Tutorial week 3

Question 1
Consider the following probability distribution for a discrete random variable, X. What is
𝐸(4𝑋 ! − 3𝑋 + 5)?

X
1 4 2
f(x)
0.15 0.50 0.35

Answer:
𝐸(4𝑋 ! − 3𝑋 + 5) = 4𝐸(𝑋 ! ) − 3𝐸(𝑋) + 5

#
𝐸(𝑋) = , 𝑥" 𝑓(𝑥" ) = 1(0.15) + 4(0.5) + 2(0.35) = 2.85
"$%

#
𝐸(𝑋 ! ) = , 𝑥"! 𝑓(𝑥" ) = 1(0.15) + 16(0.5) + 4(0.35) = 9.55
"$%

𝐸(4𝑋 ! − 3𝑋 + 5) = 4𝐸(𝑋 ! ) − 3𝐸(𝑋) + 5 = 4(9.55) − 3(2.85) + 5 = 34.65

Question 2
The following joint pdf table represents the marks obtained (out of 10) by students (X) in a
class of an undergraduate econometrics course and their weekly hours of study (Y).

Y
3 5
5 0.02 0.18
X
7 0.18 0.14
8 0.98 0.40

What are the conditional probabilities P(Y=3|X=7) and P(X=5|Y=5)?

Answer:
P(Y=3|X=7)=f(x=7,y=3)/f(x=7)=0.18/0.32=0.5625
P(X=5|Y=5) = f(x=5,y=5)/f(y=5)=0.18/0.72=0.25

Question 3
You are organizing an outdoor concert for next week and believe attendance will depend on
the weather. You consider the following possibilities as appropriate:

Weather Probability=f(x) Attendance=X

Terrible weather 0.2 500
Mediocre weather 0.6 1000
Great weather 0.2 2000
 a) Let X denote the attendance. Why is X a random variable?
b) What is the expected attendance?
c) Suppose that each ticket costs $5 and that the total cost of giving the concert is a
fixed $2000. Let Y=profit=total sales revenue-total cost=5X-2000. What is the
expected profit?
d) If the variance of attendance is 𝜎&! = 240,000, find the variance of profit Y.

Answer:

Question 4
As you walk into your econometrics exam, a friend bets you $10 that she will outscore you
on the exam. Let X be a random variable denoting your winnings. X can take the values 10, 0
if there is a tie, or -10. You know that the probability distribution for X, f(x), depends on
whether she studied for the exam or not. Let Y=0 if she studied and Y=1 if she did not study.
Consider the following joint distribution table.

Y
f(x,y) 0 1 f(x)
-10 0.18 ? ?
X 0 0 ? 0.3
10 ? 0.45 ?
f(y) ? 0.75

a) Fill in the missing elements in the table.

b) Compute E(X). Should you take the bet?
c) What is the probability distribution of your winnings if you know that she did not
study?
d) Find your expected winnings given that she did not study.
Answer:

Question 5
Suppose that X and Y are random variables with expected values 𝜇& = 𝜇' = 𝜇 and variances
𝜎&! = 𝜎'! = 𝜎 ! . Let 𝑍 = (𝑋 + 𝑌)⁄2.

a) Find E(Z).
b) Find var(Z) assuming X and Y are statistically independent.
c) Find var(Z) assuming that 𝑐𝑜𝑣(𝑋, 𝑌) = 0.5𝜎 !
Answer:

Question 6
Let X take 4 values 𝑥% = 1, 𝑥! = 3, 𝑥# = 5, 𝑥( = 3.
a) Calculate the arithmetic average 𝑥̅ = ∑("$% 𝑥% ⁄4
b) Calculate ∑("$%(𝑥" − 𝑥̅ )
c) Calculate ∑("$%(𝑥" − 𝑥̅ )!
d) Calculate (∑("$% 𝑥"! ) − 4𝑥̅ !
e) Show algebraically that ∑)"$%(𝑥" − 𝑥̅ )! = (∑)"$% 𝑥"! ) − 𝑛𝑥̅ !

Answer:
Simple Linear regression

Question 5 Ch4
Review question 5 from the end of Chapter 4

SOLUTION
C2 p148

C2 iii) answer
C7

C7 Solution
Solution tutorial week 5

Question 1

(i) The constant elasticity equation is

∑salary) = 4.62 + .162 log(sales) + .107 log(mktval )

log(

n = 177, R 2 = .299.

(ii) We cannot include profits in logarithmic form because profits are negative for nine
of the companies in the sample. When we add it in levels form, we get

∑salary) = 4.69 + .161log(sales) + .098 log(mktval ) + .000036 profits

log(

n = 177, R2 = .299.

The coefficient on profits is very small. Here, profits are measured in millions, so if profits
increase by $1 billion, which means Dprofits = 1,000 – a huge change – predicted salary
increases by about only 3.6%. However, remember that we are holding sales and market
value fixed.
Together, these variables (and we could drop profits without losing anything) explain
almost 30% of the sample variation in log(salary). This is certainly not “most” of the
variation.

(iii) Adding ceoten to the equation gives

∑salary) = 4.56 + .162 log(sales) + .102 log(mktval ) + .000029 profits + .012ceoten

log(

n = 177, R 2 = .318.

This means that one more year as CEO increases predicted salary by about 1.2%.

