a business case for financial education and microsavings
TRANSCRIPT
A Business Case for Financial Education and Microsavings
Promotion: Experimental Evidence from a For-Pro�t Program in
Rural Peru*
Chris M. Boyd�and Sandro Díez-Amigo�
Preliminary version
Abstract
This paper intends to contribute to the mixed body of evidence on the e�ectiveness of �nancial educa-
tion to promote savings among poor rural populations, and to rigorously test the business case for �nancial
education in rural areas provided by private �nancial institutions on a for-pro�t basis. In order to do so
we experimentally evaluate a for-pro�t �nancial education program, paired with savings account market-
ing, targeting poor rural women in the Apurímac region of Peru. We �nd that the program signi�cantly
increased the familiarity and trust in formal �nancial institutions, and that this e�ect was particularly
large for the implementing institution. Moreover, we �nd that the program signi�cantly increased the
probability of saving with the implementing institution. However, we only observe an incipient e�ect on
credit uptake with the implementing institution. Our �ndings suggest that �nancial education programs
can be a successful marketing tool for the promotion of microsavings among the poor even in remote rural
areas, and that their provision by �nancial institutions on a for-pro�t basis can be commercially viable if
their focus is expanded beyond savings, in order to promote and generate cross-selling of credit products
with higher pro�t margins.
*The authors would like to thank Martín Naranjo, Cecilia Marquina and their colleagues at Financiera Con�anza and theBBVA Micro�nance Foundation for their time and support, Claudia Martínez, Yuri Soares, Claudia Gutiérrez, Andrea Reyesand numerous seminar participants for their useful comments, and Cecilia Vargas Yana for her outstanding research assistance.Financial support was generously supplied by the Multilateral Investment Fund of the Inter-American Development Bank,the International Development Research Centre, the Ford Foundation, and Proyecto Capital. Research was conducted whileChris M. Boyd was a�liated to the Instituto de Estudios Peruanos (IEP), and Sandro Díez-Amigo was at the Inter-AmericanDevelopment Bank (IDB).
�University of Minnesota. Corresponding author: e-mail: [email protected]; Address: 1994 Buford Ave, St. Paul, MN55108; Phone: (612) 625 5000.
�World Bank Group. e-mail: [email protected]; Address: 1818 H St NW, Washington D.C. 20433; Phone: (202)453 1001
JEL Classi�cation Codes: C93, D04, D14, M21, R51.
Keywords: Financial Education, Financial Inclusion, Micro�nance, Savings, Rural Microsavings, Impact Evaluation,
Business Case.
Copyright © 2019 Inter-American Development Bank.
This work is licensed under a Creative Commons IGO 3.0 Attribution-NonCommercial-NoDerivatives (CC-IGO BY-
NC-ND 3.0 IGO) license (http://creativecommons.org/licenses/by-nc-nd/3.0/igo/legalcode) and may be reproduced
with attribution to the IDB and for any non-commercial purpose, as provided below. No derivative work is allowed.
Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration
pursuant to the UNCITRAL rules. The use of the IDB's name for any purpose other than for attribution, and the
use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not
authorized as part of this CC-IGO license.
Following a peer review process, and with previous written consent by the Inter-American Development Bank (IDB), a
revised version of this work may also be reproduced in any academic journal, including those indexed by the American
Economic Association's EconLit, provided that the IDB is credited and that the author(s) receive no income from the
publication. Therefore, the restriction to receive income from such publication shall only extend to the publication's
author(s). With regard to such restriction, in case of any inconsistency between the Creative Commons IGO 3.0
Attribution-NonCommercial-NoDerivatives license and these statements, the latter shall prevail.
Note that link provided above includes additional terms and conditions of the license.
The opinions expressed in this publication are those of the authors and do not necessarily re�ect the views of the
Inter-American Development Bank, its Board of Directors, or the countries they represent.
1 Introduction
The ProSavings Program, led by the Multilateral Investment Fund of the Inter-American Development Bank,
promotes the development of business strategies to o�er liquid and planned savings services tailored to low-
income populations in Latin America and the Caribbean. As part of this program in 2015 the Peruvian
micro�nance institution Financiera Con�anza, a subsidiary of the BBVA Micro�nance Foundation, started
to implement the Ahorro para Todos (�Savings for All�) �nancial education program. This intervention
targeted the rural population of the Apurímac region in Southern-Central Peru, with a focus on women over
18 years old, and its four training modules were aimed at generating knowledge and trust in the formal
�nancial system and at promoting �nancial best practices. Also, during the program trainers promoted an
associated savings account o�ered by Financiera Con�anza.
The evidence regarding the ability of government- and non-pro�t-sponsored �nancial education programs
to promote savings in the formal �nancial system is mixed. Moreover, �nancial education programs also
bene�t formal �nancial institutions by expanding their pool of clients, allowing them to reach previously
underserved rural populations. Therefore, �nancial education can be thought of not only a public good, but
also as an e�ective marketing mechanism, and succesful private or public-private provision models would be
a win-win-win for the government, rural populations, and formal �nancial institutions. However, �nancial
institutions face challenges which may or may not allow them to provide for-pro�t �nancial education in a
commercially viable manner.
In this context our experimental impact evaluation of the Ahorro para Todos program intends to contribute
to the mixed body of evidence on the e�ectiveness of �nancial education to promote savings among poor
rural populations, and to rigorously test the business case for �nancial education in rural areas provided by
private �nancial institutions on a for-pro�t basis.
We use an optimal random assignment at the community level to generate comparable control and treatment
groups which are statistically indistinguishible on both observable and unobservable characteristics. This
allows us to robustly identify and estimate the impact of the program on a range of dependent outcome
variables. Also, in July 2016 we carried out an endline survey of a random sample 1801 households across
the 89 communities included in the evaluation. Information was gathered on a range of dependent outcome
variables, grouped in six categories: (i) �nancial literacy, (ii) savings, (iii) credit, (iv) income and assets,
(v) consumption, and (vi) female empowerment. Finally, we also have access to administrave information
provided by Financiera Con�anza on transactions and balances between January 2015 and July 2016, with
which we complement the self-reported data.
We �nd that the �nancial education program signi�cantly increased the familiarity and trust in formal
�nancial institutions in poor rural areas of the Andes, and that this e�ect was particularly large for Financiera
Con�anza. Moreover, we also �nd that the program signi�cantly increased the probability of saving with
Financiera Con�anza, even if we �nd no evidence of a signi�cant impact of the program on the probability
of saving at home or with other formal �nancial institutions or cooperatives. However, on the contrary
we observe no evidence of an e�ect of the program on the probability of buying other �nancial products
from Financiera Con�anza or from any other �nancial institution. On the other hand, we estimate that the
program signi�cantly reduced poverty.
3
All the above suggests that �nancial education programs can be a succesful marketing tool for the promotion
of microsavings among the poor even in remote rural areas, and that their provision by �nancial institutions
on a for-pro�t basis can be commercially viable if their focus is expanded beyond savings, in order to promote
and generate cross-selling of credit products with higher pro�t margins. Also, the observed success of Ahorro
para Todos in reducing poverty levels rea�rms the public good nature of �nancial education, and suggests a
rationale for public sector support of this type of private development programs.
The rest of the paper is organized as follows: Section 2 presents the background for the paper; Section
3 provides a description of the intervention; Section 4 outlines the methodology used in the experimental
evaluation; Section 5 presents the main �ndings; Section 6 concludes.
2 Background
Under the presidencies of Alejandro Toledo (2001-2006), Alan García (2006-2011) and Ollanta Humala (2011-
2016) Peru experienced robust economic growth. Poverty levels fell dramatically, from 54.2% in 2002 to
21.8% in 20151, and while a�ected by the 2015 global commodities slump, it continues to be one of the best
performing economies in Latin America.
However, there are great economic disparities among Peruvian regions, and particularly across rural an urban
areas: half of the rural population in Peru (45.2%) was below the poverty line in 2015 , and more than an
eighth (13.9%) was considered extremely poor. This gap was exarcerbated in the Andean regions, were 49%
of the population was below the poverty line, and 16.5% was considered extremely poor [INEI(INEI2016)].
And, as in other contexts, the higher levels of poverty were correlated with lower levels of education, and
high rates of �nancial illiteracy.
Apurímac, an Andean region in Southern-Central Peru with a particularly challenging geography2, was the
poorest region in the country in 2012, contributing 0.4% of its GDP and growing at an average rate of 4.3%
between 2007 and 2014 [INEI(INEI2016)]. These facts made it the target of numerous public and private
development iniatives in subsequent years, such as large mining investments, which succesfully accelerated
regional GDP growth. However, at the same Apurímac struggled to tackle the issue of income inequality
across urban and rural areas, motivated in great part by the lack of access to the formal �nancial system in
the latter.
The ProSavings Program, led by the Multilateral Investment Fund of the Inter-American Development Bank,
promotes the development of business strategies to o�er liquid and planned savings services tailored to low-
income populations in Latin America and the Caribbean. As part of this program the Peruvian micro�-
nance institution Financiera Con�anza, a subsidiary of the BBVA Micro�nance Foundation, implemented
the Ahorro para Todos (�Savings for All�) �nancial education program in the Apurímac region3, with a focus
on its Abancay, Antabamba, Aymaraes and Grau provinces.
Financial education programs and related initiatives to expand access promise rural populations the op-
1See [Herrera(2003)],[INEI(INEI2016)].2Mountains are worshipped as gods in Andean religion, and Apurímac means where the gods speak in Quechua language.3After its initial implementation in Apurímac the program was subsequently expanded to the Cusco region with some
modi�cations. However, here we only study the intervention in Apurímac.
4
portunity to bene�t from the formal �nancial system. However, the evidence regarding the ability of
government- and non-pro�t-sponsored programs in Peru and elsewhere to promote savings in the formal
�nancial system and improving welfare is mixed. For example, [Cole et al.(2009)Cole, Sampson, and Zia]
�nd that a �nancial education program in Indonesia has modest e�ects on the demand for bank accounts,
which is however succesfully estimulated with savings subsidies. However, while in their evaluation of a
savings promotion pilot program in Peru [Boyd and Aldana(2015)] only �nd a small impact on �nancial
literacy, they observe a large signi�cant impact on the probability of saving at a bank (16%). Similarly,
[Ashraf et al.(2010)Ashraf, Karlan, and Yin] �nd that a Philippine commitment savings program combined
with �nancial education focused on the bene�ts of saving had a large signi�cant impact (81%) on the amount
of formal savings 12 months after the intervention, but this e�ect is no longer observed 32 months after the
intervention.
Moreover, �nancial education programs also bene�t formal �nancial institutions by expanding their pool of
clients, allowing them to reach previously underserved rural populations. Therefore, �nancial education can
be thought of not only a public good, but also as an e�ective marketing mechanism, and succesful private or
public-private provision models would be a win-win-win for the government, rural populations, and formal
�nancial institutions. However, �nancial institutions face challenges which may or may not allow them to
provide for-pro�t �nancial education in a commercially viable manner.
In this context our experimental impact evaluation of the Ahorro para Todos program intends to contribute
to the mixed body of evidence on the e�ectiveness of �nancial education to promote savings among poor
rural populations, and to rigorously test the business case for �nancial education in rural areas provided by
private �nancial institutions on a for-pro�t basis.
3 Intervention
The Ahorro para Todos (�Savings for All�) �nancial education program implemented by Financiera Con�anza
from May 2015 to April 2016 targeted the rural population of the Apurímac region in Southern-Central Peru,
with a focus on women over 18 years old. This was motivated by the evidence generated by the pilot
program carried out in early 2015, which as discussed by [Boyd and Aldana(2015)] suggested that focusing
the program on women would improve its e�ectiveness, making it easier to boost savings and providing a
more creditworthy4 potential pool of clients for the �nancial institution.
The treatment of the Ahorro para Todos program consisted on two interacting components:
(i) The �rst component consisted of four modules focused on general �nancial education.
(ii) The second component focused on the promotion of the associated savings account o�ered by Financiera
Con�anza, also called Ahorro para Todos.
The four modules5 of the �nancial education component intended to generate knowledge and trust in the
4Even if the program is focused on savings promotion, creditworthiness remains a key dimension, as it is a stylized industryfact that the taking of deposits is a commercially viable activity only insofar as it leads to the cross-selling of other productswith higher pro�t margins, such as loans.
5The four modules were named (1) �Trusting my savings to Financiera Con�anza�, (2) �If we know, we win�, (3) �Let's control
5
formal �nancial system, by explaining the advantages and disadvantages of the di�erent types of �nancial
products such as savings accounts, loans and insurance policies, as well as by promoting �nancial best
practices such as the creation of a monthly budget for the household.
An important innovation included in this part of the program consisted on the use of the �edutainment�
approach used to impart the four modules, shown to achieve results in other settings6. In order to do so,
each of the four teaching modules were structured as short stories of (made-up) local women, to which the
target population could easily relate. Also, program trainers used custom-designed dolls and portable stages
to support their explanations and maintain the concentration of the audience.
Although the challenging logistics of reaching remote rural communities in a context of limited transportation
options and budgetary contstraints meant that exceptions needed to be made, in general the four �nancial
education modules were delivered at a rate of one per month. Trainers traveled to each community, and
each module was taught in a one-hour session over or immediately before/after a meeting organized by one
of the pre-established groups7. Bene�ciaries could decide whether or not to participate in each �nancial
education session, and while a mentioned the program focused on female bene�ciaries, those men who asked
to participate were allowed to do so.
At the end of each session trainers addressed the questions and doubts of the audience about the contents
taught in the corresponding module, and then spent some time promoting the Ahorro para Todos savings
account8.
This account is a simple savings account, which (a) features no administration or transfer fees, (b) o�ers an
unsbsidized but competitive interest rate of 0.75% (3) requires a minimum opening deposit of 20 Peruvian
soles, or approximately 6 U.S. dollars, (4) provides a complimentary debit card once the amount saved
reaches 50 soles, and (5) provides a life microinsurance policy active if the amount saved is at least 100 soles.
