ALAP_2014_FINAL130

Extended Proposal
The impact of Argentina’s Asignación Universal por Hijo (Universal Child
Allowance program)) Cash Transfer Program on Enrollment, Dropout Rates
and Grade Promotion
Gastón Pierri, PhD Candidate, University of Alcalá
Robert McCaa, Professor Emeritus, University of Minnesota
Abstract
The research examines the impact of the Asignación Universal por Hijo (AUH) program on children’s
enrollment and progress in school in Argentina. The AUH program, which started in November 2009,
provides monthly cash payments to informal workers, domestic workers earning below the minimum
wage, unemployed without unemployment insurance and inactive persons without a pension if their
children (between the ages of 6 and 17) are enrolled in school. The method used in this research
compares changes in enrollment, dropout rates, and grade promotion rates across individual children
whose parents or legal guardians adopted the AUH program between 2009 and 2012. The preliminary
estimation is that the AUH increased enrollment but decreased dropout and promotion rates in
Argentina.
The impact of Argentina’s Asignación Universal por Hijo (Universal Child Allowance
program) Cash Transfer Program on Enrollment, Dropout Rates and Grade Promotion.
1.
Introduction.
The AUH is a cash transfer that is paid on a monthly basis to one of the parents, a tutor, or relative up to
the third degree of consanguinity. The participants receive an amount of money for every child under 18
years of age, except for the case of disabled people for which there is no age limit. The children must be
native from Argentina or have at least three years of residence in the country. The benefit is a set amount
per child and can be claimed for up to 5 dependent children. Its initial value was $180 (USD 47) per
child and $720 for disabled dependants. In September 2010 the value was raised to $220 (USD 56) and
then again one year later to $270 (USD 64) in order to protect its purchasing power given high levels of
inflation in Argentina.
The data employed in this paper comes from the Integrated Public Use Microdata Series-International
(IPUMS-International) data from Argentina’s Census (1990-2010) and the Permanent Household Survey
(Encuesta Permanente de Hogares, EPH), which is the regular household survey of Argentina carried
out by the National Statistical Office (Instituto Nacional de Estadística y Censos, INDEC).
This paper will explore the association between Asignación Universal por Hijo participation and
outcomes linked with beneficiary behavior. More specifically, the paper will focus on potential program
effects on two groups of people and sets of activities. The analysis will examine the linkages between
program participation, school attendance and dropout rates. Given its explicit aim of increasing school
attendance rates among poor school-aged children, what does the evidence tell us about AUH and school
attendance linkages?
2.
Description of the Argentina’s Universal Child Allowance Program
In November 2009 the Argentinean government implemented a large cash transfer program for children
and adolescents –the Universal Child Allowance- that extended the coverage of the already existing
contributory family allowance program to include the children of (1) workers that are not registered in
the social security system (informal workers) or domestic workers that receive labor incomes below the
minimum wage; (2) unemployed without unemployment insurance or (3) economically inactive workers
without pensions.
The new configuration of the child allowance system in Argentina is made up of three components: the
first one consists of a contributory cash transfer program for children and adolescents (Asignación
Familiar Contributiva), the second pillar is the non-contributory Universal Child Allowance program
(AUH), and the third component are the income tax rebates for workers in the highest income group
(Asignación por Crédito Fiscal). Both the contributory component and the AUH are administered by the
National Social Security Administration (Administración Nacional de la Seguridad Social - ANSES),
while the third pillar is administered by the Federal Tax Administration (Administración Federal de
Ingresos Públicos - AFIP).
Government expenditures on the program represent approximately 0.8% of GDP, making it one of the
largest programs in the region. Receiving any other type of social benefit from the government is not
allowed with the AUH. Therefore, all previous programs with similar targets were gradually eliminated.
3. Program Characteristics
3.1. Program Requirements
The AUH is a cash transfer that is paid on a monthly basis to one of the parents, a tutor or relative up to
the third degree of consanguinity, for every child under 18 years of age, except for the case of disabled
people for which there is no age limit. The children must be native from Argentina or have at least three
years of residence in the country. The benefit is a set amount per child and can be claimed for up to 5
children for whom the recipient is in charge. Its initial value was $180 (USD 471) per child and $720 for
disabled (four times the value of the regular benefit). In September 2010 it was raised to $220 (USD 56)
and then again one year later to$270 (USD 54) in order to limit the erosion of its purchasing power due
to inflation. Finally, in June 2013 the benefit was raised to $420 (USD 52)
3.2. Semi-conditionality of the Program
The AUH is a semi-conditional cash transfer, with 80% of its value paid on a monthly basis to the
benefit holders, whereas the remaining 20% is deposited into a savings account in the name of the
holder. Then, the latter sum is made available for withdrawal once the holder has certified the fulfillment
of the vaccination plan and other sanitary controls in the case of children under 5 years old, and has also
presented the certificate of school year completion in the case of school-aged children.
