PISA: What Makes the Difference?

Discussion of “Misreported Schooling,
Multiple Measures and Returns to
Educational Qualifications” by Erich
Battistin and Barbara Sianesi
Andreas Ammermueller
ZEW, Mannheim
Outline of discussion
• Contributions
• Assumptions
• Empirical application
Contributions
• General approach for correcting measurement
error in treatment effects framework: extend
measurement error correction (IV) to nontrivial categorical cases
• Can be extended to multiple treatments and
heterogeneous invidual returns
• Application in a policy-relevant field, estimates
of accuracy of qualification data for the UK
Assumptions
• Unconfoundedness: Few data sources provide
sufficient information to be credible, even
questionable in case of NCDS
• Non-differential misclassification given X: Even
stronger than above; what drives
missclassification and its direction?
• Independent reports of DA and DB: Difficult,
regional/school level correlation in NCDS case
Empirical application: Returns to higher
education, NCDS
• Distinguishing between only two groups invites
no less criticism than using years of schooling:
very heterogeneous groups, what is estimated?
• Restricted sample with full educational
information: might trade measurement error
against sample selection
• Biases have same direction but different size
compared to OLS/years of schooling case
main story of application
Application continued
• Do assumptions for measurement error
correction apply here?
• Test scores as controls for ability: imperfect
measure, just another educational outcome
• Not convinced of control for ability bias: no
classical approach
show control for ability bias in OLS case and
test assumptions as far as possible