Participants in the �nancial education sessions were explicitly informed that their attendance did not entail
any commitment to contract the associated savings account. Also, in order to open an account and make
deposits prospective clients needed to visit the Financiera Con�anza branch at the regional capital, Abancay.
Although it was originally conceived as a �commitment� savings account with a minimum recurrent deposit,
this requirement proved very unpopular among prospective clients during the pilot phase, and was subse-
quently dropped. However, coinciding with the main holidays Financiera Con�anza ra�ed a (non-perishable)
food basket, valued at 120 soles, among clients who had made a deposit during the previous month. This
ra�e was advertised and used during the promotion of the savings account at the end of each teaching session.
our savings and plan our expenses�, and (4) �I need insurance, and I need credit! Let's get to know �nancial services�.6For example, see [La Ferrara et al.(2012)La Ferrara, Chong, and Duryea] for an analysis of the impact of soap operas on
fertility in Brazil, or [Valdivia and Chong(2013)] for a proposal on how to use an analogous approach to promote savings amongwomen in rural Peru.
7In order to maximize female participation in the program selected groups were generally those with a majority of femaleattendance. Also, the majority of groups were �pre-established� in the sense that they had already been set up by governmentalprograms (e.g. Juntos, Health Center meetings) or another institution working in the community.
8Trainers were hired by Financiera Con�anza as �Savings Promotion Agents�.
6
4 Evaluation Methodology
4.1 Experimental Design
We use random assignment at the community level that maximizes the balance of observable characteristics
prior to the intervention. This allows us to robustly identify and estimate the impact of the program on a
range of dependent variables, as described more in detail in Section 4.3 below.
During the pilot phase of the program a �eld test was conducted by Financiera Con�anza in order to calculate
the average travel time from the regional capital of Abancay, where its nearest branch is located, to each of the
prospective intervention communities. Based on the results of this exercise Financiera Con�anza decided to
limit the intervention to communities at a distance of under six hours of Abancay, since reaching communities
further away was deemed ex-ante commercially unviable. Similarly, Financiera Con�anza determined that
Ahorro para Todos would only be implemented in communities located at least one hour and half away
from Abancay, in order to guarantee that the program maintained its focus on providing �nancial education
and access to formal �nancial services for the rural population. Moreover, some remote communities within
the six hour radius were grouped together and treated as one, in order to better accomodate the logistical
constraints arising from the limited transportation options and challenging geography, and to minimize the
potential for contamination of the control group whenever two communities were very close to each other.
Also, communities with fewer than 15 bene�ciaries of the JUNTOS conditional cash transfer program were
also excluded to ensure that the program focused on communities with a high percentage of poor households.
Similarly, additional communities were excluded because it was determined that they had no contact with the
regional capital, or because they were deemed potentially dangerous for program trainers to visit. Finally,
Financiera Con�anza also agreed to exclude the districts where communities had received or were scheduled
to receive any �nancial education treatment provided by the government or by Proyecto Capital as part of
its paralell activities in the region 9.
After the above described pre-selection process was completed there were 89 eligible communities or groups
of communities left across 22 districts in the Abancay, Antabamba, Aymaraes and Grau provinces of the
Apurímac region.
In order to maximize the balance between treatment and control groups we carried out 3,000 random assign-
ments, and chose the randomization which maximized the balance across all available variables 10. Due to
budgetary restrictions, a relevant baseline survey could not be carried out, thus we used information from
the Peruvian Social Programs Targeting O�ce (SISFOH) in order to carry out this analysis11.
9In particular, we excluded districts which had participated in the governmental Haku Wiñay program, and those which werescheduled to participate in another �nancial inclusion program to be piloted by Proyecto Capital.
10In particular, we tested for di�erences in means across 22 balance variables in each randomization, and selected the ran-domization with the highest minimum p-value.
11We used 19 SISFOH variables at the household level, collected in 2013 from all households in the evaluation districts andcompared their means at the community level to carry out the balance maximization: (i) percentage of men in the household,(ii) percentage of pregnant women in the household, (iii) head of household age, (iv) average age of household members, (v)gender of household head, (vi) bene�ciary status in ESSALUD health insurance program, (vii) bene�ciary status in SIS healthinsurance program, (viii) percentage of household members without health insurance, (ix) participation in the Vaso de Leche
program, (x) bene�ciary status in the school breakfast/lunch program, (xi) bene�ciary status in JUNTOS conditional cashtransfer program, (xii) bene�ciary status in social programs, (xiii) bene�ciary status in Pension 65 program, (xiv) percentage ofhousehold members with Spanish as native tongue, (xv) percentage of household members with Quechua as a native tongue, (xvi)
7
This exercise yielded a balanced ex-ante random assignment of 44 communities in the control group and 45
communities in the treatment group .
After the completion of the evaluation Financiera Con�anza o�ered the Ahorro para Todos program also in
communities assigned to the control group.12
4.2 Data Collection
In July 2016 we carried out an endline survey of a random sample of 1801 rural households across the 89
communities included in the evaluation, which had been randomly assigned to the control and treatment
group prior to the start of the intervention13. Information was gathered on a range of dependent outcome
variables, grouped in six categories: (i) �nancial literacy, (ii) savings, (iii) credit, (iv) income and assets, (v)
consumption, and (vi) female empowerment.
In the absence of a population census the random sampling strategy for the endline survey was as follows:
(a) for each community we calculated the female population above 18 years old according to the information
available from the Peruvian Social Programs Targeting O�ce (SISFOH); (b) for communities with less than
90 females over 18 years old we assigned a maximum target number of 30 surveys to be collected, allowing
for a maximum of two replacements; (c) for communities with more than 90 females over 18 years old we
assigned a minimum target number of 30 surveys to be collected, and then randomly added observations to
each community until reaching the target of 1800 observations split evenly between treatment and control.
In addition, on arrival to each community the surveyors followed a �eld protocol to ensure the random selection
of households to be interviewed: (1) �rst, they looked for the community president, informed her/him about
the study and requested permission to do it; (2) then, with the help of the community president they mapped
the community, noting the main buildings or geographical characteristics, and divided the community into
four quadrants; (3) next, surveyors divided each quadrant in blocks of 3-4 houses and administered the
survey to a randomly selected household in each block, leaving the other 2-3 as replacements; and (4) �nally,
surveyors applied this protocol until the target number of observations for the community was reached,
randomly choosing replacements within each block as necessary.
We supervised the �eld operations in several communities and veri�ed GPS geodata for each household in
order to validate the randomness of the sampling. The results of this validation were positive, however
percentage of household members with a national identity document, (xvii) percentage of household members who know howto read and write, (xviii) household involved in agriculture, and (xix) household involved in the service sector. In addition, wealso looked at 3 additional community level variables for balance: (xx) distance to regional capital, (xxi) number of bene�ciaresof JUNTOS conditional cash transfer program, and (xxii) percentage of JUNTOS recipients receiving the transfer in a bankaccount.
12As of January 2017 Financiera Con�anza had started to implemented the Ahorro para Todos program in some of the controlcommunities located nearer to the regional capital, expanding it to include activities to promote group credit products.
13Note that the 89 communities included four intermediate cities, two in the control group and two in the tratment group.Financiera Con�anza invited people from the urban areas of these cities to training meetings, but none attended the meetings,although a few opened the promoted savings account. Thus, we excluded urban areas from the endline survey to keep the focuson rural population, but this does not seem to a�ect the balance between treatment and control groups . Also, note that a fewrural communities close to each other were regrouped since Financiera Con�anza decided to do one bigger meeting for all ofthem. As a result of these two changes, we have 84 communities (clusters in the following regressions) instead of the designed89, but the targeting was done as originally planned: in all of the 89 communities assigned to the treatment, people were invitedto the �nancial education training meetings, and none in the communities originally assigned to control was invited to themeetings.
8
in some small communities the target number of observations could not be reached, due to migration or
reluctance to take the survey. In these instances the remaining observations were randomly reassigned to
other communities, resulting in the �nal endline survey tally of 907 observations in the control group and
894 observations in the treatment group.
Also, we check the ex-post balance between treatment and control at endline by looking at characteristics
considered to be orthogonal to the intervention, and therefore unlikely to have changed in a systematically
di�erent manner in both groups before and after the program was implemented: (i) distance to regional
capital; (ii) frequency of travel to provincial capital by respondent; (iii) number of household members;
(iv) gender of respondent; (v) age of respondent; (vi) maximum educational level of household; and (vii)
bene�ciary status of JUNTOS conditional cash transfer program.14
As detailed in the appendix, the balance at endline is good, with no signi�cant di�erences in balance variables
between treatment and control at the 1% or 5% level15.
Finally, we also have access to administrave information provided by Financiera Con�anza on transactions
and balances between January 2015 and July 2016, with which we complement the self-reported data obtained
in the endline survey.
4.3 Model Speci�cations
Using the econometric models described below we estimate the e�ect of the program on a range of dependent
outcome variables, which are grouped in six categories: (i) �nancial literacy, (ii) savings, (iii) credit, (iv)
income and assets, (v) consumption, and (vi) female empowerment16.
All reported standard errors are clustered at the community level since this was our randomization unit .
Intent-to-Treat (ITT)
Let Zi ∈ {0, 1}denote the random assignment of household i to a control or treatment area in a causal model
á la Rubin17. Then if Yi is the dependent outcome variable, the intent-to-treat (ITT) estimate of the e�ect
of the program can be represented as:
δITT = E(Yi | Zi = 1)− E(Yi | Zi = 0)
This reduced form estimate captures the relative average e�ect of the program among all those households in
communities randomly assigned to the treatment group compared to households in communities randomly
14Due to SISFOH anonimization of individual households, we could not check ex-post balance in the 22 variables used tomaximmize balance at the communidity level. However, using the Endline survey data, we checked that balance remains amongall the variables used for maximizing balance that were supposed to be time-invariant and orthogonal to the intervention.
15The only signi�cant di�erence detected was only so at the 10% level: sampled treatment households live on average 9minutes further away from the regional capital than sampled control households (2 hours and 57 minutes versus 2 hours and 48minutes).
16For simpli�cation purposes only a subset of most relevant variables are presented in each category.17See [Imbens and Rubin(2010)].
9
assigned to the control group, and can be calculated using an ordinary least squares (OLS) regression model:
yi = α0 + δITTZi + ei (1)
where i indexes households, yi is the relevant outcome variable, and Zi is a dummy variable which takes a
value of 1 if the household belongs to a community assigned to the treatment group which was o�ered to
participate in the program and 0 otherwise.
We subsequently augment this simple di�ference in means OLS regression model to include additional co-
variates:
yi = α′
0 + δ′
ITTZi +
6∑h=1
α′
hxhi+e′
i (2)
where x1i is the distance from the household's community to the regional capital in average hours of travel18,
x2i is the number of household members, x3i is a dummy variable which takes a value of 1 if the household
head is a female, x4i is the maximum educational level among all members of the household in years of
education, x5i is the age of the respondent in years, x6i is a dummy variable which takes a value of 1 if the
household is a bene�ciary of the JUNTOS conditional cash transfer program, and the rest of the notation is
as above.
Finally, we again augment the OLS regression model to include �xed e�ects by district:
yi = α′′
0 + δ′′
ITTZi +
6∑h=1
α′′
hxhi+dj + e′′
i (3)
where j indexes districts, dj are district �xed e�ects, and the rest of the notation is as above.
No additional assumptions beyond random assignment are necessary for the validity of the ITT estimate of
the impact of the program.
Treatment-on-the-Treated (TOT)
Take-up of the program was imperfect in the treatment communities in which it was o�ered, as some house-
holds chose not to participate in it. Similarly, some households from control communities which were not
o�ered the program did participate in it by attending meetings held in treatment communities.Therefore, we
also estimate treatment-on-the-treated (TOT) e�ects of the program.
TOT estimates address imperfect assignment by scaling up the ITT to re�ect how the random assignment in-
�uenced participation rates in treatment and control communities. As proposed by [Angrist et al.(1996)Angrist, Imbens, and Rubin]
18The city of Abancay is the capital of both the Abancay province and the Apurímac region. Also, as mentioned a �eldtest was conducted by Financiera Con�anza in the pilot phase of the program to calculate the average travel time from eachcommunity to Abancay.
10
this is achieved by correcting actual treatment status Ti ∈ {0, 1}, a endogenous variable in the context of
imperfect compliance, by using the original random assignment statusZi as an instrumental variable (IV).
The validity of this approach rests on two assumptions:
(i) (Yi, Ti) ⊥ Zi
(ii) Ti | Zi = 1 ≥ Ti | Zi = 0 ∀i
The �rst exclusion restriction requires that the dependent outcome variableYi and actual treatment Ti are
jointly independent from assigned treatment Zi, so that the fact that households in treatment communities
were assigned to participate in the program only a�ected the outcome indirectly by increasing their probability
to actually participate in it. The second monotonicity assumption requires that being assigned to participate
in the program makes it more likely that household in treatment communities actually participate in it.
Because in the context of our experimental design we randomly assigned communities to control and treat-
ment groups and participation in the program was signi�cantly higher in the latter19, both assumptions
seem resonable IV Wald estimator will consistently estimate the TOT e�ect. However, as discussed by
[Imbens and Angrist(1994)] this estimated e�ect will di�er from the ITT in that it is a local average treat-
ment e�ect (LATE), so that it only captures the e�ect of the program on those households which actually
participated in it because they were randomly assigned to do so, but would not have participated in it other-
wise. This is, the TOT estimate captures the e�ect of the program on the compliers, a subset of the actually
treated for which (D | Z = 1) = 1and (D | Z = 0) = 0).
