Credit for Low-Income Students and Access to and Academic Performance in Higher Education in Colombia: A Regression Discontinuity Approach

(1)

18

ABRIL DE 2015

Documentos CEDE

C E D E

ISSN 1657-7191 Edición electrónica.

Tatiana Melguizo

Fabio Sanchez

Tatiana Velasco

(2)

Serie Documentos Cede, 2015-18 ISSN 1657-7191 Edición electrónica. Abril de 2015

Bogotá, D. C., Colombia

Teléfonos: 3394949- 3394999, extensiones 2400, 2049, 3233

infocede@uniandes.edu.co http://economia.uniandes.edu.co

Ediciones Uniandes

Carrera 1ª Este No. 19 – 27, edificio Aulas 6, A. A. 4976 Bogotá, D. C., Colombia

Teléfonos: 3394949- 3394999, extensión 2133, Fax: extensión 2158

infeduni@uniandes.edu.co

Edición y prensa digital: Cadena S.A. • Bogotá Calle 17 A Nº 68 - 92 Tel: 57(4) 405 02 00 Ext. 307 Bogotá, D. C., Colombia

www.cadena.com.co

Impreso en Colombia – Printed in Colombia

La serie de Documentos de Trabajo CEDE se circula con propósitos de discusión y divulgación. Los artículos no han sido evaluados por pares ni sujetos a ningún tipo de evaluación formal por parte del equipo de trabajo del CEDE.

(3)

1

C E D E

Centro de Estudios sobre Desarrollo Económico

Credit for Low-Income Students and Access to and Academic

Performance in Higher Education in Colombia: A Regression

Discontinuity Approach

Tatiana Melguizo*_{Fabio Sanchez}**_{Tatiana Velasco}***

Abstract

This study evaluates the impact of a national level subsidized loan program, ACCES, on a number of higher education outcomes (i.e., increase in enrollment rates, decrease in dropout rates, and increase in academic performance) of low-income students in Colombia. We use national level data along with a regression discontinuity design to estimate the impact of the program. The results confirm that the program was effective in terms of increasing the potential number of low-income students at the margin who would have enrolled in college, decreasing the number of students who dropped out, and increasing their academic outcomes.

Key Words: Higher Education, Credit, Access, Academic Performance, ACCES

JEL Classification: I22, I23

*_{Associate Professor, Rossier School of Education, University of Southern California, melguizo@usc.edu} **_{Professor, Department of Economics, Universidad de los Andes, fasanchez@uniandes.edu.co}

***_{Junior Researcher, Department of Economics, Universidad de los Andes, t-velasc@uniandes.edu.co}

This study was supported by a small grant by the Spencer Foundation (2011-00128). Opinions are those of the authors alone and do not necessarily reflect those of the granting agencies or of the authors’ home institutions.

Acknowledgements: We would like to thank Juliana Marquez who participated as research assistant for an earlier version of the present paper. Special thanks for the comments received at Seminario CEDE and to Juan Esteban Saavedra, Judy Scott-Clayton, and Gema Zamarro for comments on earlier versions of the paper.

Credit for Low-Income Students and Access to and Academic

Performance in Higher Education in Colombia: A Regression

Discontinuity Approach

Abstract

Key Words: Higher Education, Credit, Access, Academic Performance, ACCES

JEL Classification: I22, I23

Acknowledgements: We would like to thank Juliana Marquez who participated as research assistant for an earlier version of the present paper. Special thanks for the comments received at Seminario CEDE and to Juan Esteban Saavedra, Judy Scott-Clayton, and Gema Zamarro for comments on earlier versions of the

paper. 1

Credit for Low-Income Students and Access to and Academic

Performance in Higher Education in Colombia: A Regression

Discontinuity Approach

Abstract

Key Words: Higher Education, Credit, Access, Academic Performance, ACCES

JEL Classification: I22, I23

(4)

2

Impacto del Crédito sobre el Acceso y Rendimiento Academico de los

Estudiantes de Bajos Ingresos en la Educación Superior en Colombia:

Un Enfoque de Regresión Discontinua.

Tatiana Melguizo*_{Fabio Sanchez}_{Tatiana Velasco}***

Resumen

Este estudio evalúa el impacto de ACCES, un programa de préstamos subsidiados a nivel nacional, sobre diferentes resultados de educación superior (tasas de matrícula, tasas de deserción escolar y rendimiento académico) de los estudiantes de bajos ingresos en Colombia. Utilizando datos a nivel nacional junto con un enfoque de regresión discontinua se estima el impacto del programa. Los resultados confirman que el programa fue efectivo: aumentando el número potencial de estudiantes de bajos ingresos que se hubieran inscrito en la universidad, disminuyendo el número de estudiantes que desertan y aumentando su rendimiento académico.

Palabras Clave: Educación Superior, Crédito, Acceso, Rendimiento Académico, ACCES

Clasificación JEL: I22, I23

*_{Profesora Asociada, Escuela de Educación de Rossier, University of Southern California,}

melguizo@usc.edu

_{Profesor Titular, Facultad de Economía, Universidad de los Andes, fasanchez@uniandes.edu.co} ***_{Investigadora Junior, Facultad de Economía, Universidad de los Andes, t_velasc@uniandes.edu.co}

2

Impacto del Crédito sobre el Acceso y Rendimiento Academico de los

Estudiantes de Bajos Ingresos en la Educación Superior en Colombia:

Un Enfoque de Regresión Discontinua.

Tatiana Melguizo*_{Fabio Sanchez}_{Tatiana Velasco}***

Resumen

Este estudio evalúa el impacto de ACCES, un programa de préstamos subsidiados a nivel nacional, sobre diferentes resultados de educación superior (tasas de matrícula, tasas de deserción escolar y rendimiento académico) de los estudiantes de bajos ingresos en Colombia. Utilizando datos a nivel nacional junto con un enfoque de regresión discontinua se estima el impacto del programa. Los resultados confirman que el programa fue efectivo: aumentando el número potencial de estudiantes de bajos ingresos que se hubieran inscrito en la universidad, disminuyendo el número de estudiantes que desertan y aumentando su rendimiento académico.

