Exercise I: Institutions; Statistical tables and IV
- Consider Table 2 of Acemoglu, Johnson and Robinson (2001). Explain what all the numbers mean, how we should interpet them, and why we care about them!
- It is a probelm that a regression of income on institutions may suffer from an endogeneity problem. If we let yi denote log of GDP per capita and si institutional quality, (old fashioned) econometricians would write the problem as
yi=α+βsi+εi
si=γ+δyi+νi
Explain in words what the problem is. Solve the system of equations and use this to show that OLS can't yield meaningful estimates. - Explain what an instrument is. How should we think of an instrument in the system above? Explain intuitively and as formally as you can why an instrument can solve the problem of endogeneity.
- Discuss the validity of settler mortality as an instrument in the above relationship. What can make the instrumment invalid, how can we check this, and what can be done to fix it?
- Discuss the alternative explanation that human capital, not institutions, is the true fundamental cause of growth. Why is it difficult to distinguish between the two explanations?
Exercise II: Institutional persistence; regression discontinuity designs (RDDs)
- Explain what a regression discontinuity design is in the case of a one-dimensional forcing variable (the standard case), what it can do, and why it may be helpful. Yo may for instance consider the case of electoral support and public spending discussed in class.
What is the distinction between the forcing variable and the discontinuity? - What can be potential problems with RDDs in general?
- Explain what the "mining mita" studied by Dell (2010) was and why it could matter for development today
- What is the difference between using the mita as and RDD and the usual design studied in question 1? How does Dell solve this?
- Explain Dell's various attempts at controlling for the forcing variable in her case. Which design is in your opinion the best to achieve i) low variance, ii) low bias, and iii) generally insightful results?
- What potential problems can occur in her RDD design? How can she test for them?