(iv) The sample correlation between log(mktval) and profits is about .78, which is fairly
high. As we know, this causes no bias in the OLS estimators, although it can cause their
variances to be large. Given the fairly substantial correlation between market value and
firm profits, it is not too surprising that the latter adds nothing to explaining CEO salaries.
Also, profits is a short term measure of how the firm is doing, while mktval is based on past,
current, and expected future profitability.
Question 2

(i) Twenty students have received a perfect score for the course. The average score is
72.600. The mean and standard deviation of actmth are 23.211 and 3.773, respectively. The
mean and standard deviation of acteng are 22.595 and 3.789, respectively.

(ii) The estimated linear equation of score on colgpa, actmth, and acteng is:

𝑠𝑐𝑜𝑟𝑒
& = 27.4277 + 13.802𝑜𝑙𝑔𝑝𝑎 + 0.5382𝑎𝑐𝑡𝑚𝑡ℎ − 0.2586𝑎𝑐𝑡𝑒𝑛𝑔
𝑛 = 856 𝑅! = .357.

(iii) Since, the English ACT score has a negative effect on score, the math ACT score is a
better predictor of performance in the economic course.

(iv) From part (ii), we can say that about 36% of the variation in score is explained by college
grade point average, math ACT, and English ACT scores.
Solutions to review questions
1 (i) Yes. Because of budget constraints, it makes sense that, the more siblings there are in a
family, the less education any one child in the family has. To find the increase in the
number of siblings that reduces predicted education by one year, we solve 1 = .094(Δsibs),
so Δsibs = 1/.094 ≈ 10.6.

(ii) Holding sibs and feduc fixed, one more year of mother’s education implies .131 years
more of predicted education. So if a mother has four more years of education, her son is
predicted to have about a half a year (.524) more years of education.

(iii) Since the number of siblings is the same, but meduc and feduc are both different, the
coefficients on meduc and feduc both need to be accounted for. The predicted difference
in education between B and A is .131(4) + .210(4) = 1.364.

(iv) Given that the standard deviation of educ is given, we can calculate SST and SSR as

∑ ( educ − educ )
722 2

σˆ educ = i =1
= 2.237
722 − 1
SST
follows, Or , = 2.237 2
721
Or , SST = 3608.006

Now given that SST = 3608.006, SSE = R 2 * SST = .214*3606.006 = 772.113 .

Hence, SSR = SST – SSE = 2835.893

SSR 2835.893
σˆ 2 = = = 3.95
n − k − 1 722 − 3 − 1

2 (i) If adults trade off sleep for work, more work implies less sleep (other things equal),
so > 1 < 0.

(ii) The signs of > 2 and > 3 are not obvious, at least to us. One could argue that more educated
people like to get more out of life, and so, other things equal, they sleep less (> 2 < 0). The
relationship between sleeping and age is more complicated than this model suggests, and
economists are not in the best position to judge such things.

(iii) Since totwrk is in minutes, we must convert five hours into minutes:
Δtotwrk = 5(60) = 300.
Then sleep is predicted to fall by .148(300) = 44.4 minutes. For a week, 45 minutes less
sleep is not an overwhelming change.

(iv) More education implies less predicted time sleeping, but the effect is quite small. If we
assume the difference between college and high school is four years, the college graduate
sleeps about 45 minutes less per week, other things equal.

(v) Not surprisingly, the three explanatory variables explain only about 11.3% of the variation
in sleep. One important factor in the error term is general health. Another is marital
status, and whether the person has children. Health (however we measure that), marital

39 © Cengage Learning 2021. All rights reserved.

 status, and number and ages of children would generally be correlated with totwrk. (For
example, less healthy people would tend to work less.)

3 (i) it is expected that both > 1 and > 2would have positive signs, i.e. > 1 > 0, > 2 > 0. Obviously,
both ATAR and MAR are measures of the quality of the entering class and hence brighter
students are expected to earn more, on average. The number of volumes in the law library
and the tuition cost are both measures of school quality. Higher values of these variables
entail better quality of the law schools and hence better pay of their graduates. We
therefore expect > 3, > 4 > 0 . Note that cost is less obvious than library volumes, but
should reflect quality of the faculty, physical plant, and so on. Finally, a larger rank of any
school entails lower prestige and recognition of that school. So it is expected that the
higher is the school rank, the lower is the salary of the graduates.
Therefore, > 5 < 0.

(ii) Other things remaining the same, as the median MAR changes by 1 unit, the median salary
changes by 0.99*100= .99% or approximately 1%.

(iii) This is an elasticity: a one percent increase in library volumes implies a .095% increase in
predicted median starting salary, other things being equal.

(iv) It is definitely better to attend a law school with a higher rank. If law school A has a
ranking 20 less than law school B, the predicted difference in starting salary is
100(.0033)(20) = 6.6% higher for law school A.

4 (i) No, it does not make sense to hold sleep, work, and leisure fixed, while changing study.
The reason is, by definition, study + sleep + work + leisure = 168. Therefore, if we change
study, we must change at least one of the other categories so that the sum is still 168.

(ii) Assumption MLR.3 entails that there should be no perfect collinearity among the
independent variables. Now, from part (i), we can write, say, study as a perfect linear
function of the other independent variables: study = 168 − sleep − work − leisure. This
holds for every observation, and so MLR.3 is violated.

(iii) To reformulate the model, we can simply drop one of the independent variables, say
leisure:

metrc = β 0 + β1 study + β 2 sleep + β 3 work + u.

Now, for example, > 1 is interpreted as the change in metrc when study increases by one
hour, where sleep, work, and u are all held fixed. In this case, as we are holding sleep and
work fixed but increasing study by one hour, then we must be reducing leisure by one
hour. The other slope parameters have a similar interpretation.

