Problem 1 (1p): Assume we want to model variable Y on X 2 and X3, data is cross sectional.

Propose
the model when: a) Y is continuous; b) Y is binary (no order)
Problem 2 (6p): Variable Names: S: annual salary ($1000), Male = 1 if male, 0 otherwise,
Exp=Years of working, MBA = 1 if having an MBA degree, 0 otherwise; (standard error of
respective coefficient is within brackets, n=25). Estimated model is:
B2 B3 B4 B5 B6 B7

(1.2) (0.015) (1.35) (1.82) (0.065) (0.42)

1. Write the sample estimated model, explain the meaning of the coefficients. Why we use the
square of Exp in the model, what is its economic meaning?
2. Given R2=0.92. Explain its meaning and test for the significance of the model.
3. Does the rate of change of salary with pespect to Exp decrease when Exp increases?
4. Check that the rate of change of Salary with respect to Exp is different between Male and Female,
between people with and without MBA degree.
5. If you want to consider the interaction between Gender and MBA, propose the model.
6. Test if we need Qualitative factors (Gender and MBA) in the model if we estimate the model of S
on Exp and Exp2 having R2=0.84.
7. Do we have perfect multicollinearity in the above model. What are the consequences of imperfect
multicollinearity and how to correct the problem?

Y= beta1+beta2 X2+beta3 X3+u

X2= alpha1+alpha 2* X+ u
 Problem 3 (3p): Given the logistic regression in which Coupon is the dependent variable (1 if
customers use coupon for payment and 0 otherwise); Card is binary, whether not the Customers have
the Credit card from store; Spending ($1000) is the total payment.

Coupon Coef. Std. Err. z P>|z| [95% Conf. Interval]

Spending .3416434 .1286726 2.66 0.008 .0894498 .593837

Card 1.098732 .4446963 2.47 0.013 .227143 1.97032
_cons -2.146373 .5772454 -3.72 0.000 -3.277753 -1.014993

1. Write the estimation of logistic regression model, logit model. Explain the meaning of coefficients.
2. Estimate the Probability of using coupon for a customer who holds the store’ Credit card and
spending of $1500. If we fix Spending, how the Probability of using coupon change when customer
is change from not holding Card to holding Card
Problem 1:

a. When Y of data is cross sectional Propose the model when Y is continuous:

Yi=β1+β2X2+ β3X3+u
b. Binary: probability of Y=1 (Y =0 or 1)
β 1+ β 2 X 2+ β 3 X 3 +u

P(Y=1)= e
1+ e
β 1+ β 2 X 2+β 3 X 3+u

Problem 2:
1.

- sample estimated model

2
Ś=27.3+3.4 exp−0.02 Ex p +3.68 Male+6.2 MBA−0.06 Male∗Exper+ 1.04 MBA∗Exper
- the meaning of the coefficients:
o If other independent variables are fixed, when exp increases by 1 year, wages increase
by $3.4 per hour
o If other independent variables are fixed, the annual salary of male is higher than salary
of female by 3.68*$1000=$3680
o
o Β6=-0.06: If others are fixed, the rate of change of the annual salary with the respect to
experience of males is lower than that of females by 0.06
B hat.:(SE B hat) -t 25 5%/2<x<t 25 5%/2
Exp-exp^2
Exp^2
2.

R2=0.92  It means that 92% of variation of dependent variable that is explained by the model
Test for significane of the model
Hypothesis:
Ho: beta2=beta3=…=beta7=0
H1: not Ho
o R2=0.92, n=25, k=7,
o F= [R2/(k-1)]/[(1-R2)/(n-k)] =(0.92/(7-1))/((1-0.92)/(25-7))=34.5
- F(6,18)=3.22
- F> F(6.18) (34.5>3.22) Reject Ho  model is significant
3. Does the rate of change of salary with pespect to Exp decrease when Exp increases?

Rate of change of salary with respect to Exp: Y’=dY/dexp=beta2-2beta3*exp

To test the rate of change of salary with pespect to Exp decrease when Exp increases, we have the
hypothesis:
Ho: beta3 >= 0
H1: beta3 <0
t-test : t= beta3_hat/se(beta3_hat)= -0.02/0.015= -4/3
t25-70.05= 1.734
t>-t180.05=-1.734
 Do not reject H0
 -beta3>=0: constant
 the rate of change of salary with pespect to Exp is decrease when Exp increases

Rate of change of “the rate of change of salary”

Ho:
linear-linear model: Y=beta 1+beta2*exp+ u
rate of change :beta2
1x rate of change>0
x rate of change <0
prabol
4.
S=B1 +B2exp-B3exp2+B4Male+B5MBA-B6Male*exp+B7MBA*exp

- Male vs Female:
S=B1 +-B3exp2+B5MBA+B7MBA*exp + (B2-B6)*exp +B4
S=B1 -B3exp2+B5MBA+B7MBA*exp +B2exp
The difference in the rate of changes of salary respect to Exp between male and female is beta6
We test the hypothesis:
Ho: β6=0
H1: β6#0