Palabras Clave: Educación Superior, Crédito, Acceso, Rendimiento Académico, ACCES

Clasificación JEL: I22, I23

*_{Profesora Asociada, Escuela de Educación de Rossier, University of Southern California,}

melguizo@usc.edu

(5)

Despite a substantial increase in college enrollment rates in Colombia over the last decade, there are still wide disparities in access by socio-economic status (Gaviria & Toro, 2012; World Bank, 2012). According to the World Bank one of the possible reasons for these disparities may be the lack of a well-developed financial aid system (World Bank, 2003). As a result, the Colombian government has made an effort to increase the financial aid available to low- and middle-income students in the country. In 2002, with support from a World Bank loan, the government designed and funded the program Acceso con Calidad a la Educación Superior (Access with Quality to Higher Education, or ACCES) to expand the opportunity to enter higher education to low- and middle-income students.

The main objective of the present study is to estimate the effect of the ACCES a loan program, on a number of educational outcomes (i.e., increase in enrollment rates, decrease in dropout rates, and increase in academic performance) of low-income students in Colombia. Thus, the overall research question that guides this study is: has the ACCES loan program been effective in increasing a number of educational outcomes for the beneficiaries? We use a regression discontinuity design (RDD) model to take advantage of the fact that the beneficiaries of the program were selected based on a government mandated procedure to rank individuals at the “department” (similar to geographic regions in the U.S.) level based on their high school exit exams.1_{We provide intent to treat (ITT) and the local average} treatment effect (LATE) estimates of the program on enrollment and dropout rates, and on academic performance. As part of our RDD strategy we restricted the sample to students who were either right below (non-beneficiary) or right above (beneficiary) according to the respective departmental cutoffs established by the Colombian Institute for Educational Loans and Studies Abroad (ICETEX).2_{Given that the program used a} government-determined exogenous set of rules to identify eligible students, there is a rigorous identification strategy with strong internal validity. This rigorous methodological strategy, coupled with access to complete national data3_{, suggests that the results of the}

1_{We provide a detailed explanation of the selection criteria below.}

2_{ICETEX, the acronym in Spanish stands for: Instituto Colombiano de Crédito Educativo y Estudios en el}

Exterior.

3_{We would like to thank ICETEX for providing us with the complete list of all the applicants to the}

ACCES program between 2002 and 2012.

Despite a substantial increase in college enrollment rates in Colombia over the last decade, there are still wide disparities in access by socio-economic status (Gaviria & Toro, 2012; World Bank, 2012). According to the World Bank one of the possible reasons for these disparities may be the lack of a well-developed financial aid system (World Bank, 2003). As a result, the Colombian government has made an effort to increase the financial aid available to low- and middle-income students in the country. In 2002, with support from a World Bank loan, the government designed and funded the program Acceso con Calidad a la Educación Superior (Access with Quality to Higher Education, or ACCES) to expand the opportunity to enter higher education to low- and middle-income students.

The main objective of the present study is to estimate the effect of the ACCES a loan program, on a number of educational outcomes (i.e., increase in enrollment rates, decrease in dropout rates, and increase in academic performance) of low-income students in Colombia. Thus, the overall research question that guides this study is: has the ACCES loan program been effective in increasing a number of educational outcomes for the beneficiaries? We use a regression discontinuity design (RDD) model to take advantage of the fact that the beneficiaries of the program were selected based on a government mandated procedure to rank individuals at the “department” (similar to geographic regions in the U.S.) level based on their high school exit exams.1_{We provide intent to treat (ITT) and the local average} treatment effect (LATE) estimates of the program on enrollment and dropout rates, and on academic performance. As part of our RDD strategy we restricted the sample to students who were either right below (non-beneficiary) or right above (beneficiary) according to the respective departmental cutoffs established by the Colombian Institute for Educational Loans and Studies Abroad (ICETEX).2_{Given that the program used a} government-determined exogenous set of rules to identify eligible students, there is a rigorous identification strategy with strong internal validity. This rigorous methodological strategy, coupled with access to complete national data3_{, suggests that the results of the}

1_{We provide a detailed explanation of the selection criteria below.}

2_{ICETEX, the acronym in Spanish stands for: Instituto Colombiano de Crédito Educativo y Estudios en el}

Exterior.

3_{We would like to thank ICETEX for providing us with the complete list of all the applicants to the}

(6)

study have both strong internal and external validity. The LATE results of the study confirm that the program was effective in increasing the probability that a beneficiary student would enroll in college by about 20 percentage points and being this impact larger for students from the lowest socio-economic strata. In addition, for the students who benefitted from the program, ACCES was effective in terms of decreasing dropout rates by about 7 percent and increasing academic performance for the students at the margin (i.e., those right above and below the department’s cutoff).

This study contributes to the literature by estimating the effect of a national level subsidized loan program targeted to low-income students. As described in detail below, there is very little evidence regarding the impact of financial aid programs in the form of loans (Avery & Hoxby, 2004; Chen & DesJardins, 2010; DesJardins, Ahlburg & McCall, 2006; Singell, 2004). A number of descriptive studies in Colombia suggest a positive association between receiving financial aid and college access and persistence (Cerdán-Infantes & Blom, 2007; Melguizo, Sanchez & Jaime, 2011). However, these estimates might be biased given that application to financial aid programs is voluntary, and one would expect more academically prepared and motivated students to apply to these programs (Dynarski, 2002). Our study improves on the previous literature by using an appropriate estimation strategy that reduces the threat of having biased estimates.

