Econ 140 - Spring 2016

Section 8
GSI: Fenella Carpena
March 17, 2016

Additional Exercises
Question 1. For each of the following functions, state whether it can be linearized. If yes, write the
resulting regression function in a form that can be estimated using OLS. If no, explain why.
β1 β2 X2 i
1. Yi = β0 X1i e
Xi
2. Yi = β0 +β1 Xi

eβ0 +β1 Xi
3. Yi = 1+eβ0 +β1 Xi

β1 β2
4. Yi = β0 X1i X2i + ui

Answer: (1) Yes, ln(Yi ) = ln(β0 ) + β1 ln(X1i ) + β2 X2i ; (2) Yes, (1/Yi ) = β0 (1/Xi ) + β1 ; (3)
Yes, ln(Yi /(1 − Yi )) = β0 + β1 Xi ; (4) No, it cannot be linearized due to the additive term ui .

Question 2. Consider the following two regressions of wages on age and gender.

\ = 323.70 + 5.15 · Age − 169.78 · F emale

Earn
(21.18) (0.55) (13.06)

R2 = 0.13, SER = 274.75

and

\ = 5.44 + 0.015 · Age − 0.421 · F emale

ln(Earn)
(0.08) (0.002) (0.036)

2
R = 0.17, SER = 0.75
where Earn is weekly earnings in dollars, Age is measured in years, and F emale is a dummy variable equal
to 1 if the individual is female, and 0 otherwise.
(a) Interpret each regression carefully. For a given age, how much less do females earn on average? Should
you choose the second specification on grounds of the higher regression R2 ?

Answer: The first regression (which has a linear specification) suggests that every additional
year in age is associated with $5.15 more weekly earnings, holding gender constant. In this
regression, women on average also earn $169.78 less wages for any given age. The intercept
has no useful interpretation since there are likely no observations who are male and have zero
age. The regression explains 13% of the variation in weekly earnings.

The second regression (which has a log-linear specification) suggests that for every additional
year in age, earnings increase by 1.5%. Women on average earn 42.1% less than men for
any given age. Again, the intercept has no useful interpretation since there are likely no

1
 observations who are male and have zero age. The regression explains 17% of the variation in
log earnings.

Even if the R2 in the second regression is higher, we should not choose the second specification
since the dependent variable in the two regressions is different, thus the R2 cannot be compared.

(b) Suppose that your professor points out to you that age and ln(earn) profiles typically take on an inverted
U-shape. How would you extend the previous regression to test this idea?

Answer: You can add Age2 in the regression, specifically, regress ln(Earn) on Age and Age2 .
Then, an inverted U-shape would mean that the coefficient on Age2 would be negative.

(c) Now, consider the regression where you add the square of age to your log-linear regression in part (a).

\ = 3.04 + 0.147 · Age − 0.42 · F emale − 0.0016 Age2

ln(Earn)
(0.18) (0.009) (0.033) (0.0001)

R2 = 0.28, SER = 0.68

Interpret the results from the above regression. Why is the Age coefficient here so large relative to its
value in the regression in part (a)?

Answer: The coefficient on the variable that was added, Age2 , is statistically significant and
has resulted in a substantial increase in the regression R2 . The increase in the Age coefficient is
due to the fact that earnings increase more initially than later in life or, mathematically speak-
ing, it compensates for the negative coefficient on Age2 , which lowers earnings as individuals
become older.

Question 3. Suppose you have data on weight, age, height and gender for 100 male and female children,
between the ages of 9 and 12, who are all at least 4 feet tall. Using this data, you estimate the following
relationship

eight = 45.59 + 4.32 · Height4

W\
(3.81) (0.46)

2
R = 0.55, SER = 15.69
where W eight is in pounds, and the Height4 variable is inches above 4 feet (so for a child who is 4 feet tall,
Height4 takes on the value 0, while for a child who is 4 feet and 5 inches tall, Height4 takes on the value 5).

(a) Interpret the results of this regression.

Answer: The average weight of children in the sample who are exactly 4 feet tall is 45.59.
For every inch above 4 feet, children in the sample gain roughly 4.32 pounds. The regression
explains 55 percent of the weight variation for children in the sample.

