Happiness index
1.566***
2.884***
0.078
-0.081
-0.124
(0.275)
(0.286)
(0.275)
(0.313)
(0.286)
Note: Treatment effects are estimated from OLS at the endline. A-Only and A&T are awareness only and awareness
& training treatment groups, respectively. All outcomes are standardized indices so that the control group has mean
zero and standard deviation one. The standardized baseline index/value of the outcome variables is controlled for
in all regressions except for the Knowledge index. Column (1) & (2): the treatment effect of A-Only (awareness
only) & A&T (awareness & training), respectively. Column (3) - (5): Social desirability score is a baseline measure
of the participant’s propensity to give socially desirable answers. A low social desirability score (LSDS) refers to
having a score below or equal to the median for the sample (Refer to section C.1 of Appendix C for details). Refer
to Table A1 for the details of all outcome variables. All regressions also control for all individual characteristics
(refer to footnote of Table 1). We also include region fixed effects. Robust standard errors clustered at the region
level are in parentheses in columns (1) - (5). *** p<0.01, ** p<0.05, * p<0.10
28
Table 5: Spill-over effects of awareness among friends & siblings
Participants
Siblings
Friends
Variables of interest
(1)
(2)
(3)
A-Only
3.732***
3.313***
2.084***
(0.120)
(0.111)
(0.063)
A&T
3.969***
3.867***
2.095***
(0.070)
(0.172)
(0.089)
Control mean
3.792
4.383
4.150
(1.852)
(1.836)
(2.601)
Observations
690
197
504
Note: Treatment effects are estimated from OLS. A-Only and A&T are awareness only and awareness & training
treatment groups, respectively. Knowledge Index is a standardized index so that the control group has mean zero and
standard deviation one. All regressions also control for all individual characteristics (refer to footnote of Table 1).
29
Table 6: Heterogeneity analysis
Outcome variables
Knowledge
Index
Defend
physical
assault
Defend
sexual
assault
Fight back &
neutralize
assailant
Intention to
complete
schooling
Intention to
graduate
college
Highest
level of
education
Intention
to PLF
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
A-Only
3.431***
2.268***
1.592***
2.319***
0.735*
1.203***
0.202
0.879***
(0.138)
(0.320)
(0.361)
(0.312)
(0.408)
(0.256)
(0.313)
(0.238)
A&T
3.725***
2.749***
2.846***
3.566***
2.385***
2.910***
0.520**
2.274***
(0.195)
(0.330)
(0.345)
(0.407)
(0.220)
(0.233)
(0.232)
(0.201)
Mother non-graduate
-0.189
-0.341
-0.313
-0.309
-0.278**
-0.505***
-0.450
-0.206
(0.155)
(0.250)
(0.361)
(0.371)
(0.099)
(0.153)
(0.273)
(0.163)
Mother non-graduate x A-
Only
0.371**
0.208
0.463
0.221
0.413**
0.742***
0.606**
0.404**
(0.165)
(0.240)
(0.367)
(0.371)
(0.186)
(0.203)
(0.281)
(0.186)
Mother non-graduate x A&T
0.145
0.335
0.205
0.125
0.458***
0.612***
0.831***
0.324*
(0.187)
(0.238)
(0.361)
(0.373)
(0.129)
(0.180)
(0.269)
(0.173)
Feel unsafe
0.043
-0.024
0.041
0.150
-0.340***
-0.220***
-0.115
-0.173
(0.131)
(0.162)
(0.139)
(0.203)
(0.104)
(0.059)
(0.117)
(0.100)
Feel unsafe x A-Only
0.056
0.055
-0.139
-0.149
0.430***
0.181**
0.055
0.102
(0.136)
(0.178)
(0.220)
(0.206)
(0.146)
(0.084)
(0.131)
(0.120)
Feel unsafe x A&T
-0.042
0.123
0.021
-0.118
0.458***
0.393**
0.054
0.260*
(0.162)
(0.182)
(0.183)
(0.221)
(0.151)
(0.148)
(0.127)
(0.135)
Low freedom
0.054
0.048
-0.320*
-0.044
-0.026
-0.481***
-0.102
-0.539***
(0.284)
(0.198)
(0.163)
(0.137)
(0.233)
(0.135)
(0.241)
(0.128)
Low freedom x A-Only
-0.160
-0.216
0.346
-0.062
0.110
0.617**
0.138
0.582***
(0.280)
(0.213)
(0.230)
(0.146)
(0.295)
(0.219)
(0.271)
(0.169)
Low freedom x A&T
0.005
-0.080
0.316
-0.008
-0.038
0.352**