40 © Cengage Learning 2021. All rights reserved.

5 (i) Conditioning on the outcomes of the explanatory variables, we have

E(θ! 1 ) = E( β̂1 + βˆ2 ) = E( β̂1 ) + E( βˆ2 ) = β1 + β2 = θ1 .

Hence Z 1 is an unbiased estimator of Z 1.

(ii) ( )
( ) ( )
Var(θˆ1 ) = Var βˆ1 + βˆ2 = Var βˆ1 + Var βˆ2 + 2Cov βˆ1 , βˆ2 ( )
= Var ( βˆ ) + Var ( βˆ ) + 2Corr ( βˆ , βˆ )σ σ
1 2 1 2 βˆ1 βˆ2

where σ βˆ and σ β̂ are the standard error of estimate of > 1 and > 2, respectively.
1 2

6 (i) We expect a negative impact of pollution on house prices. In other words, more pollution
can be expected to lower housing values and so we expect > 1 < 0. Note that > 1 is the
elasticity of price with respect to nox. The parameter > 2 should be positive because
number of rooms is often used to measure the size of the house. (However, it does not
allow us to distinguish homes where each room is large from homes where each room is
small.)

(ii) If we assume that rooms increases with quality of the home, then log(nox) and rooms are
negatively correlated when poorer neighbourhoods have more pollution, something that
is often true.

(iii) The estimate of > 1 is –.824, which implies that if nox increases by 1%, the price of the
house falls by .824%. The intercept is estimated as 6.15, which theoretically indicates that
for any community with zero average number of bedrooms in houses (room = 0) and with
pollution level of 1 unit, that is nox = 1, the predicted median price of the house is
log(price) = 6.15 or 468 717.38 dollars (apx). This makes sense, as prices are higher with
low levels of pollution and there is always a price for blocks of land.

(iv) The predicted median house prices are as follows:

Pre-renovation:
log(price) = 6.15 – .824 log(5.24) + .376(2) = 5.53719 or exp(5.53719) = $253 963.63.

Post-renovation:
log(price) = 6.15 – .824 log(5.24) + .376(4) = 6.28919 or exp(6.28919) = $538 717.38.

Hence, after the renovation, the median house price has increased by 538,717.38 –
253,963.63 = 284,753.74 dollars.

Note that this increase in price is due to the increase in log(price) by

∆log(price) = (6.28917 – 5.53719 = .752 or 75.2%, as expected based on the estimated
model.

7 (i) Because x1 is highly correlated with x2 and x3, and these latter variables have large partial
effects on y, the simple and multiple regression coefficients on x1 can differ by large
amounts.

(ii) Here we would expect > 1 and > 1 to be similar (subject, of course, to what we mean by
‘almost uncorrelated’). The amount of correlation between x2 and x3 does not directly
affect the multiple regression estimate on x1 if x1 is essentially uncorrelated with x2 and x3.

41 © Cengage Learning 2021. All rights reserved.

 (iii) The required calculations are shown in the table below:

(y − y) uˆi2 = ( yi − yˆi )
2
2
yˆi
(2–2)2 = 0 1.3929 (0.6071)2
(1–2)2= 1 2.0714 (–1.0714)2
(3–2)2= 1 3.1071 (–0.1071)2
(1–2)2= 1 2.5714 (–1.5714)2
(2–2)2= 0 1.4286 (0.5714)2
(0–2)2= 4 .8215 (–0.8214)2
(3–2)2= 1 1.9286 (1.0714)2
(4–2)2= 4 2.6785 (1.3214)2
8 8

∑( y − y ) ∑ uˆ
2 2
= 12 i = 7.893
i =1 i =1

Hence SST = 12 and SSR = 7.893, as shown in the above table.

SSR 7.893
σˆ 2 = = = 1.5786
n − k −1 8 − 2 −1
Standard error of regression σˆ 2 = σˆ = 1.5786 = 1.2564
SSR 7.893
R2 = 1 − = 1− = .3423
SST 12
The value of R2 shows that about 34% of the variation of the dependent variable y is
explained by the variations of the independent variables, which seems somewhat
reasonable.

9 (i) The shares, by definition, add to one. If we do not omit one of the shares then the
equation would suffer from perfect multicollinearity. The parameters would not have a
ceteris paribus interpretation, as it is impossible to change one share while holding all of
the other shares fixed.

(ii) Because each share is a proportion (and can be at most one, when all other shares are
zero), it makes little sense to increase sharep by one unit. If sharep increases by .01 – which
is equivalent to a one percentage point increase in the share of property taxes in total
revenue – holding shareI, shareS, and the other factors fixed, then growth increases by
> 1(.01). With the other shares fixed, the excluded share, shareF, must fall by .01 when
sharep increases by .01.

10 An important fact about R2 is that it never decreases, and it usually increases, when another
independent variable is added to a regression and the same set of observations is used for both
regressions. However, if two regressions use different sets of observations, we generally cannot
tell how the R-squareds will compare. It could go either way.
Here, in the two equations, the sets of observations are different (i.e. n = 142 for the first
equation and n = 99 for the second equation). Therefore, R-squared is smaller when the
variable age is added to the second equation.

44 © Cengage Learning 2021. All rights reserved.

Computer questions
C1 (i) Probably > 2 > 0, as more income typically means better nutrition for the mother and
better antenatal care.