(7)

Characteristics of the Colombian Postsecondary Education System

Colombia’s postsecondary education system expanded substantially in the early 1990s, resulting in an over-representation of private postsecondary institutions (about 70 percent).4_{The system is comprised of about 305 postsecondary institutions, the majority of} which are universities (193), and the rest are either technical or technical-professional institutes. There has been a substantial increase in access to postsecondary education in the last decade (Sanchez & Marquez, 2012). The enrollment rates increased from about 24.4 percent (the Latin American average) in 2002 to about 37.1 percent in 2010 (Gaviria & Toro, 2012; World Bank, 2012).

Despite having enrollment rates above the regional average, there are wide disparities in access to postsecondary education by socio-economic status. Colombia has had two big expansions of its higher education system in the last two decades. Before the first expansion of the system in the early 1990s, approximately two percent of the relevant age-cohort population of the poorest quintile was enrolled in tertiary education, compared to 20 percent for the best-off quintile. After the expansion, the largest gains in coverage occurred in the best-off quintile, where net coverage rose from 23 percent to 40 percent. However, it is worth noting that there was also an expansion of about 170 percent in the first quintile (World Bank, 2003). Colombia had a second large expansion in college enrollment in the last decade. Recent evidence suggests that the share of enrolled students from the lowest incomes increased from 26.64 percent to over 40 percent between 2000 and 2009. Despite the increase in enrollment rates it is noteworthy that a sizable proportion of the increase in enrollment for this group of students happened at the technical and technological institutions that traditionally have fewer resources to support students, and hence exhibit high dropout rates (about 60 percent) (Sanchez & Marquez, 2012, 2012; World Bank, 2012).

As these data suggest, glaring socio-economic inequalities have been a constant since the inception of the higher education system in Colombia. In the mid-1980s Jimenez and

4_{For a more detailed description of the tertiary education system in Colombia see World Bank (2012) and}

World Bank (2003).

Characteristics of the Colombian Postsecondary Education System

Colombia’s postsecondary education system expanded substantially in the early 1990s, resulting in an over-representation of private postsecondary institutions (about 70 percent).4_{The system is comprised of about 305 postsecondary institutions, the majority of} which are universities (193), and the rest are either technical or technical-professional institutes. There has been a substantial increase in access to postsecondary education in the last decade (Sanchez & Marquez, 2012). The enrollment rates increased from about 24.4 percent (the Latin American average) in 2002 to about 37.1 percent in 2010 (Gaviria & Toro, 2012; World Bank, 2012).

Despite having enrollment rates above the regional average, there are wide disparities in access to postsecondary education by socio-economic status. Colombia has had two big expansions of its higher education system in the last two decades. Before the first expansion of the system in the early 1990s, approximately two percent of the relevant age-cohort population of the poorest quintile was enrolled in tertiary education, compared to 20 percent for the best-off quintile. After the expansion, the largest gains in coverage occurred in the best-off quintile, where net coverage rose from 23 percent to 40 percent. However, it is worth noting that there was also an expansion of about 170 percent in the first quintile (World Bank, 2003). Colombia had a second large expansion in college enrollment in the last decade. Recent evidence suggests that the share of enrolled students from the lowest incomes increased from 26.64 percent to over 40 percent between 2000 and 2009. Despite the increase in enrollment rates it is noteworthy that a sizable proportion of the increase in enrollment for this group of students happened at the technical and technological institutions that traditionally have fewer resources to support students, and hence exhibit high dropout rates (about 60 percent) (Sanchez & Marquez, 2012, 2012; World Bank, 2012).

As these data suggest, glaring socio-economic inequalities have been a constant since the inception of the higher education system in Colombia. In the mid-1980s Jimenez and

4_{For a more detailed description of the tertiary education system in Colombia see World Bank (2012) and}

(8)

Tang (1987) conducted a survey of high school seniors in Colombia and found that the socio-economic distribution of students in postsecondary institutions would substantially change if selection was based purely on merit. Most recently, Gaviria and Toro (2012) also found evidence that the private selective institutions are mostly enrolling students from the highest socio-economic status of the country. In summary, the country has currently an enrollment rate above the Latin American region and despite recent gains in access for students from low-income households, substantial inequalities in terms of access by level of income and type of institution (i.e., elite private) still remain.

Enrollment and Drop Out Rates

Despite a substantial increase in the enrollment rates of students in the system, the dropout rates are high and even increased during the years of the expansion. The dropout rates have risen substantially in the last decade; the five-year dropout rate for the 2000 students’ cohort was 50.9 percent and it increased to 55.1 percent for the 2005 cohort. The Ministry of Education (MEN) in collaboration with the University of Los Andes, developed a system to collect and analyze data related to college dropout and persistence, el Sistema de Prevención y Análisis de la Deserción en Educación Superior,5_(SPADIES). This system has collected information about enrollment and dropout for every student in every postsecondary institution in the country since 1998. According to the SPADIES, the first semester dropout rates have been increasing steadily in the last decade. Specifically, the rate increased from about 18 to 24 percent between 2000 and 2009 (Sanchez & Marquez, 2012). The explanation for the increase in the dropout rates is related to the expansion of the system during the same period. As mentioned above, the gross enrollment rate increased from 24 to more 30 percent from 1998 to 2008 (above the regional mean), and most of the access growth came from the students of the lowest income levels, and from students with relatively low scores on the national high school exit exam (SABER 11),

5_{SPADIES is similar to U.S.’s National Student Clearinghouse (NSC). Like the NSC, the SPADIES}

coverage of institutions is not complete and some institutions have imperfect reporting.

Tang (1987) conducted a survey of high school seniors in Colombia and found that the socio-economic distribution of students in postsecondary institutions would substantially change if selection was based purely on merit. Most recently, Gaviria and Toro (2012) also found evidence that the private selective institutions are mostly enrolling students from the highest socio-economic status of the country. In summary, the country has currently an enrollment rate above the Latin American region and despite recent gains in access for students from low-income households, substantial inequalities in terms of access by level of income and type of institution (i.e., elite private) still remain.