0.203
0.510***
(0.300)
(0.227)
(0.196)
(0.170)
(0.251)
(0.143)
(0.258)
(0.172)
Observations
690
690
690
690
690
690
690
690
R-squared
0.893
0.826
0.821
0.890
0.770
0.798
0.761
0.822
Note: Treatment effects are estimated from OLS. A-Only and A&T are awareness only and awareness & training treatment groups, respectively. All outcomes are standardized indices so that the
control group has mean zero and standard deviation one. The standardized baseline index/value of the outcome variables is controlled for in regression except for the Knowledge index. Refer to
Table A1 for the details of all outcome variables. All regressions also control for all individual characteristics (refer footnote of Table 1). We also include region fixed effects. Mother non-
graduate is equal to 1 if the mother’s maximum education is less than a college degree. Feel unsafe equals 1 if the perception of safety is less than the median at baseline. Low freedom is equal
to 1 if the degree of freedom at home is lower than the median at baseline. Degree of freedom is the sum of whether the participant was allowed to go to four different kinds of places (marketplaces,
outside house after dark, friend’s place and school/college) alone, with someone accompanying them, or not allowed at all. It takes the value from 3 to 12 where higher scores denote less the
freedom.
30
Table 7: Causal mediation analysis
Panel A: Intention to complete schooling
Panel B: Intention to graduate college
Variable of interest
(1)
(2)
(3)
(4)
(5)
(6)
A-Only
1.956***
1.409***
1.253***
2.192***
1.620***
1.455***
(0.104)
(0.287)
(0.289)
(0.112)
(0.270)
(0.273)
A&T
3.067***
2.275***
1.901***
3.315***
2.399***
2.000***
(0.095)
(0.350)
(0.360)
(0.111)
(0.349)
(0.362)
Knowledge index
0.030
0.008
-0.014
-0.039
(0.053)
(0.060)
(0.042)
(0.051)
Defend sexual assault
0.194**
0.114**
0.277***
0.191***
(0.070)
(0.051)
(0.068)
(0.048)
Hope for the future
0.184***
0.197***
(0.050)
(0.040)
Panel C: Intention to PLF
Panel D: Happiness index
Variable of interest
(1)
(2)
(3)
(4)
(5)
(6)
A-Only
1.531***
1.393***
1.279***
1.450***
0.496
0.071
(0.138)
(0.197)
(0.211)
(0.130)
(0.296)
(0.189)
A&T
2.378***
2.058***
1.787***
2.873***
1.581***
0.554**
(0.122)
(0.225)
(0.248)
(0.100)
(0.344)
(0.257)
Knowledge index
-0.053
-0.068
0.094
0.031
(0.042)
(0.042)
(0.063)
(0.044)
Defend sexual assault
0.148**
0.090*
0.268***
0.048
(0.055)
(0.046)
(0.071)
(0.043)
Hope for the future
0.132***
0.509***
(0.030)
(0.035)
Note: Treatment effects are estimated from OLS at the endline. A-Only and A&T are awareness only and awareness & training
treatment groups, respectively. All outcomes are standardized indices so that the control group has mean zero and standard
deviation one. Knowledge index, Defend sexual assault and Hope for the future are also standardized indices, like the
outcome variables. The standardized baseline index/value of the outcome variables is controlled for in all regressions. Refer
to Table A1 for the details of all the outcome variables. All regressions also control for all individual characteristics (refer to
the footnote of Table 1). We also include region fixed effects. Robust standard errors clustered at the region level are in
parentheses. *** p<0.01, ** p<0.05, * p<0.10
31
Figures
Figure 1: Project timeline 2020-21
Figure 2: Effect size at the endline survey
Note: This figure shows the estimated treatment effects (same as in columns (1) – (2) of Table 1) in standard
deviations, where the control group has mean zero and standard deviation one. Effects are reported with 95%
confidence intervals. Refer to Table A1 for the detailed description of all the outcome variables.