3.3 Particularity of the Program
The AUH is not an ad-hoc program designed to alleviate the situation of families with children in social
vulnerability, like the cases of Bolsa Familia in Brazil or Oportunidades in Mexico; rather, as mentioned
above, it is an extension of the already existing contributory child allowance program for the children of
formal workers, unemployed with unemployment insurance or retirees. In fact, the amount of the benefit
is the same in both systems.
This aspect is important because, unlike the means-tested CCT, the restriction of the AUH is not directly
related to family incomes but rather to the labor condition of the adults in charge of the children, as well
as to their labor incomes if they are employed in informal jobs. However, the difficulties in monitoring
labor incomes that arise in a context of informality weaken the enforcement of such restriction in the
latter case.
4 Data and methods
4.1. Data
Integrated Public Use Microdata Series- International (IPUMS-International) data related to Argentina’s
Census (1990-2010) and Permanent Household Survey (Encuesta Permanente de Hogares, EPH) by
National Statistical Office (INDEC) represents about 70% of the urban population and 60% of the
country’s total population.
IPUMS- International: IPUMS- International Censuses provide population numbers, household or
family size and composition, and information on sex and age distribution. The data also includes other
demographic, economic, and health-related topics. The 2001 and 2010 Argentina de facto census
collected data through face-to-face interviews. The census dates were November 17-18, 2001 and
October 27-28, 2010. A 10% self-weighted sample of households from the census is available through
IPUMS International at the University of Minnesota.
1
*PThe prices shown where US the value in dollar conversion at thate timestage of updating the
amount.
EPH is a longitudinal survey that includes retrospective questions. Selected households are interviewed
for two successive quarters, followed by a two-quarter break and they appear again in the sample of the
two successive quarters. (See graphic below.)
PERMANET HOUSEHOLD SURVEY
YEAR 2009
T1
T2
X
X
X
T3
YEAR 2010
T4
T1
T2
X
X
X
X
X
X
X
T3
T4
X
T2
T3
T4
T1
T2
T3
T4
X
X
X
X
T1
YEAR 2012
X
X
X
YEAR 2011
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Panel data build: Quarter I2009-Quarter IV2012 includes information before and after the
implementation of the AUH in November 2009.
4.2 Methodology proposal:
A non-experimental method based on application of matching techniques to define a control group that
allows estimating what the situation of beneficiaries would have been had they not gained access to the
program. To estimate the ATT parameter, a Difference-in-Difference Matching Estimator (DID) will be
implemented based on the available information from before and after the policy implementation,
through a comparison of the temporal changes of the outcome variable in the beneficiary group with the
changes in the same variable in the control group. The advantage of this strategy lies in the possibility to
control for biases derived from time invariant unobserved characteristics.
Difference in Difference implementation:
DID will compare treatment and comparison groups in terms of how their outcome changes over time in
relation to the outcomes observed for a pre-intervention baseline. That means, given a two-period setting
where t=0 before program and t=1 after program implementation, letting YtT and YtC be the respective
outcomes for a program beneficiary and non-treated units in time t, the DID method will estimate the
average program impact as follows:
DID=E(YtT – Y0T | T1=1) – E(Y1C – Y0C || T1=0)
Where T1 =1 denotes treatment or the AUH at t=1, where T=0 denotes untreated participants. The DID
alone allows for unobserved heterogeneity that may lead to selection bias.
A possible estimator of the impact of treatment is the change in outcome variable over time and is called
the Before-After estimator or the Differences estimator.
The Differences estimator is:
∆ˆd =Y1,D=1,t0 −Y0,D=1,t0−1
Where Y denotes the appropriate mean and D = 1 indicates a treated person. The treated people receive
the treatment services at time t0. In this case, the treated group is use before they received treatment as
the comparison group, or our estimate of the missing counterfactual. This method is different than the
cross-sectional estimator, in which we compare the outcomes of treated individuals with that of
untreated people in the comparison group at the same point in time.
