ECON 6202 Policy Evaluation Methods
UNSW, School of Economics

FINAL EXAM
Read the following instructions carefully before attempting the exam:
The exam is due on the 5th of August at 17:00. Please send me by email your structured but as
succinct as possible answers and do files (or equivalent in other programs). Do not send them in
compressed zip folders (only as attachments). Make sure you present your results properly.
Please also make sure you are attempting the exam paper for the right course code. Include your
course code, name and zID in the filename of your submission. The data necessary to answer
the empirical questions in the exam is in the ‘Take Home Exam – DataFile’ (data.dta) on
Moodle.
Food aid is one of the most common forms of humanitarian aid but, in recent years, it has
received increasing criticism. Aid workers, human rights observers, and journalists have
accused humanitarian aid of being not only ineffective, but of actually promoting conflict. In
this take home exam, we will study the effect of food aid distributed by the US (in the form of
wheat) on conflict.
Data.dta is a panel dataset with information related to aid and conflict for 127 countries
(ID_country) between 1970 and 2006 (year). The countries are classified in six regions: South
Asia, East Asia and Pacific, Europe and Central Asia, Latin America and Caribbean, Middle
East and North Africa, and sub-Saharan Africa. The variable ID_region takes values 1 to 6
depending on the region of the country. The variable wheat_aid is the amount (in 1000 metric
tons) of wheat coming from the US received by a country in a specific year (notice that data is
missing in some year/countries). The variable any_war is a dummy variable equal to 1 if there
was any conflict in a country that year and is 0 otherwise (a conflict is defined as the use of
armed force between two parties that results in at least 25 battle deaths in a year).
1) Using OLS, regress the conflict dummy on the amount of wheat aid a country received that
year from the US. Control for country and year fixed effects. Explain why the estimated
parameter from this regression is biased. Pick one omitted variable and predict the direction of
the omitted variable bias. Provide also a mathematical explanation for the expected direction
of the bias.
In order to estimate the effect of food (wheat) aid on conflict, we will exploit two features of
the US wheat aid system. First, US agricultural price stabilization policy requires the
government to purchase wheat from US farmers, causing the government to accumulate excess
reserves in high production years. Much of the government surplus is then shipped the next
year to developing countries as food aid. Second, the US tends to give this surplus wheat to the
same group of countries. We use this information in order to build an instrument for wheat_aid
by multiplying the production of wheat in the US the previous year by the propensity of a
country in receiving wheat aid from the US. The variable US_wheat_production contains the
production of wheat by the US in a year. The production is measured in millions of tons.
Remember to use the one year lag production for the instrument. Our proxy for the propensity
of a country to receive wheat aid from the US will be the share of years in our sample where
wheat_aid > 0. Notice that data is missing in some year/countries. Do not take into account
those year when you calculate the propensity of a country to receive wheat aid.
2) Estimate the first stage equation where the multiplication between the production of wheat
in the US the previous year and the propensity of a country to receive wheat aid from the US
is the instrument, and wheat_aid is the endogenous variable. Include country and year fixed
effects in the first stage. Interpret the results.
3) Using the proposed instrument, estimate the effect of wheat_aid on any_war. Interpret the
results.
4) Under which assumptions is the proposed IV setting valid? Estimate the effect in question
3) controlling for region fixed effects multiplied by year fixed effects. How do the identification
assumptions change?
5) Estimate the reduced form of question 3). Under which assumptions are these estimates
valid? Write down mathematically and explain each assumption.
6) Estimate the reduced form of question 3) using two alternative instruments. Alternative
instrument 1: Instead of using the lagged US wheat production use the lagged US carrot
production. Alternative instrument 2: Instead of using the lagged US wheat production use the
lagged US lettuce production. You can find the values for the production of carrots and lettuce
in the US for a specific year in the variables USprod_Carrots and USprod_Lettuce. Do these
estimates provide any encouraging results for the validity of the assumption you stated in
question 4)? For answering this question remember that the US production of carrots or lettuce
should have no influence on US wheat aid.
Wheat production in the US was extremely high between 1981 and 1985. Because of this we
should expect the following two results:
a) Countries with a high propensity to receive wheat aid from the US should receive high levels
of wheat aid the year after the production of wheat in the US is extremely high (between 1982
and 1986).
b) If food aid has any effect on conflict, we should also observe differences in the variable
any_war for countries that have a high propensity to receive wheat aid from the US the year
after the production of wheat in the US is extremely high (between 1982 and 1986).
7) Follow a Differences-in-Differences methodology to test expected results a) and b). Under
which conditions are Differences-in-Differences a valid methodology to test these expected
results?