Figure 3: Effect size on educational outcomes
Note: This figure shows estimated treatment effects (same as in columns (1) – (2) of Table 3) in standard
deviations, where the control group has mean zero and standard deviation one; except STEM and Professional.
STEM is a binary variable which takes the value 1 if the participant has opted for science as a major for pre-
college education (11th grade) and 0 otherwise. Professional is a categorical variable which takes the value 1 if
the participant has opted for science or commerce as major for pre-college education (11th grade), and 0 otherwise.
Effects are reported with 95% confidence intervals. Refer to Table A1 for the detailed description of all the
outcome variables.
Online Appendix
She can fight her own battles:
Experimental evidence on the effects of increasing awareness about rights
and self-defense training on female empowerment in India
Chitwan Lalji
‡‡‡‡‡‡‡
, Debayan Pakrashi
§§§§§§§
, Sarani Saha
********
, Soubhagya Sahoo
††††††††
and Russell Smyth
‡‡‡‡‡‡‡‡
‡‡‡‡‡‡‡
Economics Area, Indian Institute of Management Kozhikode. Email: chitwan@iimk.ac.in
§§§§§§§
Department of Economic Sciences, Indian Institute of Technology Kanpur. Email: pakrashi@iitk.ac.in
********
Department of Economic Sciences, Indian Institute of Technology Kanpur. Email: sarani@iitk.ac.in
††††††††
Department of Economic Sciences, Indian Institute of Technology Kanpur. Email: ssahoo@iitk.ac.in
‡‡‡‡‡‡‡‡
Department of Economics, Monash University, Clayton, Australia. Email: russell.smyth@monash.edu
1
Contents
Appendix A: Additional Tables and Figures ....................................................................................................... 2
A.1 Tables ........................................................................................................................................................... 2
A.2 Figures ........................................................................................................................................................ 14
Appendix B: Data ................................................................................................................................................. 21
B.1 Index construction ...................................................................................................................................... 21
B.2 Primary outcomes ....................................................................................................................................... 21
B.3 Educational outcomes & Career choice ...................................................................................................... 23
Appendix C: Social Desirability Bias ................................................................................................................. 25
C.1 Social desirability score .............................................................................................................................. 25
C.2 Bias check by trimming .............................................................................................................................. 26
Appendix D: Details of the Interventions
List of Tables
Table A1: Description of outcome variables ........................................................................................................... 2
Table A2: Summary statistics of all outcomes ......................................................................................................... 3
Table A3: Summary statistics of educational outcomes .......................................................................................... 4
Table A4: Balance test of individual characteristics ................................................................................................ 5
Table A5: Balance test of baseline outcomes .......................................................................................................... 6
Table A6: Balance test of follow-up sample ............................................................................................................ 7
Table A7: Balance test of baseline individual characteristic in follow-up survey ................................................... 8
Table A8: Balance test of baseline outcomes in follow-up survey .......................................................................... 9
Table A9: Treatment effects on all outcomes (Mean standardized) ...................................................................... 10
Table A10: Robustness check for social desirability bias in follow up survey ..................................................... 11
Table A11: Heterogeneity analysis of spill-over effects of awareness .................................................................. 12
Table A12: Impact of hope for the future on aspirations at endline: OLS & 2SLS ............................................... 13
List of Figures
Figure A1: Map of the study area .......................................................................................................................... 14
Figure A2: Treatment effects on awareness, self-defense & confidence ............................................................... 15
Figure A3: Treatment effects on intention & hope ................................................................................................ 16
Figure A4: Treatment effects on health & happiness............................................................................................. 17
Figure A5: Treatment effects on educational outcomes ........................................................................................ 18
Figure A6: Spillover effect of awareness on knowledge index ............................................................................. 19
Figure A7: Effect size at the follow-up survey ...................................................................................................... 20
2
Appendix A: Additional Tables and Figures
A.1 Tables
Table A1: Description of outcome variables
Outcome Variables
Measure
Definition
B
E
F
Panel A. Awareness
Knowledge Scale
Based on 11 questions testing knowledge
(True/False): score 0-11
Continuous: 0-11
✕
✓
✓
Panel B. Ability & Confidence to Defend Oneself
Defend physical assault
Ability to defend oneself against physical
assault from an acquaintance or stranger
scale: score 1-7
Continuous: 1-7
✓
✓
✕
Defend sexual assault
Confidence to defend oneself against sexual
assault scale: score 0-10
Continuous: 0-10
✓
✓
✓
Fight back & neutralize an
assailant
Confidence to fight back & neutralize an
assailant in 6 different ways (Hollander, 2014;
Weitlauf et al., 2001) scale: score 1-10
Continuous: 1-10
✓
✓
✕
Panel C. Intention & Hope
Current enrolment Status
Continued education in next academic year in
school or college six months after end line
survey (Only students in follow-up survey)
Indicator: = 1 if
continued
✕
✕
✓
Hope for the future
Subjective hope for the future scale: score 0-
10
Continuous: 0-10
✓
✓
✓
Intention to complete schooling
Intention to complete schooling scale: score
0-10
Continuous: 0-10
✓
✓
✓
Intention to graduate college
Intention to graduate college scale: score 0-10
Continuous: 0-10
✓
✓
✓
Highest level of education
Highest level of education participant wishes
to pursue: score 1-5 (grade 12th to Ph.D.)