(ii) On the one hand, an increase in income generally increases the consumption of a food,
and cigs and faminc could be positively correlated. On the other, family incomes are also
higher for families with more education, and more education and cigarette smoking tend
to be negatively correlated. The sample correlation between cigs and faminc is about
−.173, indicating a negative correlation.

(iii) The regressions without and with faminc are

bwght = 3.395 – .015 cigs
n = 1388, R2 = .023

and
bwght = 3.316 – .013 cigs + .003 faminc
n = 1388, R2 = .030
The effect of cigarette smoking is slightly smaller when faminc is added to the regression,
but the difference is not great.

C2 (i) The estimated equation is

price = –19.32 + .128 sqrft + 15.20 bdrms
n = 88, R2 = .632.

(ii) Holding square footage constant, ∆price = 15.20 ∆bdrms and so price increases by 15.20,
which means $15 200.

(iii) Now ∆price = .128 ∆sqrft + 15.20 ∆bdrms = .128(140) + 15.20 = 33.12, or $33 120. Because
the size of the house is increasing, this is a much larger effect than in (ii).

C3 (i) Probably > 2> 0, as expect that the larger the size of the university the greater the
remuneration holding the rank of the university constant.

(ii) The regressions without and with studnum are

remuneration = 1032.55 – 0.687rank
n = 37, R2 = 0.335, A 2 = 0.316
remuneration = 950.19 – 0.668rank + 0.00198studnum
n = 37, R2 = 0.521, A 2 = 0.493
The effect of rank is basically the same when studnum is added to the regression.
However, there is a large increase in the R-squared.

(iii) Now re-estimate the equation but use the log of each variable. That is, estimate the model
log(remuneration) = 5.915 – 0.138log(rank) + 0.148log(studnum)
n = 37, R2 = 0.417, A 2 = 0.382
The estimated elasticity of remuneration with respect to studnum is the estimated
coefficient on log(studnum), so a 1% increase in studnum will result in a .148% increase in
remuneration, other things equal.

47 © Cengage Learning 2021. All rights reserved.

C14 (i) turnout = 50.86 − 0.043rain + 0.293turnout15 + 0.253ukip14 _ pct − 0.08 femperc
(7.75) (0.022) (0.028) (0.014) (0.164)
− 0.48logpop + 0.116 percdegree + 0.138medianage − 0.322 percde
(0.111) (0.028) (0.046) (0.03)
n=398, R 2 =.883

⎛ −0.042627 ⎞
(ii) 100 × ⎜ ⎟ = −0.06%
⎝ 73.2 ⎠
This is around a fall of 0.06% in turnout. This does not seem to be a substantive effect on
the referendum.

(iii) Turnout vote is higher with higher median age holding other factors constant. It is lower as
the percentage of residents with lower social grade increases, holding other factors
constant.

(iv) leave = 99.98 − 0.042rain − 0.176turnout15 + 0.498ukip14 _ pct − 1.063 femperc

12.85) (0.036) (0.047) (0.022) (0.272)
− 0.017logpop − 0.529 percdegree + 0.345medianage − 0.354 percde
(0.185) (0.046) (0.076) (0.051)
n=398, R 2 =.864

(v) Brexit support was higher in districts with greater proportion of older voters. Support for
Ukip in the 2014 parliament elections increased leave vote share. The percentage of
higher education decreases leave vote share, as does the percentage of females. The
higher the percentage of lower social grade, the lower the leave vote share.

C15 (i) The variable educ ranges from 6 to 20. Out of 1230 men, 512, or 41.63% of the sample,
completed 12th grade, but no higher. The average years of education for the men in the
sample is about 13.04, which is higher than the average of motheduc (12.18) and fatheduc
(12.45).

(ii) The regression results are

ijEN = 6.96 + .304 lO1ℎijEN + .190 no1ℎijEN
5 = 1230 A 9 = .249
About 25% of the variation in educ is explained by parents’ education. The coefficient on
motheduc means that, holding father’s education fixed, another year of mother’s
education is associated with about .304 additional years of education for the child, or
slightly less than one-third of a year.

(iii) When abil is added to the regression we get

ijEN = 8.45 + .189 lO1ℎijEN + .111 no1ℎijEN + .502 opqP
5 = 1230 A 9 = .428
The three explanatory variables together explain almost 43% of the variation in educ. This
is much more than in part (ii); clearly abil is helping.

56 © Cengage Learning 2021. All rights reserved.

 (iv) When abil2 is added to the regression, the estimated equation is
ijEN = 8.24 + .190 lO1ℎijEN + .109 no1ℎijEN + .401 opqP + .051 opqP 9
5 = 1230 A 9 = .444
The derivative with respect to abil is .401 + .102 abil. Setting equal to zero and solving
gives
. 401
opqP ∗ = − ≈ −3.93
. 102
so about −4. The second derivative is .102, and so we know we have found the global
minimum.

(v) Out of 1230 men, only 15 have abil <–3.93, or only about 1.2 percent of the sample. This is
reassuring because it means we can effectively ignore what is happening to the left of
–3.93. The important story is that the level of education increases with ability at an
increasing rate.

(vi) We used the equation from part (iv) and plugged in the mean values for motheduc and
fatheduc. Thus, we used the equation
ijEN = 8.24 + .190 12.18 + .109 12.45 + .401 opqP + .051 opqP 9
which has an intercept of about 11.9. We generated 2000 values of abil, equally spaced
between −5 and 6, to generate the following graph:
16
educhat

15
14
13
12
11

-5 -4 -3 -2 -1 0 1 2 3 4 5
abil

Its minimum is at roughly −4, as the calculation in part (iv). The slope of the function is
increasing.