Since we are comparing the same people, we avoid many of the heterogeneity problems that arise from
comparing two different groups of people. The differences estimator relies on the identifying assumption
that the outcome of interest would remain unchanged in the absence of treatment.
∆ˆd =Y1,D=1,t0 −Y0,D=1,t0−1
=(Y1,D=1,t0 −Y0,D=1,t0−1)+Y0,D=1,t0 −Y0,D=1,t0
=(Y1,D=1,t0 −Y0,D=1,t0)+(Y0,D=1,t0 −Y0,D=1,t0−1) = impact + trend in absence of treatment
Whenever the difference estimator is used as an estimate of program impact, it is implicitly imposing the
assumption that the second term above is zero.
The study estimates of how the outcome would have changed for the treated people in the absence of
them receiving treatment, this is a missing counterfactual trend. A possible estimate of this missing
counterfactual trend could be the actual trend for the untreated group over this same time period.
The differences-in-differences (DID) estimator is:
∆ˆDID =∆ˆdD=1−∆ˆdD=0
=(Y1,D=1,t0 −Y0,D=1,t0−1)−(Y0,D=0,t0 −Y0,D=0,t0−1)
The key identifying assumption of the DID estimator is that the observed trend in the outcome of the
comparison group is the same as the counterfactual trend of the treated group. In other words, that’s
means that:
∆ˆDID=(Y1,D=1,t0−Y0,D=1,t0−1)−(Y0,D=0,t0−Y0,D=0,t0−1)=(Y1,D=1,t0−Y0,D=1,t0−1)−(Y0,D=0,t
0 −Y0,D=0,t0−1) +Y0,D=1,t0 −Y0,D=1,t0
=(Y1,D=1,t0 −Y0,D=1,t0)−(Y0,D=1,t0 −Y0,D=1,t0−1) +Y0,D=0,t0 −Y0,D=0,t0−1
=(Y1,D=1,t0 −Y0,D=1,t0)
This estimator allows for difference in the levels of the outcome between the treatment and the
comparison group but does not allow a differential trend. The model put this two-period model in a
regression framework as follows:
Yt =β0+β1It=t0 +β2D+δD×It=t0 +εt where It=t0 is an indicator variable for t = t0, D is an indicator for
belonging to the treatment group, and δ = ∆DID . The parameters correspond to:
Y0,D=0,t0−1 = βˆ0 Y0,D=0,t0 =βˆ0+βˆ1
Y0,D=0,t0 − Y 0,D=0,t0−1 = βˆ1 Y0,D=1,t0−1 =βˆ0+βˆ2
Y0,D=1,t0 =βˆ0+βˆ1+βˆ2+δˆ Y 1,D=1,t0 − Y 0,D=1,t0−1 = βˆ1 + δˆ
Thus, we have ∆ˆDID =(Y1,D=1,t0 −Y0,D=1,t0−1)−(Y0,D=0,t0 −Y0,D=0,t0−1)=δˆ
With the naïve, cross-sectional estimator, the model needs to believe that people selecting into treatment
did not have systematically different observables (in levels) than those who did not get treated. With
DID, the study selects people selecting into treatment that do not have systematically different trends
than those who choose not to be treated. This simple DID model assumes that the treatment effect is
constant across all groups, or ∆ATE = ∆ATET = ∆ATEN = δ
Starting with the comparison group. The study estimate the following pre-treatment and post-treatment
period models separately:
Y0,D=0,t0−1 = XD=0,t0−1βD=0,t0−1 + ε0,t0−1 Y0,D=0,t0 = XD=0,t0 βD=0,t0 + ε0,t0
We can decompose the change in the mean of the outcome over time as
Y0,D=0,t0 −Y0,D=0,t0−1 = (XD=0,t0 −XD=0,t0−1)βˆD=0,t0−1 +XD=0,t0(βˆD=0,t0 −βˆD=0,t0−1)
The first term represents changes in the outcome resulting from changes in the observables. This term is
often zero because many covariates simply do not (or cannot) change, like race, sex, education, etc. The
second term represents the trend that is allowed to vary by covariates
Following the traditional terminology of this approach, D is defined as a variable that indicates the
receipt of the transfer (D=1 if the household/person receives the transfer; D=0 on the contrary), and Y is
the outcome of interest (being Y1 the outcome in the presence of the benefit, and Y0 in its absence).
