Discussion
What This Paper Adds
This paper contributes in three ways:
1. Explicit selection modeling. By using a binary Roy-type model, the paper accounts for the fact that CI does not select children randomly, and that the families most likely to be chosen may also be those whose children benefit most.
2. Subjective income expectations in a developing-country context. Collecting individual-level beliefs about future earnings and using them to estimate perceived returns to education is uncommon in the CI literature and rare in developing-country education research. The finding, that these beliefs are realistic on average but do not predict aspirations, is itself informative: it suggests that other constraints bind first.
3. The MTE as a diagnostic tool. The marginal treatment effect curve provides information that ATE and ATT do not: it tells us whether the right children are being reached. The finding that the MTE is positive and declining means CI’s targeting is, in this sense, well-designed.
Why Not Statistically Significant?
The ATT of 17–20 percentage points is substantively meaningful but imprecisely estimated. There are two complementary explanations:
Statistical. The sample is small (403 children total, 163 sponsored), and the model is complex: five parameters governing the error structure alone, estimated by numerical integration with bootstrap standard errors. Each of these choices adds uncertainty. This is not a flaw in the design, it is the cost of using a more credible identification strategy.
Limitations
The outcome is a stated aspiration, not an enrollment decision. Children may say they aspire to university but not act on it when the time comes. Conversely, some who say they do not aspire may go on to attend. The relationship between stated aspirations and actual educational attainment is an open question in this context.
The survey is a single cross-section. We cannot observe the same child before and after sponsorship, which means we cannot rule out pre-existing differences not captured by the model. The identification relies on the exclusion restrictions being valid, a credible but ultimately untestable assumption.
With 163 sponsored and 240 non-sponsored children, the sample is too small to detect the likely effect size with the complex Roy model at conventional significance levels. A larger study using the same methodology could produce more precise estimates.
The large overlap between CI and Prospera participation (85–86% of both groups) means the two programs cannot be fully separated. The robustness analysis assuming additive Prospera effects is reassuring, but the independent effect of CI cannot be perfectly isolated from the context of Prospera.
Policy Implications
Despite the imprecision, the findings have a clear message for CI and similar organizations:
CI does not face an efficiency-equity trade-off in the context of Mexico. The children most in need, who are the most likely to be selected, also appear to gain the most from the program. This is not guaranteed — many targeting rules select the most vulnerable but also the least likely to respond. The MTE result is reassuring.
For policymakers more broadly, the finding that perceived returns to education are not significant predictors of aspiration at ages 12–15 has implications for information-based interventions. Providing information about returns to schooling, popular in development economics since Jensen (2010), may not be effective at this age in this context. Interventions that work directly on internal constraints (self-efficacy, role models, attainable short-term goals) may be more productive.
Open Questions
The paper raises several directions for future work:
- Long-run outcomes. Do the sponsorship-induced aspirations observed at age 12–15 translate into actual enrollment at secondary and tertiary level?
- Mechanisms. Is the positive correlation between selection and treatment effect driven by family motivation, by the school-attendance requirement, or by something else?
- External validity. The results are from two states in rural Mexico. Do they hold in other CI contexts (Kenya, Indonesia, the Philippines)?
- Complementarity with Prospera. Given the near-universal Prospera overlap, it would be valuable to study CI’s effects in settings where Prospera is absent.