Enrollment and Drop Out Rates

Despite a substantial increase in the enrollment rates of students in the system, the dropout rates are high and even increased during the years of the expansion. The dropout rates have risen substantially in the last decade; the five-year dropout rate for the 2000 students’ cohort was 50.9 percent and it increased to 55.1 percent for the 2005 cohort. The Ministry of Education (MEN) in collaboration with the University of Los Andes, developed a system to collect and analyze data related to college dropout and persistence, el Sistema de Prevención y Análisis de la Deserción en Educación Superior,5_(SPADIES). This system has collected information about enrollment and dropout for every student in every postsecondary institution in the country since 1998. According to the SPADIES, the first semester dropout rates have been increasing steadily in the last decade. Specifically, the rate increased from about 18 to 24 percent between 2000 and 2009 (Sanchez & Marquez, 2012). The explanation for the increase in the dropout rates is related to the expansion of the system during the same period. As mentioned above, the gross enrollment rate increased from 24 to more 30 percent from 1998 to 2008 (above the regional mean), and most of the access growth came from the students of the lowest income levels, and from students with relatively low scores on the national high school exit exam (SABER 11),

5_{SPADIES is similar to U.S.’s National Student Clearinghouse (NSC). Like the NSC, the SPADIES}

(9)

administered by the Colombian Institute for the Promotion of Postsecondary Education (ICFES) (MEN-CEDE, 2009; Sanchez & Marquez, 2012).

The previous results suggest that despite the government’s success in increasing access for low-income students, the dropout rates have increased dramatically over the last decade and in particular for low-income students (Sanchez & Marquez, 2012).

Theoretical Framework and Review of the Literature

Becker’s (1967) human capital model predicts that individuals invest in education until the present value of benefits equals costs. Individuals who have no access to government assistance must rely either on family resources or credit to cover the costs. Individuals who need to rely on lenders would need to provide assets as collateral. The resulting constraint on borrowing implies that educational outcomes will be determined not only by the costs and benefits of the investment, but also by preexisting inequalities in family resources (Dynarski, 2002, 2003; Long, 2004).

(10)

Financial aid and student’s college enrollment and persistence/dropout rates

A substantial body of literature explores the impact between receiving financial aid (mostly in the form of grants) and college enrollment, persistence, and attainment (Angrist, Lang, & Oreopoulos, 2009; Angrist, Autor, Hudson & Pallais, 2104; Alon, 2011; Avery & Hoxby, 2004; Bettinger, 2010; Bettinger, Long, Oreopoulos & Sanbonmatsu, 2011; Bruce & Carruthers, 2011; Castleman & Long, 2012; Cornwell, Lee & Mustard, 2005; DesJardins, Ahlburg & McCall, 2006; Deming & Dynarski , 2009; Dynarski, 2002, 2003, 2008; Goldrick-Rab, Harris, Benson & Kelchen, 2011; Hansen, 1983; Heller, 1997; Kane, 1994; Long, 2004; Scott-Clayton, 2011; Stinebrickner & Stinebrickner, 2008). A review of the literature that summarized around twenty studies published from the mid-1980s to the mid-1990s exploring the association of financial aid on student’s enrollment concluded that with the exception of two studies6_{(Hansen, 1983; Kane, 1994), most of the studies} reviewed found that decreases in financial aid led to declines in enrollment. Heller (1997) concluded that in most of the studies reviewed financial aid was positively associated with enrollment and that low-income and minority students were more sensitive to changes in financial aid than their white peers. A limitation of the majority of the studies reviewed above is that they did not acknowledge the problem of simultaneity between aid eligibility and college enrollment, persistence, and attainment. Dynarski (2003) argues that aid eligibility is correlated with many observed and unobserved characteristics that affect college decisions. In her study she used a change in aid policy in the early 1980s to identify the effect of financial aid on college attendance and completion. Using a difference-in-difference (DID) estimation strategy, she concluded that the elimination of the aid program reduced college attendance probability by more than one-third. The estimates showed that an offer of $1,000 in grant aid increased the probability of attending college by about 3.6 percentage points and also appeared to increase school completion. More recent studies have used rigorous methods to estimate the impact of different merit-aid grant programs in

6_{Hansen tested the impact of the Pell grant on college enrollment and found no effect. He provided a}

number of explanations for this counter-intuitive finding such as that the subsidy might not have been targeted enough, or that it was not large enough, or that in the absence of this subsidy the enrollment rates of low-income students might have been even lower. Kane later replicated the study using a different set of years and confirmed the finding. He argues that given that the eligibility criteria when the Pell grant started were not clear, the students who managed to understand it would have attended college in the absence of the subsidy.

Financial aid and student’s college enrollment and persistence/dropout rates

A substantial body of literature explores the impact between receiving financial aid (mostly in the form of grants) and college enrollment, persistence, and attainment (Angrist, Lang, & Oreopoulos, 2009; Angrist, Autor, Hudson & Pallais, 2104; Alon, 2011; Avery & Hoxby, 2004; Bettinger, 2010; Bettinger, Long, Oreopoulos & Sanbonmatsu, 2011; Bruce & Carruthers, 2011; Castleman & Long, 2012; Cornwell, Lee & Mustard, 2005; DesJardins, Ahlburg & McCall, 2006; Deming & Dynarski , 2009; Dynarski, 2002, 2003, 2008; Goldrick-Rab, Harris, Benson & Kelchen, 2011; Hansen, 1983; Heller, 1997; Kane, 1994; Long, 2004; Scott-Clayton, 2011; Stinebrickner & Stinebrickner, 2008). A review of the literature that summarized around twenty studies published from the mid-1980s to the mid-1990s exploring the association of financial aid on student’s enrollment concluded that with the exception of two studies6_{(Hansen, 1983; Kane, 1994), most of the studies} reviewed found that decreases in financial aid led to declines in enrollment. Heller (1997) concluded that in most of the studies reviewed financial aid was positively associated with enrollment and that low-income and minority students were more sensitive to changes in financial aid than their white peers. A limitation of the majority of the studies reviewed above is that they did not acknowledge the problem of simultaneity between aid eligibility and college enrollment, persistence, and attainment. Dynarski (2003) argues that aid eligibility is correlated with many observed and unobserved characteristics that affect college decisions. In her study she used a change in aid policy in the early 1980s to identify the effect of financial aid on college attendance and completion. Using a difference-in-difference (DID) estimation strategy, she concluded that the elimination of the aid program reduced college attendance probability by more than one-third. The estimates showed that an offer of $1,000 in grant aid increased the probability of attending college by about 3.6 percentage points and also appeared to increase school completion. More recent studies have used rigorous methods to estimate the impact of different merit-aid grant programs in

