Question 1

a) EViews Steps: Import → Import from file → ces2013.xlsx → Finish → No Link →

Quick → Estimate Equation → exp01 c size

Regression output: 𝐸𝑋𝑃 = 5 363. 83 + 953. 69 𝑆𝐼𝑍𝐸

Regression output: (50. 8973) (148. 4712)

Regression output (3 sig fig): 𝐸𝑋𝑃 = 5 360. 00 + 954. 00 𝑆𝐼𝑍𝐸

Regression output: (50. 8973) (148. 4712)

b) The regression estimation coefficient for SIZE is 953.69, meaning as the number of
people within a household increases by one unit, ie: one person, then the total
household expenditure in USD, EXP, increases by $953.69 on average. In a literal
sense, this interpretation does make sense due to the positive relationship, as more
people in the household would increase household expenditure. However, the
intercept is 5 363.83, thus if there were no persons in the household, the expenditure
would be $5 363.83, which does not make literal sense.

c) 𝐸𝑋𝑃 = 5 363. 83 + 953. 69 (5) = $10 132. 28

𝐸𝑋𝑃 = $10 100. 00 (3 𝑠𝑖𝑔 𝑓𝑖𝑔)

d) The p-value for SIZE equals 0, thus the null hypothesis of 𝐻0: β2 = 0 where β2 = 0 is
rejected and concluded that size is a statistically significant predictor of EXP at a
significance level of 1%.
2
 2
e) The 𝑅 = 0. 052535 indicating that the change in the number of persons in the
household explains 5.25% of the total variation in EXP.
2 𝐸𝑆𝑆 𝑅𝑆𝑆
𝑅 = 𝑇𝑆𝑆
=1− 𝑇𝑆𝑆
11
2.34𝑒
0. 052535 = 1 − 𝑇𝑆𝑆
11
2.34𝑒
𝑇𝑆𝑆
= 1 − 0. 052535
11
2.34𝑒
𝑇𝑆𝑆
= 0. 947465
𝑇𝑆𝑆 = 147 874. 0551
𝐸𝑆𝑆
0. 052535 = 1 − 147 874.0551
𝐸𝑆𝑆 = 7 768. 5365
𝐸𝑆𝑆 = 7 770. 0000 (3 𝑠𝑖𝑔 𝑓𝑖𝑔)

Question 2
a) EViews Steps: Import → Import from file → ces2013.xlsx → Finish → No Link →
Quick → Estimate Equation → exp01 c size sizeam

Regression output: 𝐸𝑋𝑃 = 4 902. 28 + 645. 11 𝑆𝐼𝑍𝐸 + 1 317. 96 𝑆𝐼𝑍𝐸𝐴𝑀

Regression output: (154. 2778) (59. 0733) (130. 9117)

Regression output (3 sig fig): 𝐸𝑋𝑃 = 4 900. 00 + 645. 00 𝑆𝐼𝑍𝐸 + 1 320. 00 𝑆𝐼𝑍𝐸𝐴𝑀
Regression output: (154. 2778) (59. 0733) (130. 9117)
3
 2
b) The 𝑅 has increased as the additional variable of SIZEAM may better explain the
changes in EXP. The addition of adult males variables explain the change in total
household expenditure better than SIZE alone.

c) SIZEAM coefficient is 1 317.96, meaning that as the number of males in the

household increases by one unit, ie: one male, then the total household expenditure
in USD, EXP, increases by $1 317.96 on average.

Two tailed t-test: where 𝑆𝐼𝑍𝐸𝐴𝑀 = β3

𝐻0: β3 = 0
𝐻1: β3 ≠ 0
α = 5%
𝑡0.025,6334 ≈ 1. 960
𝑡 = 10. 0676
𝑡𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 < 𝑡𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 → 𝐻0 𝑖𝑠 𝑟𝑒𝑗𝑒𝑐𝑡𝑒𝑑 𝑎𝑡 5% 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑐𝑒 𝑙𝑒𝑣𝑒𝑙, 𝑡ℎ𝑢𝑠 β3 𝑖𝑠 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡

d) Joint significance test, f-test: where 𝑆𝐼𝑍𝐸 = β2 and 𝑆𝐼𝑍𝐸𝐴𝑀 = β3

𝐻0: β2 = 0, β = 0
3
𝐻1: 𝑎𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝑠𝑙𝑜𝑝𝑒 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟 β ≠ 0
α = 5%
𝑝 𝑣𝑎𝑙𝑢𝑒 = 0. 000
𝑝 𝑣𝑎𝑙𝑢𝑒 < 0. 05 → 𝐻0 𝑖𝑠 𝑟𝑒𝑗𝑒𝑐𝑡𝑒𝑑 𝑎𝑡 5% 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑐𝑒 𝑙𝑒𝑣𝑒𝑙
𝐹𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 ≈ 3. 00
𝐹 (2, 6331) = 229. 01
𝐹𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 < 𝐹𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 → 𝐻0 𝑖𝑠 𝑟𝑒𝑗𝑒𝑐𝑡𝑒𝑑 𝑎𝑡 5% 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑐𝑒 𝑙𝑒𝑣𝑒𝑙
𝑇ℎ𝑢𝑠 𝑆𝐼𝑍𝐸 𝑎𝑛𝑑 𝑆𝐼𝑍𝐸𝐴𝑀 𝑖𝑠 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡𝑙𝑦 𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐𝑎𝑙𝑙𝑦 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑡

e) The model in Question 2 is preferred as the tests and an overall regression output
indicates the additional variables in Q2 model is a better explanation for the changes
in EXP. The t-test indicates the SIZEAM is a significant explanatory variable and the
f-test indicates the SIZE and SIZEAM is significantly different, therefore it is beneficial
to use both variables in the regression to better explain the changes in EXP.

