THE CHINESE UNIVERSITY OF HONG KONG, SHENZHEN

2022 - 2023 TERM 2

ECO 3121 Introductory Econometrics

ASSIGNMENT 3

TOPIC: Multiple Linear Regression Model.

INSTRUCTIONS:
• Please label clearly each answer with the appropriate question number and letter. Securely
staple all answer sheets together, and make certain that your name(s) and student number(s)
are printed clearly at the top of each answer sheet.
• Please use STATA to do Question 1, and report your STATA commands and results
together with your answers to the questions.
• Hand-written answers must be legible. Illegible assignments will be returned unmarked.
• Please combine your answers with supporting documents into one Adobe PDF file and
submit.

DUE DATE: 5PM Friday April 21, 2023

Please submit your work on Blackboard. Late submissions will receive a 0 with no excuses.

MARKING: Marks for each question are indicated in parentheses. Total marks for the assignment
equal 140. Marks are given for both content and presentation.
Question 1 (50 marks)

Data file: 3121A3.dta

Data Description: A random sample of 935 employees drawn from the 1980 U.S. population of
all employed paid workers.

Variable Definitions:
𝑤𝑎𝑔𝑒𝑖 : average monthly earnings of worker 𝑖 in 1980, in dollars.
𝑒𝑑𝑢𝑐𝑖 : years of formal education completed by worker 𝑖, in years.
𝑡𝑒𝑛𝑢𝑟𝑒𝑖 : years of worker 𝑖 with the current employer, in years.
𝑚𝑎𝑟𝑟𝑖𝑒𝑑𝑖 : an indicator variable equal to 1 if worker 𝑖 is married, and 0 if worker 𝑖 is single.
𝑏𝑙𝑎𝑐𝑘𝑖 : an indicator variable equal to 1 if worker 𝑖 is black, and 0 if worker 𝑖 is nonblack.
𝑠𝑜𝑢𝑡ℎ𝑖 : an indicator variable equal to 1 if worker 𝑖 lives in south, and 0 otherwise.
𝑢𝑟𝑏𝑎𝑛𝑖 : an indicator variable equal to 1 if worker 𝑖 lives in metropolitan area, and 0 otherwise.

Compute and present OLS estimates of the following population regression equation for the full
sample of 935 paid workers:

log(𝑤𝑎𝑔𝑒𝑖 ) = 𝛽0 + 𝛽1 𝑒𝑑𝑢𝑐𝑖 + 𝛽2 𝑒𝑥𝑝𝑒𝑟𝑖 + 𝛽3 𝑡𝑒𝑛𝑢𝑟𝑒𝑖 (1)

+𝛽4 𝑚𝑎𝑟𝑟𝑖𝑒𝑑𝑖 + 𝛽5 𝑏𝑙𝑎𝑐𝑘𝑖 + 𝛽6 𝑠𝑜𝑢𝑡ℎ𝑖 + 𝛽7 𝑢𝑟𝑏𝑎𝑛𝑖 + 𝑢𝑖

where 𝑢𝑖 is a random error term that is assumed to satisfy all the assumptions of the classical linear
regression model.

(10 marks)
(a) What is the approximate difference in monthly salary between blacks and nonblacks? Is this
difference statistically significant?

(10 marks)
(b) Add the variables 𝑒𝑥𝑝𝑒𝑟 2 and 𝑡𝑒𝑛𝑢𝑟𝑒 2 to the equation and show that they are jointly
insignificant at even the 20% level.

(10 marks)
(c) Extend the original model to allow the return to education to depend on race (black or nonblack)
and test whether the return to education does depend on race.

(10 marks)
(d) Again, start with the original model (1), but now allow wages to differ across four groups of
people: married and black, married and nonblack, single and black, and single and nonblack. What
is the estimated wage differential between married blacks and married nonblacks? Is this difference
statistically significant?

(10 marks)
(e) Start with the original model (1), use the Chow Test to test whether there is any difference in
regression models between blacks and nonblacks
Question 2 (54 marks)

You are conducting an econometric investigation into the hourly wage rates of female and male
employees. The sample data consist of observations for 526 employees on the following variables:
𝑤𝑎𝑔𝑒𝑖 : average hourly earnings of employee 𝑖, in dollars per hour.
𝑒𝑑𝑖 : years of formal education completed by employee 𝑖, in years.
𝑒𝑥𝑝𝑖 : the number of years of work experience accumulated by employee 𝑖.
𝑓𝑒𝑖 : an indicator variable equal to 1 if employee 𝑖 is female, and 0 if employee 𝑖 is male.