Continuous: 1-5
✓
✓
✓
Intention to participate in labor
force
Intention to participate in labor force scale:
score 0-10
Continuous: 0-10
✓
✓
✓
Panel D. Health & Happiness
Mental health
Mental health in general in the past 7 days:
score 1-5
Continuous: 1-5
✓
✓
✓
Happiness scale
Feeling of happiness in general in the past 7
days: score 0-10
Continuous: 0-10
✓
✓
✓
Panel F. Educational Outcomes (Participants who appeared Board examination)
Hindi score (Only for grades 10 &
12)
Marks secured in Hindi in the board
examination for the academic year 2020-21.
Continuous: 0-100
✕
✕
✓
English score (Only for grades 10
& 12)
Marks secured in English in the board
examination for the academic year 2020-21.
Continuous: 0-100
✕
✕
✓
Overall academic performance
(Only for grades 10 & 12)
Average marks in all subjects in the board
examination for the academic year 2020-21.
Continuous: 0-100
✕
✕
✓
Mathematics score (Only for grade
10)
Marks secured in Mathematics in the board
examination for the academic year 2020-21.
Continuous: 0-100
✕
✕
✓
Science score (Only for grade 10)
Marks secured in Science in the board
examination for the academic year 2020-21.
Continuous: 0-100
✕
✕
✓
STEM (Only for grade 10)
Choice of science stream as a major for pre-
college education (for grade 11) in next
academic year.
Indicator: = 1 if
stream is Science
✕
✕
✓
Professional (Only for grade 10)
Choice of commerce or science as a major for
pre-college education (for grade 11) in next
academic year.
Indicator: = 1 if
stream is Science
or Commerce
✕
✕
✓
Note: The first three columns list the outcome variables and how they are measured and defined. The last three columns
indicate whether outcomes are measured at the baseline (B), end line (E), and/or follow-up survey (F).
✓
corresponds to
Yes and ✕ corresponds to No.
3
Table A2: Summary statistics by treatment types
Control
A-Only
A&T
Endline
Follow-up
Endline
Follow-up
Endline
Follow-up
Variables
(1)
(2)
(3)
(4)
(5)
(6)
A. Awareness
Knowledge scale
3.79
3.98
10.67
8.53
10.73
8.97
(1.85)
(2.46)
(0.66)
(2.05)
(0.63)
(1.76)
B. Ability & Confidence to Defend Oneself
Defend physical assault
3.97
-
6.28
-
6.68
-
(0.80)
(0.46)
(0.46)
Defend sexual assault
4.56
4.65
7.31
6.88
8.89
7.38
(1.24)
(1.81)
(0.86)
(1.64)
(0.61)
(1.48)
Fight back & neutralize
assailant
3.36
-
7.19
-
8.93
-
(1.43)
(0.54)
(0.63)
C. Intention & Hope
Current enrolment Status
-
0.95
-
0.98
-
1.00
(0.22)
(0.13)
(0.00)
Hope for the future
5.75
6.38
7.60
7.41
8.98
7.68
(0.81)
(1.12)
(0.71)
(1.11)
(0.66)
(1.02)
Intention to complete
schooling
5.74
6.32
7.79
7.36
8.84
7.50
(1.02)
(1.20)
(0.82)
(1.24)
(0.59)
(1.09)
Intention to graduate college
5.72
5.80
7.84
7.66
8.82
7.81
(0.95)
(1.27)
(0.74)
(1.43)
(0.61)
(1.20)
Highest level of education
3.22
3.17
4.39
4.41
4.55
4.48
(1.15)
(1.01)
(0.63)
(0.66)
(0.54)
(0.59)
Intention to PLF
5.57
5.79
7.74
7.59
8.86
7.71
(1.37)
(1.37)
(0.71)
(1.08)
(0.52)
(0.95)
D. Health & Happiness
Mental health
3.66
3.09
4.53
3.33
4.67
3.68
(0.47)
(0.73)
(0.50)
(0.85)
(0.48)
(1.21)
Happiness scale
6.13
6.51
7.57
7.56
8.97
7.77
(0.99)
(1.28)
(0.79)
(1.12)
(0.66)
(1.10)
Observations
202
102
243
122
245
124
Note: A-Only and A&T are the awareness only and awareness & training treatment groups, respectively.