Question 3
a) EViews Steps: Import → Import from file → ces2013.xlsx → Finish → No Link →
Quick → Estimate Equation → exp01 c size sizeam sizeaf

4
Regression output:
𝐸𝑋𝑃 = 5 654. 38 + 485. 72 𝑆𝐼𝑍𝐸 + 1 495. 16 𝑆𝐼𝑍𝐸𝐴𝑀 + 470. 79 𝑆𝐼𝑍𝐸𝐴𝐹
: (174. 4115) (78. 9527) (143. 2222) (154. 8536)

Regression output (3 sig fig):

𝐸𝑋𝑃 = 5 650. 00 + 486. 00 𝑆𝐼𝑍𝐸 + 1 500. 00 𝑆𝐼𝑍𝐸𝐴𝑀 + 471. 00 𝑆𝐼𝑍𝐸𝐴𝐹
: (174. 4115) (78. 9527) (143. 2222) (154. 8536)

Testing for Collinearity

Method 1: Variance Inflation Factor
EViews Steps: View → Coefficient Diagnostic → Variance Inflation Factor

5
With the values of VIF being below 2.5, this suggests there is no multicollinearity.

Method 2: Correlation Matrix

EViews Steps: control exp size sizeam sizeaf → open as group → covariance
analysis → correlation

6
 With the values of the correlation being below 0.58, this suggests there is moderate
correlation, but no strong basis to indicate multicollinearity.

b) Creating linear restriction:

EViews Steps: genr sizea =sizeam + sizeaf → Quick →Estimate Equation → exp01 c
size sizea

Two tailed t-test: where β3= β4

𝐻0: β3,4 = 0
𝐻1: β3,4 ≠ 0
α = 1%
𝑡0.005,6334 ≈ 2. 576
𝑡 = 8. 4032
𝑡𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 < 𝑡𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 → 𝐻0 𝑖𝑠 𝑟𝑒𝑗𝑒𝑐𝑡𝑒𝑑 𝑎𝑡 5% 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑐𝑒 𝑙𝑒𝑣𝑒𝑙
𝑇ℎ𝑢𝑠, 𝑆𝐼𝑍𝐸𝐴 𝑖𝑠 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡𝑙𝑦 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑡 𝑡𝑜 𝑡ℎ𝑒 𝑟𝑒𝑠𝑡𝑟𝑖𝑐𝑡𝑖𝑜𝑛

c) Joint significance test, f-test: where β3= β4

𝐻0: β3= β4 = 0
𝐻1: 𝑎𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝑠𝑙𝑜𝑝𝑒 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟 β ≠ 0
α = 1%
𝐹𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 ≈ 4. 61
𝐹 (2, 6331) = 55. 366
𝐹𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 < 𝐹𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 → 𝐻0 𝑖𝑠 𝑟𝑒𝑗𝑒𝑐𝑡𝑒𝑑 𝑎𝑡 1% 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑐𝑒 𝑙𝑒𝑣𝑒𝑙
𝑇ℎ𝑢𝑠 β3 𝑎𝑛𝑑 β4 𝑖𝑠 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡𝑙𝑦 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑡, 𝑡ℎ𝑒𝑟𝑒𝑓𝑜𝑟𝑒 𝑡ℎ𝑒 𝑟𝑒𝑠𝑡𝑟𝑖𝑐𝑡𝑖𝑜𝑛 𝑖𝑠 𝑛𝑜𝑡 𝑣𝑎𝑙𝑖𝑑

2 2
d) The 𝑅 has increased with the additional variables, as 𝑅 measures the proportion of
the dependent variable variance that is explained by the independent variables. The
addition of more variables to the model may explain more of the variance of the
dependent variables, ie: the change in expenditure.