6_{Hansen tested the impact of the Pell grant on college enrollment and found no effect. He provided a}

(11)

a number of states in the U.S. (Angrist, Lang, & Oreopoulos, 2009; Angrist, Autor, Hudson & Pallais, 2104; Bettinger, 2010; Bettinger, Long, Oreopoulos & Sanbonmatsu, 2011; Bruce & Carruthers, 2011; Castleman & Long, 2012; Cornwell, Lee & Mustard, 2005; Goldrick-Rab, Harris, Benson & Kelchen, 2011; Scott-Clayton, 2011a). Findings suggest that the impact of the grants varied according to the merit-based criteria used in the different states. The Georgia Hope program that required all recipients to maintain a minimum 3.0 GPA but had no full-time enrollment condition reduced the fraction of freshmen completing a full course load by six percentage points, decreased full-load enrollments and increase course withdrawals among resident freshmen (Cornwell, Lee & Mustard, 2005). On the other hand, the PROMISE scholarship in West Virginia that offers free tuition to students

who maintain a minimum GPA and course load found that PROMISE recipients were 25

percent more likely to have earned 30 or more credits in the first years and six to eight percent more likely to exceed the cumulative GPA requirement of the program. The results of the Florida State Access Grant program also found a positive impact on a number of outcomes such as increasing the likelihood of attending a four-year college versus a two-year college by 3.2 percent, maintaining continuous enrollment by 4.3 percent, and increasing bachelor degree attainment rates within six year by 4.6 percent (Castleman & Long, 2012).

(12)

recently Angrist, Autor, Hudson & Pallais (2104) evaluated the early impact of a randomly assigned scholarships for postsecondary studies in Nebraska. They found a positive impact of the program in terms of enrollment and persistence in particular for disadvantaged students (i.e., nonwhite, low ACT scores, targeting lower-tier colleges). They found the largest impact on enrollment, 3.4 percentage points, and persistence, 10.6 percentage points for disadvantaged students. In contrast, the impact on applicants with previous high academic achievement was smaller, 1.4 percentage points for enrollment and 5.7 for persistence.

Absent from the literature are evaluations on the effect of financial aid in the form of loans on a number of educational outcomes. Desjardins, Ahlburg & McCall (1999) used an event history analysis strategy to explore the association between receiving different forms of financial aid on student stop out and drop out for a sample of students attending the University of Minnesota at Minneapolis. They concluded that students who had access to loans had a smaller and negative statistically significant probability of stopping out in every year except the first one. They found no relationship between receiving loans and dropping out. Chen & Desjardins (2010) and Singell (2004) also found a positive effect of subsidized loans on student retention. Avery & Hoxby (2004) conducted a survey of high-ability students applying to college between 1999 and 2000 and they found that an additional $1,000 in loans raised a student probability of matriculating by 7 percent. Again, with the exception of Singell (2004), these studies on loans do not account for differential selection into different loan programs. Thus, the results of this study provides much needed evidence related to the effects of loans on enrollment, dropout, and academic performance.

(13)

Description of the financial aid program-ACCES

One of the factors that has allegedly contributed to the relatively low enrollment and high dropout rates of low-income students in Colombia is the lack of access to financial aid. Before 2002, the government could only offer financial aid to one in ten enrolled students. In addition, access to student loans for low-income students in the country was extremely limited because in order to secure a loan, students or their families had to provide real estate or other assets as collateral (Cerdan-Infantes & Blom, 2007). Before the establishment of ACCES, ICETEX had a loan program that was not designed to serve low-income populations. They financed only 50 percent of the tuition cost, required families to come up with two warrantors that owned real estate, expected students to start repayment immediately following graduation, charged unsubsidized rates (between 22-24 percent interest rates), and offered a relatively short re-payment period. To address these inequalities, in 2002 the government, with support from a $200 million dollar loan from the World Bank plus $87 million that the country committed to the program as part of the requirements of loan, created the ACCES program with the goal of increasing access while maintaining quality. The main objective of ACCES was to increase equity in access by increasing the proportion of high-achieving low-income students enrolling in higher education (ICETEX, 2005; ICETEX, 2011).7

The ACCES program removes collateral requirements and simplifies application procedures. ACCES primarily targets low-income students. In Colombia the government divides the population into six socio-economic strata (1 being the poorest and 6 the richest). The ACCES program provides subsidized loans (up to 75 percent of the tuition) for low-income students belonging to strata 1 and 28_{. Thus, low-income eligible applicants} can receive a loan -with 0 percent real interest rate- for up to 75 percent of tuition costs

7_{For a more detailed explanation of the ACCES program in Spanish see}

https://www.icetex.gov.co/dnnpro5/es-co/créditoeducativo/estudiostécnicostecnológicosyuniversitarios/largoplazoacces.aspx

8_{Students belonging to strata 1 and 2 who are also classified as being part of SISBEN (System of Selection}

of Beneficiaries for Social Programs) 1 and 2 receive a subsidy of 25 percent of the tuition. SISBEN identifies more accurately needy populations for different social programs and services. SISBEN is a proxy means-test that combines assets, household characteristics and socio-economic demographic information to create a continuous score – the SISBEN score. The SISBEN level is determined based on the score.