C16 (i) Regressing colGPA on PC yields YOPQR@ = 2.989 + 0.170RY. Students without a PC have
(&.&U&) (&.&VW)
on average a 2.989 GPA, while students who own a PC have college GPAs on average
0.17 points higher.

57 © Cengage Learning 2021. All rights reserved.

Solutions to review questions
1 (i) Yes. Because of budget constraints, it makes sense that, the more siblings there are in a
family, the less education any one child in the family has. To find the increase in the
number of siblings that reduces predicted education by one year, we solve 1 = .094(Δsibs),
so Δsibs = 1/.094 ≈ 10.6.

(ii) Holding sibs and feduc fixed, one more year of mother’s education implies .131 years
more of predicted education. So if a mother has four more years of education, her son is
predicted to have about a half a year (.524) more years of education.

(iii) Since the number of siblings is the same, but meduc and feduc are both different, the
coefficients on meduc and feduc both need to be accounted for. The predicted difference
in education between B and A is .131(4) + .210(4) = 1.364.

(iv) Given that the standard deviation of educ is given, we can calculate SST and SSR as

∑ ( educ − educ )
722 2

σˆ educ = i =1
= 2.237
722 − 1
SST
follows, Or , = 2.237 2
721
Or , SST = 3608.006

Now given that SST = 3608.006, SSE = R 2 * SST = .214*3606.006 = 772.113 .

Hence, SSR = SST – SSE = 2835.893

SSR 2835.893
σˆ 2 = = = 3.95
n − k − 1 722 − 3 − 1

2 (i) If adults trade off sleep for work, more work implies less sleep (other things equal),
so > 1 < 0.

(ii) The signs of > 2 and > 3 are not obvious, at least to us. One could argue that more educated
people like to get more out of life, and so, other things equal, they sleep less (> 2 < 0). The
relationship between sleeping and age is more complicated than this model suggests, and
economists are not in the best position to judge such things.

(iii) Since totwrk is in minutes, we must convert five hours into minutes:
Δtotwrk = 5(60) = 300.
Then sleep is predicted to fall by .148(300) = 44.4 minutes. For a week, 45 minutes less
sleep is not an overwhelming change.

(iv) More education implies less predicted time sleeping, but the effect is quite small. If we
assume the difference between college and high school is four years, the college graduate
sleeps about 45 minutes less per week, other things equal.

(v) Not surprisingly, the three explanatory variables explain only about 11.3% of the variation
in sleep. One important factor in the error term is general health. Another is marital
status, and whether the person has children. Health (however we measure that), marital

39 © Cengage Learning 2021. All rights reserved.

 (iii) The required calculations are shown in the table below:

(y − y) uˆi2 = ( yi − yˆi )
2
2
yˆi
(2–2)2 = 0 1.3929 (0.6071)2
(1–2)2= 1 2.0714 (–1.0714)2
(3–2)2= 1 3.1071 (–0.1071)2
(1–2)2= 1 2.5714 (–1.5714)2
(2–2)2= 0 1.4286 (0.5714)2
(0–2)2= 4 .8215 (–0.8214)2
(3–2)2= 1 1.9286 (1.0714)2
(4–2)2= 4 2.6785 (1.3214)2
8 8

∑( y − y ) ∑ uˆ
2 2
= 12 i = 7.893
i =1 i =1

Hence SST = 12 and SSR = 7.893, as shown in the above table.

SSR 7.893
σˆ 2 = = = 1.5786
n − k −1 8 − 2 −1
Standard error of regression σˆ 2 = σˆ = 1.5786 = 1.2564
SSR 7.893
R2 = 1 − = 1− = .3423
SST 12
The value of R2 shows that about 34% of the variation of the dependent variable y is
explained by the variations of the independent variables, which seems somewhat
reasonable.

9 (i) The shares, by definition, add to one. If we do not omit one of the shares then the
equation would suffer from perfect multicollinearity. The parameters would not have a
ceteris paribus interpretation, as it is impossible to change one share while holding all of
the other shares fixed.

(ii) Because each share is a proportion (and can be at most one, when all other shares are
zero), it makes little sense to increase sharep by one unit. If sharep increases by .01 – which
is equivalent to a one percentage point increase in the share of property taxes in total
revenue – holding shareI, shareS, and the other factors fixed, then growth increases by
𝛽 1(.01). With the other shares fixed, the excluded share, shareF, must fall by .01 when
sharep increases by .01.

10 An important fact about R2 is that it never decreases, and it usually increases, when another
independent variable is added to a regression and the same set of observations is used for both
regressions. However, if two regressions use different sets of observations, we generally cannot
tell how the R-squareds will compare. It could go either way.
Here, in the two equations, the sets of observations are different (i.e. n = 142 for the first
equation and n = 99 for the second equation). Therefore, R-squared is smaller when the
variable age is added to the second equation.

44 © Cengage Learning 2021. All rights reserved.

11 In this example, each additional year of education is associated with one less year of
experience. Thus, when considering the effect of an additional year of education on wages, we
must also consider the effect of one less year of experience. As education increases by one
year, log wage rises by 0.094, but then experience falling by one year will reduce log wage by
0.026. The net effect is a 0.068 increase in log wage. As this is a semi-log model, we estimate
that an additional year of education will increase wages by about 6.8%.