Question 4
a) EViews Steps: Import → Import from file → oecd.xlsx → Finish → No Link → Quick
→ Estimate Equation → GDP c unemploy

7
 Regression output: 𝐺𝐷𝑃 = 3. 1083 − 0. 0403 𝑈𝑁𝐸𝑀𝑃𝐿𝑂𝑌
Regression output: (0. 6703) (0. 08065)

Regression output (3 sig fig): 𝐺𝐷𝑃 = 3. 1100 − 0. 040 𝑈𝑁𝐸𝑀𝑃𝐿𝑂𝑌

Regression output: (0. 6703) (0. 08065)

b) The regression estimation coefficient for UNEMPLOY is -0.0403, meaning as the

average annual rate of growth of real GDP increases by one unit, ie: percentage,
then the average annual unemployment rate would decrease by -0.0403, ie: -4.03%.
In a literal sense, this interpretation does make sense due to the inverse relationship
of real GDP and the unemployment rate, as rising real GDP indicates a growing
economy and thus labour, ie: employment is needed. The intercept is 3.1083, thus if
there were no growth or decline in the average annual unemployment rate, the
average annual rate of growth of real GDP would be 3.1083%, which does not make
literal sense as the average annual rate of growth of real GDP would not be able to
change without a change in the average annual unemployment rate as
unemployment rate is a key factor in the economy which affects GDP.

c) 𝐺𝐷𝑃 = 3. 1083 − 0. 0403 (0. 028) = 3. 1071716

𝐺𝐷𝑃 = 3. 11% (3 𝑠𝑖𝑔 𝑓𝑖𝑔)

8
d) One tailed t-test: where 𝑈𝑁𝐸𝑀𝑃𝐿𝑂𝑌 = β2
𝐻0: β2 = 0
𝐻1: β2 > 0
α = 5%
𝑝 𝑣𝑎𝑙𝑢𝑒 = 0. 6215
α = 0. 05 < 𝑝 𝑣𝑎𝑙𝑢𝑒 → 𝐻0 𝑖𝑠 𝑛𝑜𝑡 𝑟𝑒𝑗𝑒𝑐𝑡𝑒𝑑 𝑎𝑡 5% 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑐𝑒 𝑙𝑒𝑣𝑒𝑙, 𝑡ℎ𝑢𝑠 β2 𝑖𝑠 𝑛𝑜𝑡 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡

e) Unemployment may be underestimated because of discouraged workers, who have

attempted to look for employment for an extensive amount of time, in which they are
so unsuccessful that they stop looking for employment. A nation may include
discouraged workers in the unemployment rate, when they should have their own
separate category or be considered not a part of the workforce anymore. Another
factor not considered by the unemployment rate is the difference between part-time
and full-time workers and hidden unemployment which may cause the unemployment
rate to be underestimated.

f) Creating linear restriction:

EViews Steps: genr newunemploy = unemploy+0.5 → Quick → Estimate Equation →
GDP c newunemploy

Regression output: 𝐺𝐷𝑃 = 3. 1284 − 0. 04034 𝑁𝐸𝑊𝑈𝑁𝐸𝑀𝑃𝐿𝑂𝑌

Regression output: (0. 7063) (0. 08065)

Regression output (3 sig fig): 𝐺𝐷𝑃 = 3. 1300 − 0. 040 𝑁𝐸𝑊𝑈𝑁𝐸𝑀𝑃𝐿𝑂𝑌

Regression output: (0. 7063) (0. 08065)

The NEWUNEMPLOY coefficient is -0.04034, which is the same as the previous

regression. The intercept is now 3.1284, thus if there were no growth or decline in the
average annual unemployment rate, the average annual rate of growth of real GDP
would be 3.1284%.
2
Both the RSS is 60.1639 and the 𝑅 = 0. 010317 remain constant. As the values shift
by 0.5, the difference between the estimated and fitted model remains constant.
Initial Model:
2 𝐸𝑆𝑆 𝑅𝑆𝑆
𝑅 = 𝑇𝑆𝑆
=1− 𝑇𝑆𝑆
By adding 0.5, there is a shifting of the model by this 0.5. If 0.5 is added to every
predicted value, the variance of the predicted values, RSS will remain the same,
because adding a constant to each value doesn't change the spread of the data.
Therefore, RSS remains constant.
Again, adding 0.5 to every actual value does not change the data spread, so the
variance of the actual values, TSS remains constant.
2
Since both RSS and TSS remain constant when shifting the model by 0.5, the 𝑅
2 𝐸𝑆𝑆 𝑅𝑆𝑆
value calculated using the formula 𝑅 = 𝑇𝑆𝑆
=1− 𝑇𝑆𝑆
will also remain constant.

9
Additionally, under the following formula:
2 2
∑((𝑌𝑖+0.5)−𝑌) ∑𝑒𝑖
2
𝑅 = =1−
2 2
∑((𝑌𝑖+0.5)−𝑌) ∑((𝑌𝑖+0.5)−𝑌)

2 2
∑(𝑌𝑖−𝑌) ∑𝑒𝑖
2
𝑅 = =1−
2 2
∑(𝑌𝑖−𝑌) ∑(𝑌𝑖−𝑌)

Shifting the model by 0.5 when it is added to every predicted value results in:
𝑌𝑖 + 0. 5

Shifting the predicted values by 0.5 also shifts the residuals. The original residuals 𝑒𝑖

become 𝑒𝑖 − 0. 5 after shifting.

Therefore, there is a balancing effect with the addition of the 0.5 with the predicted
2
values, but the removal of 0.5 with the residuals, thus the 𝑅 value remains constant.

10