Description of the financial aid program-ACCES

One of the factors that has allegedly contributed to the relatively low enrollment and high dropout rates of low-income students in Colombia is the lack of access to financial aid. Before 2002, the government could only offer financial aid to one in ten enrolled students. In addition, access to student loans for low-income students in the country was extremely limited because in order to secure a loan, students or their families had to provide real estate or other assets as collateral (Cerdan-Infantes & Blom, 2007). Before the establishment of ACCES, ICETEX had a loan program that was not designed to serve low-income populations. They financed only 50 percent of the tuition cost, required families to come up with two warrantors that owned real estate, expected students to start repayment immediately following graduation, charged unsubsidized rates (between 22-24 percent interest rates), and offered a relatively short re-payment period. To address these inequalities, in 2002 the government, with support from a $200 million dollar loan from the World Bank plus $87 million that the country committed to the program as part of the requirements of loan, created the ACCES program with the goal of increasing access while maintaining quality. The main objective of ACCES was to increase equity in access by increasing the proportion of high-achieving low-income students enrolling in higher education (ICETEX, 2005; ICETEX, 2011).7

The ACCES program removes collateral requirements and simplifies application procedures. ACCES primarily targets low-income students. In Colombia the government divides the population into six socio-economic strata (1 being the poorest and 6 the richest). The ACCES program provides subsidized loans (up to 75 percent of the tuition) for low-income students belonging to strata 1 and 28_{. Thus, low-income eligible applicants} can receive a loan -with 0 percent real interest rate- for up to 75 percent of tuition costs

7_{For a more detailed explanation of the ACCES program in Spanish see}

https://www.icetex.gov.co/dnnpro5/es-co/créditoeducativo/estudiostécnicostecnológicosyuniversitarios/largoplazoacces.aspx

8_{Students belonging to strata 1 and 2 who are also classified as being part of SISBEN (System of Selection}

(14)

(subject to a program maximum). Applicants from strata 3 can receive a loan for up to 50 percent of the tuition -with 0 percent real interest rate- and applicants from strata 4 through 6 can receive a loan to cover up to 50 percent of the cost of the tuition -with 4 percent real interest rate. It is important to stress that the focus of this study is on students from strata 1 and 2, and within these two strata we explore heterogeneous effects. Upon graduation, students have a repayment grace period of one year after which they start repaying the loan balance. Approximately 83 percent of loans are repaid (Shen & Ziderman, 2008; ICETEX, 2011).9_{Most recently, it has been determined that ICETEX will} forgive 100 percent of the loans to students from the poorest backgrounds who attained an outstanding score on SABER PRO, the Colombian college exit exam (World Bank, 2012). Since the establishment of the program, about 291,306 students from strata 1, 2, and 3 had benefitted from the program (35.8 percent from strata 1, 47.8 percent from strata 2, and 13.9 from strata 3). Most of the students, about 80 percent, enrolled in private universities (ICETEX, 2008).

Criteria to Select ACCES Beneficiaries

Since the inception of the program, ICETEX defined and has continued to refine a procedure that is used by “departments” (geographic regions equivalent to U.S. states) to rank applicants and select the beneficiaries to the ACCES program. In order to apply for the ACCES subsidized loan the student should register in the ICETEX webpage10 providing information on the spot and score obtained in SABER 11, on the higher education institution in which the student was admitted and on the future guarantor of the student´s loan. Once the registrations process is closed ICETEX proceeds to select the potential beneficiaries of the credit. First of all, ICETEX removes all applicants whose loan guarantors do not comply with the established requisites11_{. Then ICETEX select the}

9_{For a more detailed explanation of repayment options and interest rates see page 72 of World Bank, 2012.} 10_{The registration process through the webpage is relatively recent. Some years ago the student filled a}

form which was then transcript into a database containing the information of all applicants.

11_{The study and approval of the loan guarantor of the educational loan is undertaken by the CIFIN (Central}

Financial Reporting), an organization that processes and supplies to ICETEX information related to customers and users of financial, real, cooperative and insurance sectors. For the requirements of the loan (subject to a program maximum). Applicants from strata 3 can receive a loan for up to 50 percent of the tuition -with 0 percent real interest rate- and applicants from strata 4 through 6 can receive a loan to cover up to 50 percent of the cost of the tuition -with 4 percent real interest rate. It is important to stress that the focus of this study is on students from strata 1 and 2, and within these two strata we explore heterogeneous effects. Upon graduation, students have a repayment grace period of one year after which they start repaying the loan balance. Approximately 83 percent of loans are repaid (Shen & Ziderman, 2008; ICETEX, 2011).9_{Most recently, it has been determined that ICETEX will} forgive 100 percent of the loans to students from the poorest backgrounds who attained an outstanding score on SABER PRO, the Colombian college exit exam (World Bank, 2012). Since the establishment of the program, about 291,306 students from strata 1, 2, and 3 had benefitted from the program (35.8 percent from strata 1, 47.8 percent from strata 2, and 13.9 from strata 3). Most of the students, about 80 percent, enrolled in private universities (ICETEX, 2008).

Criteria to Select ACCES Beneficiaries

Since the inception of the program, ICETEX defined and has continued to refine a procedure that is used by “departments” (geographic regions equivalent to U.S. states) to rank applicants and select the beneficiaries to the ACCES program. In order to apply for the ACCES subsidized loan the student should register in the ICETEX webpage10 providing information on the spot and score obtained in SABER 11, on the higher education institution in which the student was admitted and on the future guarantor of the student´s loan. Once the registrations process is closed ICETEX proceeds to select the potential beneficiaries of the credit. First of all, ICETEX removes all applicants whose loan guarantors do not comply with the established requisites11_{. Then ICETEX select the}

9_{For a more detailed explanation of repayment options and interest rates see page 72 of World Bank, 2012.} 10_{The registration process through the webpage is relatively recent. Some years ago the student filled a}

form which was then transcript into a database containing the information of all applicants.