Treatment and Control columns show the mean of the corresponding variables. Standard Error (SD) is
reported in parenthesis.
4
Table A3: Summary statistics of educational outcomes
Control
A-Only
A&T
Variables
(1)
(2)
(3)
A. Students in grades 10 & 12
Hindi score
51.46
66.25
62.28
(6.47)
(6.30)
(6.52)
English score
54.92
64.87
76.82
(6.71)
(7.51)
(5.82)
Overall academic performance
60.30
69.42
71.68
(3.84)
(4.48)
(3.45)
Observations
52
68
50
B. Students in grade 10
Mathematics score
59.16
73.75
79.22
(8.21)
(9.55)
(8.07)
Science score
64.89
76.88
73.66
(7.90)
(7.39)
(7.27)
STEM
0.55
0.71
0.78
(0.50)
(0.46)
(0.42)
Professional
0.74
0.88
0.88
(0.45)
(0.32)
(0.33)
Observations
38
52
41
Note: A-Only and A&T are the awareness only and awareness & training treatment groups, respectively.
Treatment and Control columns show the mean of the corresponding variables. Standard Error (SD) is
reported in parenthesis. Refer to footnote to Table 3.
5
Table A4: Balance test of individual characteristics
Control
A-Only
A&T
t-test p-values
Ref: Control
Ref: A-Only
(SD)
(SD)
(SD)
A-Only
A&T
A&T
Variables
(1)
(2)
(3)
(4)
(5)
(6)
Age (in years)
14.58
14.30
14.38
0.18
0.42
0.76
(1.85)
(1.75)
(1.82)
Grade
9.20
9.09
8.96
0.53
0.18
0.50
(1.54)
(1.53)
(1.51)
Caste=General
0.30
0.23
0.18
0.43
0.11
0.41
(0.46)
(0.42)
(0.38)
Caste =OBC
0.62
0.71
0.73
0.30
0.18
0.82
(0.49)
(0.46)
(0.45)
Caste=SC/ST
0.08
0.06
0.10
0.48
0.81
0.47
(0.28)
(0.23)
(0.30)
Birth order
1.32
1.27
1.34
0.55
0.75
0.35
(0.51)
(0.44)
(0.50)
Joint family dummy
0.31
0.37
0.33
0.26
0.50
0.46
(0.46)
(0.48)
(0.47)
Rural dummy
0.50
0.43
0.57
0.83
0.80
0.63
(0.50)
(0.50)
(0.50)
Father graduate dummy
0.29
0.39
0.29
0.65
0.97
0.67
(0.45)
(0.49)
(0.46)
Mother graduate dummy
0.20
0.28
0.20
0.67
0.99
0.68
(0.40)
(0.45)
(0.40)
Father employed dummy
0.89
0.86
0.89
0.37
0.76
0.22
(0.32)
(0.34)
(0.31)
Mother employed
dummy
0.01
0.00
0.02
0.58
0.61
0.20
(0.10)
(0.06)
(0.13)
Household size
5.91
6.19
6.08
0.23
0.40
0.72
(2.31)
(2.61)
(2.43)
Income status
0.85
0.80
0.82
0.50
0.68
0.69
(0.36)
(0.40)
(0.38)
Observations
202
243
245
690
690
488
Note: A-Only and A&T are Awareness Only and Awareness & Training treatment groups, respectively. Treatment
and Control columns show the mean of the corresponding variables. The variables are as follows: age (in years),
grade (7th to 12th), rural dummy (rural versus other regions of residence), joint family versus other types of family
structure, caste category of the participant (General, OBC (other backward castes) and SC/ST (Scheduled
Caste/Scheduled Tribe)), father’s education (graduate and non-graduate), mother’s education (graduate and non-
graduate), father’s employment (employed and unemployed), mother’s employment (employed and unemployed),
birth order (among siblings), income level of family (relatively rich) compared to others in the village (which
takes a value 1 if prosperous, very comfortable and reasonably comfortable versus others, i.e., just getting along,
poor or very poor), and household size (number of members in the family). t-test p-values are derived from linear
regression, with the variable of interest as the dependent variable and the treatment indicator as an independent
variable with standard errors clustered at the region level. (Control group is base for column (4) & (5) and A-Only
is base for column (6)). Standard Error (SD) is reported in parenthesis.