The TOT e�ect is therefore given by the IV Wald estimate
δTOT =E(Yi | Zi = 1)− E(Yi | Zi = 0)
E(Ti | Zi = 1)− E(Ti | Zi = 0)
which we calculate using a two stage least squares (2SLS) regression model:
yi = β0 + δTOT T̂i + ui (4)
Ti = γ0 + γ1Zi + vi (5)
where Ti is a dummy variable which takes a value of 1 if the household actually participated in the program20
and 0 otherwise, and the rest of the notation is as above.
As before, we subsequently augment this simple di�ference in means 2SLS regression model to include addi-
tional covariates:
19As mentioned before, at endline the take-up in treatment group was estimated at 57.7%, and the contamination in thecontrol group was estimated at 5.1%.
20Participation in the program is de�ned as having attended one or more �nancial education session.
11
yi = β′
0 + δ'TOT
T̂i +
6∑h=1
β′
hxhi + u′
i (6)
Ti = γ′
0 + γ′
1Zi +
6∑h=1
γ′
hxhi + v′
i (7)
where the notation is as above.
Finally, we again augment the 2SLS regression model to include �xed e�ects by district:
yij = β′′
0 + δ′′
TOTT̂i +
6∑h=1
β′′
hxhi + dj + u′′
i (8)
Tij = γ′′
0 + γ′′
1Zi +
6∑h=1
γ′′
hxhi + dj + v′′
i (9)
where the notation is as above.
5 Results
According to endline survey data we estimate take-up of the program in treatment areas at 57.7%. Conversely,
5.1% of surveyed control households participated in the program despite the fact that it was not o�ered in
their communities.21 Therefore, we have a �rst stage with which to calculate the TOT impact of the program,
which given the take-up and contamination rates is generally approximately double the observed ITT e�ect.
Both sets of results are presented in Tables 2 to 8. Below we divide the results into direct and indirect
impacts since we assume that the �nancial education program �rst impacts �nancial literacy and savings�
direct e�ect�, and then through savings other variables�assets, female empowerment�can be impacted by
the �nancial education program.
5.1 Direct Impact
Financial Literacy
As presented in Table 2 we observe a mild but signi�cant e�ect of the program on �nancial knowledge
variables like the familiarity with formal �nancial institutions, and identifying lending and borrowing as roles
of formal �nancial institutions. However, we observe impacts of the program on practical knowledge and
trust in �nancial institutions. Speci�cally, we observe sign�cant impacts of the program on the likelihood of
21Participation in the program is de�ned as having attended one or more teaching sessions.
12
saving in a formal �nancial institution (∼ 7.5 pp ITT, ∼ 15.1 pp ToT) and the awareness about the deposit
guarantee they o�er and the additional governmental supervision to which they are subject (∼ 8.5 pp ITT,
∼ 17.1 pp ToT). Moreover, these results on awareness and trust are magni�ed when looking at familiarity
with and trust in Financiera Con�anza in particular. We estimate a large and signi�cant positive impact
(∼ 4.6 pp ITT, ∼ 9.2 pp ToT) of the program on the probability of being familiar with Financiera Con�anza,
raising familiarity with the institution from a control average of 2.1%. Similarly, we �nd a very large and
signi�cant positive impact (∼ 21.1 pp ITT, ∼ 42.2 pp ToT) of the program on the probability of claiming to
trust Financiera Con�anza, raising trust in the institution from a control average of 5.4% .
However, we only �nd very mild or no evidence of program impact on other dimensions in this category,
including some which received substantial attention during the teaching sessions (e.g. interest in other
�nancial products like insurance, or the correct use of a budget in the household). This suggests the �nancial
education program was succesful in transmitting the target population the �rules of thumb� that allowed
them to save in the formal system, but it was less successful in transmitting more complex concepts such
as �nancial roles of formal �nancial institutions, or insurance. Moreover, this re�ects the priorities in the
design and implementation of the program, which were promoting savings and generating trust in the formal
�nancial system and the implementing micro�nance institution.
Savings
As presented in Table 3, we �nd a signi�cant positive impact of the program on the probability of saving
with Financiera Con�anza. The observed e�ect is moderate (∼ 0.31σ ITT, ∼ 0.61σ TOT) when looking at
self-reported savings with Financiera Con�anza, raising the probability of having saved with the institution
since January 2015 by ∼ 2.3 pp ITT, ∼ 4.5 pp ToT, from a control average of 0.6%. However, when looking
at the administrative information provided by Financiera Con�anza (in Table 4) we �nd that the program
actually had a considerably larger positive e�ect (∼ 1.30σ ITT, ∼ 2.60σ TOT), raising the probability of
having saved with the institution since January 2015 by ∼ 4.3 pp ITT, ∼ 8.6 pp ToT, from the control
average of 0.1%.22
Moreover, we �nd mild to no evidence of a signi�cant impact of the program on the probability of saving at
home or with other formal �nancial institutions or cooperatives, suggesting that the observed positive e�ect
of the program is limited to increasing the probability of saving with Financiera Con�anza.23 Although not
signi�cant, the point estimates for savings amounts in Table 3 suggest savings were moved from cooperatives
(negative point estimates) to homes and the formal �nancial system.24
22The observed underrreporting of savings with Financiera Con�anza at the endline survey could be simply due to recall bias.However, it could also be due to respondents being afraid to report any savings in a non-governmental �nancial institution (e.g.while enforcement of the rule is unclear, since December 2015 bene�ciaries of the JUNTOS program could lose their conditionalcash transfer if they save in an account other than the one provided by the program at Banco de la Nación, or if they obtaincredit from a formal �nancial institution).
23Since we do not have administrative information from other institutions, it is possible that the probability of savings withthem is underreported, as it is the case with Financiera Con�anza.
24Note that the dispersion of these continuous variables is considerably larger than those of the binary probability of savingvariables, so that the statistical power to detect e�ects of the same size is greatly reduced. For example, while only signi�cantwith 90% con�dence, the observed impact on the maximum amount saved with Financiera Con�anza since January 2015 isabout 20 soles, corresponding with the minimum deposit required to open a savings account at the institution. Transactionalinformation provided by Financiera Con�anza suggests that the average number of deposits was similar to that of withdrawals,with a total of PEN 6,245,725 deposited and PEN 5,817,236 withdrawn as of February 2017, for a net increase in deposits ofPEN 428,489.
13
Credit
As presented in Table 5, we �nd little evidence of an e�ect of the program on having a credit with Financiera
Con�anza (less than 1pp increase, signi�cant only at the 10% level), measured with administrative data. At
the same time, we observe a small but positive impact�only for our preferred speci�cation�on obtaining
a credit with a formal �nancial institution and a cooperative through the use of saving, measured with self-
reported data. This suggests there is room for Financiera Con�anza to o�er competitive credit products to
the target population, where cooperatives are present and have had the monopoly of the rural micro�nance
market. Table 5 also shows that other credit-related variables (e.g., considering themselves creditworthy)
were not singi�cantly a�ected by the �nancial education intervention.25
5.2 Indirect Impact
Income and Assets
As presented in Table 6, we do not �nd consistent signi�cant impacts of the program on agricultural activities.
However, the negative point estimates, some of them signi�cant for our preferred speci�cation, suggest an
incipient departure from livestock activities due to the intervention. For instance, the likelihood of performing
veterenary procedures decreased by ∼ 9.3 pp ITT, ∼ 18.6 pp ToT, for our preferred speci�cation. Moreover,
the total number of income-generating activities or the involvement on new activities aside agriculture and
livestock do not seem to have been a�ected by the program.
Household assets as well as agriculture-related assets were not consistently and signi�cantly a�ected by the
program, but the negative point estimates raise concerns about wether investments on assets were made from
savings, or whether savings were increased due to a decrease in assets.
We also measured the impact of the program on the probability of being poor, as measured by the JUNTOS
conditional cash transfer program poverty score, and �nd no signi�cant e�ects although the related point
estimates were always negative.26 This suggests the �nancial education program, in any case, was not harmful
to the target population.
Consumption
As presented in Table 7, we �nd no signi�cant evidence of an e�ect of the program on many di�erent measures
of food consumption. The program did not improve any good consumption habits, neither it changed any
25Note that there are more strict requirements to qualify for a credit product than to open a savings account, and thatwhile the Ahorro para Todos savings account was actively promoted during program activities, that was not the case for creditproducts such as loans. However, in October 2016 Financiera Con�anza linked the Ahorro para Todos savings account withits Palabra de Mujer program o�ering credit to groups of women, resulting in 19 groups created and PEN 51,350 worth ofcredit extended to 140 customers as of February 2017. This anecdotical evidence suggests that with additional re�nements theprogram may indeed have the potential to generate cross-selling opportunities beyond savings.
26This score was created as a targetting instrument by the JUNTOS conditional cash transfer program using National Surveydata from 2001 and 2004. It estimates the probability that a rural household is poor based on its characteristics: (i) percentageof illiterate adult females in the household, (ii) percentage of children under 18 (legal age) attending school, (iii) type of fuelused for cooking, (iv) appliances in the household, (v) services the household has access to, and (vi) dwelling type.
14
tempation goods consumption. This suggests that, at least, the program positive e�ects on savings did not
come from reducing food consumption.
Female Empowerment
As presented in Table 8, we �nd some evidence of a positive impact of the program on female community
participation. In particular, we �nd a modest e�ect ( ∼ 19.5 pp ITT, ∼ 38.9 pp ToT) on the number of
groups or organizations to which women belong. This result probably comes directly from the fact that the
�nancial education was delivered to groups of women. As the very least, this result suggests the provision
of �nancial education in groups made women believe they belong to a new group of peers. However, we �nd
no evidence of a signi�cant impact of the program on a variety of measures of female empowerment at the
household (e.g., women decides freely how to spend her income) nor at the community level (e.g., the number
of positions that women hold in the groups to which they belong).
Overall, it is possible that the lack of indirect impacts of the program is due to the little timethat passed
between the end of the invention and the survey data collection (nearly three months). We recognize that
important social transformation processes, like female empowerment, require more than a few months and
more than one program to occur.
It is worth mentioning that spillovers cannot be estimated with our design. If spillovers are present, they
would be reducing the magnitude of our impacts, not in�ating them. From the follow-up of training sessions,
we know that people treated by the Program actually received the training, since trainers registered all
participants every session, and that knowledge was hardly transmitted outside the sessions or to people
outside the training groups. Thus, we think spillovers may not the biasing our results importantly .
5.3 Heterogeneous e�ects
Heterogeneous e�ects regressions show that knowledge and trust in Financiera Con�anza do not change with
transaction costs. However, the positive impact on savings with Financiera Con�anza (administrative data)
decreases with the distance from the capital Abancay�where the main branch is located�for our preferred
speci�cations (with control variables and district �xed e�ects).
5.4 Robustness Checks
Our results are robust to Bonferroni adjustment. Moreover, alternative measures of the self-reported savings
in Table 3 do not change the main results on savings.
15
6 Conclusion
We �nd that Ahorro para Todos, a for-pro�t �nancial education program, succesfully increased the familiarity
and trust in formal �nancial institutions in poor rural areas of the Andes. This e�ect was particularly
large for Financiera Con�anza, the Peruvian micro�nance institution which implemented the program in
the Apurímac region of Peru from May 2015 to April 2016: we estimate that it increased familiarity with
Financiera Con�anza by approximately 9.2 percentage points with respect to a control average of 1.2%, and
that it increased trust in the institution by aproximately 42 percentage points with respect to a control average
of 5.4% for the target population that e�ectively received at least one module of the �nancial education and
savings promotion program.
Moreover, we �nd that the �nancial education program increased the probability of saving with Financiera
Con�anza by 8.6 percentage points from the control average of 0.06%, for those e�ectively treated. Although
not sign�cant, self-reported data about savings at home and other institutions suggest the target population
might have been moving their savings from (informal) cooperatives to home and formal �nancial institu-
tions.27This very large e�ect, particularly in the context of a very competitive micro�nance market, suggests
that Ahorro para Todos was a very e�ective marketing tool for Financiera Con�anza which did not generate
�free-riding� from other �nancial institutions. 28
Moreover, we observe very incipient evidence of an e�ect of the program on the probability of buying other
�nancial products from Financiera Con�anza (or from any other �nancial institution), but at the same time
the possibility of expansion with these products in rural areas, still captured by cooperatives. Given that
it is a stylized industry fact that the taking of deposits is a pro�table business only insofar as it leads to
the cross-selling of other products with higher pro�t margins (e.g. loans), this suggests that in order to be
commercially viable the �nancial education program should expand its focus beyond savings. It is possible
that we do not �nd a large e�ect on having a credit with Financiera Con�anza due to the brief amount of
time elapsed between the intervention and the endline survey. Note that we are not able to follow up later
with administrative data since Financiera Con�anza intervined in the control communities right after the
endline survey. However, Financiera Con�anza decided to do the expansion linking the Ahorro para Todos
Program to their group credit for women program (Palabra de Mujer) in order to enhance the cross-selling
of savings and credit products.
Regarding the indirect e�ects of the program, we �nd no signi�cant evidence of the program inducing changes
in assets acquisition, either for the household or for income-generating activities. Moreover, we �nd no
evidence that the program signi�cantly reduced poverty, nor that it caused changes in food consumption.
Similar to to the �ndings of [Boyd and Aldana(2015)] we observe that program did not have any e�ects
on female empowerment at the household level or at the community level�as it is usually expected from
programs targeting women�, except for the positive impact on female community participation, suggesting
that additional e�orts are necessary on this front. The overall lack of indirect e�ects con�rms that the
27Cooperatives have been one of the most important providers of credit and savings in the rural areas of Apurimac. However,as they are currently unregulated in Peru, they have defaulted and left with people's savings. This created an initial mistrust inthe Program, which was addressed di�erentiating cooperatives from formal �nancial institutions. However, people in the controlgroup, or people receiving only a few sessions might still be saving at them.