The impact of the transfer is measured by the Average Treatment Effect on the Treated (ATT), which is
conditional on a Propensity Score model, P(X), where X represents a vector of observable
characteristics:
ATT(X)=E[Y1-Y0/P(X),D=1 where E [.] is the expectation of the difference between the two
outcomes, with and without the treatment, over the population receiving the transfer (D=1).
Since the counterfactual, E [Y0 ⁄P(X), D=1], is not an observable situation, Propensity Score Matching
techniques are employed to estimate it. Given that only the ATT needs to be identified, it is sufficient to
verify the assumptions suggested by Heckman:
One “Ignorability of Treatment in a Conditional Mean Sense Condition”; and two “Matching
Condition”. The first condition implies that the selection of treated and control groups is made based on
the Propensity Score solely, and then, after accounting for it, the assignment to treatment is independent
of mean outcomes; the second condition ensures that for every possible value of Propensity Score there
exists beneficiary and non- beneficiary control cases.
5. Strategies for identification.
The base of this study is the correct identification of the AUH’s beneficiary households (treated group)
and of those that will constitute the control group. Unfortunately, the EPH does not directly inquire
about this matter so the identification has to be addressed in an indirect way. In order to identify the
households receiving the AUH in 2010, 2011 and 2012 we resorted to a question2 that captures the
totality of cash transfers received by household members, both from government and private institutions,
church, etc. Given that the question involves a rather wide group of institutions, it is not possible to
assume that the answers refer exclusively to this program. Therefore, the households were initially
classified as AUH beneficiaries only when the declared amounts matched the values established by the
program. In other words, the amount of the transfer was used as treatment indicator.
Considering the frequency of the cash values appearing in this question it is possible to assume that
some households declared the amount that was actually received on a monthly basis as benefit (80% of
2
The V5M variable captures cash transfers received by household members, both from government and
private institutions, church, etc
the sum of the benefit), whereas others declared the full amount. In order to minimize the potential
misclassification, the frequency of each of these values in 2010 was compared to data from 2009 (before
the AUH) so as to verify that those values considered to be AUH were not present in the year before the
implementation.
Because of the incompatibility of the AUH with all other types of social benefits of any government
level, when the values of the AUH transfers started to appear, the payments of other national programs
began to disappear (this is the case, for instance, of the Plan Jefes y Jefas de Hogares Desocupados,
Plan Familias or Seguro de Capacitacióny Empleo). In addition, when the values observed suggested
that there was more than one person per household receiving the AUH, the total amount of the benefit
received by the household was compared to the number of children in the household. The study
excluded from the analysis the households with more than one recipient member and whose total AUH
incomes suggested the presence of more children than the actual number of children living in the
household. This is to avoid cases were the amount of the benefit is repeated mistakenly for more than
one adult member.
In addition, excluding those without children further cut down the group of households classified as
beneficiaries. This was done to reduce the heterogeneity of this group in relation to the control group
that is comprised of AUH eligible households, and therefore have children living in them. In effect,
these eligible households (control group) are those that fulfill all the requirements to obtain an AUH and
have not yet received it.
As it was already mentioned, the potential beneficiaries are households with children under 18 years of
age whose heads of household or spouses are non-registered wage earners or domestic workers receiving
incomes below a minimum wage; unemployed persons without unemployment insurance or
economically inactive workers without pensions.
Therefore, the analysis will be limited to the households (and its members) with children that were
eligible for the AUH in 2009, differentiating them according to whether they gained access to the benefit
in 2010 (treated group) or not (control group). Thus, whereas the eligibility condition corresponds to
2009, the recipient condition corresponds to 2010.
6. Target Population
Since we are interested in evaluating the effects of the AUH on the school attendance of primary and
secondary students in the program, the sample in the analysis of individuals is comprised for school
attendance. Since the gap necessarily depends on children’s age, the estimations are done for two
different age groups: students from six to twelve years of age (Primary school) and students form
thirteen to seventeen years of age (High school).
People in economically active ages: men between 18 and 64 years old, and women between 18 and 59
years old. In both cases, the maximum age limits are set by the legal retirement age.
7. Preliminary Results
The preliminary results suggest that the allowance has a mostly positive impact with increased school
attendance of primary and secondary students and a reduction in the dropout rate, although there appears
to be a negative impact on grade promotion outcomes.