12 (i) Since the policy variable is independent of the error term 𝑣D , each observation 𝑦D can be
expressed as a function of the policy variable and error term. Let 𝑦D = 𝛼 + 𝛽𝑤D + 𝑣D . This
framework captures different levels of the policy variable.
For example, for 𝑦 0 = 𝛼 + 𝛽 ∗ 0 + 𝑣 0 = 𝛼 + 𝑣(0). For 𝑦 1 = 𝛼 + 𝛽 + 𝑣(1).

(ii) We can estimate 𝛼 and 𝛽 by regressing 𝑦D on 𝑤D . We can do this since 𝑤D is independent

of 𝑣D , therefore satisfying MLR.4 given that 𝐸 𝑣D 𝑤D = 0.

(iii) We are given 𝐸 𝑣D 𝑤D , 𝑥D/ , … , 𝑥Da = 𝐸 𝑣D 𝑥D/ , … , 𝑥Da = 𝜂 + 𝛾/ 𝑥D/ + ⋯ + 𝛾a 𝑥Da . If we

define 𝑢D = 𝑣D − (𝜂 + 𝛾/ 𝑥D/ + ⋯ + 𝛾a 𝑥Da ), then 𝐸 𝑢D 𝑤D , 𝑥D/ , … , 𝑥Da = 0. Therefore, we
can run a regression of 𝑦D = 𝜓 + 𝛽𝑤D + 𝛾/ 𝑥D/ + ⋯ + 𝛾a 𝑥Da + 𝑢D where 𝜓 = 𝛼 + 𝜂.

(iv) We can estimate 𝛽, 𝜓, and 𝛾f by regressing 𝑦D on 𝑤D , 𝑥D/ , … , 𝑥Da . This wil yield unbiased
estimates since the error term has a zero conditional mean: 𝐸 𝑢D 𝑤D , 𝑥D/ , … , 𝑥Da = 0.

n
13 (i) For notational simplicity, define szx = ∑ ( z − z ) x ; this is not quite the sample covariance
i =1
i i

between z and x because we do not divide by n – 1, but we are only using it to simplify
notation. Then we can write 𝛽 1 as
n

∑ (z − z ) y i i
β!1 = i =1
.
szx
This is clearly a linear function of the yi: take the weights to be wi = (zi − z )/szx. To show
unbiasedness, as usual we plug yi = 𝛽 0 + 𝛽 1xi + ui into this equation, and simplify:
n

∑ ( z − z )( β i 0 + β1 xi + ui )
β!1 = i =1

szx
n n
β 0 ∑ ( zi − z ) + β1szx + ∑ ( zi − z )ui
= i =1 i =1

szx
n

∑ ( z − z )u
i i
= β1 + i =1

szx
n
where we use the fact that ∑ ( z − z ) = 0 always. Now szx is a function of the zi and xi and the
i =1
i

expected value of each ui is zero conditional on all zi and xi in the sample. Therefore,
conditional on these values,
n

∑ (z i − z )E(ui )
E( β!1 ) = β1 + i =1
= β1
szx

because E(ui) = 0 for all i.

45 © Cengage Learning 2021. All rights reserved.

Computer questions
C1 (i) Probably 𝛽 2 > 0, as more income typically means better nutrition for the mother and
better antenatal care.

(ii) On the one hand, an increase in income generally increases the consumption of a food,
and cigs and faminc could be positively correlated. On the other, family incomes are also
higher for families with more education, and more education and cigarette smoking tend
to be negatively correlated. The sample correlation between cigs and faminc is about
−.173, indicating a negative correlation.

(iii) The regressions without and with faminc are

bwght = 3.395 – .015 cigs
n = 1388, R2 = .023

and
bwght = 3.316 – .013 cigs + .003 faminc
n = 1388, R2 = .030
The effect of cigarette smoking is slightly smaller when faminc is added to the regression,
but the difference is not great.

C2 (i) The estimated equation is

price = –19.32 + .128 sqrft + 15.20 bdrms
n = 88, R2 = .632.

(ii) Holding square footage constant, ∆price = 15.20 ∆bdrms and so price increases by 15.20,
which means $15 200.

(iii) Now ∆price = .128 ∆sqrft + 15.20 ∆bdrms = .128(140) + 15.20 = 33.12, or $33 120. Because
the size of the house is increasing, this is a much larger effect than in (ii).

C3 (i) Probably 𝛽 2> 0, as expect that the larger the size of the university the greater the
remuneration holding the rank of the university constant.

(ii) The regressions without and with studnum are

remuneration = 1032.55 – 0.687rank
n = 37, R2 = 0.335, 𝑅 2 = 0.316
remuneration = 950.19 – 0.668rank + 0.00198studnum
n = 37, R2 = 0.521, 𝑅 2 = 0.493
The effect of rank is basically the same when studnum is added to the regression.
However, there is a large increase in the R-squared.

(iii) Now re-estimate the equation but use the log of each variable. That is, estimate the model
log(remuneration) = 5.915 – 0.138log(rank) + 0.148log(studnum)
n = 37, R2 = 0.417, 𝑅 2 = 0.382
The estimated elasticity of remuneration with respect to studnum is the estimated
coefficient on log(studnum), so a 1% increase in studnum will result in a .148% increase in
remuneration, other things equal.

47 © Cengage Learning 2021. All rights reserved.