28Note that the impacts on total savings or savings at other institutions were not signi�cant, but had large variances. Aswell, it is possible that the �free-riding� appears later, since the impacts on savings may vanish over time, as it was found by[Ashraf et al.(2010)Ashraf, Karlan, and Yin] for a Philippine commitment savings account.
16
�nancial education intervention achieved its goal of increasing savings among rural people in the formal
�nancial system, specially in the implementing insitution, without harmful side e�ects.
Similarly, we �nd that Ahorro para Todos had no impact on some dimensions of �nancial education which
were featured prominently in the program, such as the use of a budget in the household. This sug-
gests that despite the observed positive impacts on several fronts, additional e�orts are necessary to maxi-
mize the e�ciency of the program, for example by complementing the �edutainment� approach with other
methodologies with potential to improve learning of best practices such as �rules of thumb� proposed by
[Drexler et al.(2014)Drexler, Fischer, and Schoar]. 29
Moreover, when comparing survey and administrative data we �nd that savings levels are signi�cantly un-
derreported. While this could simply be due to recall bias, it could also be due to respondents being afraid
to report any savings in a non-governmental �nancial institution for fear of losing their bene�ciary status
in social programs. In any case, irrespective of the underlying reason this underreporting has important
implications for the interpretation and validity of any results based on self-reported savings data from poor
populations in Peru.
In conclusion, all the above suggests that �nancial education programs can be a succesful marketing tool
for the promotion of microsavings among the poor even in remote rural areas, and that their provision by
�nancial institutions on a for-pro�t basis can be commercially viable if their focus is expanded beyond savings,
in order to promote and generate cross-selling of credit products with higher pro�t margins. However, this
requires the formal �nancial institutions to design �nancial products suitable to the rural population demand
and being competitive in an enviroment dominated by the informal cooperatives.
29Although budget notebooks were designed as part of the program, they were not implemented due to logistic issues. Theywere supposed to reinforce the lessons learnt on the budgeting module, so as to make doing budgets a habit. If we considerbudgeting as a �nancial ability, it is not surprising that the sole fact of having training sessions did not e�ectively promote thepractice of budgeting.
17
References
[Anagol et al.(2014)Anagol, Etang, and Karlan] Santosh Anagol, Alvin Etang, and Dean Karlan. Continued
existence of cows disproves central tenets of capitalism? Economics Department Working Paper 122,
Yale University, January 2014.
[Angrist et al.(1996)Angrist, Imbens, and Rubin] Joshua D. Angrist, Guido W. Imbens, and Donald B. Ru-
bin. Identi�cation of causal e�ects using instrumental variables. Journal of the American Statistical Asso-
ciation, 91(434):444�455, 1996. doi: 10.1080/01621459.1996.10476902. URL http://www.tandfonline.
com/doi/abs/10.1080/01621459.1996.10476902.
[Ashraf et al.(2010)Ashraf, Karlan, and Yin] Nava Ashraf, Dean Karlan, and Wesley Yin. Female empow-
erment: Impact of a commitment savings product in the philippines. World Development, 38(3):
333 � 344, 2010. ISSN 0305-750X. doi: http://dx.doi.org/10.1016/j.worlddev.2009.05.010. URL
//www.sciencedirect.com/science/article/pii/S0305750X09001910.
[Berry et al.(2015)Berry, Karlan, and Pradhan] James Berry, Dean Karlan, and Menno Pradhan. The im-
pact of �nancial education for youth in ghana. Working Paper 21068, NBER, April 2015.
[Boyd and Aldana(2015)] Chris Boyd and Ursula Aldana. Social Protection, Entrepreneurship and Labour
Market Activation, volume 12 of Policy in Focus, chapter The Impact of Financial Education on Condi-
tional Cash Transfer Bene�ciaries in Peru, pages 26�27. IPC-UNDP, 2015.
[Cole et al.(2009)Cole, Sampson, and Zia] Shawn Cole, Thomas Sampson, and Bilal Zia. Prices or knowl-
edge? what drives demand for �nancial services in emerging markets? Working Paper 09-117, Harvard
Business School, 2009.
[Drexler et al.(2014)Drexler, Fischer, and Schoar] Alejandro Drexler, Greg Fischer, and Antoinette Schoar.
Keeping it simple: Financial literacy and rules of thumb. American Economic Journal: Applied Eco-
nomics, 6(2):1�31, April 2014. doi: 10.1257/app.6.2.1. URL http://www.aeaweb.org/articles?id=
10.1257/app.6.2.1.
[Herrera(2003)] Javier Herrera. La pobreza en el peru 2002. Technical report, Instituto Nacional de Estadis-
tica e Informatica, September 2003.
[Imbens and Angrist(1994)] Guido W. Imbens and Joshua D. Angrist. Identi�cation and estimation of local
average treatment e�ects. Econometrica, 62(2):467�475, 1994. ISSN 00129682, 14680262. URL http:
//www.jstor.org/stable/2951620.
[Imbens and Rubin(2010)] Guido W. Imbens and Donald B. Rubin. Rubin Causal Model, pages 229�241.
Palgrave Macmillan UK, London, 2010. ISBN 978-0-230-28081-6. doi: 10.1057/9780230280816_28. URL
http://dx.doi.org/10.1057/9780230280816_28.
[INEI(INEI2016)] INEI. Evolucion de la pobreza monetaria 2009-2015. Technical report, Instituto Nacional
de Estadistica e Informatica, INEI2016.
[Karlan and Linden(2014)] Dean Karlan and Leigh L. Linden. Loose knots: Strong versus weak commitments
to save for education in uganda. Working Paper 19863, NBER, January 2014.
18
[La Ferrara et al.(2012)La Ferrara, Chong, and Duryea] Eliana La Ferrara, Alberto Chong, and Suzanne
Duryea. Soap operas and fertility: Evidence from brazil. American Economic Journal: Applied Eco-
nomics, 4(4):1�31, July 2012. doi: 10.1257/app.4.4.1. URL http://www.aeaweb.org/articles?id=
10.1257/app.4.4.1.
[MIF(2015)] MIF. Herramientas efectivas para el ahorro inclusivo en america latina y el caribe. Technical
report, Multilateral Investment Fund, March 2015.
[Miller et al.(2014)Miller, Reichelstein, Salas, and Zia] Margaret Miller, Julia Reichelstein, Christian Salas,
and Bilal Zia. Can you help someone become �nancially capable? a meta-analysis of the literature.
Policy Research Working Paper 6745, World Bank, January 2014.
[OECD(2013)] OECD. Oecd/infe toolkit to measure �nancial literacy and �nancial inclusion: Guidance,
core questionnaire and supplementary questions. Technical report, OECD, September 2013.
[Valdivia and Chong(2013)] Martin Valdivia and Alberto Chong. Mini-novelas, tecnologias de las informacion
y la comunicacion (tic), educacion �nanciera y fomento de ahorros. enbreve, (36), February 2013.
T0
- T
1T
0 -
T1
T0
- T
1
(1)
(2)
(3)
-0.0
00
17
70
.00
92
40
.01
04
(0.0
05
24
)(0
.00
81
8)
(0.0
08
22
)
0.0
00
14
60
.00
10
30
.00
11
2
(0.0
02
25
)(0
.00
36
4)
(0.0
03
65
)
0.3
79
-1.0
83
-1.2
03
(0.3
28
)(1
.00
1)
(1.0
16
)
0.8
88
*-1
.44
8-1
.66
4
(0.4
16
)(1
.35
0)
(1.3
63
)
0.0
05
60
-0.0
22
9-0
.02
91
(0.0
08
22
)(0
.01
62
)(0
.01
59
)
-0.0
20
9***
-0.0
09
17
-0.0
11
7
(0.0
05
64
)(0
.01
36
)(0
.01
14
)
0.0
11
10
.01
44
0.0
20
4
(0.0
07
38
)(0
.02
09
)(0
.01
86
)
0.0
07
52
-0.0
02
65
-0.0
05
55
(0.0
05
18
)(0
.01
15
)(0
.01
12
)
0.0
06
45
-0.0
09
64
-0.0
10
6
(0.0
03
60
)(0
.01
30
)(0
.01
34
)
0.0
03
62
-0.0
12
2-0
.01
20
(0.0
02
72
)(0
.01
71
)(0
.01
79
)
0.0
02
42
0.0
07
58
0.0
07
82
(0.0
02
70
)(0
.00
76
2)
(0.0
07
46
)
-0.0
08
57
-0.0
38
3*
-0.0
40
9*
(0.0
05
29
)(0
.01
65
)(0
.01
70
)
10
89
98
98
4
xii
. H
ouse
ho
ld m
emb
ers
rece
ivin
g P
ensi
on 6
5 P
rogra
m (
%)
v.
Fem
ale
ho
use
ho
ld h
ead
(=
1)
vi.
Ho
use
ho
ld m
emb
ers
wit
h E
ssal
ud
Hea
lth I
nsu
rance
(%
)
vii
. H
ouse
ho
ld m
emb
ers
wit
h S
IS H
ealt
h I
nsu
rance
(%
)
vii
i. H
ouse
ho
ld m
emb
ers
wit
ho
ut
Hea
lth I
nsu
rance
(%
)
ix.
Ho
use
ho
ld m
emb
ers
rece
ivin
g V
aso
de
Lec
he
Pro
gra
m (
%)
Ob
serv
atio
ns
*p
<0
.1;
**p
<0
.05
; ***p
<0
.01
;
x.
Ho
use
ho
ld m
emb
ers
rece
ivin
g S
cho
ol
Mea
ls P
rogra
m (
%)
TA
BL
E 1
Bal
ance
Bet
wee
n C
ontr
ol
and
Tre
atm
ent
Gro
up
s
i. M
ale
mem
ber
s in
the
ho
use
ho
ld (
%)
ii.
Pre
gnan
t w
om
en i
n t
he
ho
use
ho
ld (
%)
iii.
Age
of
the
ho
use
ho
ld h
ead
iv.
Age
of
the
ho
use
ho
ld m
emb
ers
(mea
n)
xi.
Ho
use
ho
ld m
emb
ers
rece
ivin
g J
unto
s C
CT
Pro
gra
m (
%)
T0
- T
1T
0 -
T1
T0
- T
1
(1)
(2)
(3)
-0.0
11
40
.04
20
0.0
43
9
(0.0
06
58
)(0
.02
70
)(0
.02
70
)
0.0
19
2**
0.0
06
35
0.0
11
8
(0.0
06
71
)(0
.02
87
)(0
.02
70
)
-0.0
14
4*
-0.0
06
94
-0.0
12
6
(0.0
06
39
)(0
.02
76
)(0
.02
61
)
0.0
00
76
4-0
.00
71
9-0
.00
82
4*
(0.0
01
77
)(0
.00
41
7)
(0.0
04
13
)
-0.0
21
9***
-0.0
02
77
-0.0
04
13
(0.0
06
50
)(0
.02
03
)(0
.02
02
)
0.0
28
2***
0.0
18
00
.02
08
(0.0
05
16
)(0
.02
19
)(0
.02
14
)
-0.0
12
2***
-0.0
00
66
7-0
.00
08
93
(0.0
02
83
)(0
.00
51
8)
(0.0
04
52
)
-56
.04
***
5.2
38
0.6
08
(2.6
89
)(1
4.0
2)
(13
.21
)
0.0
29
4-0
.01
08
-0.0
54
2
(0.0
18
3)
(0.1
89
)(0
.19
2)
0.0
60
5***
0.1
55
0.1
56
(0.0
09
41
)(0
.09
48
)(0
.09
43
)
10
89
98
98
4O
bse
rvat
ions
*p
<0
.1;
**p
<0
.05
; ***p
<0
.01
;
Note
s: C
olu
mn (
1)
rep
ort
s th
e b
alan
ce b
etw
een c
ontr
ol
and
tre
atm
ent
com
munit
ies
(T0
- T
1)
for
the
rand
om
izat
ion t
hat
max
imiz
ed b
alan
ced
pri
or
to
the
inte
rven
tio
n,
usi
ng h
ouse
ho
ld l
evel
dat
a fr
om
SIS
FO
H f
or
var
iab
les
i to
xix
. V
aria
ble
s x
x t
o x
xii
wer
e ca
lcula
ted
at
the
com
munit
y l
evel
usi
ng t
he
Junto
s R
oll
of
Use
rs.
Co
lum
n (
2)
rep
ort
s th
e b
alan
ce b
etw
een c
ontr
ol
and
tre
atm
ent
com
munit
ies
usi
ng a
ver
ages
of
each
var
iab
le a
t th
e co
mm
unit
y
level
, fo
r th
e 8
9 c
om
munit
ies
assi
gned
to
tre
atm
ent
or
contr
ol
pri
or
to t
he
inte
rven
tio
n.
Co
lum
n (
3)
rep
ort
s th
e b
alan
ce b
etw
een c
ontr
ol
and
tre
atm
ent
com
munit
ies
usi
ng a
ver
ages
of
each
var
iab
le a
t th
e co
mm
unit
y l
evel
, w
her
e co
mm
unit
ies
are
regro
up
ed a
cco
rdin
g t
o t
he
actu
al i
nte
rven
tio
n (
trai
nin
g
sess
ions)
, an
d t
hey
co
mp
rise
the
bal
ance
of
var
iab
les
at t
he
com
munit
y l
evel
use
d i
n t
he
regre
ssio
ns.
Sta
nd
ard
err
ors
rep
ort
ed i
n p
aren
thes
es.
Sourc
e: S
ISF
OH
20
13
and
Junto
s R
oll
of
Use
rs b
y S
epte
mb
er 2
01
4.
xvii
. H
ouse
ho
ld m
emb
ers
who
kno
w h
ow
to
rea
d (
%)
xvii
i. H
ouse
ho
ld m
emb
ers
in a
gri
cult
ura
l ac
tivit
ies
(%)
xix
. H
ouse
ho
ld m
emb
ers
in s
ervic
es a
ctiv
itie
s (%
)
xx
. N
um
ber
of
Junto
s C
CT
rec
ipie
nts
in t
he
com
munit
y
xx
i. D
ista
nce
fro
m t
he
com
munit
y t
o t
he
regio
nal
cap
ital
xx
ii.
