Education Choices in Mexico by Attanasio, Meghir

1
Education Choices in Mexicoby Attanasio, Meghir
Santiago

• Comments by
• Francisco H.G. Ferreira

2
Summary

• Question What was the impact of PROGRESA on
  school enrolment?

3
Summary

• Question What was the impact of PROGRESA on
  school enrolment?
• Answer 1 Randomization at village level permits
  use of simple differences on post-treatment
  sample. Finds significant positive impact, but
  both for eligible and non-eligible households.
  (Schultz, 2001)

4
Summary

• Problem 1 Randomization did not balance sample
  on pre-treatment enrollment (Behrman Todd,
  2000).
• Solution 1 Differences-in-differences reveal
  significant positive impact for eligible
  households, and insignificant for non-eligible.
• Problem 2 Do announcement effects affect
  credibility of control villages as appropriate
  counterfactual?
• Problem 3 How can we investigate the likely
  effects of changes in program design?
• Solution 2 Estimate structural model for the
  occupational choice of children.

5
Main Comment/Question

• Combination of randomized quasi-experimental data
  and structural model definitely the right
  direction.
• But I wonder whether use of a full dynamic
  programming model of occupational choice was not
  an overshooting in imposition of structure
• Two occupational categories only.
• Terminal values as quadratic function of final
  schooling.
• Discount factor of 0.9.
• More flexible structures available through
  current value multinomial logits, with
  means-testing and conditionality restrictions.

6
Other questions

1. Does allowing for announcement effects go in the
   right direction? Shouldnt the difference in the
   impact of the grant between treatment and control
   fall as control knows it will receive treatment
   next year? It seems to rise (Table 14).
2. How is the individual wage variable constructed?
   Was there an Appendix?
3. Overall balance main reason to estimate a model
   of behaviour is to investigate impact of changes
   in program design. Perhaps a more flexible model,
   with more simulation exercises?