Public-Private Partnerships: Causal Evidence from Colombia

2023-02-13

2025-01-31

- Student score on PPVT
- Student score on MODEL quantitative instrument modules (Number Identification)
- Student score on MODEL executive function instrument modules (Head, Toes, Knees and Shoulders, Forward Digit Span, Backward Digit Span, Pencil Tap)
- Student score on MODEL socioemotional development modules (Perspective-Taking/Empathy, Understanding Feelings)

- Parental expenditure/year on educational materials (Question H7_14 in household survey)
- Index of parental time investment in child (see explanation below)
- Index of parental involvement with school (see explanation below)
- Number of parent-reported school absences (Questions H4_2 and H4_4 in household survey)
- Index of parent-reported perception of the school environment (see explanation below)

Indices will be constructed following the procedure in Anderson (2008) and consist of the indicated survey measures:

Measures to be included in "Index of parental time investment in child":
- Parental frequency of talking with child about school (Question H7_10 in household survey)
- Parental amount of time spent with child on reading (Question H7_11 in household survey)
- Parental amount of time spent with child on numbers (Question H7_12 in household survey)

Measures to be included in "Index of parental involvement with school":
- Frequency with which parent reports contacting the school (Question H7_1 in household survey)
- Frequency with which parent reports being contacted by the school (Question H7_3 in household survey)
- Indicator for whether parent reports belonging to school's PTA (Question H7_8 in household survey)

Measures to be included in "Index of parent-reported perception of the school environment"
- Number of times parent reports being contacted by school due to problems between child and another student (Question H8_1 in household survey)
- Number of times parent reports contacting school due to problems between child and another student (Question H8_2 in household survey)
- Indicator for perceived existence of physical and verbal mistreatment inside the school (Question H8_3 in household survey)
- Satisfaction of child with attending school (Question H8_4 in household survey)
- Frequency of child being reluctant to attend school (Question H8_5 in household survey)
- Satisfaction of parent with child's education at school (Question H8_6 in household survey)
- Indicator for whether parent would have changed child's school if they could (Question H8_7 in household survey)
- Indicator for whether child was socializing well with classmates (Question H8_9 in household survey)

We use rank-ordered lists of applicants to oversubscribed schools in administration to define our sample. The Secretary of Education of the District (SED) in Bogota conducted a pairwise randomization for each of the applicant lists. Students were sorted in a descending order by a priority index, which was designed to give certain weights to particular criteria such as closeness to the school, victim status, or household wealth. Then, pairs within each lists were formed. One of the students of the pair was assigned to a school in administration and the other was assigned to a regular public school.

The way the student pairs were constructed implies that the two students should be similar to each other in terms of sociodemographic characteristics. Taking advantage of that similarity within a pair, we will estimate the causal effect of attending a school in administration by comparing treated and control students within their respective peers.

The uncontrolled estimating equation is as follows:

y_{ip} = constant + β*(treat)_{ip} + u_{ip}

The controlled estimating equation follows:

y_{ip} = constant + β*(treat)_{ip} + Γ'X_{ip} + u_{ip},

where X are the components of the priority index for student i in pair p, recorded prior to treatment assignment in the administrative database of the school assignment system, as well as student age and gender.

As a robustness check, we will also estimate a model with pairwise fixed effects. We note that the power of this specification depends on the rate of missingness in the student-level data, as pairs with one missing student will necessarily be dropped (in contrast to the previous specification without pairwise fixed effects):

y_{ip} = β*(treat)_{ip} + Γ'X_{ip} + λ_{p} + u_{ip}

where y_{ip} denotes an outcome of interest for student i in pair p, treat is an indicator variable that takes a value of 1 if the student attends a school in administration, X_{ip} is a vector containing student age and gender, and λ_{p} denote pair fixed effects.

We will explicitly estimate heterogeneity with respect to two dimensions:
- Gender
- Identity of provider operating the school in administration.

Not available

The Secretary of Education of the District (SED) used a computer to implement an algorithm that sorted students by priority order, formed adjacent pairs in that priority ordering, and determined admission offers randomly within that pair.

Students who are applying for admissions to the first grade offered by schools in administration.

No

1,502 students.

1,502 students.

The sample of 1,502 students is divided evenly: 751 students are assigned to a school in administration and the remaining 751 are assigned to a regular public school.

Main outcomes are test scores that have been standardized, so that MDEs are in terms of standard deviations. The MDE for each main outcome is 0.148 SD.