Junto
s C
CT
rec
ipie
nts
in t
he
com
munit
y r
ecei
ven
g t
he
tran
sfer
at
a
BN
(%
)
xvi.
Ho
use
ho
ld m
emb
ers
wit
h a
Nat
ional
Id
enti
ty D
ocu
men
t (%
)TA
BL
E 1
(C
onti
nued
)
Bal
ance
Bet
wee
n C
ontr
ol
and
Tre
atm
ent
Gro
up
s
xii
i. H
ouse
ho
ld m
emb
ers
no
t re
ceiv
ing a
ny S
oci
al P
rogra
m (
%)
xiv
. H
ouse
ho
ld m
emb
ers
wit
h S
pan
ish a
s fi
rst
tongue
(%)
xv.
Ho
use
ho
ld m
emb
ers
wit
h Q
uec
hua
as f
irst
to
ngue
(%)
(1.1
)(1
.2)
(1.3
)(2
.1)
(2.2
)(2
.3)
0.0
46
50
.06
42
*0
.07
54
**
0.0
88
30
.12
34
*0
.15
12
**
0.2
74
51
,80
1
(0.0
36
7)
(0.0
35
5)
(0.0
31
9)
(0.0
68
9)
(0.0
68
4)
(0.0
65
2)
(0.4
46
5)
0.1
23
80
.14
40
0.1
30
1*
0.2
35
20
.27
67
*0
.26
09
*1
.05
40
1,8
01
(0.1
00
9)
(0.0
87
3)
(0.0
72
4)
(0.1
91
6)
(0.1
66
6)
(0.1
45
5)
(0.8
20
8)
0.0
56
1*
**
0.0
53
0*
**
0.0
45
9*
**
0.1
06
6*
**
0.1
01
9*
**
0.0
92
0*
**
0.0
12
11
,80
1
(0.0
20
4)
(0.0
13
4)
(0.0
14
5)
(0.0
38
3)
(0.0
25
7)
(0.0
27
4)
(0.1
09
5)
0.0
11
0-0
.01
57
-0.0
23
40
.02
09
-0.0
30
3-0
.04
69
0.3
72
71
,80
1
(0.0
57
6)
(0.0
47
5)
(0.0
41
4)
(0.1
08
6)
(0.0
90
2)
(0.0
81
7)
(0.4
83
8)
0.0
70
20
.06
54
**
0.0
85
1*
**
0.1
33
40
.12
57
**
0.1
70
7*
**
0.2
15
01
,80
1
(0.0
45
6)
(0.0
32
4)
(0.0
31
7)
(0.0
84
9)
(0.0
60
7)
(0.0
61
3)
(0.4
11
0)
0.1
85
3*
**
0.1
80
5*
**
0.2
10
7*
**
0.3
52
1*
**
0.3
47
0*
**
0.4
22
4*
**
0.0
54
01
,80
1
(0.0
53
9)
(0.0
35
7)
(0.0
36
4)
(0.0
98
5)
(0.0
71
1)
(0.0
73
2)
(0.2
26
2)
0.0
59
50
.07
48
*0
.07
12
**
0.1
13
10
.14
39
*0
.14
28
**
0.1
72
01
,80
1
(0.0
50
1)
(0.0
40
0)
(0.0
30
9)
(0.0
95
2)
(0.0
76
6)
(0.0
61
0)
(0.3
77
6)
0.0
17
4-0
.01
04
-0.0
15
60
.03
31
-0.0
20
1-0
.03
12
0.5
07
21
,80
1
(0.0
51
6)
(0.0
43
7)
(0.0
43
3)
(0.0
97
1)
(0.0
82
7)
(0.0
84
9)
(0.5
00
2)
0.0
38
80
.03
15
0.0
62
3*
*0
.07
38
0.0
60
50
.12
50
**
0.1
32
31
,80
1
(0.0
26
2)
(0.0
24
0)
(0.0
26
2)
(0.0
50
6)
(0.0
46
4)
(0.0
55
6)
(0.3
39
0)
0.0
27
1-0
.00
85
-0.0
33
00
.05
15
-0.0
16
3-0
.06
62
0.7
08
91
,80
1
(0.0
34
4)
(0.0
32
1)
(0.0
35
7)
(0.0
64
9)
(0.0
61
1)
(0.0
71
7)
(0.4
54
5)
0.0
31
90
.03
25
0.0
48
5*
**
0.0
60
50
.06
25
0.0
97
2*
**
0.1
13
61
,80
1
(0.0
32
1)
(0.0
25
3)
(0.0
17
4)
(0.0
60
0)
(0.0
48
5)
(0.0
35
2)
(0.3
17
5)
0.0
17
30
.02
10
0.0
03
10
.03
29
0.0
40
30
.00
61
0.1
12
51
,80
1
(0.0
22
8)
(0.0
21
3)
(0.0
20
4)
(0.0
42
6)
(0.0
40
3)
(0.0
40
1)
(0.3
16
1)
f. T
rust
s F
inan
cier
a C
on
fian
za (
1 =
Yes
)
g.
Kn
ow
s d
epo
sits
at
form
al f
inan
cial
in
stit
uti
on
s ar
e
gu
aran
teed
in
cas
e o
f b
ankru
ptc
y (
1 =
Yes
)
Note
s:
This
tab
le r
epo
rts
dif
fere
nce
s fi
nan
cial
lit
erac
y i
n b
etw
een t
he
rand
om
ly a
ssig
ned
tre
atm
ent
and
co
ntr
ol
gro
up
s at
end
line.
Eac
h c
ell
in c
olu
mns
(1.x
) an
d (
2.x
) re
po
rts
resu
lts
for
a si
ngle
reg
ress
ion o
f th
e in
dep
end
ent
var
iab
le i
n t
he
corr
esp
ond
ing r
ow
, w
ith c
lust
ered
sta
nd
ard
err
ors
at
the
com
unit
y l
evel
in p
aren
thes
es b
elo
w e
ach e
stim
ate.
Co
lum
ns
(1.x
)
pre
sent
OL
S I
nte
nt-
to-T
reat
(IT
T)
esti
mat
es.
Co
lum
ns
(2.x
) p
rese
nt
2S
LS
Tre
atm
ent-
on-t
he-
Tre
ated
(T
OT
) es
tim
ates
, w
ith a
ctual
tre
atm
ent
as t
he
end
ogen
ous
var
iab
le a
nd
ran
do
m
assi
gnm
ent
as t
he
inst
rum
ent.
Co
lum
ns
(x.0
) p
rese
nt
the
dif
fere
nce
in m
eans
bet
wee
n t
reat
men
t an
d c
ontr
ol
wit
ho
ut
add
itio
nal
co
var
iate
s. C
olu
mns
(x.1
) p
rese
nt
the
dif
fere
nce
in
mea
ns
wit
h a
dd
itio
nal
co
var
iate
s: (
i) d
ista
nce
to
reg
ional
cap
ital
; (i
i) n
um
ber
of
ho
use
ho
ld m
emb
ers;
(ii
i) g
end
er o
f ho
use
ho
ld h
ead
; (i
v)
age
of
resp
ond
ent;
(v)
max
imum
ed
uca
tio
nal
level
of
ho
use
ho
ld;
and
(vi)
ben
efic
iary
sta
tus
of
JUN
TO
S c
ond
itio
nal
cas
h t
ransf
er p
rogra
m,
bes
ides
all
bal
ance
var
iab
les
in T
able
1.
Co
lum
ns
(x.2
) p
rese
nt
the
dif
fere
nce
in m
eans
wit
h a
dd
itio
nal
co
var
iate
s an
d d
istr
ict
fixed
eff
ects
. C
olu
mn (
3)
rep
ort
s th
e m
ean v
alue
of
each
ind
epen
den
t var
iab
le a
t en
dli
ne,
wit
h s
tand
ard
dev
iati
ons
in p
aren
thes
es.
Co
lum
n (
4)
rep
ort
s th
e num
ber
of
ob
serv
atio
ns
in e
ach r
egre
ssio
n. Sourc
e:
End
line
surv
ey o
f 1
80
1 h
ouse
ho
lds
carr
ied
out
in J
uly
20
16
, S
ISF
OH
20
13
and
Junto
s R
oll
of
Use
rs b
y S
epte
mb
er
20
14
.
k.
Mak
es a
bu
dget
(1
= Y
es)
h.
If i
n d
ou
bt
asks
per
son
nel
of
form
al f
inan
cial
in
stit
uti
on
s fo
r
info
rmat
ion
ab
ou
t sa
vin
gs
(1 =
Yes
)
i. H
as r
equ
este
d i
nfo
rmat
ion
ab
ou
t fi
nan
cial
pro
du
cts
fro
m a
ny
form
al f
inan
cial
in
stit
uti
on
(1
= Y
es)
j. I
f s/
he
cou
ld,
s/h
e w
ou
ld b
uy i
nsu
ran
ce f
or
pea
ce o
f m
ind
(1
=Y
es)
a. W
ou
ld s
ave
in a
fin
anci
al i
nst
itu
tio
n (
1 =
Yes
)
*p
<0
.1;
**p
<0
.05
; *
**p
<0
.01
;
l. M
akes
a b
ud
get
co
rrec
tly (
1 =
Yes
)
b.
Wit
h h
ow
man
y f
orm
al f
inan
cial
in
stit
uti
on
s is
fam
ilia
r
c. I
s fa
mil
iar
wit
h F
inan
cier
a C
on
fian
za (
1 =
Yes
)
Est
imat
ed I
mp
act
of
the
Pro
gra
m o
n F
inan
cial
Lit
erac
y
Co
ntr
ol
Mea
nN
TA
BL
E 2
d.
Iden
tifi
es l
end
ing a
nd
tak
ing d
epo
sits
as
role
s o
f fo
rmal
fin
anci
al i
nst
itu
tio
ns
(1 =
Yes
)
e. K
no
ws
form
al f
inan
cial
in
stit
uti
on
s ar
e su
per
vis
ed b
y t
he
go
ver
nm
ent
(1 =
Yes
)
Inte
nt-
to-T
reat
(IT
T)
Tre
atm
ent-
on
-th
e-T
reat
ed (
TO
T)
(1.1
)(1
.2)
(1.3
)(2
.1)
(2.2
)(2
.3)
0.0
24
70
.01
38
0.0
22
40
.04
69
0.0
26
60
.04
48
0.5
45
81
,80
1
(0.0
72
8)
(0.0
61
9)
(0.0
23
9)
(0.1
37
5)
(0.1
17
2)
(0.0
46
8)
(0.4
98
2)
0.0
48
10
.03
53
0.0
49
8*
0.0
91
40
.06
79
0.0
99
9*
0.2
30
41
,80
1
(0.0
31
1)
(0.0
28
0)
(0.0
25
2)
(0.0
58
6)
(0.0
53
7)
(0.0
51
6)
(0.4
21
3)
55
.28
39
25
8.0
34
5*
41
4.2
77
1*
*1
05
.03
22
49
6.2
74
4*
83
0.9
60
3*
**
97
4.3
49
71
,79
9
(15
5.4
97
4)
(15
3.3
22
9)
(17
1.6
81
0)
(29
7.4
98
2)
(29
0.0
80
3)
(32
2.2
58
4)
(17
06
.98
20
)
0.0
20
2*
*0
.02
16
**
*0
.02
27
**
*0
.03
84
**
0.0
41
5*
**
0.0
45
4*
**
0.0
05
51
,80
1
(0.0
08
3)
(0.0
07
1)
(0.0
06
2)
(0.0
16
0)
(0.0
13
8)
(0.0
12
4)
(0.0
74
1)
0.1
63
71
.82
01
2.8
23
20
.31
09
3.4
98
65
.66
03
14
.33
30
1,8
01
(11
.62
41
)(8
.71
36
)(7
.84
25
)(2
1.9
43
8)
(16
.54
52
)(1
5.5
47
4)
(28
8.3
18
7)
0.1
02
1*
*0
.10
20
**
*0
.11
48
**
*0
.19
39
**
0.1
96
0*
**
0.2
30
2*
**
0.0
27
91
,80
1
(0.0
45
0)
(0.0
36
2)
(0.0
34
9)
(0.0
86
5)
(0.0
70
6)
(0.0
71
2)
(0.4
88
7)
0.0
19
1*
**
0.0
19
2*
**
0.0
20
8*
**
0.0
36
2*
**
0.0
37
0*
**
0.0
41
7*
**
0.0
03
31
,80
1
(0.0
07
1)
(0.0
06
1)
(0.0
06
3)
(0.0
13
7)
(0.0
11
8)
(0.0
12
5)
(0.0
57
4)
0.0
17
9*
**
0.0
18
0*
**
0.0
18
4*
**
0.0
34
1*
**
0.0
34
6*
**
0.0
36
9*
**
0.0
03
31
,80
1
(0.0
06
7)
(0.0
05
6)
(0.0
05
8)
(0.0
12
8)
(0.0
10
9)
(0.0
11
4)
(0.0
57
4)
0.0
14
6*
*0
.01
37
**
*0
.01
33
**
*0
.02
77
**
0.0
26
4*
**
0.0
26
7*
**
0.0
03
31
,80
1
(0.0
06
8)
(0.0
05
2)
(0.0
04
8)
(0.0
13
0)
(0.0
10
1)
(0.0
09
9)
(0.0
57
4)
0.0
02
60
.00
32
0.0
02
60
.00
50
0.0
06
20
.00
52
0.0
27
61
,80
1
(0.0
09
6)
(0.0
08
0)
(0.0
07
8)
(0.0
18
2)
(0.0
15
2)
(0.0
15
3)
(0.1
63
8)
24
.90
29
41
.18
22
11
6.1
71
4*
47
.30
21
79
.16
08
23
2.9
16
3*
10
7.9
38
31
,80
1
(73
.11
14
)(6
8.3
40
5)
(68
.04
20
)(1
38
.40
03
)(1
30
.08
07
)(1
34
.72
47
)(1
26
7.1
09
0)
-0.0
08
7-0
.01
44
0.0
03
0-0
.01
66
-0.0
27
60
.00
60
0.0
92
61
,80
1
(0.0
19
9)
(0.0
19
9)
(0.0
18
5)
(0.0
37
6)
(0.0
37
6)
(0.0
36
3)
(0.2
90
0)
-23
6.5
20
3-1
10
.67
39
-78
.65
50
-44
9.2
61
5-2
12
.73
84
-15
7.6
98
35
40
.75
96
1,8
01
(19
7.4
44
6)
(14
3.4
20
6)
(17
2.6
07
1)
(37
0.6
64
2)
(27
1.7
94
5)
(34
0.2
10
0)
(42
70
.64
40
)
-25
8.2
77
0-9
4.3
88
75
5.1
69
1-4
90
.69
21
-18
1.5
36
51
10
.65
85
78
6.3
49
21
,79
9
(19
6.4
72
2)
(12
9.2
51
6)
(14
1.8
75
3)
(36
4.6
57
8)
(24
4.7
84
8)
(27
7.4
86
1)
(33
94
.84
80
)
Inte
nt-
to-T
reat
(IT
T)
Tre
atm
ent-
on
-th
e-T
reat
ed (
TO
T)
f.3
. H
as m
ore
th
an S
/.2
0 s
aved
wit
h F
C a
t th
e ti
me
of
the
surv
ey
f.4
. H
as m
ore
th
an S
/.5
0 s
aved
wit
h F
C a
t th
e ti
me
of
the
surv
ey
Est
imat
ed I
mp
act
of
the
pro
gra
m o
n S
avin
gs
(Su
rvey
Dat
a)
Co
ntr
ol
Mea
nN
TA
BL
E 3
d.