6
Table A5: Balance test of baseline outcomes
Control
A-Only
A&T
t-test p-values
Ref: Control
Ref: A-Only
(SD)
(SD)
(SD)
A-Only
A&T
A&T
Variables
(1)
(2)
(3)
(4)
(5)
(6)
B. Ability & Confidence to Defend Oneself
Defend physical assault
0.52
0.54
0.53
0.19
0.40
0.43
(0.11)
(0.09)
(0.10)
Defend sexual assault
0.52
0.54
0.53
0.34
0.91
0.20
(0.15)
(0.12)
(0.11)
Fight back & neutralize
assailant
0.46
0.48
0.47
0.33
0.62
0.20
(0.11)
(0.11)
(0.12)
C. Intention & Hope
Hope for the future
0.59
0.60
0.59
0.42
0.72
0.36
(0.12)
(0.13)
(0.13)
Intention to complete schooling
0.60
0.62
0.59
0.42
0.55
0.03
(0.09)
(0.08)
(0.10)
Intention to graduate college
0.61
0.62
0.60
0.44
0.67
0.09
(0.10)
(0.08)
(0.09)
Highest level of education
0.65
0.71
0.67
0.28
0.76
0.30
(0.23)
(0.24)
(0.19)
Intention to PLF
0.58
0.61
0.60
0.56
0.71
0.74
(0.15)
(0.14)
(0.11)
D. Health & Happiness
Mental health
0.87
0.83
0.83
0.14
0.19
0.94
(0.12)
(0.12)
(0.12)
Happiness scale
0.59
0.60
0.61
0.72
0.25
0.31
(0.15)
(0.15)
(0.15)
Observations
202
243
245
690
690
488
Note: A-Only and A&T are the awareness only and awareness & training treatment groups, respectively.
Treatment and Control columns show the mean of the corresponding variables. Variables are averages of
responses to survey questions associated with the outcomes, such that the value of each variable is between 0 and
1. Refer to Table A1 for the details of all the outcome variables. t-test p-values are derived from linear regression,
with the variable of interest as the dependent variable and the treatment indicator as an independent variable with
standard errors clustered at the region level (control group is base for column (4) & (5) and A-Only is base for
column (6)). Standard Error (SD) is reported in parenthesis.