-test : t= beta6_hat/se(beta6_hat)=0.06/0.065=0.923
T(18, 0.025)= 2.101
|t|< t(18,0.025) do not reject Ho 

- MBA vs no MBA
S= B1 +(B2+B7)exp-B3exp2+B4Male+B5-B6Male*exp
S=B1 +B2exp-B3exp2+B4Male-B6Male*exp
The difference in the rate of changes of salary respect to Exp between people with and without MBA
degree is beta7
We test the hypothesis:
Ho: β7=0
H1: β7#0

-test : t= beta7_hat/se(beta7_hat)=1.04/0.42=2.47
T(18, 0.025)= 2.101
t>t (18, 0,025)  reject Ho there is no difference in the rate of changes of salary respect to Exp
between people with and without MBA degrees

5.

to consider the interaction between Gender and MBA, the model:

S=27.3+ ^β 2 exp− ^β 3 Ex p + ^β 4 Male+ ^β 5 MBA− β^ 6 Male∗Exper+ ^β7 MBA∗Exper + ^β 8 Male∗MBA+ ú

2

6.
Snew=27.3 3.4Exp -0.02Exp2
H0: B4=B5=B6=B7=0
H1: not Ho
M=4, n=25, k=7
[(0.92-0.84)/4]/[(1-0.92)/(25-7)]=4.5 > 3.01 reject h0 we cannot drop the variable=> we need the qualitative
factors in the model
7.

Consequences: Estimates of se of coefficients can be bias (large)

 The t-ratio in the test of significance of coefficient are small, the
independent variable(s) will be not significant (it is supposed to be
significant), eventhought the R2 is still very high
Note: it is normally multicollinearity will exist in multiple regression.
However, we are concerning in the level of seriousness of it. It means if
multi happens but it does not cause serious consequences, it will be fine
to live with it. But if makes serious consequences, we need to consider
this problem seriously

P3:

Coupon Coef. Std. Err. z P>|z| [95% Conf. Interval]

Spending .3416434 .1286726 2.66 0.008 .0894498 .593837

Card 1.098732 .4446963 2.47 0.013 .227143 1.97032
_cons -2.146373 .5772454 -3.72 0.000 -3.277753 -1.014993

1.Write the estimation of logistic regression model, logit model. Explain the meaning of coefficients.
logistic regression model
−2.146+0.3416∗S +1.098∗C
e
p= −2.146+0.3416∗S+1.098∗C
1+e
Logit model:

Ln(p/(1-p)) = −2.146+ 0.3416∗S+1.098∗C +u

-The meaning of coefficient:
The coefficient of Spending is 0.3416: If Card fixed, when Spending increases 1 point the odd of
getting increases e^(0.3416)= 1.407 times which means 140,7%
The coefficient of Card is 1.098: If Spending fixed, when customer use the card, the odd of
getting increases e^(1.098)= times which means

Estimate the Probability of using coupon for a customer who holds the store’ Credit card and spending of $1500. If
we fix Spending, how the Probability of using coupon change when customer is change from not holding Card to
holding Card

e−2.146+0.3416∗1.5+1.098∗1
p= =0.369
1+e−2.146+0.3416∗1.5+1.098∗1
If we fix Spending, how the Probability of using coupon change when customer is change from not holding Card
to holding Card increases 1.098*p (1-p) =0.25572
 1.
Meaning of the coefficients:

- Beta2_hat: If other variables fixed, when the annual total sales of firms I increases by
1 % the annual salary of CEO increases 0.182%
-Beta3_hat: If other variables fixed, when the market value of firms I increases by 1 %
the annual salary of CEO increases 0.102%
2.
R^2=ESS/TSS= (TSS-RSS)/TSS= (64.646-42.06)/ 64.646=34.3%
It means that 34.3% of variation of dependent variable that is explained by the model
3.
Ho: Beta2=beta3 beta2-beta3=0
H1: not Ho
t-statistic=(beta2_hat-beta3_hat)/ se(beta2_hat-beta3_hat)
Se(beta2_hat-beta3_hat)=sqrt(var(beta2)+var(beta3)-2cov(beta2,beta3))
=sqrt(0.0412^2+0.0493^2-2*-0.001473)=0.08411
T=(0.182-0.102)/ 0.08411=0.95117
|T| <t(60-7, 0.05/2)=2.007 do not reject the coefficients of AR and MV are equal

4 Dropping AGE and AGE^2

Restriction:
Ho: beta6=beta7=0
H1:not H0
F=[(RSS new- RSS old)/m]/ (RSSold/(n-k))=(42.474-42.060)/2/[42.06/(60-7)]=0.26084
F0.05(2,60-7)=3.1
F< F0.05(2,60-7)=3.1 do not reject Ho choose model 1

5.
H0: Beta5=beta7=0
H1: not Ho

5.