Sin
ce J
anu
ary 2
01
5 h
as s
aved
wit
h F
inan
cier
a C
on
fian
za (
1 =
Yes
)
a. C
urr
enty
sav
es (
1 =
Yes
)
b.
Sin
ce t
he
star
t o
f 2
01
5 h
as s
aved
fo
r a
spec
ific
go
al (
1 =
Yes
)
c. M
axim
um
am
ou
nt
saved
at
ho
me
sin
ce t
he
star
t o
f 2
01
5
k.
To
tal
savin
gs
(at
ho
me,
fo
rmal
an
d i
nfo
rmal
fin
anci
al
inst
itu
tio
ns)
*p
<0
.1;
**p
<0
.05
; *
**p
<0
.01
;
e. M
axim
um
am
ou
nt
saved
wit
h F
inan
cier
a C
on
fian
za s
ince
Jan
uar
y 2
01
5
f.1
. L
og o
f am
ou
nt
saved
wit
h F
C a
t th
e ti
me
of
the
surv
ey
g.
Sin
ce J
anu
ary 2
01
5 h
as s
aved
wit
h o
ther
fo
rmal
fin
anci
al
inst
itu
tio
ns
(1 =
Yes
)
h.
Max
imu
m a
mo
un
t sa
ved
wit
h o
ther
fo
rmal
fin
anci
al i
nst
itu
tio
ns
sin
ce J
anu
ary 2
01
5
i. S
ince
Jan
uar
y 2
01
5 h
as s
aved
wit
h a
fin
anci
al c
oo
per
ativ
e (1
=
Yes
)
j. M
axim
um
am
ou
nt
saved
wit
h a
co
op
erat
ive
sin
ce J
anu
ary 2
01
5
f.2
. H
as m
ore
th
an S
/.5
sav
ed w
ith
FC
at
the
tim
e o
f th
e su
rvey
(1.1
)(1
.2)
(1.3
)(2
.1)
(2.2
)(2
.3)
0.0
55
9*
**
0.0
50
6*
**
0.0
43
0*
**
0.1
06
3*
**
0.0
97
3*
**
0.0
86
2*
**
0.0
01
11
,80
1
(0.0
14
4)
(0.0
10
8)
(0.0
11
1)
(0.0
27
6)
(0.0
21
6)
(0.0
21
2)
(0.0
33
2)
25
.01
17
**
16
.55
33
**
3.5
25
64
7.5
08
8*
*3
1.8
18
9*
*7
.06
86
0.1
26
31
,80
1
(11
.42
50
)(7
.35
53
)(7
.22
20
)(2
1.6
68
6)
(14
.32
51
)(1
4.1
72
8)
(3.8
02
9)
14
.31
42
**
9.8
58
7*
*3
.39
91
27
.18
93
**
18
.95
05
**
6.8
15
00
.02
48
1,8
01
(6.5
44
6)
(4.0
10
0)
(4.1
35
4)
(12
.46
75
)(7
.84
84
)(8
.09
13
)(0
.74
83
)
0.2
55
5*
**
0.2
28
2*
**
0.1
75
0*
**
0.4
85
3*
**
0.4
38
6*
**
0.3
50
9*
**
0.0
04
21
,80
1
(0.0
72
1)
(0.0
50
6)
(0.0
51
4)
(0.1
38
3)
(0.1
01
5)
(0.0
98
8)
(0.1
26
5)
0.0
54
8*
**
0.0
50
4*
**
0.0
43
0*
**
0.1
04
1*
**
0.0
97
0*
**
0.0
86
3*
**
0.0
01
11
,80
1
(0.0
14
5)
(0.0
10
9)
(0.0
11
1)
(0.0
27
8)
(0.0
21
8)
(0.0
21
2)
(0.0
33
2)
0.0
44
8*
**
0.0
41
6*
**
0.0
29
9*
**
0.0
85
0*
**
0.0
80
0*
**
0.0
60
0*
**
0.0
01
11
,80
1
(0.0
13
5)
(0.0
09
9)
(0.0
10
1)
(0.0
25
8)
(0.0
19
7)
(0.0
19
5)
(0.0
33
2)
0.0
25
7*
**
0.0
22
1*
**
0.0
13
1*
*0
.04
89
**
*0
.04
25
**
*0
.02
63
**
0.0
00
01
,80
1
(0.0
08
4)
(0.0
05
4)
(0.0
05
3)
(0.0
16
1)
(0.0
10
7)
(0.0
10
4)
(0.0
00
0)
e. H
as h
ad m
ore
th
an S
/.5
sav
ed w
ith
FC
(=
1)
on
aver
age
f. H
as h
ad m
ore
th
an S
/.2
0 s
aved
wit
h F
C (
=1
) o
n a
ver
age
g.
Has
had
mo
re t
han
S/.
50
sav
ed w
ith
FC
(=
1)
on
aver
age
Note
s:
This
tab
le r
epo
rts
dif
fere
nce
s fi
nan
cial
lit
erac
y i
n b
etw
een t
he
rand
om
ly a
ssig
ned
tre
atm
ent
and
co
ntr
ol
gro
up
s at
end
line.
Eac
h c
ell
in c
olu
mns
(1.x
) an
d (
2.x
) re
po
rts
resu
lts
for
a si
ngle
reg
ress
ion o
f th
e in
dep
end
ent
var
iab
le i
n t
he
corr
esp
ond
ing r
ow
, w
ith c
lust
ered
sta
nd
ard
err
ors
at
the
com
unit
y l
evel
in p
aren
thes
es b
elo
w e
ach e
stim
ate.
Co
lum
ns
(1.x
)
pre
sent
OL
S I
nte
nt-
to-T
reat
(IT
T)
esti
mat
es.
Co
lum
ns
(2.x
) p
rese
nt
2S
LS
Tre
atm
ent-
on-t
he-
Tre
ated
(T
OT
) es
tim
ates
, w
ith a
ctual
tre
atm
ent
as t
he
end
ogen
ous
var
iab
le a
nd
ran
do
m
assi
gnm
ent
as t
he
inst
rum
ent.
Co
lum
ns
(x.0
) p
rese
nt
the
dif
fere
nce
in m
eans
bet
wee
n t
reat
men
t an
d c
ontr
ol
wit
ho
ut
add
itio
nal
co
var
iate
s. C
olu
mns
(x.1
) p
rese
nt
the
dif
fere
nce
in
mea
ns
wit
h a
dd
itio
nal
co
var
iate
s: (
i) d
ista
nce
to
reg
ional
cap
ital
; (i
i) n
um
ber
of
ho
use
ho
ld m
emb
ers;
(ii
i) g
end
er o
f ho
use
ho
ld h
ead
; (i
v)
age
of
resp
ond
ent;
(v)
max
imum
ed
uca
tio
nal
level
of
ho
use
ho
ld;
and
(vi)
ben
efic
iary
sta
tus
of
JUN
TO
S c
ond
itio
nal
cas
h t
ransf
er p
rogra
m,
bes
ides
all
bal
ance
var
iab
les
in T
able
1.
Co
lum
ns
(x.2
) p
rese
nt
the
dif
fere
nce
in m
eans
wit
h a
dd
itio
nal
co
var
iate
s an
d d
istr
ict
fixed
eff
ects
. C
olu
mn (
3)
rep
ort
s th
e m
ean v
alue
of
each
ind
epen
den
t var
iab
le a
t en
dli
ne,
wit
h s
tand
ard
dev
iati
ons
in p
aren
thes
es.
Co
lum
n (
4)
rep
ort
s th
e num
ber
of
ob
serv
atio
ns
in e
ach r
egre
ssio
n. Sourc
e:
End
line
surv
ey o
f 1
80
1 h
ouse
ho
lds
carr
ied
out
in J
uly
20
16
, S
ISF
OH
20
13
and
Junto
s R
oll
of
Use
rs b
y S
epte
mb
er
20
14
.
a. S
ince
Jan
uar
y 2
01
5 h
as s
aved
wit
h F
inan
cier
a C
on
fian
za (
1
= Y
es)
*p
<0
.1;
**p
<0
.05
; *
**p
<0
.01
;
b.
Max
imu
m a
mo
un
t sa
ved
wit
h F
inan
cier
a C
on
fian
za s
ince
Jan
uar
y 2
01
5
c. A
ver
age
amo
un
t sa
ved
wit
h F
inan
cier
a C
on
fian
za (
Ap
ril
20
14
-Ju
ly 2
01
6)
d.
Lo
g o
f av
erag
e am
ou
nt
saved
wit
h F
inan
cier
a C
on
fian
za
(Ap
ril
20
14
-Ju
ly 2
01
6)
Inte
nt-
to-T
reat
(IT
T)
Tre
atm
ent-
on
-th
e-T
reat
ed (
TO
T)
Est
imat
ed I
mp
act
of
the
pro
gra
m o
n S
avin
gs
at F
inan
cier
a C
on
fian
za (
Ad
min
istr
ativ
e D
ata)
Co
ntr
ol
Mea
nN
TA
BL
E 4
(1.1
)(1
.2)
(1.3
)(2
.1)
(2.2
)(2
.3)
-0.0
25
30
.02
10
0.0
17
9-0
.04
80
0.0
40
40
.03
58
0.5
56
41
,80
1
(0.0
45
9)
(0.0
35
8)
(0.0
36
0)
(0.0
86
5)
(0.0
68
3)
(0.0
70
8)
(0.4
97
)
0.0
02
60
.02
20
0.0
36
70
.00
49
0.0
42
40
.07
35
0.1
78
81
,80
1
(0.0
32
9)
(0.0
32
1)
(0.0
34
7)
(0.0
62
1)
(0.0
61
4)
(0.0
69
1)
(0.3
83
3)
0.0
01
60
.02
18
0.0
41
40
.00
31
0.0
41
90
.08
30
0.1
88
21
,80
1
(0.0
36
0)
(0.0
35
4)
(0.0
39
9)
(0.0
68
0)
(0.0
67
7)
(0.0
79
2)
(0.4
14
5)
0.0
03
4*
0.0
02
6*
0.0
02
60
.00
64
*0
.00
49
*0
.00
52
*0
.00
17
1,8
01
(0.0
01
8)
(0.0
01
3)
(0.0
01
6)
(0.0
03
4)
(0.0
02
5)
(0.0
03
1)
(0.0
40
8)
0.0
01
20
.00
47
0.0
01
40
.00
23
0.0
09
10
.00
27
0.0
08
31
,80
1
(0.0
04
8)
(0.0
05
6)
(0.0
05
4)
(0.0
09
1)
(0.0
10
7)
(0.0
10
6)
(0.0
90
9)
-0.0
00
80
.01
25
0.0
26
6*
*-0
.00
15
0.0
24
10
.05
33
**
0.0
28
41
,79
8
(0.0
09
6)
(0.0
09
7)
(0.0
11
0)
(0.0
18
1)
(0.0
18
7)
(0.0
22
9)
(0.1
66
1)
0.0
13
90
.01
94
0.0
32
2*
*0
.02
63
0.0
37
30
.06
47
**
0.0
50
11
,79
7
(0.0
16
4)
(0.0
16
7)
(0.0
15
2)
(0.0
31
1)
(0.0
32
3)
(0.0
31
9)
(0.2
18
2)
e. B
elie
ves
th
at s
avin
g i
n a
fo
rmal
fin
anci
al i
nst
itu
tio
n m
akes
it
easi
er t
o o
bta
in a
lo
an (
1 =
Yes
)
f. I
n t
he
pas
t h
as o
bta
ined
a l
oan
usi
ng s
avin
gs
in a
fo
rmal
fin
anci
al i
nst
itu
tio
n a
s co
llat
eral
(1
= Y
es)
Note
s:
This
tab
le r
epo
rts
dif
fere
nce
s fi
nan
cial
lit
erac
y i
n b
etw
een t
he
rand
om
ly a
ssig
ned
tre
atm
ent
and
co
ntr
ol
gro
up
s at
end
line.
