# Econ 403: Topics in Development Economics and Policy

1
Econ 403: Topics in Development Economics and Policy
Assignment 1
Due: January 25th, 2016, 2:10pm (beginning of class)
Location: Lecture room
I. Robustness of the Cross-Country Income – Poverty Relationship
The objective of this question is to assess whether modifications to the regression model
in Besley & Burgess (2003) affect the conclusions regarding the sufficiency of promoting
national income per capita growth to achieve the MDG poverty goal.
• In order to carry this exercise, you will use the Stata dataset titled
“povertygoals7_bis.dta”, made available by the authors. it is available on the
course’s Blackboard webpage.
• This dataset is composed of repeated observations for a sample of countries. We
will employ the sample of countries used for the analysis by the authors, those
that include extreme poverty measures (which use a poverty of approximately
\$1.09 per person/day).
Instructions
(a) Estimate the bivariate regression:
ln Pit = ? + ?ln µit + eit
where
ln Pit = the natural logarithm of the poverty headcount ratio (headcoun)
ln µit = the natural logarithm of the GNP per capita, PPP (gnppc)
Hints: Construct variables in natural logs using ‘generate’ command.
For the regression analysis, use ‘regress’ command, with option ‘if povline < 50’.
Also, include the option ‘cluster(ccode)’ to allow the error terms to be correlated
for all observations for each country.
The command should have the following structure:
regress lny lnx if povline < 50, cluster(ccode)
[2 points]
(b) Estimate the multivariate regression allowing for country fixed effects (which
control for unobserved characteristics that are fixed at the country level).
ln Pit = Si ?i1(ccode = i) + ?ln µit + eit
where
1(ccode = i) are indicator variables for each one of all countries i
2
Hints: The country fixed effects specification can be estimated in Stata using the
‘areg’ command, with the option ‘absorb(ccode)’
The command should have the following structure:
areg lny lnx if povline < 50, absorb(ccode) cluster(ccode)
[2 points]
following controls:
(1) year fixed effects:
ln Pit = Si ?i1(ccode = i) + ?ln µit + Stdt1(year = t) + eit
where 1(year = t) are indicator variables for each one of the years (denoted t)
in the dataset.
(2) the natural logarithm of population size
ln Pit = Si ?i1(ccode = i) + ?ln µit + Stdt1(year = t) + ß ln POPit + eit
where
ln POPit = the natural logarithm of the country population (pop)
Hint: The year fixed effects should be added manually to the country fixed effects
specification, by creating year dummy variables. The other options mentioned
above should remain the same.)
[3 points]
(d) Test for the statistical significance of each estimate of the ? (simple/partial)
correlation in each model. Interpret the magnitudes of the coefficients in each
model, and relate these, in the context of the discussion of omitted variables (or
unobserved heterogeneity) bias. [5 points]
(e) Calculate the predicted annual growth rate of national income per capita necessary
to halve world poverty by 2015 (from 1990). Interpret the results. What
conclusions can we reach regarding the main analysis conducted by the authors?
[5 points]
(f) Repeat steps (a) – (e) using as dependent variable the natural logarithm of the
poverty gap. What conclusions can we reach in this case? [3 points] 