The regression model you propose to use is the log-level regression equation

log(𝑤𝑎𝑔𝑒𝑖 ) = 𝛽0 + 𝛽1 𝑒𝑑𝑖 + 𝛽2 𝑒𝑥𝑝𝑖 + 𝛽3 𝑒𝑑𝑖2 + 𝛽4 𝑒𝑥𝑝𝑖2 + 𝛽5 𝑒𝑑𝑖 𝑒𝑥𝑝𝑖 + 𝛽6 𝑓𝑒𝑖 + 𝛽7 𝑓𝑒𝑖 𝑒𝑑𝑖 +
+𝛽8 𝑓𝑒𝑖 𝑒𝑥𝑝𝑖 + 𝛽9 𝑓𝑒𝑖 𝑒𝑑𝑖2 + 𝛽10 𝑓𝑒𝑖 𝑒𝑥𝑝𝑖2 + 𝑢𝑖 (1)

Using the sample data described above, your research assistant computes OLS estimates of
regression equation (1) and of three restricted versions of equation (1). For each of the sample
regression equations estimated on the sample of n = 526 observations, the following table contains
the OLS coefficient estimates (with estimated standard errors in parentheses below the coefficient
estimates) and the summary statistics SSR (residual sum-of-squares), SST (total sum-of-squares),
and number of sample observations.

Dependent variable: 𝐥𝐨𝐠(𝒘𝒂𝒈𝒆𝒊 )

Independent variables (1) (2) (3) (4)
𝑒𝑑𝑖 -0.6286 -0.8608 -0.9978 -1.078
(0.4460) (0.4454) (0.3234) (0.3410)
𝑒𝑥𝑝𝑖 0.2827 0.1011 0.1011 0.1388
(0.07958) (0.07212) (0.07048) (0.07419)
𝑒𝑑𝑖2 0.04526 0.05083 0.05677 0.06305
(0.01583) (0.01591) (0.01120) (0.01179)
𝑒𝑥𝑝𝑖2 -0.006402 -0.003908 -0.003961 -0.004291
(0.001093) (0.0008178) (0.0008129) (0.0008565)
𝑒𝑑𝑖 𝑒𝑥𝑝𝑖 0.008016 0.01069 0.01085 -0.009264
(0.004226) (0.004307) (0.004131) (0.004354)
𝑓𝑒𝑖 3.490 -0.8498 -2.005 ---
(2.841) (2.728) (0.2591)
𝑓𝑒𝑖 𝑒𝑑𝑖 -0.4276 -0.2438 --- ---
(0.4478) (0.4489)
𝑓𝑒𝑖 𝑒𝑥𝑝𝑖 -0.2988 --- --- ---
(0.06737)
𝑓𝑒𝑖 𝑒𝑑𝑖2 0.01577 0.01156 --- ---
(0.01836) (0.01866)
𝑓𝑒𝑖 𝑒𝑥𝑝𝑖2 0.004996 --- --- ---
(0.001507)
𝐼𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡 3.821 7.235 7.955 6.796
 (3.196) (3.150) (2.410) (2.537)
SSR 4185.31 4428.28 4432.23 4943.72
SST 7160.41 7160.41 7160.41 7160.41
Observations 526 526 526 526

(10 marks)
(a) Compare the goodness-of-fit to the sample data of the four sample regression equations (1),
(2), (3) and (4) in the table. Calculate the value of an appropriate goodness-of-fit measure for each
of the sample regression equations (1), (2), (3) and (4) in the table. Which of the four sample
regression equations provides the best fit to the sample data?

(10 marks)
(b) Use the estimation results for regression equation (3) in the above table to perform a test of the
proposition that female employees of any given education and experience have lower average
wage than male employees of the same education and experience. Perform the test at the 5 percent
significance level (i.e., for significance level 𝛼 = 0.05). State the null hypothesis and the
alternative hypothesis. Show how you calculate the required test statistic. State the decision rule
you use, and the inference you would draw from the test.

(10 marks)
(c) State the coefficient restrictions that regression equation (2) in the table imposes on regression
equation (1). Explain in words what the restrictions mean. Use the estimation results given in the
table to perform a test of these coefficient restrictions at the 5 percent significance level (i.e., for
significance level 𝛼 = 0.05). State the null and alternative hypotheses, show how you calculate the
required test statistic. State the decision rule you use, and the inference you would draw from the
test. Would your inference be the same at the 1 percent significance level (i.e., for significance
level 𝛼 = 0.01)? Based on the outcome of the test, which of the two regression equations would
you choose, equation (1) or equation (2)?