Eac
h c
ell
in c
olu
mns
(1.x
) an
d (
2.x
) re
po
rts
resu
lts
for
a si
ngle
reg
ress
ion o
f th
e in
dep
end
ent
var
iab
le i
n t
he
corr
esp
ond
ing r
ow
, w
ith c
lust
ered
sta
nd
ard
err
ors
at
the
com
unit
y l
evel
in p
aren
thes
es b
elo
w e
ach e
stim
ate.
Co
lum
ns
(1.x
)
pre
sent
OL
S I
nte
nt-
to-T
reat
(IT
T)
esti
mat
es.
Co
lum
ns
(2.x
) p
rese
nt
2S
LS
Tre
atm
ent-
on-t
he-
Tre
ated
(T
OT
) es
tim
ates
, w
ith a
ctual
tre
atm
ent
as t
he
end
ogen
ous
var
iab
le a
nd
ran
do
m
assi
gnm
ent
as t
he
inst
rum
ent.
Co
lum
ns
(x.0
) p
rese
nt
the
dif
fere
nce
in m
eans
bet
wee
n t
reat
men
t an
d c
ontr
ol
wit
ho
ut
add
itio
nal
co
var
iate
s. C
olu
mns
(x.1
) p
rese
nt
the
dif
fere
nce
in
mea
ns
wit
h a
dd
itio
nal
co
var
iate
s: (
i) d
ista
nce
to
reg
ional
cap
ital
; (i
i) n
um
ber
of
ho
use
ho
ld m
emb
ers;
(ii
i) g
end
er o
f ho
use
ho
ld h
ead
; (i
v)
age
of
resp
ond
ent;
(v)
max
imum
ed
uca
tio
nal
level
of
ho
use
ho
ld;
and
(vi)
ben
efic
iary
sta
tus
of
JUN
TO
S c
ond
itio
nal
cas
h t
ransf
er p
rogra
m,
bes
ides
all
bal
ance
var
iab
les
in T
able
1.
Co
lum
ns
(x.2
) p
rese
nt
the
dif
fere
nce
in m
eans
wit
h a
dd
itio
nal
co
var
iate
s an
d d
istr
ict
fixed
eff
ects
. C
olu
mn (
3)
rep
ort
s th
e m
ean v
alue
of
each
ind
epen
den
t var
iab
le a
t en
dli
ne,
wit
h s
tand
ard
dev
iati
ons
in p
aren
thes
es.
Co
lum
n (
4)
rep
ort
s th
e num
ber
of
ob
serv
atio
ns
in e
ach r
egre
ssio
n. Sourc
e:
End
line
surv
ey o
f 1
80
1 h
ouse
ho
lds
carr
ied
out
in J
uly
20
16
, S
ISF
OH
20
13
and
Junto
s R
oll
of
Use
rs b
y S
epte
mb
er
20
14
.
a. C
on
sid
ers
her
/him
self
cre
dit
wo
rth
y (
1 =
Yes
)
*p
<0
.1;
**p
<0
.05
; *
**p
<0
.01
;
g.
In t
he
pas
t h
as o
bta
ined
a l
oan
usi
ng s
avin
gs
in a
fin
anci
al
coo
per
ativ
e as
co
llat
eral
(1
= Y
es)
b.
Sin
ce J
anu
ary 2
01
5 h
as m
ade
pay
men
ts t
ow
ard
s o
ne
or
mo
re
loan
s (1
= Y
es)
c. N
um
ber
of
loan
s si
nce
Jan
uar
y 2
01
5 i
n a
ny f
inan
cial
inst
itu
tio
n
d.
Has
a l
oan
wit
h F
inan
cier
a C
on
fian
za (
1 =
Yes
)
(Ad
min
istr
ativ
e D
ata)
Inte
nt-
to-T
reat
(IT
T)
Tre
atm
ent-
on
-th
e-T
reat
ed (
TO
T)
Est
imat
ed I
mp
act
of
the
pro
gra
m o
n C
red
it
Co
ntr
ol
Mea
nN
TA
BL
E 5
(1.1
)(1
.2)
(1.3
)(2
.1)
(2.2
)(2
.3)
0.2
03
30
.01
25
-0.0
91
20
.38
61
0.0
24
0-0
.18
29
3.4
65
91
,80
1
(0.1
67
4)
(0.1
07
7)
(0.0
69
1)
(0.3
07
8)
(0.2
04
1)
(0.1
37
1)
(1.1
36
3)
-0.0
55
4*
*0
.00
03
-0.0
02
9-0
.10
52
**
0.0
00
5-0
.00
58
0.5
61
81
,80
1
(0.0
27
2)
(0.0
18
1)
(0.0
16
5)
(0.0
50
9)
(0.0
34
3)
(0.0
32
3)
(0.2
53
7)
-35
9.8
11
66
69
.32
06
16
7.8
07
6-6
84
.89
67
1,2
87
.82
92
33
5.7
45
15
17
6.1
32
01
,78
6
(87
7.6
81
2)
(73
6.8
70
3)
(74
2.9
39
9)
(1,6
57
.06
36
)(1
,41
1.1
32
1)
(1,4
56
.42
11
)(1
09
96
.31
)
-0.0
85
60
.05
85
0.0
21
0-0
.16
25
0.1
12
40
.04
22
0.7
79
61
,79
7
(0.1
31
5)
(0.0
68
1)
(0.0
87
1)
(0.2
44
8)
(0.1
29
8)
(0.1
71
0)
(1.1
56
6)
-0.0
84
0-0
.00
43
0.0
45
2-0
.15
97
-0.0
08
40
.09
08
0.2
65
11
,79
9
(0.0
71
7)
(0.0
30
5)
(0.0
27
8)
(0.1
32
6)
(0.0
57
8)
(0.0
55
7)
(0.4
41
5)
-0.0
22
1-0
.00
20
-0.0
31
9-0
.04
19
-0.0
03
9-0
.06
40
0.6
26
31
,80
1
(0.0
40
4)
(0.0
32
3)
(0.0
42
7)
(0.0
76
4)
(0.0
61
1)
(0.0
83
6)
(0.4
83
9)
0.0
29
6-0
.07
22
-0.0
92
8*
*0
.05
62
-0.1
38
8-0
.18
61
**
*0
.51
19
1,8
01
(0.0
64
6)
(0.0
45
8)
(0.0
36
3)
(0.1
21
3)
(0.0
87
8)
(0.0
71
0)
(0.5
00
0)
-0.0
85
9*
-0.0
52
8-0
.05
50
-0.1
63
1*
-0.1
01
4-0
.11
03
*0
.35
87
1,8
01
(0.0
47
4)
(0.0
42
1)
(0.0
33
6)
(0.0
89
1)
(0.0
79
5)
(0.0
65
3)
(0.4
79
7)
-0.0
54
1-0
.04
84
**
-0.0
28
1-0
.10
28
-0.0
93
1*
*-0
.05
63
0.4
81
41
,80
1
(0.0
38
0)
(0.0
20
7)
(0.0
24
5)
(0.0
71
0)
(0.0
37
8)
(0.0
47
8)
(0.4
99
8)
0.0
03
6-0
.03
36
-0.0
39
8*
0.0
06
8-0
.06
46
-0.0
79
8*
0.2
48
81
,80
1
(0.0
35
7)
(0.0
28
5)
(0.0
23
0)
(0.0
67
4)
(0.0
54
2)
(0.0
45
3)
(0.4
71
7)
-0.0
24
8-0
.00
82
-0.0
05
7-0
.04
72
-0.0
15
8-0
.01
15
0.5
94
71
,80
1
(0.0
21
1)
(0.0
16
0)
(0.0
13
2)
(0.0
40
9)
(0.0
30
5)
(0.0
26
1)
(0.1
80
5)
Est
imat
ed I
mp
act
of
the
pro
gra
m o
n A
sset
s an
d I
nco
me
Co
ntr
ol
Mea
nN
d.
Nu
mb
er o
f m
arket
cro
ps
h.
Co
nst
ruct
ed n
ew a
gri
cult
ura
l, l
ives
tock
or
fore
stry
infr
astr
uct
ure
sin
ce J
anu
ary 2
01
5 (
1 =
Yes
)
g.
Per
form
ed v
eter
inar
y p
roce
du
res
sin
ce J
anu
ary 2
01
5 (
1 =
Yes
)
f. A
dd
ed o
ne
or
mo
re t
yp
es o
f li
ves
tock
sin
ce J
anu
ary 2
01
5 (
1 =
Yes
)
Inte
nt-
to-T
reat
(IT
T)
Tre
atm
ent-
on
-th
e-T
reat
ed (
TO
T)
i. B
ou
gh
t o
ne
or
mo
re e
lect
rica
l ap
pli
ance
s si
nce
Jan
uar
y 2
01
5
TA
BL
E 6
e. U
sed
fer
tili
zer
in t
he
last
agri
cult
ura
l se
aso
n (
1 =
Yes
)
j. N
um
ber
of
inco
me-
gen
erat
ing b
usi
nes
ses
no
t re
late
d t
o
agri
cult
ure
, li
ves
tock
or
fore
stry
sin
ce J
anu
ary 2
01
5
Note
s:
This
tab
le r
epo
rts
dif
fere
nce
s fi
nan
cial
lit
erac
y i
n b
etw
een t
he
rand
om
ly a
ssig
ned
tre
atm
ent
and
co
ntr
ol
gro
up
s at
end
line.
Eac
h c
ell
in c
olu
mns
(1.x
) an
d (
2.x
) re
po
rts
resu
lts
for
a
single
reg
ress
ion o
f th
e in
dep
end
ent
var
iab
le i
n t
he
corr
esp
ond
ing r
ow
, w
ith c
lust
ered
sta
nd
ard
err
ors
at
the
com
unit
y l
evel
in p
aren
thes
es b
elo
w e
ach e
stim
ate.
Co
lum
ns
(1.x
) p
rese
nt
OL
S I
nte
nt-
to-T
reat
(IT
T)
esti
mat
es.
Co
lum
ns
(2.x
) p
rese
nt
2S
LS
Tre
atm
ent-
on-t
he-
Tre
ated
(T
OT
) es
tim
ates
, w
ith a
ctual
tre
atm
ent
as t
he
end
ogen
ous
var
iab
le a
nd
ran
do
m a
ssig
nm
ent
as
the
inst
rum
ent.
Co
lum
ns
(x.0
) p
rese
nt
the
dif
fere
nce
in m
eans
bet
wee
n t
reat
men
t an
d c
ontr
ol
wit
ho
ut
add
itio
nal
co
var
iate
s. C
olu
mns
(x.1
) p
rese
nt
the
dif
fere
nce
in m
eans
wit
h a
dd
itio
nal
covar
iate
s: (
i) d
ista
nce
to
reg
ional
cap
ital
; (i
i) n
um
ber
of
ho
use
ho
ld m
emb
ers;
(ii
i) g
end
er o
f ho
use
ho
ld h
ead
; (i
v)
age
of
resp
ond
ent;
(v)
max
imum
ed
uca
tio
nal
lev
el o
f ho
use
ho
ld;
and
(vi)
ben
efic
iary
sta
tus
of
JUN
TO
S c
ond
itio
nal
cas
h t
ransf
er p
rogra
m,
bes
ides
all
bal
ance
var
iab
les
in T
able
1.
Co
lum
ns
(x.2
) p
rese
nt
the
dif
fere
nce
in m
eans
wit
h a
dd
itio
nal
co
var
iate
s
and
dis
tric
t fi
xed
eff
ects
. C
olu
mn (
3)
rep
ort
s th
e m
ean v
alue
of
each
ind
epen
den
t var
iab
le a
t en
dli
ne,
wit
h s
tand
ard
dev
iati
ons
in p
aren
thes
es.
Co
lum
n (
4)
rep
ort
s th
e num
ber
of
ob
serv
atio
ns
in e
ach r
egre
ssio
n. Sourc
e:
End
line
surv
ey o
f 1
80
1 h
ouse
ho
lds
carr
ied
out
in J
uly
20
16
, S
ISF
OH
20
13
and
Junto
s R
oll
of
Use
rs b
y S
epte
mb
er 2
01
4.
a. N
um
ber
of
acti
vit
ies
gen
erat
ing i
nco
me
for
the
ho
use
ho
ld
*p
<0
.1;
**p
<0
.05
; *
**p
<0
.01
;
k.
JUN
TO
S c
on
dit
ion
al c
ash
tra
nsf
er p
rogra
m p
over
ty s
core
(0
-
1)
b.
