CHAPTER 3

THE TWO-VARIABLE
MODEL: HYPOTHESIS
TESTING

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 1

 3-1. The assumptions underlying the method of least squares

 we cannot make testing statistical hypotheses based on the

SRF, unless we make some specific assumptions about how
ui are generated) we call it classical linear regression
model (CLRM)
 Assumptions are :
A3.1>The regression model is linear in the parameters.
A3.2>The explanatory variable(s) X is uncorrelated with the disturbance term u.
A3.3>The expected, or mean, value of the disturbance term u is zero E(u|X ) = 0 i

A3.4>The variance of each u is constant, or homoscedastic (homo means equal

i

and scedastic means variance)

 If this is not the case, then we have heteroscedasticity, or unequal variance

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 2

 classical linear regression model (CLRM)
A3.3

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 3

classical linear regression model (CLRM)
A3.4

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 4

 classical linear regression model (CLRM)

A3.5>There is no correlation between two error terms. This is the

assumption of no autocorrelation. cov (ui uj) = 0 , i≠j

Here cov stands for covariance

A3.6>The regression model is correctly specified. Alternatively, there is

no specification bias or specification error in the model used in empirical
analysis.

A3.7>the error term ui follows the normal distribution with mean zero
and variance

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 5

 U= -1, 2, 3 , -.5, .3
 Uj=2, 3, -0.5, 0.3
 Ui =-1,2, 3, -0.5
 Cov(Ui,Uj) =0 no autocorrelation
 Cov(Ui,Uj) ><0  autocorrelation

Dr. Usama Alqalawi - Department

of Economic, Hashemite University 2/10/2018 6
 classical linear regression model (CLRM)
A3.5

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 7

 3.2 VARIANCES AND STANDARD ERRORS OF
ORDINARY LEAST SQUARES ESTIMATORS
 The CLRM enable us to estimate the variances and
standard errors of the ordinary least squares (OLS)
estimators b1 and b2.
 their values
 estimators' values of b1 and b2 ,will change from

sample to sample

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 8

 3.2 VARIANCES AND STANDARD ERRORS OF
ORDINARY LEAST SQUARES ESTIMATORS

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 9

  is the variance of the disturbance term ui that can be estimated
by

 standard error of the regression (SER) which is simply the

standard deviation of the Y values about the estimated
regression line
 RSS = residual Sum Square==
 df= degree of freedom = n-k = n-2, where k = number of

parameter

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 10

 Return to our example
GPAi Hi hi=
2 1 -1 1
3 2 0 0
2 1 -1 1
4 6 4 16
1 0 -2 4
sum sum
= =
0 22

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 11

 Return to our example
 = =0.218182

 Var(b1)= *0.218182=0.0833
 SE(b1)=0.2886
 Var(b2)==0.0099
 SE(b2)=0.0995

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 12

 3.3 WHY OLS? THE PROPERTIES OF OLS
ESTIMATORS
 Gauss-Markov Theorem
 the OLS estimators have minimum variance in the class of linear estimators; that is,

they are BLUE

 1. b1 and b are linear estimators; that is, they are linear functions of the random
2

variable Y,

 2.They are unbiased; that is, E(b1)=B1 and E(b2)= B2.

 3. = that is, the OLS estimator of the error variance is unbiased. In repeated
applications.

 4. b1 and b2 are efficient estimators; that is, var (b1) is less than the variance of any
other linear unbiased estimator of B1, and var (b2) variance of any other linear
unbiased estimator of B2.

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 13

 3.4 THE SAMPLING, OR PROBABILITY,
DISTRIBUTIONS OF OLS ESTIMATORS
 Central Limit Theorem
 If there is a large number of independent and

identically distributed random variables, then, with a

few exceptions, the distribution of their sum tends to
be a normal distribution as the number of such
variables increases indefinitely.