7
Table A6: Balance test of follow-up sample
Non-Follow-up
Follow-up
t-test p-values
Baseline
Endline
Baseline
Endline
µ(1) = µ(3)
µ(2) = µ(4)
(1)
(2)
(3)
(4)
(5)
(6)
A. Outcome variables
Knowledge scale
-
8.65
-
8.71
-
0.812
(3.30)
(3.40)
Defend physical assault
3.71
5.73
3.71
5.76
0.968
0.791
(0.70)
(1.31)
(0.69)
(1.28)
Defend sexual assault
5.30
7.08
5.34
7.06
0.715
0.871
(1.29)
(1.94)
(1.23)
(2.00)
Fight back & neutralize assailant
4.72
6.71
4.71
6.66
0.938
0.816
(1.13)
(2.44)
(1.10)
(2.45)
Hope for the future
5.90
7.57
6.00
7.53
0.278
0.743
(1.24)
(1.49)
(1.26)
(1.48)
Intention to complete schooling
6.06
7.58
6.06
7.55
0.922
0.851
(0.91)
(1.50)
(0.92)
(1.50)
Intention to graduate college
6.08
7.58
6.09
7.55
0.887
0.808
(0.94)
(1.44)
(0.93)
(1.51)
Highest level of education
3.39
4.15
3.38
4.06
0.884
0.249
(1.12)
(0.93)
(1.13)
(1.02)
Intention to PLF
5.97
7.58
5.96
7.42
0.902
0.193
(1.30)
(1.50)
(1.34)
(1.70)
Mental health
4.20
4.32
4.20
4.33
0.990
0.859
(0.62)
(0.65)
(0.61)
(0.65)
Happiness scale
5.99
7.66
6.01
7.62
0.840
0.707
(1.51)
(1.40)
(1.51)
(1.41)
B. Individual characteristics
Age (in years)
14.41
14.41
0.975
(1.80)
(1.81)
Grade
9.09
9.06
0.795
(1.51)
(1.55)
Caste =General
0.23
0.23
0.900
(0.42)
(0.42)
Caste = OBC
0.69
0.69
0.943
(0.46)
(0.46)
Caste= SC/ST
0.08
0.08
0.942
(0.27)
(0.27)
Birth order
1.29
1.32
0.381
(0.47)
(0.50)
Joint Family dummy
0.34
0.34
0.933
(0.48)
(0.47)
Rural dummy
0.50
0.50
0.939
(0.50)
(0.50)
Father graduate dummy
0.32
0.32
0.997
(0.47)
(0.47)
Mother graduate dummy
0.23
0.22
0.902
(0.42)
(0.42)
Father employed dummy
0.88
0.88
0.880
(0.32)
(0.33)
Mother employed dummy
0.01
0.01
0.722
(0.09)
(0.11)
Household size
6.05
6.08
0.858
(2.43)
(2.49)
Income status
0.82
0.82
0.837
(0.39)
(0.38)
Treatment group
1.06
1.06
0.976
(0.80)
(0.80)
Observations
342
342
348
348
-
-
Note: Refer to Table A1 and footnote of Table A4 for the details of all the outcome variables and individual characteristics,
respectively. Columns (1) and (3) report the mean of the variables for follow-up and non-follow-up sample, respectively, in
the baseline survey. Columns (2) and (4) report the mean of the variables for follow-up and non-follow-up sample,
respectively, in the end-line survey. Columns (5) & (6) report the test of equality (t-test p-values) between means of follow-
up and non-follow-up samples at baseline and end line, respectively. Standard Error (SD) is reported in parenthesis.
8
Table A7: Balance test of baseline individual characteristic in follow-up survey
Control
A-Only
A&T
t-test p-values
Ref: Control
Ref: A-Only
(SD)
(SD)
(SD)
A-Only
A&T
A&T
Variables
(1)
(2)
(3)
(4)
(5)
(6)
Age (in years)
14.51
14.47
14.27
0.89
0.40
0.54
(1.81)
(1.81)
(1.80)
Grade
9.12
9.25
8.83
0.62
0.25
0.13
(1.50)
(1.60)
(1.51)
Caste=General
0.25
0.25
0.19
0.99
0.35
0.41
(0.44)
(0.44)
(0.39)
Caste=OBC
0.68
0.67
0.72
0.97
0.66
0.68
(0.47)
(0.47)
(0.45)
Caste=SC/ST
0.07
0.07
0.10
0.93
0.66
0.71
(0.25)
(0.26)
(0.30)
Birth order
1.26
1.31
1.39
0.63
0.26
0.41
(0.51)
(0.47)
(0.52)
Joint Family dummy
0.32
0.33
0.36
0.94
0.48
0.47
(0.47)
(0.47)
(0.48)
Rural dummy
0.52
0.43
0.56
0.75
0.90
0.64
(0.50)
(0.50)
(0.50)
Father graduate dummy
0.24
0.35
0.37
0.56
0.51
0.93
(0.43)
(0.48)
(0.49)
Mother graduate dummy
0.16
0.27
0.23
0.52
0.67
0.85
(0.37)
(0.45)
(0.43)
Father employed dummy
0.88
0.87
0.89
0.77
0.91
0.59
(0.32)
(0.34)
(0.32)
Mother employed dummy
0.02
0.01
0.01
0.57
0.57
0.99
(0.14)
(0.09)
(0.09)
Household size
5.95
5.98
6.29
0.94
0.30
0.42
(2.49)
(2.55)
(2.44)
Income status
0.85
0.79
0.84
0.26
0.79
0.34
(0.36)
(0.41)
(0.37)
Observations
102
122
124
348
348
246
Note: See footnote to Table A4.
