EC3303 Econometrics I Homework 1

(Note: Please type your answers and hand in only 1 copy per group. Write all group members’
names (as in matric card) in your submission. Deposit your submission in Kelvin Seah’s Mailbox
(mailbox 59) at the Economics Dept, Level 6 of AS2, before October 5 2018, 2359hrs)

1. The Trim and Fit (TAF) programme was a weight loss programme introduced by the Ministry
of Education that was meant to tackle child obesity in Singapore schools between 1992 and
2007. Suppose the percentage of obese school children in Singapore increased from 10% in
1990 to 14% in 2009. What can you say about the policy? Did it work to produce its intended
effect? Elaborate.

(3 Marks)

Answer:
Whether the policy worked or not is unclear from the information provided. Many
observers, seeing the increase in obesity prevalence, would conclude that the policy did not
work. But the causal effect of the TAF programme on obesity is not measured by the
increase in obesity over the years. Instead, to know the causal effect, we need to ask: what
would obesity rates have been had the TAF programme not been implemented? Suppose
that if the TAF policy had not been implemented, the obesity rate in 2009 would have been
16%. Then the causal effect of the TAF programme would be the difference between the
actual obesity rate of 14% and the counterfactual rate of 16%. In other words, the effect of
the TAF programme is to reduce obesity by 2 percentage points. Hence, we will need to
know “what would have been” in order to say whether the policy actually worked.

2. You postulate a linear relationship between two variables, X and Y of the form:

𝑌𝑖 = 𝛽0 + 𝛽1 𝑋𝑖 + 𝑢𝑖

where 𝑢𝑖 denotes all other factors apart from 𝑋𝑖 affecting the value of Y for observation i.

1 𝑉𝑎𝑟[(𝑋𝑖 −𝜇𝑋 )𝑢𝑖 ]

Suppose LSA #1 is satisfied so that 𝐸(𝑢𝑖 |𝑋𝑖 ) = 0 and 𝑉𝑎𝑟(𝛽̂1 ) = 𝑛 × [𝑉𝑎𝑟(𝑋𝑖 )]2
.
Show that, if 𝑉𝑎𝑟(𝑢𝑖 |𝑋𝑖 = 𝑥) = 𝜎𝑢2 , where 𝜎𝑢2 is a constant, then the expression for
𝜎2
𝑉𝑎𝑟(𝛽̂1 ) simplifies to 𝑛𝜎𝑢2 . Show all steps clearly.
𝑋

(4 Marks)

Answer:

1
2
3
3. You postulate a linear relationship between two variables, X and Y of the form:

𝑌𝑖 = 𝛽0 + 𝛽1 𝑋𝑖 + 𝑢𝑖

where 𝑢𝑖 denotes all other factors apart from 𝑋𝑖 affecting the value of Y for observation i.

Show the assumption that:

𝐸(𝑢𝑖 |𝑋𝑖 = 𝑥) = 0

(i.e. least squares assumption 1) is equivalent to the assumption that

𝐸(𝑌|𝑋) = 𝛽0 + 𝛽1 𝑋

(2 Marks)

Answer:

4
5
4. You are interested in examining the relationship between earnings and height. Accordingly
you run a regression of Earn on Height using a sample of American workers (where the
variable Earn represents annual labour earnings in US dollars in 2015; and where Height
represents the height of the worker in inches in 2015). The height of individuals in your
sample ranges from 48 inches to 84 inches.

You obtained the following regression output:

. regress Earn Height, robust

Linear regression Number of obs = 17870

F( 1, 17868) = 197.19
Prob > F = 0.0000
R-squared = 0.0109
Root MSE = 26777

Robust
Earn Coef. Std. Err. t P>|t| [95% Conf. Interval]

Height 707.6716 50.39502 14.04 0.000 608.8924 806.4507

_cons -512.7336 3379.864 -0.15 0.879 -7137.594 6112.126

For all questions below, provide your final answers to 2 decimal places.

(a) How much more or less do we expect a person whose height is 72 inches to earn in annual
earnings compared to a person whose height is only 62 inches?
(1 Mark)

(b) Suppose a person, Jane, actually earns $33,712.97 per week. Jane is 59 inches tall. How large
is the residual specific for Jane?
(1 Mark)

(c) Is the association between height and annual earnings statistically significant at the 5%
level? Explain.
(1 Mark)

(d) Construct a 90% confidence interval for the coefficient on Height.

(1 Mark)

(e) Construct a 97.5% confidence interval for the coefficient on Height.

(1 Mark)

(f) The mean height of workers in the sample is 65 inches. What is the mean earnings of
workers in the sample?
(1 Mark)

(g) Does height account for a large fraction of the variance in earnings across individuals?
Explain.