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 14

3.4 THE SAMPLING, OR PROBABILITY,
DISTRIBUTIONS OF OLS ESTIMATORS

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 15

 3.5 HYPOTHESIS TESTING

 null hypothesis is deliberately chosen to find out whether Y is

related to X at all.
 if the zero null hypothesis is sustainable, there is no point at all in

including X in the model.

 if X really belongs in the model, you would fully expect to reject

the zero null hypothesis H0 in favor of the alternative hypothesis

H1
 we cannot look at the numerical results alone, for we know that

because of sampling fluctuations, the numerical value will change

from sample to sample

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 16

3.5 HYPOTHESIS TESTING

 To test any hypotheses about B2 as well as B1 we can

use either:
◦ 1. The confidence interval approach or
◦ 2. The test of significance approach
 Since we know that ≈ 95% of the area of the normal
distribution lies within two standard deviation units of the
mean value
 Therefore, if our null hypothesis is B2 = 0 and the computed b2
= 0.0013, we can find out the probability of obtaining such a
value from the Z, or standard normal, distribution

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 17

 3.5 HYPOTHESIS TESTING

 If this probability is very small, we can reject the null

hypothesis, but if it is large, say, greater than 10
percent, we may not reject the null hypothesis.
 Knowing Z value can be calculated by

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 18

  But since is unknown we use t distribution
and calculate t value using

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 19

 Testing H0:B2 = 0 versus H1: B2 ≠ 0:
The Confidence Interval Approach
 Confidence interval Approach:
 H0: B2=0
 H1: B2≠0
 P (-tt ≤ ≤ tt ) =1-α %
 P: Probability
 tt: t tabulated ( take it from t table) depend on α and

df using two tail t table

 α : is significant level
 1- α : is confidence level

 if this interval includes the null-hypothesized value of B , we 2

do not reject the (fail to reject H0) hypothesis.

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 20

t table

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 21

 Return to our example
 H0: B2=0
 H1: B2≠0
 P (-3.182≤ ≤ 3.182 ) =1-5 %

 P (- 0.3166 ≤ ≤ 0.3166) =95%

 P (-0.7711≤ ≤ -0.1379) =95% * -1

 P (0.771≥ ≥ 0.1379) =95% Rearrange

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 22

 Return to our example
 P (0.1379 ≤ ≤ 0.771) =95%

 Now since 0 is not in the above range then we

reject H0. so hour of study (Hi) dose effect
(GPAi)
 such a confidence interval is known as the

region of acceptance (of H0) and the area

outside the confidence interval is known as
the rejection region (of H0).

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 23

Return to our example

0.1379
2/10/2018 0.771

Dr. Usama Alqalawi - Department of Economic, Hashemite University 24

 Return to our example
 We can do the same procedure for B1.
 now what if we want to test whether B2= 2?
 answer:
 H0: B2=2
 H1: B2≠2

P (0.1379 ≤ ≤ 0.771) =95% since B2 =2 is not

in the interval we reject H0. we say that with
95% confidence we are sure that population
slope (B2) dose not equal 2 .

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 25

 The Test of Significance Approach to
Hypothesis Testing(T-test)
 We need to calculate the test statistics that
we will call it t calculated (tc) and then
compare it to critical value of t that we will
call it t tabulated (tt)
 Then we compare between them such that:

 If |tc |>tt then we reject H0

 Else
 We fail to reject H0

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 26

Return to our example
 H0: B2=0
 H1: B2≠0
 tc= =4.5678
 tt=3.182
 Since |4.567|>3.182 we reject H0

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 27

 T-TEST
 we need to know three things to perform t-
test:
◦ 1. The d.f.
◦ 2. The level of significance, , which is a matter of
personal choice, although 1, 5, or 10 percent levels
are usually used in empirical analysis. Instead of
arbitrarily choosing the value, you can find the p
value (the
◦ 3. Whether we use a one-tailed or two-tailed test

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 28

 P-value
 P-value stands for Probability value
 Although 1, 5, and 10 percent are the

commonly used values of α, there is nothing

holy about these values.
 In practice, it is preferable to find the
 p value, also known as the exact significance

level, of the test statistic. This may be defined

as the lowest significance level at which a null
hypothesis can be rejected

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 29

P-value
 As we noticed earlier we reject H0 at α=5% so
for sure we will reject it at 10% ( try it !) but
what about 1%.
 At 1% tt=5.841 > tc So we will fail to reject

H0 at 1% so P-value should be between 0.01

and 0.05 and since P-value is certain α we
need to find a specific value
 So lets choose a value between 5% and 1%
 So if α =2%, tt = 4.541 tc>tt, so P-value

should be between 1% and 2% and so on …..