(12 marks)
(d) Write the expression (or formula) for the marginal effect of 𝑒𝑥𝑝𝑖 on log (𝑤𝑎𝑔𝑒𝑖 ) for male
employees implied by regression equation (1). Use regression equation (1) to compute a test of the
proposition that the marginal effect of 𝒆𝒙𝒑𝒊 on 𝐥𝐨𝐠 (𝒘𝒂𝒈𝒆𝒊 ) for male employees is equal to
zero for male employees with any given values of 𝑒𝑑𝑖 and 𝑒𝑥𝑝𝑖 . State the coefficient restrictions
on regression equation (1) implied by this proposition; that is, state the null hypothesis and the
alternative hypothesis. Write the restricted regression equation implied by the null hypothesis.
OLS estimation of this restricted regression equation yields a Residual Sum-of-Squares value of
SSR = 5086.07. Use this information, together with the results from OLS estimation of equation
(1), to calculate the required test statistic. State the decision rule you use, and the inference you
would draw from the test. Choose an appropriate significance level for the test.
(12 marks)
(e) Write the expression (or formula) for the marginal effect of 𝑒𝑑𝑖 on log (𝑤𝑎𝑔𝑒𝑖 ) for female
employees implied by regression equation (1). Use regression equation (1) to compute a test of the
proposition that the marginal effect of 𝒆𝒅𝒊 on 𝐥𝐨𝐠 (𝒘𝒂𝒈𝒆𝒊 ) for female employees is constant
for female employees with any given values of 𝑒𝑑𝑖 and 𝑒𝑥𝑝𝑖 . State the coefficient restrictions on
regression equation (1) implied by this proposition; that is, state the null hypothesis and the
alternative hypothesis. Write the restricted regression equation implied by the null hypothesis.
OLS estimation of this restricted regression equation yields a Residual Sum-of-Squares value of
SSR =4348.85. Use this information, together with the results from OLS estimation of equation
(1), to calculate the required test statistic. State the decision rule you use, and the inference you
would draw from the test. Choose an appropriate significance level for the test.

Question 3 (36 marks)

Consider a dummy variable regression model 𝑦𝑖 = 𝛽0 + 𝛽1 𝛿𝑖 + 𝑢𝑖 , where 𝛿𝑖 = 1 if the

observation belongs to group 1 and 𝛿𝑖 = 0 otherwise (e.g., group 1 refers to female). Let the total
sample size be 𝑁 and the group 1 sample size be 𝑁1 , so that the base group has a size of 𝑁 − 𝑁1 .
Let the OLS estimators be 𝛽̂0 and 𝛽̂1 .
1 1
(a) Show that 𝛽̂1 = ∑𝛿𝑖 =1 𝑦𝑖 − ∑𝛿𝑖 =0 𝑦𝑖 , which is the difference in the group averages of
𝑁1 𝑁−𝑁1
𝑦 between group 1 and the base group.
1
(b) Show that 𝛽̂0 = ∑𝛿𝑖 =0 𝑦𝑖 , which is the sample average of 𝑦 within the base group.
𝑁−𝑁1
①⑥

On averages the monthly salary of blacks Is approximately 18.83% lower then

for non blacks & the difference is statistically significant at 190 Significant level

The P Value of F-test

-

is 01226 , which is greater than 2090 .

thence expert
and tenure are Jointly insignificant at even 2090 level .
⑥

Since P Value of the coefficient

-

on the interaction term equals

0,263 the
, Interaction between education & race is States cake

Insignificant Hong the return to education doesn't depend

, on race

⑤
 On average the month salary of married blacks
,
is
approximately
1795% lower then that for married non -
blacks The
. difference Is statically
significant

[ 127,158 -
(13,055-1 108/875) / (61-1)
f- =
=
5.6414
(1%055-11081875) / [935-216+11]
Critical value at 5% Significance Kool Is Coos =
201

Since F- 5,64147201 = Ciaos ,

blacks & non blacks follow different

wage models

②⑨ EACH
1-418%31/515 0,4041

§
=

pzz =
7160411525 bell
Ega) Provide the
6914 fit while Gala) Pronto
,

=
' -

¥É¥% =
*•
*•

E-9137

=
1-
¥%?,¥÷ = 03739

8914)

RT = 1- 49%1%4,1%-5=03029
⑤ Equation(3) → log lwagei )

ftp.yexp?tBsediekPitRofeitUiGl oglwagei)1edi,expi,fei
=
Both edit Bzeapi + Red

-41 -

E- ( loglwagei ) / edi ,
expi , fei=o / =
Bo
 Nall hypothesis Ho :P . =o for Po 20 )

Alternative typo thesis Hi ! Roco ( left tail test )

F- test statistic center Ho

HRH
-_Ñ% =?;;→= -
1- it

Degree of freedom =
D- 7--526-7=519

Decision rate : let tail f- test . Compare sample t statistic with 2 tend critical

value of tag distribution .

