ECON 399 Fall 2019

Introductory Econometrics Jeffrey Penney

Assignment 1
due in class on October 1st, 2019
#1 Consider a dataset of 180 countries that contains their GDP growth rates for the year 2015 (call this
variable 𝐺𝐷𝑃) and a variable for whether or not they had an active civil war in the country in that same
year (call this variable 𝑤𝑎𝑟).
a) What kind of dataset is this?
b) Suppose you wanted to test the effect of having an active civil war on GDP growth. What regression
would you run?
c) What is the expected sign on the coefficient of interest? Explain.
d) Can you argue that the coefficient of interest has a causal interpretation? Whether you claim yes or no,
make a mathematical argument as to why (hint: talk about which SLR are satisfied or violated and why).

#2 Consider the following questions in the simple regression framework.

a) What is a regression residual? Define it mathematically. How is it different from an error term?
b) What is the difference between the population regression function (PRF) and the fitted regression line?
Define each mathematically.
c) What is SLR3? Give an intuitive example of a case where it can fail if we are regressing wages on
education.
d) Taken together, SLR1-SLR4 say something about the unbiasedness of OLS. What do these
assumptions imply for the expected values of 𝛽̂0 and 𝛽̂1 ?

#3 A person wishes to model the determinants of winning in sumo wrestling matches. The following
regression is modelled: 𝑤𝑖𝑛𝑠𝑖 = 𝛽0 + 𝛽1 𝑤𝑒𝑖𝑔ℎ𝑡𝑖 + 𝛽2 ℎ𝑒𝑖𝑔ℎ𝑡𝑖 + 𝛽3 𝑐ℎ𝑎𝑚𝑝𝑖 + 𝑢𝑖 where 𝑤𝑒𝑖𝑔ℎ𝑡 is the
weight of the wrestler, ℎ𝑒𝑖𝑔ℎ𝑡 is the height of the wrestler, 𝑐ℎ𝑎𝑚𝑝 is the number of championships that
the wrestler has previously won as an amateur (before turning professional), and 𝑤𝑖𝑛𝑠 is the number of
wins a wrestler has in their career at the professional level in their first 25 matches only. The sample is
obtained by randomly picking 20 wrestlers. Out of these 20 wrestlers, two have zero wins, and three have
the maximum of 25.

a) What is the expected sign on 𝛽̂3 ? Explain.

b) Suppose you are paired with another student to estimate the model, and they suggest estimating the
regression log⁡(𝑤𝑖𝑛𝑠𝑖 ) = 𝛽0 + 𝛽1 𝑤𝑒𝑖𝑔ℎ𝑡𝑖 + 𝛽2 ℎ𝑒𝑖𝑔ℎ𝑡𝑖 + 𝛽3 𝑐ℎ𝑎𝑚𝑝𝑖 + 𝑢𝑖 instead. What are the
implications of this?
c) Name one factor that we would expect to find in the error term 𝑢.
d) Suppose that wrestlers are randomly assigned to fight at different weights when competing (for
example, the first fighter is told to weigh 156 pounds, the second 250 pounds, and so forth) instead of
choosing their weight themselves. Consider the simple linear regression 𝑤𝑖𝑛𝑠𝑖 = 𝛽0 + 𝛽1 𝑤𝑒𝑖𝑔ℎ𝑡𝑖 + 𝑢𝑖 .
Does the estimate of 𝛽̂1 have a causal interpretation? Why?

#4 Suppose that the population model for the dependent variable 𝑦 is

𝑦 = 𝛽0 + 𝛽1 𝑥1 + 𝛽2 𝑥2 + 𝛽3 𝑥3 + 𝑢
and the model satisfies assumptions MLR1-MLR4. However, we instead estimate a different model
𝑦 = 𝛽0 + 𝛽1 𝑥1 + 𝛽2 𝑥2 + 𝑢.

Define the estimated coefficients from this regression as 𝛽̃0 , 𝛽̃1 , and 𝛽̃2 . Show that the expected value of
𝛽̃1 is equal to
∑𝑛𝑖=1 𝑟̂𝑖1 𝑥𝑖3
𝐸(𝛽̃1 ) = 𝛽1 + 𝛽3
∑𝑛𝑖=1 𝑟̂𝑖12

where the 𝑟̂𝑖1 are the OLS residuals from the regression of 𝑥1 on 𝑥2 .
Hints: Begin by using
∑𝑛𝑖=1 𝑟̂𝑖1 𝑦𝑖
𝛽̃1 =
∑𝑛𝑖=1 𝑟̂𝑖1
2

and then substitute the population model for 𝑦. Recall the algebraic properties of OLS and make use of
them, and use the fact that ∑𝑛𝑖=1 𝑟̂𝑖1 𝑥𝑖1 = ∑𝑛𝑖=1 𝑟̂𝑖1
2
. After simplifying, take the expected value of beta,
treating 𝑥𝑖3 and 𝑟̂𝑖1 as non-random.