(1 Mark)

6
Answer:
(a) A person whose height is 72 inches is predicted to earn 707.6716(72 − 62) = $7,076.72
more in annual earnings than a person whose height is 62 inches.

(b) Jane’s predicted annual earnings is −512.7336 + 707.6716 × 59 = $41,239.89. The

residual specific for Jane is $33,712.97 - $41,239.89 = - $7,526.92.

(c) The absolute value of the t-statistic is 14.04, which is greater than 1.96 (the critical value
under a 5% 2 sided test). Hence the association between height and earnings is statistically
significant at the 1% level.

(d) The 90% confidence interval is (707.6716 ± 1.64(50.39502))= (625.02, 790.32) (some
rounding error is acceptable).

(e) The 97.5% confidence interval is (707.6716 ± 2.24(50.39502))= (594.79, 820.56) (some
rounding error is acceptable).

(f) -512.7336+707.6716(65) = $45,485.92

(g) No, the regression R-squared is 0.0109. This means that height explains only 1% of the
variation in earnings across individuals.

7
 5. The stata data file TeachingRatings.dta contains data on course evaluations, course
characteristics, and professor characteristics for 463 courses at the University of Texas at
Austin. A detailed description is given in TeachingRatings_Description. One of the
characteristics is an index of the professor’s “beauty” as rated by a panel of 6 judges. In this
exercise, you will investigate how course evaluations are related to professor’s beauty.

(a) Construct a scatterplot of average course evaluations (𝐶𝑜𝑢𝑟𝑠𝑒_𝐸𝑣𝑎𝑙) on the professor’s

beauty (𝐵𝑒𝑎𝑢𝑡𝑦). Does there appear to be a relationship between the variables? Attach the
scatterplot in your answer.
(2 Marks)

Answer:
5
4
course_eval

3
2

-2 -1 0 1 2
beauty

There appears to be a weak positive relationship between beauty and course evaluation.

8
(b) Run a regression of average course evaluations (𝐶𝑜𝑢𝑟𝑠𝑒_𝐸𝑣𝑎𝑙) on the professor’s beauty
(𝐵𝑒𝑎𝑢𝑡𝑦). Report your regression results using the format taught in lecture. Attach the
regression output in your answer.
(2 Marks)
. reg course_eval beauty, robust

Linear regression Number of obs = 463

F( 1, 461) = 16.94
Prob > F = 0.0000
R-squared = 0.0357
Root MSE = .54545

Robust
course_eval Coef. Std. Err. t P>|t| [95% Conf. Interval]

beauty .1330014 .0323189 4.12 0.000 .0694908 .1965121

_cons 3.998272 .0253493 157.73 0.000 3.948458 4.048087

Answer:

Estimated slope is 0.133 and estimated intercept is 3.998 (or 4.00). So the OLS regression
line is
̂
𝐶𝑜𝑢𝑟𝑠𝑒_𝐸𝑣𝑎𝑙 = 3.998 + 0.133 × 𝐵𝑒𝑎𝑢𝑡𝑦
Note: Here, I will accept both the heteroskedasticity-robust and homoskedasticity-only
output here. This is because we have not studied the differences between them yet. Do,
however, note that going forth, we will only be using heteroskedasticity-robust standard
errors (that is, in Stata, we include a “, robust” option next to the regress command).

(c) Professor Nick Young has an average value of 𝐵𝑒𝑎𝑢𝑡𝑦, while Professor Rachel Chu’s value of
𝐵𝑒𝑎𝑢𝑡𝑦 is two standard deviations above the average. Predict Professor Young’s and
Professor Chu’s course evaluations.
(2 Marks)

Answer:
. sum beauty

Variable Obs Mean Std. Dev. Min Max

beauty 463 4.75e-08 .7886477 -1.450494 1.970023

Professor Young’s value of 𝐵𝑒𝑎𝑢𝑡𝑦 is 0. Professor Chu’s 𝐵𝑒𝑎𝑢𝑡𝑦 is 2×0.789=1.577.

Therefore, Professor Stock’s predicted course evaluation is
̂
𝐶𝑜𝑢𝑟𝑠𝑒_𝐸𝑣𝑎𝑙 = 3.998 + 0.133 × 1.577 = 4.208
while Professor Watson’s predicted course evaluation is
̂
𝐶𝑜𝑢𝑟𝑠𝑒_𝐸𝑣𝑎𝑙 = 3.998 + 0.133 × 0 = 3.998

9
(d) Does Beauty explain a large fraction of the variance in evaluations across courses? Explain.
(2 Marks)

Answer:
No, Beauty explains only 3.6% of the variance in evaluations across courses. This is because
evaluations are influenced by other important factors as well (such as professors’ ability to
explain, patience, communication skills, etc).

10