I ,
1ft ≥ -
Ca , do not refer to of significance lend 2
2. If + < -

Cd
, reject Ho at the significance here/ &

Critical Value of tag at 5% Significance level

Laos =/ 645 i

Inference level 5%
since -1=-717382
-

1645 = -

Coos feted Ho of significance

,

\
⑥ Equation b) Imposes on Gambon Uthe exclusion restriction PEO Rio

Test ! Ho ! D8 -0 Bio -0
,

Hi ! Ho is not true

Interpretation of Ho : The marginal log lwagd effect of expi is equal lidenticall formale &female
workers

F statistics

14428,28-418%31112
F =

%{%%!%→ =

4185,31 1$26 -
to y
-
= 1495

Decision Rate
if F KG , do not roofed Ho at significance
level 12--005) is loos = 3.00
1 ,

If F > on toted Ho at significance lead (2--0101) is cool

=
4161
2. ,

Inference

Since 7--14,957300 =
Coos , retell to at 5% Significance kod
Since F =
149574161 =
Cool , reject Ho at 190 Significance lad
/

Choose equation G) : restriction incorporated in equation 121 are rejected at a sufficient4100

Significance tend
⑦
For male workers , the marginal effect of edi is

2-tlloskggei.gl?diiexPiifei-#--pz-zPaexPitPsedo
Ted

Pz Rs
'

Ho ! =
O ,
Ba = 0 20
,

Hi ! Ho ISnot the

Restricted motel implied by the null hypothesis

B- Uni edit D8 Uni emit Bg Uni ed? + MounierPita i

1091 wage i ) = Both edit Redi t Po Uni +

F-Statistics

(508607-4185.31)/3
F
1%11%1%1,1 =

4185131 / 1526-10-1 ,
=
36195

Decision Rule
If F ≤ On donot total Hoot significance kid
✗
It
,

2 . If F > Ca , resect Ho at the significance level 2

Critical Value of 1=3,515 at 5% Significance level 12=005) IS 1005=260

critical Value of 7314s at 1% Significance lad 1220101 ) is cool = 7178

Inference
since F =
36,957260 =
Ciaos petal Ho, we the 5% Significance herd
,

since 7--361957 3178 =

Cool
, Afore Ho , at the 1% Significance HM
② For female workers , the marginal effect of edits

28110g lwagei ) / edi ,e✗Piife ) ( p + pp) -121Mt Pg) edit Rse✗ Pi

,
=

2 edi

Test :

Ho ? the Pg -0 ,
Rs -0
true
Hi 1 Ho is not

Sd Ps-0 , Bg = - B] restricted model is therefore !

,

Both edit this ex Pit Ps G Fei) edit Pa exp

ftp.of-ei-P-tfeiedi-P8/-eiexPi-PiofeiexPi2ta
loglwagei) =
-

F-Statistics

= (ssRr-SSRar =
(4348185-4185131)/2--10,06
SSRur / Ch -
k -7
4185,31 / (526-10-1)

Decision Mk

Ii If f- 2- Cd , do not resect Ho at the significance level ✗

2 .
If F > ca , retted Ho are the significance level ✗

Critical Value of 1=3,3274 at 5% Significance level (2--0105) is loos = 3.00

Critical value of Fbi 3274 At 1% Significance level Carroll Is Cool = 4,61

Inference
since 7--10106 23100 -
Coos , Med Ho at the 5% Significance had

since P = 1010674.61 = Cool , resect Ho H the 1% Significance level

③
⑨ The first order condition for Ñ & Ñ are respected
,

Tiki PT Ñ8i 1--0

-
-

& T.21-ii-PT-P.si/8i--0

Since Si = 1 for group1 and Si = 0 forthe base group ,

the FOC for Ñ can be written as

¥ ( Yi
,
-
poi B) ⑦-

which implies

pi
Ñ=¥¥;i
-

Furthermore ,

pi ¥ ? ti k too Isi
=
-
=
# {✗ i -

p? ¥
Substituting for PT ,
we get

¥ §÷ Yi ¥ / ¥ Yi )
+ Iti
N-¥ ±§
-

A ↓ Eti Ii
- =
=
, , Si-0
-

,
,
Si-4

=
'
-2 Yi -
¥ ¥ .
Yi
Wi N Si

? ↓ -2g ¥ Esri Yi
binding be
NjM- get P
Yi -
=

,
we , , ,
⑥
pi = ±? ti
- -

¥ I #¥ ,
Yi -

¥¥ ,
◦
Yi )

1T¥ % E"
'

=
¥ / E. ,
'
" +
§
"
) -

,
Yi +

=
¥ E. Yi , ◦