(b) You remember from the medical literature that females in the adult population are, on average, shorter
than males and weigh less. You also seem to have heard that females, controlling for height, are supposed
to weigh less than males. To see if this relationship holds for children, you add a binary variable (DFY)
that takes on the value one for girls and is zero otherwise. You estimate the following regression function:

eight = 36.27 + 17.33 · DF Y + 5.32 · Height4 − 1.83 · DF Y · Height4

W\
(5.99) (7.36) (0.80) (0.90)

2
R = 0.58, SER = 15.41
Are the signs on the new coefficients as expected? Are the new coefficients individually statistically
significant? Write down and sketch the regression function for boys and girls separately.

2
 Answer: The regression results show that on average, short girls weigh more than short
boys, and also that tall boys weight more than tall girls. If we think that the findings from
the medical literature for adults also applies to children, then these findings are perhaps
unexpected. The coefficient on DF Y is statistically significant at the 95% level, and the same
is true for the coefficient on DF Y · Height. The regression function for boys is W\ eight =
36.72 + 5.32 · Height4, and for girls it is W eight = 53.60 + 3.49 · Height4. So if you sketched
\
the regression line, the one for girls would have a higher intercept and flatter slope, relative to
the regression line for boys.
(c) Using the regression in part (b), state the hypothesis that the regression function is identical for boys
and girls. What test statistic would you use to this hypothesis?
Answer: If we write the population regression function as W eight = β0 +β1 DF Y +β2 Height4+
β3 DF Y · Height4 + u, the hypothesis that the regression function is the identical for boys and
girls can be written as H0 : β1 = β3 = 0. To test this hypothesis, we could use an F-test.
(d) Consider the regression in part (b) but now assume that in addition to testing whether the relationship
between height and weight changes by gender, you also wanted to test if the relationship between height
and weight changes by age. Briefly outline how you would specify the regression to test this relationship,
where the regression includes the gender binary variable (DF Y ) and an age binary variable (call it
Older) that takes on a value of one for eleven to twelve year olds and is zero otherwise. How would the
estimated relationship vary between the following four groups: younger girls, older girls, younger boys,
and older boys?
Answer: W eight = β0 + β1 DF Y + β2 Height4 + β3 DF Y · Height4 + β4 Older + β5 Older ·
Height4 + u. The estimates of W\
eight would vary as follows.
Younger Older
Boys β̂0 + β̂2 Height4 β̂0 + β̂4 + (β̂2 + β̂5 )Height4
Girls β̂0 + β̂1 + (β̂2 + β̂3 )Height4 β̂0 + β̂1 + β̂4 + (β̂2 + β̂3 + β̂5 )Height4

Question 4. Consider the scatterplot of y and x below. Explain what transformation you would use, and
what regression you would estimate to model this pattern. Can you think of two variables that might have
an economic relationship shaped like this?

Answer: Note that there is some curvature in the shape of the scatterplot; that is, if we were
to fit curve through these points, it would be steep for small values of x, but less steep for large
values of x, so we could transformthe x variable into ln(x). The sample regression would then
be yb = β0 + β1 · ln(x).

There are many economic relationships with this shape. For example, this shape might represent
the decreasing marginal product of labor in production, where x is labor and y is output (e.g,
if we have only 10 sewing machines in our clothing factory, then going from 9 to 10 workers is
more valuable to production than going from 100 to 101 workers).

3
Question 5. A regression of wage (hourly wage, measured in dollars per hour) and educ (years of schooling)
using data from a random sample of 526 American workers yields the following:

[ = −0.90 + 0.54 · educ

wage

(a) Interpret the intercept of this regression.

Answer: The intercept of -0.90 literally means that a person with 0 years of education has a
predicted hourly wage of -90 cents an hour.
Suppose that using ln(wage) instead as the response variable, we obtain the following regression:

\ = 0.584 + 0.083 · educ

ln(wage)

n = 526, R2 = 0.186

(b) Interpret the slope. Compare the interpretation of the slope in the two regressions when the response
variable is ln(wage) vs. wage.
Answer: To interpret the slope 0.083, we say that an additional year of education is associated
with a 8.3% increase in hourly wage. In part (a), the slope obtained was 0.54, which means
that each additional year of schooling is associated with an increase in hourly wage of 54 cents.
This 54 cent increase is the same whether it is for the 1st year of education or the 20th year
of education. In contrast, the regression above instead imposes a constant percentage effect of
education on wage.
(c) Interpret the R2 in the regression where ln(wage) is the dependent variable.
Answer: The R2 shows that the variable educ explains about 18.6% of the variation in
ln(wage) (NOT wage).