You are on page 1of 9



A) Sample Size:

Sample Size = d.f (TSS) + 1

= 14 + 1

= 15

B) Value of RSS:

Value of RSS = TSS - ESS

= 66042 - 65965

= 77

C) d.f of ESS and RSS:

D.f of ESS = No. Of variables - 1




D.f of RSS = d.f (TSS) - d.f (ESS)

= 14 - 2

= 12


D) value of R2 and R2


R2 = 66042 / 65965

R2 = 0.9988


R2 = 1 - (1- R2) x (n-1)/( n-k)
= 1 - (1 - 0.9988) x (15-1)/ (15-3)

= 0.9986

E) Test the hypothesis that X2 and X3 have zero influence on Y. Which test do you use and

To test the co-efficients of X2 and X3 are simultaneously 0.We should use a F test.

F = ESS/d.f (ESS)


RSS/d.f (RSS)

F = 65965/2



F = 5140.13

This F value is highly significant, leading to the rejection of the null hypothesis

F) From the preceding information, can you determine the individual contribution of X2
and X3 toward Y?

No. We need the results of the two-variable regression models.


A) Interpret the regression results.

Ceteris paribus, if the BTU rating of an air conditioner goes up by a unit, the average price of
the air conditioner goes up by about 2.3 cents. Other partial slope coefficients should be
interpreted similarly. The intercept value has no viable economic meaning in the present
B) Do the results make economic sense?

Yes. A prier, each X variable is expected to have a positive impact on the price.

C) At , test the hypothesis that the BTU rating has no effect on the price of an air
conditioner versus that it has a positive effect.

Degree of Freedom = n - k - 1

= 19 - 3 - 1

= 15

For 15 d.f. the 5% one-tailed critical t value is 1.753. The observed t value of 0.023 / 0.005 =
4.6 exceeds this critical t value. Hence, we reject the null hypothesis.

D) Would you accept the null hypothesis that the three explanatory variables explain a
substantial variation in the prices of air conditioners? Show clearly all your calculations.

H0: R2 = 0 and H1 : R2 > 0. Using the F test, we obtain

F= R2/k


(1- R2)/d.f

F= 0.84/3


(1-0.84)/ 15

F= 26.25

This F value is significant beyond the 0.01 level of significance. So, reject the null


Marginal Propensity to consume is the change in consumption to the changes in
income.MPC is coefficient of disposable income i.e, X2t.Thus, the

MPC is 0.93.

B) Is the MPC statistically different from 1? Show the appropriate testing procedure

t = (b2 - 1 )/ SE (b2)

t= 0.93 -1



= - 18.7466

For 73 d.f.(n-k),(76-3) this t value is highly significant. Hence reject the null hypothesis that
the true MPC is unity (Note: The se is obtained as 0.93 / 249.06 = 0.003734).

C) What is the rationale for the inclusion of the prime rate variable in the model? A priori,
would you expect a negative sign for this variable?

Since expenditure on items such as automobiles, washers and dryers, etc., is often financed,
the cost of borrowing becomes an important 5 determinant of consumption expenditure.
Therefore, the interest rate, representing the cost of borrowing, is expected to have a
negative impact on consumption expenditure.

D) Is b3 significantly different from zero?
Yes. The t value is -3.09, which is significant at about the 0.01 level of significance (two-tailed

E) Test the hypothesis that R2 = 0.

F= R2/k


(1- R2)/d.f

F= 0.9996/2



F = 91,213.5

This F value is obviously very high, leading to the rejection of the null hypothesis that 2 R = 0.
(Note: The F value reported by the authors is different because of rounding.)

F) Compute the standard error of each coefficient.

SE(b)= b/t

SE(b) = -10.96/-3.33

se( b) = 3.2913

SE(b1) = 0.93/249.06

se( b1 ) = 0.003734

SE(b2) = -2.09/-3.09

se( b2) = 0.6764

A) Develop a multiple regression model to explain the average starting pay of MBA
graduates, obtaining the usual regression output.

As a first pass, consider the following results obtained from EViews. The dependent variable
is average starting pay (ASP). Note: In this regression output, we present the adjusted 2 R for
the first time.

As these results suggest, GPA, tuition and recruiter perception have

statistically significant positive impact on average starting salaries at the

0.1% or lower level of significance. The percentage of employed graduates

also has a positive effect, indicating that higher demand for the graduates of

a particular school translates into a higher salary. The 2 R value is

reasonably high.

B) If you include both GPA and GMAT scores in the model, a priori, what problem(s) may
you encounter and why?

Since GPA and GMAT are likely to be collinear, if we introduce them

both in the model, as in (a), we would not expect both the variables to be

individually statistically significant. This is borne out by the results given in

C)If the coefficient of the tuition variable is positive and statistically signifi- cant, does that
mean it pays to go to the most expensive business school? What might the tuition variable
be a proxy for?

If the tuition variable is a proxy for the quality of education, higher

tuition may well have a positive impact on ASP, ceteris paribus. The results

in (a) may support such a hypothesis.

(D)Suppose you regress GMAT score on GPA and find a statistically significant positive
relationship between the two. What can you say about the problem of multicollinearity?

Regressing GMAT on GPA, we obtain the following EViews output:

From these results it seems that GMAT and GPA are collinear.

E)Set up the ANOVA table for the multiple regression in part (a) and test the hypothesis
that all partial slope coefficients are zero.

The Excel Analysis of Variance output is as follows (EViews does not

automatically provide an ANOVA table in regressions):

Source of

Variation SS d.f MSS F P-value

Regression 10376408086 5 2075281617 52.21057924 3.576E-17

Residual 1709176777 43 39748297.13

Total 12085584863 48

Note: In the source of variation, Regression is ESS, Residual is RSS, and

Total is TSS.

Since the p value of the estimated F value is so virtually zero, we can

conclude that collectively all the slope coefficients are not equal to zero,
multicollinearity among some variables notwithstanding.

F)Do the ANOVA exercise in part (e), using the R2 value.

Following the format of Table 4.3, we obtain

Source of

Variation SS d.f MSS F P-valu

Regression 0.8586 (∑ y2i) 5 0.8586(∑y2i)/5 52.21 0.0000

Residual (1- 0.8586)(∑ y2i) 43 (1-0.8586)(∑y2i)/

Total (∑ y2i) 48

Note: ∑y2i = 550,977,068,808.00 The conclusion is the same as before.


Here are the raw data for calculations:

Dependent Variable Explanatory Variable(s) RSS

Auction Price None 4830,756.7

Auction Price Age 2,245,713.7

Auction Price Number of bidders 4,059,311.8

Auction Price Age, Number of bidders 525,462.2

In all the cases the total sum of squares is 4,803,756.7. Note: The RSS can easily be obtained
from the EViews regression outputs for the above regressions. We compare the first model
that has no explanatory variables since price is regressed only on the intercept (RSSr =
4,803,756.7) with the model with all the explanatory variables (RSS = 525,462.2). Applying
the ur F formula given in this question, we obtain:

F = (RSSr - RSSur)/m


RSSur/(n - k) ' Fm,n-k
F= (4803,756.7-525,462.2)/2



F = 2,139,147.25



≈ 118.058

This F value is about the same as in Equation (4.57), save the rounding errors.