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 30

 P-value
 But since we cant continue because the table
have no value between 1% and 2% so we stop
at this point and leave it to computer to solve
it
 But in general P-value is the value of α that

make tt= tc .

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 31

A One-Tailed Test
 If hypothesis contain the simple < or > then
we use one tail test
 Since it’s expected that H effect GPA Positively

the hypothesis may be that

 H0: B2≤0.

 H1: B20. (here the alternative hypothesis is one sided).

 the probability of committing a type I error is not

divided equally between the two tails of the t

distribution but is concentrated in only one tail, either
left or right. In the present case it will be the right tail.

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 32

 A One-Tailed Test
 Back to our example
 tc=4.5678 ( as before)
 df=3

α tt

10% 1.638

5% 2.353

1% 4.541

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 33

 A One-Tailed Test
 So we can say
 We reject the H0 at 10% , 5% and 1% since tc

> from tt at 10% , 5% and 1%

4.5678

1.638 2.353 4.541

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 34

 3.6 HOW GOOD IS THE FITTED REGRESSION LINE:
THE COEFFICIENT OF DETERMINATION,

 How well the estimated regression line, fits the actual

Y values?
 The measure has been developed and is known as the

coefficient of determination, denoted by the symbol

 Recall that: ( subtract from both side)

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 35

3.6 HOW GOOD IS THE FITTED REGRESSION LINE:
THE COEFFICIENT OF DETERMINATION,

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 36

 3.6 HOW GOOD IS THE FITTED REGRESSION LINE:
THE COEFFICIENT OF DETERMINATION,

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 37

 3.6 HOW GOOD IS THE FITTED REGRESSION LINE:
THE COEFFICIENT OF DETERMINATION,

 :the total variation of the actual Y values about their

sample mean, which may be called the total sum of
squares (TSS).
 :the total variation of the estimated Y values about

their mean value, which may be called the sum of

squares due to regression or simply the explained
sum of squares (ESS).
 : as before, the residual sum of squares (RSS) or

residual or unexplained variation of the Y values

about the regression line

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 38

 3.6 HOW GOOD IS THE FITTED REGRESSION LINE:
THE COEFFICIENT OF DETERMINATION,

Then we can say

 the total variation in the observed Y values about their

mean value can be partitioned into two parts, one
attributable to the regression line and the other to
random forces, because not all actual Y observations
lie on the fitted line.
 And :

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 39

 3.6 HOW GOOD IS THE FITTED REGRESSION LINE:
THE COEFFICIENT OF DETERMINATION,

 Let define


measures the proportion or percentage of the total
variation in Y explained by the regression model.
 Two properties of may be noted:
 1. It is a non-negative quantity.
 2. Its limits are (0≤ ≤ 1) since a part (ESS) cannot be

greater than the whole (TSS)

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 40

 3.6 HOW GOOD IS THE FITTED REGRESSION LINE:
THE COEFFICIENT OF DETERMINATION,

 Other formula

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 41

 3.6 HOW GOOD IS THE FITTED REGRESSION LINE:
THE COEFFICIENT OF DETERMINATION,

 And

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 42

 Return to example
 Since we compute RSS before
 RSS== 0.6546

 We need to calculate TSS==

+ i=Yi yi
GPA
2 -0.4 0.16 1-(0.6546/5.2)=
3 0.6 0.36 = 1- 0.1259
2 -0.4 0.16 = 0.8741
4 1.6 2.56
1 -1.4 1.96
sum sum sum
12 0 5.2
Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 43
Return to example
 Hour of study Hi explains about 87.4 % of the
variation in GPAi scores

 We can present all the result as follow

 =1.491+0.4545
 SE=
 t= (5.1663) (4.5678) =87.4 %
 P-Value= (0.0141) (0.0197) d.f=3

Dr. Usama Alqalawi - Department of Economic, Hashemite University 2/10/2018 44