 Introductory econometrics

Q0.10Question 4
1. No, it does not make sense to hold sleep, work, and leisure fixed, while
changing study. The reason is, by definition, study + sleep + work +
leisure = 168. Therefore, if we change study, we must change at least
one of the other categories so that the sum is still 168.

2. Assumption MLR.3 entails that there should be no perfect collinearity

among the independent variables. Now, from part (i), we can write, say,
study as a perfect linear function of the other independent variables:
study = 168 − sleep − work − leisure. This holds for every observation,
and so MLR.3 is violated

3. To reformulate the model, we can simply drop one of the independent

variables, say leisure:

metrc = β0 + β1 study + β2 sleep + β3 work + u. (1)

Now, for example, β1 is interpreted as the change in metrc when study

increases by one hour, where sleep, work, and u are all held fixed. In
this case, as we are holding sleep and work fixed but increasing study
by one hour, then we must be reducing leisure by one hour. The other
slope parameters have a similar interpretation.

Q0.20Question 5
1. Conditioning on X and using properties of expected value
( ) ( ) ( ) ( )
E|X θ̂1 = E β̂1 + β̂2 = E β̂1 + E β̂2 = β1 + β2 = θ1 . (2)

Ergo, θ̂1 is an unbiased estimator of the population parameter θ1 .

2.
( ) ( ) ( ) ( ) ( )
Var θ̂1 = Var β̂1 + β̂2 = Var β̂1 + Var β̂2 + 2 Cov β̂1 , β̂2
( ) ( ) ( )
= Var β̂1 + Var β̂2 + 2 Corr β̂1 , β̂2 σβ̂1 σβ̂2
(3)
where σβˆ1 and σβˆ2 are the standard error of estimate of β1 and βˆ1
ˆ
respectively.

Emiliano A. Carlevaro Page 1

 Introductory econometrics

Q0.30Question 7
1. Because x1 is highly correlated with x2 and x3 , and these latter vari-
ables have large partial effects on y, the simple and multiple regression
coefficients on x1 can differ by large amounts.

2. Here we would expect β˜1 and βˆ1 to be similar (subject, of course, to

what we mean by ‘almost uncorrelated’). The amount of correlation
between x2 and x3 does not directly affect the multiple regression esti-
mate x1 if x1 is essentially uncorrelated with other covariates.

3. In this case we are (unnecessarily) introducing multicollinearity into

the regression: x2 and x3 have small partial effects on y and yet x2
and x3 are highly correlated with x1. Adding x2 and x3 increases the
standard error of the coefficient on x1 substantially, so se(βˆ1 ) is likely
to be much larger than se(β̃).

4. In this case, adding x2 and x3 will decrease the residual variance with-
out causing much collinearity (because x1 is almost uncorrelated with
x2 and x3), so we should see se(βˆ1 ) smaller than se(β̃). The amount of
correlation between x2 and x3 does not directly affect se(βˆ1 ).

Emiliano A. Carlevaro Page 2

Solution to Tutorial week 9 on
Chapter 6 Inference

Q0.10Question 1
 
β̂ ∼ N ormal β, σβ̂2

but because we don’t know σβ̂2 but we estimate it, we resort to β̂ ∼ T -Student.

H0 : βcomm = 0 (1)
H1 : βcomm < 0 (2)

α = 0.05, tcritical = 1.65, dgf = 163

−0.0726 − 0
t∗ =
.059
We do not reject H0 at 0.05 significance level.

Q0.20Question 2
1.

H0 : β3 = θ (3)
H1 : β3 > θ (4)

2.

∆ ros = 50 (5)
∆ salary % ∼
= (.00024 × 100)∆ ros (6)
∆ salary % ∼
= 1.02% (7)

1
 ECON2515 & 7242 Introductory econometrics

H0 : β3 = 0 (8)
H1 : β3 > 0 (9)

tcritical = 1.28 at 10% significance level with df = 205. One can also use
zcritical = 1.28 instead of t-Student.

0.00021 − θ
t∗ = = 0.44
.00054
We cannot reject H0 . There is not enough evidence to reject it.
Based on this sample, the estimated ros coefficient appears to be different
from zero only because of sampling variation. On the other hand, including
ros may not be causing any harm; it depends on how correlated it is with
the other independent variables (although these are very significant even with
ros in the equation).

Q0.30Question 9
y=βb + β1 x1 + β2 x2 + β3 x3 + u

 
Var β̂1 − 3β̂2 = (10)
     
= Var β̂1 + Var −3β̂2 + 2 Cov β̂1 , −3β̂2
(11)
   
= Var β̂1 + (−3)2 Var β̂2 + 2(1)(−3) Cov (12)
     
= Var β̂1 + 9 Var β̂2 − 6 Cov β̂1 , β̂2 (13)

β̂1 −3β̂2 −1
t∗ = q
Var(β̂1 )+9 Var(β̂2 )−6(Cov(β̂1 ,β̂2 )

The model is
y = β0 + β1 x1 + β2 x2 + β3 x3 + u (14)
and the restriction we want to test is

θ = β1 − 3β2 (15)

Emiliano A. Carlevaro Page 2

 ECON2515 & 7242 Introductory econometrics

The goal is to combine both equations to have a new model. The new
model will have a new variable whose coefficient, surprisepsurprise, is
θ. Then, the software will give you the standard error (= ( Var(θ̂)).
Let’s see.

