Question 1

A
Eviews:

OLS regression:
^
lnearnings=5.443+ 0.092∗education+ 0.078∗dgovt+1.067∗ln hours
se: (0.710) (0.010) (0.066) (0.188)
t-statistic: (7.666) (9.081) (1.189) (5.689)
p-value: (0.000) (0.000) (0.235) (0.000)

R2 = 0.246
n = 494

B
 ^
β 0 = 5.443  if education, dgovt and lnhours are all equal to 0, the predicted value of
lnearnings is 5.443
 ^
β 1 = 0.092  ceteris paribus, if education increases by one year of schooling, earnings
are predicted to increase by 9.2%
 ^
β 2 = 0.078  ceteris paribus, if a worker is employed in the public sector, their
predicted earnings are 7.8% higher than otherwise
 ^
β 3 = 1.067  ceteris paribus, if hours increase by 1%, earnings are predicted to increase
by 1.067%

C
 The intercept, education, and lnhours variables all have p-values of 0.000
o 0.000 < 0.001 (p-value < a)
o Therefore, these coefficients are statistically significant at the 1% level
 The dgovt variable has a p-value of 0.235
o 0.235 < 0.001 (p-value < a)
o Therefore, this coefficient is not statistically significant at the 1% level
 D
H0: ^
β 2= ^
β 3=0
H1: either ^β 2 ≠ 0 or ^
β 3 ≠ 0 or both

E
In large samples, the F-statistic follows an F-
distribution with q numerator degrees of
freedom (number of restrictions) and ∞
denominator degrees of freedom.

Numerator dof = q (number of restrictions)

=2

Denominator dof = ∞

Therefore, the critical value associated with

a significance level of 1% is 4.61 (as obtained
from the given table)

F
Step 1: null & alternative hypothesis
H0: ^
β 2= ^
β 3=0
H1: either ^β 2 ≠ 0 or ^
β 3 ≠ 0 or both

Step 2: unrestricted model

^
ln earnings=5.443+ 0.092∗education+ 0.078∗dgovt+1.067∗ln hours

Step 3: restricted model

^
lnearnings=5.443+ 0.092∗education
Step 4: F-statistic: (the test statistic associated with the test in (e))

Step 5: Critical value: (from question (e))

Critical value associated with a significance level of 1% = 4.61

Step 6: Decision:
Null is rejected if the F-statistic > critical value:
16.848 > 4.61
AND/OR
Null is rejected if the p-value < a:
0.000 < 0.001

Therefore, we reject the null and accept that dgovt & lnhours are jointly significant at the 1%
significance level.

Bonus question
^
lnearnings=5.442576+ 0.092392∗education +0.078110∗dgovt +1.067111∗ln hours

Education = 15

Working 40 hours Working 50 hours

Working in the public sector 10.843 11.081
(dgovt =1)
Not working in the public 10.765 11.003
sector

Question 2
 A
^ 2
 β 0 = 41.977  if diffexpend, diffexp and prtystrA all equal 0, then predicted percentage
of the vote received by candidate A is 41.977%
 ^
β 3 = 0.149  ceteris paribus, if prtystrA increases by 1 percentage points, predicted
percentage of the vote received by candidate A will increases by 0.149 percentage
points

 The intercept coefficient has a p-value of 0, whilst the prtystrA coefficient has a p-value
of 0.008
o 0 and 0.008 are both < 0.1 (10% significance level)
o Therefore, we can conclude that these coefficients are individually significant

B
We cannot directly interpret the diffexp coefficient. I.e. for example, say that a $1 million
increase in diffexp results in a 6.370 percentage point increase in the percentage of the vote
received by candidate A.

This is due to the fact that this statement doesn’t consider that the diffexp variable is also
contained in the quadratic term. Therefore, the coefficient of the diffexp variable is not the only
impact that diffexp has on voteA.

C
The marginal effect of the difference in campaign expenditures = ^
β 1+ 2∗^
β 2∗diffexp
= 6.370+2∗0.120∗diffexp
= 6.370+0. 24∗diffexp

E.g., If diffexp increases by 1 million, voteA will increase by 6.61 percentage points

D
2
voteA=41.97735+6.369680∗diffexpend+ 0.120385 ¿ diffexpend +0.148541∗prtystrA

prtystrA = 55

Difference in campaign expenditures Predicted vote share of candidate A

Candidate A spends 2 million more than B 63.368%
Candidate A spends 1 million more than B 56.637%
Candidate A & B spend the same amount 50.147%
Candidate A spends 1 million less than B 43.898%

E
H0: ^
β 2=0
H1: ^
β2≠ 0

F
T-statistic:

0.120385−0
¿
0.135144
¿ 0.8907 9 …
¿ 0.891(3 dp)

Critical value:
At a 10% significance level the critical value for a two-
sided test is t* = 1.64

Given a large sample (n = 173), the t-statistic

asymptotically follows a normal distribution. Hence, we
can use the corresponding critical values as seen in the
z-score table & graph to find t* for this question.

G
We reject the null if the absolute value of the t-statistic > critical value

0.891 < 1.64

Therefore, we fail to reject the null, and hence accept that ^

β2≠ 0

As a result, I do not agree with the claim of the researcher. It is clear that the difference in
campaign expenditures does not have a constant marginal effect on A’s vote share and hence a
linear model (in which ^ β 2=0 ) would not be sufficient.