Per
cen
tage
of
inco
me
ori
gin
atin
g f
rom
agri
cult
ure
, li
ves
tock
or
fore
stry
sel
f-em
plo
ym
ent
acti
vit
ies
c. L
and
ow
ned
(m
2)
(1.1
)(1
.2)
(1.3
)(2
.1)
(2.2
)(2
.3)
0.0
12
60
.00
44
0.0
12
60
.02
39
0.0
08
40
.02
53
0.9
48
91
,80
1
(0.0
22
5)
(0.0
14
6)
(0.0
13
5)
(0.0
42
5)
(0.0
27
6)
(0.0
26
5)
(0.2
20
2)
0.0
25
60
.02
16
0.0
51
70
.04
86
0.0
41
50
.10
37
0.6
94
11
,80
1
(0.0
54
0)
(0.0
41
6)
(0.0
34
2)
(0.1
02
1)
(0.0
78
9)
(0.0
67
6)
(0.4
60
9)
-0.0
03
0-0
.02
55
-0.0
25
1-0
.00
57
-0.0
49
0-0
.05
03
0.9
47
81
,80
1
(0.0
20
5)
(0.0
24
0)
(0.0
20
2)
(0.0
38
7)
(0.0
45
1)
(0.0
39
0)
(0.2
22
5)
0.0
12
4-0
.04
50
-0.0
65
7*
0.0
23
6-0
.08
64
-0.1
31
7*
0.7
07
41
,80
1
(0.0
45
4)
(0.0
39
2)
(0.0
39
1)
(0.0
85
5)
(0.0
74
1)
(0.0
77
1)
(0.4
55
1)
0.0
40
20
.04
60
0.0
16
10
.07
63
0.0
88
40
.03
23
0.7
04
61
,80
1
(0.0
58
3)
(0.0
40
4)
(0.0
33
6)
(0.1
10
9)
(0.0
76
2)
(0.0
66
6)
(0.4
56
3)
0.0
26
40
.02
01
-0.0
23
10
.05
01
0.0
38
6-0
.04
63
0.4
43
11
,80
1
(0.0
68
9)
(0.0
47
1)
(0.0
39
9)
(0.1
30
4)
(0.0
88
9)
(0.0
78
5)
(0.4
96
9)
0.0
09
5-0
.01
50
0.0
08
00
.01
80
-0.0
28
80
.01
60
0.5
80
21
,80
1
(0.0
54
0)
(0.0
35
8)
(0.0
34
3)
(0.1
02
2)
(0.0
68
1)
(0.0
67
5)
(0.4
93
7)
-0.0
30
5-0
.05
05
-0.0
41
3-0
.05
80
-0.0
97
2-0
.08
29
0.2
69
31
,80
1
(0.0
49
6)
(0.0
35
5)
(0.0
30
1)
(0.0
92
8)
(0.0
69
2)
(0.0
60
5)
(0.4
43
7)
0.0
28
90
.01
63
0.0
04
80
.05
50
0.0
31
40
.00
96
0.2
34
91
,80
1
(0.0
40
5)
(0.0
36
7)
(0.0
27
8)
(0.0
76
1)
(0.0
69
8)
(0.0
54
6)
(0.4
24
0)
0.0
14
60
.02
01
0.0
23
90
.02
77
0.0
38
60
.04
80
0.0
86
61
,80
1
(0.0
17
3)
(0.0
18
8)
(0.0
21
0)
(0.0
32
8)
(0.0
35
9)
(0.0
41
4)
(0.2
81
4)
h.
Ho
use
ho
ld d
rin
ks
sod
a in
a b
ad m
on
th (
1 =
Yes
)
g.
Ho
use
ho
ld d
rin
ks
sod
a in
a g
oo
d m
on
th (
1 =
Yes
)
f. H
ou
seh
old
eat
s o
ut
in a
bad
mo
nth
(1
= Y
es)
Inte
nt-
to-T
reat
(IT
T)
Tre
atm
ent-
on
-th
e-T
reat
ed (
TO
T)
Est
imat
ed I
mp
act
of
the
Pro
gra
m o
n C
on
sum
pti
on
Co
ntr
ol
Mea
nN
i. H
ou
seh
old
dri
nks
alco
ho
l in
a g
oo
d m
on
th (
1 =
Yes
)
TA
BL
E 7
e. H
ou
seh
old
eat
s o
ut
in a
go
od
mo
nth
(1
= Y
es)
Note
s:
This
tab
le r
epo
rts
dif
fere
nce
s fi
nan
cial
lit
erac
y i
n b
etw
een t
he
rand
om
ly a
ssig
ned
tre
atm
ent
and
co
ntr
ol
gro
up
s at
end
line.
Eac
h c
ell
in c
olu
mns
(1.x
) an
d (
2.x
) re
po
rts
resu
lts
for
a
single
reg
ress
ion o
f th
e in
dep
end
ent
var
iab
le i
n t
he
corr
esp
ond
ing r
ow
, w
ith c
lust
ered
sta
nd
ard
err
ors
at
the
com
unit
y l
evel
in p
aren
thes
es b
elo
w e
ach e
stim
ate.
Co
lum
ns
(1.x
) p
rese
nt
OL
S I
nte
nt-
to-T
reat
(IT
T)
esti
mat
es.
Co
lum
ns
(2.x
) p
rese
nt
2S
LS
Tre
atm
ent-
on-t
he-
Tre
ated
(T
OT
) es
tim
ates
, w
ith a
ctual
tre
atm
ent
as t
he
end
ogen
ous
var
iab
le a
nd
ran
do
m a
ssig
nm
ent
as
the
inst
rum
ent.
Co
lum
ns
(x.0
) p
rese
nt
the
dif
fere
nce
in m
eans
bet
wee
n t
reat
men
t an
d c
ontr
ol
wit
ho
ut
add
itio
nal
co
var
iate
s. C
olu
mns
(x.1
) p
rese
nt
the
dif
fere
nce
in m
eans
wit
h a
dd
itio
nal
covar
iate
s: (
i) d
ista
nce
to
reg
ional
cap
ital
; (i
i) n
um
ber
of
ho
use
ho
ld m
emb
ers;
(ii
i) g
end
er o
f ho
use
ho
ld h
ead
; (i
v)
age
of
resp
ond
ent;
(v)
max
imum
ed
uca
tio
nal
lev
el o
f ho
use
ho
ld;
and
(vi)
ben
efic
iary
sta
tus
of
JUN
TO
S c
ond
itio
nal
cas
h t
ransf
er p
rogra
m,
bes
ides
all
bal
ance
var
iab
les
in T
able
1.
Co
lum
ns
(x.2
) p
rese
nt
the
dif
fere
nce
in m
eans
wit
h a
dd
itio
nal
co
var
iate
s
and
dis
tric
t fi
xed
eff
ects
. C
olu
mn (
3)
rep
ort
s th
e m
ean v
alue
of
each
ind
epen
den
t var
iab
le a
t en
dli
ne,
wit
h s
tand
ard
dev
iati
ons
in p
aren
thes
es.
Co
lum
n (
4)
rep
ort
s th
e num
ber
of
ob
serv
atio
ns
in e
ach r
egre
ssio
n. Sourc
e:
End
line
surv
ey o
f 1
80
1 h
ouse
ho
lds
carr
ied
out
in J
uly
20
16
, S
ISF
OH
20
13
and
Junto
s R
oll
of
Use
rs b
y S
epte
mb
er 2
01
4.
a. H
ou
seh
old
eat
s m
eat
in a
go
od
mo
nth
(1
= Y
es)
*p
<0
.1;
**p
<0
.05
; *
**p
<0
.01
;
j. H
ou
seh
old
dri
nks
alco
ho
l in
a b
ad m
on
th (
1 =
Yes
)
b.
Ho
use
ho
ld e
ats
mea
t in
a b
ad m
on
th (
1 =
Yes
)
c. H
ou
seh
old
eat
s fr
uit
in
a g
oo
d m
on
th (
1 =
Yes
)
d.
Ho
use
ho
ld e
ats
fru
it i
n a
bad
mo
nth
(1
= Y
es)
(1.1
)(1
.2)
(1.3
)(2
.1)
(2.2
)(2
.3)
0.0
43
50
.02
65
0.0
41
1*
0.0
81
80
.05
05
0.0
81
5*
0.3
78
71
,74
8
(0.0
35
7)
(0.0
33
0)
(0.0
23
8)
(0.0
66
5)
(0.0
61
9)
(0.0
47
6)
(0.4
85
2)
-0.0
21
00
.00
13
-0.0
13
9-0
.03
97
0.0
02
5-0
.02
77
1.5
60
91
,78
5
(0.0
39
0)
(0.0
33
6)
(0.0
33
4)
(0.0
73
3)
(0.0
63
4)
(0.0
65
2)
(0.3
87
0)
0.2
89
2*
0.1
19
80
.19
50
**
*0
.54
71
**
0.2
29
3*
0.3
89
0*
**
1.9
36
71
,78
5
(0.1
48
4)
(0.0
74
6)
(0.0
63
2)
(0.2
66
3)
(0.1
38
5)
(0.1
25
7)
(1.2
78
9)
0.0
14
3-0
.00
61
-0.0
03
10
.02
58
-0.0
11
0-0
.00
58
0.5
42
81
,51
8
(0.0
92
4)
(0.0
74
7)
(0.0
70
1)
(0.1
65
1)
(0.1
32
7)
(0.1
29
9)
(0.8
78
5)
0.0
08
2-0
.00
23
-0.0
00
70
.01
45
-0.0
04
1-0
.00
13
0.6
97
01
,23
9
(0.0
23
9)
(0.0
19
5)
(0.0
19
3)
(0.0
41
8)
(0.0
33
8)
(0.0
34
6)
(0.1
97
6)
0.0
17
60
.00
15
-0.0
02
50
.03
01
0.0
02
6-0
.00
47
0.2
37
01
,37
8
(0.0
17
4)
(0.0
10
6)
(0.0
08
2)
(0.0
29
4)
(0.0
18
6)
(0.0
14
6)
(0.1
44
1)
0.0
08
8-0
.01
22
-0.0
25
80
.01
59
-0.0
22
5-0
.05
00
0.6
83
51
,51
0
(0.0
20
1)
(0.0
14
6)
(0.0
16
1)
(0.0
36
2)
(0.0
26
6)
(0.0
30
6)
(0.1
80
1)
0.0
25
9-0
.00
64
-0.0
11
50
.04
66
-0.0
11
5-0
.02
17
0.6
88
91
,50
2
(0.0
25
5)
(0.0
16
9)
(0.0
17
1)
(0.0
45
0)
(0.0
30
0)
(0.0
31
5)
(0.2
11
9)
TA
BL
E 8
e. I
nte
rper
son
al r
elat
ion
ship
s an
d s
elf-
stee
m i
nd
ex (
0-1
)
Note
s:
This
tab
le r
epo
rts
dif
fere
nce
s fi
nan
cial
lit
erac
y i
n b
etw
een t
he
rand
om
ly a
ssig
ned
tre
atm
ent
and
co
ntr
ol
gro
up
s at
end
line.
Eac
h c
ell
in c
olu
mns
(1.x
) an
d (
2.x
) re
po
rts
resu
lts
for
a
single
reg
ress
ion o
f th
e in
dep
end
ent
var
iab
le i
n t
he
corr
esp
ond
ing r
ow
, w
ith c
lust
ered
sta
nd
ard
err
ors
at
the
com
unit
y l
evel
in p
aren
thes
es b
elo
w e
ach e
stim
ate.
Co
lum
ns
(1.x
) p
rese
nt
OL
S I
nte
nt-
to-T
reat
(IT
T)
esti
mat
es.
Co
lum
ns
(2.x
) p
rese
nt
2S
LS
Tre
atm
ent-
on-t
he-
Tre
ated
(T
OT
) es
tim
ates
, w
ith a
ctual
tre
atm
ent
as t
he
end
ogen
ous
var
iab
le a
nd
ran
do
m a
ssig
nm
ent
as
the
inst
rum
ent.
Co
lum
ns
(x.0
) p
rese
nt
the
dif
fere
nce
in m
eans
bet
wee
n t
reat
men
t an
d c
ontr
ol
wit
ho
ut
add
itio
nal
co
var
iate
s. C
olu
mns
(x.1
) p
rese
nt
the
dif
fere
nce
in m
eans
wit
h a
dd
itio
nal
covar
iate
s: (
i) d
ista
nce
to
reg
ional
cap
ital
; (i
i) n
um
ber
of
ho
use
ho
ld m
emb
ers;
(ii
i) g
end
er o
f ho
use
ho
ld h
ead
; (i
v)
age
of
resp
ond
ent;
(v)
max
imum
ed
uca
tio
nal
lev
el o
f ho
use
ho
ld;
and
(vi)
ben
efic
iary
sta
tus
of
JUN
TO
S c
ond
itio
nal
cas
h t
ransf
er p
rogra
m,
bes
ides
all
bal
ance
var
iab
les
in T
able
1.
Co
lum
ns
(x.2
) p
rese
nt
the
dif
fere
nce
in m
eans
wit
h a
dd
itio
nal
co
var
iate
s
and
dis
tric
t fi
xed
eff
ects
. C
olu
mn (
3)
rep
ort
s th
e m
ean v
alue
of
each
ind
epen
den
t var
iab
le a
t en
dli
ne,
wit
h s
tand
ard
dev
iati
ons
in p
aren
thes
es.
Co
lum
n (
4)
rep
ort
s th
e num
ber
of
ob
serv
atio
ns
in e
ach r
egre
ssio
n. Sourc
e:
End
line
surv
ey o
f 1
80
1 h
ouse
ho
lds
carr
ied
out
in J
uly
20
16
, S
ISF
OH
20
13
and
Junto
s R
oll
of
Use
rs b
y S
epte
mb
er 2
01
4.
a. T
he
wo
man
dec
ides
fre
ely h
ow
to
sp
end
her
in
com
e (1
= Y
es)
*p
<0
.1;
**p
<0
.05
; *
**p
<0
.01
;
h.
Co
mm
un
ity p
arti
cip
atio
n i
nd
ex (
0-1
)
b.
On
a s
cale
of
1 t
o 4
, h
ow
wel
l d
oes
th
e w
om
an f
eel
wit
h h
er
acti
vit
ies,
rel
atio
nsh
ips,
etc
.
c. N
um
ber
of
gro
up
s o
r o
rgan
izat
ion
s to
wh
ich
th
e w
om
an
bel
on
gs
d.
Nu
mb
er o
f p
osi
tio
ns
the
wo
man
has
hel
d i
n t
he
gro
up
s to
wh
ich
sh
e b
elo
ngs
g.
Gen
der
ro
les
ou
tsid
e th
e h
ou
seh
old
in
dex
(0
-1)
Inte
nt-
to-T
reat
(IT
T)
Tre
atm
ent-
on
-th
e-T
reat
ed (
TO
T)
Est
imat
ed I
mp
act
of
the
Pro
gra
m o
n F
emal
e E
mp
ow
erm
ent
Co
ntr
ol
Mea
nN
f. G
end
er r
ole
s in
th
e h
ou
seh
old
in
dex
(0
-1)