θ + 3β2 = β1
then replace in the model, and rearrange
y = β0 + (θ + 3β2 ) x1 + β2 x2 + β3 x3 + u (16)
β0 + θx1 + 3β2 x1 + β2 x2 + β3 x3 + u (17)
y = β0 + θx1 +β2 (3x1 + x2 ) +β3 x3 + u. (18)
|{z} | {z }
This is your guy x′2

As it is clear from this last equation the new model has new regressors.
We are interested in the coefficient and standard error for regresor x1 .

Q0.40Question 10
1. With df = 706 – 4 = 702, we use the standard normal critical
value (df = ∞), which is 1.96 for a two-tailed test at the 5% level.
Now t⋆educ = 11.13/5.88 ≈ 1.89, so |t⋆educ | = 1.89 < 1.96, and we
fail to reject H0: βeduc = 0 at the 5% level. Also, t⋆age ≈ 1.52, so
age is also statistically insignificant at the 5% level.
2. We need to compute the R-squared form of the F-statistic for
joint significance. But F = [(.113 − .103)/(1 − .113)](702/2)
≈ 3.96. The 5% critical value in the F2702 distribution can be
obtained from Table G.3b with denominator df = ∞: cv = 3.00.
Therefore, educ and age are jointly significant at the 5% level
(3.96 > 3.00). In fact, the p-value is about .019, and so educ and
age are jointly significant at the 2% level.
3. Not really. These variables are jointly significant, but including
them only changes the coefficient on totwrk from –.151 to –.148.
4. The standard t and F-statistics that we used assume homoscedas-
ticity, in addition to the other CLM assumptions. If there is het-
eroscedasticity in the equation, the tests are no longer valid. We
will have to correct them for heteroskedasticity.

Emiliano A. Carlevaro Page 3

Solution to Tutorial week 10 on
Chapter 6 Inference

Exercises from Wooldridge, J. M. (2020) et al. [Chapter 6]. In Introductory

Econometrics: A Modern Approach (Asia Pacific Edition, pp.)

Question 12
1. In columns (2) and (3), the coefficient on profmarg is actually negative,
although its t-statistic is only about –1. It appears that, once firm sales
and market value have been controlled for, profit margin has no effect
on CEO salary.

2. We use column (3), which controls for the most factors affecting salary.
The t-statistic on log(mktval) is about 2.05, which is just significant at
the 5% level against a two-sided alternative. (We can use the standard
normal critical value, 1.96.) So log(mktval) is statistically significant.
Because the coefficient is an elasticity, a ceteris paribus 10% increase
in market value is predicted to increase salary by 1%. This is not a
huge effect, but it is not negligible, either.

3. These variables are individually significant at low significance levels,

with tceoten � 3.11 and tcomten � –2.79. Other factors fixed, another year
as CEO with the company increases salary by about 1.71%. On the
other hand, another year with the company, but not as CEO, lowers
salary by about 0.92%. This second finding at first seems surprising,
but could be related to the ‘superstar’ effect: firms that hire CEOs from
outside the company often go after a small pool of highly regarded
candidates, and salaries of these people are bid up. More non-CEO
years with a company makes it less likely the person was hired as an
outside superstar.

1
 ECON2515 & 7242 Introductory econometrics

4. Holding other factors fixed and given the estimated coefficient of the
variable comten, which is statistically significant and shows an adverse
impact on CEO salary (as discussed in part (iii) above), it is reasonable
to conclude that longer tenure with a company is associated with lower
CEO salary. However, if one would like to examine the effect of longer
tenure on employees’ salary in general, then we cannot conclude any-
thing based on the given estimates, since in that case, the dependent
variable would have to represent salary of employees, instead of that
of the CEO only.

Question C5
1. With n=70, the R2 = 0.623 and SSR=9.016.
2. The coefficient on log(EGM) is an elasticity. A correct statement is
that ‘a 1% increase in the number of EGMs is associated with about a
0.77% increase in pokies expenditure per adult.
3. H0 : β1 = 0.7 vs H1 : β1 ̸= 0.7 t = 0.7249. H0 : β2 = 0.7 vs H1 :
β2 ̸= 0.7, t = −1.0126. The two-sided critical value at the 10% level
of significance is 1.668± . Hence for both tests we can’t reject the null
hypothesis at the 10% level of significance.
4. The restricted model is
log(exp_per_adult) = 3.6923 + 0.6944{log(EGM) + log(unemployed)}
(1)
(0.2268) (0.0664) (2)
2
n = 70, R = 0.617, SSR = 9.175 (3)

5. H0 : β1 = β2 H1 : β1 ̸= β2 .
(SSRr − SSRur ) /q (9.175 − 9.0163)/1
F ≡ = = 1.179
SSRur /(n − k − 1) 9.0163/67
The 5% critical value for an F with 1 and 67 df is around 3.98. We
can’t reject the null hypothesis at the 5% level of significance.

Question C6
1. In the model
log( wage ) = β0 + β1 educ + β2 exper + β3 tenure + u

Emiliano A. Carlevaro Page 2

 ECON2515 & 7242 Introductory econometrics

The null hypothesis of interest is H0 : β2 = β3 .

2. Let θ2 = β2 − β3 . Then we can estimate the equation

log( wage ) = β0 + β1 educ + θ2 exper + β3 ( exper + tenure ) + u

to obtain the 95% CI for θ2 . This turns out to be about .0020 ±

1.96(.0047), or about -.0072 to .0112. Because zero is in this CI, θ2
is not statistically different from zero at the 5% level, and we fail to
reject H0 : β2 = β3 at the 5% level.

Emiliano A. Carlevaro Page 3