11_{The study and approval of the loan guarantor of the educational loan is undertaken by the CIFIN (Central}

(15)

potential beneficiaries –i.e. those with an approved application- of the ACCES loans based on the spot12_{on SABER 11 that defines the cutoff to be eligible. Some other variables such} as the socio-economic status (belonging to strata 1 or 2), whether the institution the applicant was admitted to is an accredited institution, and whether the institution is part of a consortium established by ICETEX13_{may marginally alter the ranking among the eligible.} Nonetheless, these additional criteria do not affect the ranking of our sample as: 1) the sample of the study exclusively comprises applicants from strata 1 and 2; 2) few of the students in our study were admitted to accredited institutions14_{since their requirement for} admission are higher SABER 11 scores than the ones of the students close to the cutoff; 3) almost all universities belong to the ICETEX consortium

Since 2008, ICETEX has established a departmental “cutoff” for eligible applicants corresponding to the 70th_{percentile of the SABER 11 spots in the department. It is} important to clarify that the departmental “cutoff” corresponds to the maximum spot (minimum SABER 11 score) at which any applicant would receive credit with the available funds, and that no applicant with a SABER 11 score below the maximum spot qualifies for a credit. Thus, once the departments have calculated the departmental cutoff they simply rank applicants in descending order and determine beneficiaries up to the point where their budget allocation for the fiscal year enables them. Thus, the spot on SABER 11 is the most important factor that determines eligibility. According to the ICETEX officials no individual would become beneficiary if her SABER 11 score is lower than the minimum score (maximum spot) below the established departmental “cutoff”. However, a small proportion of applicants who did not meet the maximum SABER 11 “cutoff” spot may

guarantor see

http://www.icetex.gov.co/dnnpro5/es-co/cr%C3%A9ditoeducativo/estudiost%C3%A9cnicostecnol%C3%B3gicosyuniversitarios/largoplazoac ces.aspx .

12_{ICFES established a complex statistical process to calculate the “spot” of each high school graduate on}

the SABER 11 exam. For a detailed description in Spanish, see Resolucion 489, October 20, 2008, (Ministerio de Educacion Nacional and Instituto Colombiano para el Fomento de la Educacion Superior, 2008). The summary of the procedure in English is available from the authors upon request.

13_{ICETEX has established an agreement or consortium with over 300 universities and technical and}

technological institutions –almost all Colombian higher education institutions. As part of the agreement these institutions agree to cover 25 percent of the tuition of the ACCES beneficiaries from strata 1 through 3. For a list of institutions with agreements see: https://www.icetex.gov.co/dnnpro5/es-co/créditoeducativo/institucionesconconvenio.aspx

14_{In fact as of 2014 there were only 18 accredited institutions out of more than 300.}

potential beneficiaries –i.e. those with an approved application- of the ACCES loans based on the spot12_{on SABER 11 that defines the cutoff to be eligible. Some other variables such} as the socio-economic status (belonging to strata 1 or 2), whether the institution the applicant was admitted to is an accredited institution, and whether the institution is part of a consortium established by ICETEX13_{may marginally alter the ranking among the eligible.} Nonetheless, these additional criteria do not affect the ranking of our sample as: 1) the sample of the study exclusively comprises applicants from strata 1 and 2; 2) few of the students in our study were admitted to accredited institutions14_{since their requirement for} admission are higher SABER 11 scores than the ones of the students close to the cutoff; 3) almost all universities belong to the ICETEX consortium

Since 2008, ICETEX has established a departmental “cutoff” for eligible applicants corresponding to the 70th_{percentile of the SABER 11 spots in the department. It is} important to clarify that the departmental “cutoff” corresponds to the maximum spot (minimum SABER 11 score) at which any applicant would receive credit with the available funds, and that no applicant with a SABER 11 score below the maximum spot qualifies for a credit. Thus, once the departments have calculated the departmental cutoff they simply rank applicants in descending order and determine beneficiaries up to the point where their budget allocation for the fiscal year enables them. Thus, the spot on SABER 11 is the most important factor that determines eligibility. According to the ICETEX officials no individual would become beneficiary if her SABER 11 score is lower than the minimum score (maximum spot) below the established departmental “cutoff”. However, a small proportion of applicants who did not meet the maximum SABER 11 “cutoff” spot may

guarantor see

http://www.icetex.gov.co/dnnpro5/es-co/cr%C3%A9ditoeducativo/estudiost%C3%A9cnicostecnol%C3%B3gicosyuniversitarios/largoplazoac ces.aspx .

12_{ICFES established a complex statistical process to calculate the “spot” of each high school graduate on}

the SABER 11 exam. For a detailed description in Spanish, see Resolucion 489, October 20, 2008, (Ministerio de Educacion Nacional and Instituto Colombiano para el Fomento de la Educacion Superior, 2008). The summary of the procedure in English is available from the authors upon request.

13_{ICETEX has established an agreement or consortium with over 300 universities and technical and}

technological institutions –almost all Colombian higher education institutions. As part of the agreement these institutions agree to cover 25 percent of the tuition of the ACCES beneficiaries from strata 1 through 3. For a list of institutions with agreements see: https://www.icetex.gov.co/dnnpro5/es-co/créditoeducativo/institucionesconconvenio.aspx

(16)

still be approved if they are officially displaced by internal conflict, or meet other requirements to gain “exceptional” status.15

Figure 1 illustrates the different departmental “cutoffs” or maximum spots used to select beneficiaries to the program during the second semester of 2010 for individuals willing to enroll in college in the first semester of 2011. On the x-axis we plotted the proportion of students from strata 1 and 2 who graduated from high school in 2010 (who are the main target of the program) and on the y-axis we plotted the maximum spot/cutoff of the different departments. For example, for students in the major city of Colombia, Bogota, the maximum spot to qualify for ACCES is 554, and about 50 percent of high school graduates are from strata 1 and 2. This spot is substantially lower (i.e., requires a higher SABER 11 spot) than the spot of 884 in SABER 11 that a student in Choco (one of the most remote and poor areas in the country) would need to qualify for the program. In Choco almost 100 percent of high school graduates belong to the strata 1 and 2. Figure 2 clearly illustrates that the departmental cutoffs were established in a progressive way to promote the enrollment of low-income students in particular from remote and poor regions of the country. Thus, the program “cutoff” rules determine that the greater the percentage of low-income students in a department, the higher the spot (i.e., lower SABER 11 spot requirement) to be eligible for the subsidized loan program.