9
Table A8: Balance test of baseline outcomes in follow-up survey
Control
A-Only
A&T
t-test p-values
Ref: Control
Ref: A-Only
(SD)
(SD)
(SD)
A-Only
A&T
A&T
Variables
(1)
(2)
(3)
(4)
(5)
(6)
B. Ability & Confidence to Defend Oneself
Defend sexual assault
0.54
0.55
0.52
0.83
0.24
0.08
(0.14)
(0.12)
(0.11)
C. Intention & Hope
Hope for the future
0.59
0.61
0.60
0.56
0.72
0.70
(0.11)
(0.13)
(0.14)
Intention to complete
schooling
0.60
0.62
0.60
0.31
0.76
0.11
(0.10)
(0.08)
(0.09)
Intention to graduate
college
0.59
0.63
0.61
0.10
0.40
0.21
(0.10)
(0.08)
(0.09)
Highest level of education
0.62
0.72
0.68
0.08
0.18
0.49
(0.22)
(0.25)
(0.19)
Intention to PLF
0.56
0.61
0.61
0.34
0.30
0.98
(0.16)
(0.14)
(0.10)
D. Health & Happiness
Mental health
0.88
0.82
0.83
0.07
0.13
0.84
(0.12)
(0.13)
(0.11)
Happiness scale
0.60
0.58
0.63
0.51
0.16
0.01
(0.15)
(0.15)
(0.15)
Observations
102
122
124
348
348
246
Note: See footnote to Table A5.
10
Table A9: Treatment effects on all outcomes (Mean standardized)
End line
Wald Test
RI p-values
Multiple
hypothesis testing
A-Only
A&T
β(1)= β(2)
A-Only
A&T
Wald
Joint
Westfall-
Young
Outcome Variables
(1)
(2)
(5)
(6)
(7)
(8)
(9)
A. Awareness
Knowledge index
2.065***
2.196***
0.127
0.000
0.000
0.000
0.000
(0.066)
(0.039)
B. Ability & Confidence to Defend Oneself
Defend physical assault
1.741***
2.203***
0.007
0.001
0.000
0.001
0.001
(0.142)
(0.047)
Defend sexual assault
1.627***
2.256***
0.000
0.000
0.000
0.001
0.001
(0.125)
(0.034)
Fight back & neutralize
assailant
1.487***
2.315***
0.000
0.000
0.001
0.001
0.001
(0.041)
(0.035)
C. Intention & Hope
Hope for future
1.287***
2.235***
0.000
0.000
0.001
0.000
0.000
(0.161)
(0.039)
Intention to complete
schooling
1.014***
2.122***
0.000
0.000
0.001
0.001
0.001
(0.222)
(0.066)
Intention to graduate
college
1.354***
2.282***
0.000
0.000
0.000
0.001
0.001
(0.101)
(0.077)
Highest level of
education
1.138***
1.695***
0.044
0.001
0.000
0.000
0.000
(0.255)
(0.060)
Intention to PLF
1.155***
2.119***
0.000
0.001
0.000
0.001
0.001
(0.113)
(0.102)
D. Health & Happiness
Mental health
1.500***
1.457***
0.917
0.000
0.046
0.001
0.001
(0.395)
(0.060)
Happiness index
1.173***
2.268***
0.000
0.000
0.001
0.000
0.000
(0.171)
(0.047)
Observations
690
690
-
-
-
-
-
Note: All outcomes are standardized indices so that the overall mean zero and standard deviation one. Refer to
footnote of Table 1.
11
Table A10: Robustness check for social desirability bias in follow up survey
Variables of interest
A-Only
A&T
LSDS
A-Only ✕ LSDS
A&T ✕ LSDS
Outcome Variables
(1)
(2)
(3)
(4)
(5)
A. Awareness
2.264***
2.022***
0.053
-0.019
-0.098
Knowledge index
(0.231)
(0.573)
(0.266)
(0.308)
(0.311)