15_{In our sample of beneficiaries we found that 3 percent of the applicants below the departmental cutoff}

received ACCES. This might be related to students getting additional points for their “exceptional” status and also because applicants who were already enrolled in the second semester could apply to ACCES using their first semester GPA instead of their SABER 11 scores.

still be approved if they are officially displaced by internal conflict, or meet other requirements to gain “exceptional” status.15

Figure 1 illustrates the different departmental “cutoffs” or maximum spots used to select beneficiaries to the program during the second semester of 2010 for individuals willing to enroll in college in the first semester of 2011. On the x-axis we plotted the proportion of students from strata 1 and 2 who graduated from high school in 2010 (who are the main target of the program) and on the y-axis we plotted the maximum spot/cutoff of the different departments. For example, for students in the major city of Colombia, Bogota, the maximum spot to qualify for ACCES is 554, and about 50 percent of high school graduates are from strata 1 and 2. This spot is substantially lower (i.e., requires a higher SABER 11 spot) than the spot of 884 in SABER 11 that a student in Choco (one of the most remote and poor areas in the country) would need to qualify for the program. In Choco almost 100 percent of high school graduates belong to the strata 1 and 2. Figure 2 clearly illustrates that the departmental cutoffs were established in a progressive way to promote the enrollment of low-income students in particular from remote and poor regions of the country. Thus, the program “cutoff” rules determine that the greater the percentage of low-income students in a department, the higher the spot (i.e., lower SABER 11 spot requirement) to be eligible for the subsidized loan program.

15_{In our sample of beneficiaries we found that 3 percent of the applicants below the departmental cutoff}

(17)

Figure 1. Proportion of High School Graduates in Strata 1 and 2 and Departmental “Cutoff” Spots in Exit High School Exam

Bogota

Antioquia

Risaralda

Santander San Andres

Caldas Meta

Quindio Valle

Cundinamar Atlantico

N. Santand Magdalena

Tolima

Boyaca Bolivar

Cesar

Casanare CaucaAmazonas

NariñoHuila SucreCordoba

La Guajira

Arauca Guaviare

Guainia CaquetaVaupes

Choco

Putumayo Vichada

500

600

700

800

900

Saber

11 c

ut

of

f b

y d

epar

tm

ent

.6 .7 .8 .9 1

Proportion of students in SES 1 & 2

Source: ICETEX, ICFES (2013). Students who took SABER 11 in the second semester of 2010.

(18)

Figure 2. Proportion of ACCES Beneficiaries and High School Exit Exam Score for Applicants from Strata 1 and 2

.2

.4

.6

.8

1

Tr

ea

tm

ent

pr

oba

bi

lity

-1 -.5 0 .5 1

Distance to the cutoff

Source: Authors' estimations over the optimal bandwidth Imbens and Kalyanaraman (2012).

Data

(19)

Methodology

Sample

ICETEX provided us with the complete list of all the applicants to the ACCES program between 2002 and 2012, a total of 105,142 individuals. We restricted this sample to include only applicants from the lowest strata because we are interested in evaluating the impact of the ACCES program among the poorest segments of the population. Besides for these two strata the same financial conditions of the loan apply. After restricting the sample to include low-income students (i.e., defined by those from the 1st_{and 2}nd socio-economic strata), who had available information in the SPADIES datasets, and who did not receive any other private or public financial aid (to avoid confounding the impact of ACCES), we reached a final sample of 46,418 individuals. We found that around 3 percent of the beneficiaries did not have the minimum SABER 11 spot to be ACCES eligible. This group might not only include applicants wrongly selected as beneficiaries, but also students who according to their department’s information, gained “exceptional status” due to their personal history (e.g., being refugees from armed conflict). The samples used to estimate the impact of the program on the three outcomes of interest (i.e., enrollment, dropout, and academic performance) are described in detail below.

Enrollment

(20)

Dropout and Academic Performance

The total sample of students used to estimate the impact of the program in terms of dropout and academic performance, defined as percentage of courses passed, consisted of 46,418 individuals from strata 1 and 2. We restricted the sample to individuals who fell into the optimal bandwidth a total of 19,435 low-income college students.

Estimation Strategy

Since the system of departmental cutoffs provides a source of exogenous variation in ACCES loan eligibility, our proposed evaluation uses a regression discontinuity design (Murnane & Willett, 2011) to estimate the effects of the program on three outcomes: 1) enrollment, 2) dropout, and 3) academic performance measured as percentages of courses successfully passed by the students. It is important to mention that even though the RDD technique is strong in terms of internal validity of the estimates, the findings only apply to the students at the margin. In other words, our findings should be restricted to applicant low-income students around each departmental cutoff. No inferences can be made about students who were well above or below the cutoff established by each department of the country.

Intent to Treat (ITT). We first estimate the ITT effect of the program on the three outcomes of interest (i.e., enrollment, dropout, and academic performance) based on comparing the applicants to ACCES who were around each departmental cutoff. Specifically, we used the SABER 11 cutoff criterion used in each department for each specific semester, and then restricted the sample to individuals who fell within the optimal bandwidth.

We specified a model (equation 1) where the value of each specific outcome of

interest Yi, is a function of a variable indicating whether the individual is eligible to the

program Di, and a function of the assignment variable based on his or her SABER 11 score