Chapter 7

Tests of hypotheses
4.1. Introduction
Hypothesis testing is a method of analysis that makes inferences about
population parameters or relationship of variables from sample data. The
procedure is that we make an assumption about the population parameter or
the relationship variables under study and then test the assumptions on the
basis of sample information or data. It is one the major techniques of
statistical inference, which is used to solve many practical problems in
business and in any field in general.

Objectives
Upon completion of this unit you will be able to

 Define hypothesis and statistical tests

 Identify the types of hypothesis and types of errors
 Know level of significance and power of a test
 Test about means
 Test about proportions
 Make analysis of variance

4.2. Definition of hypothesis

A hypothesis is a statement or assumption made about the population
parameters, which may or may not be true. It also takes the form of a
statement or assumption about relationship between two variables, an
independent variable (cause) and a dependent variable (effect).
A hypothesis is not necessarily true, it can be either right or wrong, and we
use a sample data to help us decide. If we know every thing, there is no need
for statistical hypothesis testing. When there is uncertainty, statistical
hypothesis testing will help us learn as much possible from the information
available to us. In broader view, hypothesis testing is one component of the
decision- making process.
As an example consider the following.
1. The proportion of educated people in Ethiopia is not less than 30%.
2. The average monthly income of families of AA is less or equal to Birr
800.
3. The cause of high number of Aids patients in Africa is poverty.

The first two examples are assumptions about population parameters,

proportion of educated people in Ethiopia and average family monthly
income respectively. The third example shows assumption about the
relationship of two variables, poverty and number of aids patients. In all
cases of the above examples, we need to draw conclusions or make
decisions about the specified population parameters or relationship of
variables on the basis of sample information. These decisions are called
statistical decisions. These assumptions about the population parameters
or relationship of variables are called hypotheses. Such main statements
(assumptions) are also called null hypotheses which are usually denoted
by Ho. A hypothesis accepted when Ho is rejected is called an alternative
hypothesis which is also denoted by Ha or H 1. The alternative hypothesis,
which is also called research hypothesis, is to be accepted only if there is
convincing statistical evidence that would rule out the null hypothesis as a
reasonable possibility. Hence, hypotheses are composed of both the null and
alternative hypotheses and are always formulated as opposite, so that when
one is true the other is false. Therefore, the above assumptions are not full
hypotheses, for they represent only the null hypothesis, which is the most
likely assumption. Let us now try to rewrite the above assumptions into full
flagged hypotheses.

1. Ho: The proportion of educated people in Ethiopia is not less than

30%.
 Ha: The proportion of educated people in Ethiopia is less than
30%.
Or Ho: P ≥ 0.3
Ha: P < 0.3
2. Ho: The average monthly income of families of AA is less or equal to
Birr 800.
Ha: The average monthly income of families of AA is more than
Birr 800.
Or Ho: µ ≤ 800
Ha: µ > 800
3. Ho: The cause of high number of Aids patients in Africa is poverty.
Ha: The cause of high number of Aids patients in Africa is not poverty.

Based on sample values we either accept or reject Ho. If Ho is rejected then

the alternative hypothesis, Ha, is accepted. Rejecting Ho could either be the
assumption is truly wrong or due to chance that the sample data is wrong. If
the decision cannot be explained as being probably due to solely to chance,
the difference is said to be statistically significant. Tests devised to check
whether this is so are called significant tests. The set of outcomes or values
of the random sample for which we reject Ho is known as the rejection or
critical region, while the set of values for which Ho is accepted is known as
the acceptance region. Thus, a test of hypotheses is the partitioning of the
set of outcomes into the critical and the acceptance region.

In testing hypothesis we always consider the following assumptions

1. The data set is a random sample from the population of interest, and
2. Either the quantity being measured is approximately normal, or else the
sample size is large enough that the central limit theorem ensures that the
sample average is approximately normally distributed.
Let us now consider the hypothesis about thee average monthly family
income of AA given above
Ho: µ ≤ 800
Ha: µ > 800
The objective here is to check if the claim or assumption made about the
average family income is correct or not.

The decision as to whether or not to approve the assumption will depend on

the test results from the sample data. Therefore, the decision will be based

on the level of the sample mean, X , which is referred to as the test statistic.

To make test suppose we take a random sample of 500 families and decide

to reject Ho if X > 800 and accept Ho if X ≤ 800.

The value 800 is commonly referred to as critical value. The test statistic is
then used to formulate a decision rule to translate sample evidence into a

course of action. Therefore, the decision is accept Ho X ≤ 800 and

reject Ho if X > 800.

Therefore, the set of values { X > 800} is called critical or rejection region

and the set of values { X ≤ 800} ia called acceptance region.

When testing such hypotheses we always make errors since the test
depends on sample data. There are two types of such errors, type I and type
II errors. Type I error is made when Ho is wrongly rejected, while Type II error
is made when Ho is wrongly accepted. These two types of errors can be
summarized as follow

Accept Ho Reject Ho
Ho True No error Type I error
Ho False Type II error No error
Notation
The probability of type I error is denoted by  and the probability of type II
error is denoted by .
P (Type 1 error) = P (rejecting Ho  Ho is true) = 
P (Type 11 error) = P (accepting Ho  Ho is false) = 
Level of significance. The probability of rejecting Ho when it is infact true is
denoted by  and is known as a level of significance. That is,
Level of significance =  = P (reject Ho/ Ho is true)
= P (type I error)
 is usually taken as 0.05 or 0.01.

Power of a test. The power of a test is defined as the probability of

rejecting the null hypothesis H o when it is actually false or the probability of
accepting Ho when it is actually false. That is,
Power of a test = P (rejecting Ho/ Ho is false)
= 1 – P (accept H o/ H0 is false)
= 1 – P (type II error)
=1-
where  is the probability of type II error.
In a similar manner we show that,
Power of a test = 1 - 
This shows that the smaller the two types of errors,  and , the stronger the
power of the test. Therefore, the objective of any test is to minimize the two
types of errors. However, it can be shown that we cannot simultaneously
minimize the two types of errors. Hence when testing hypothesis,

We determine the critical values by first specifying alpha (), the probability
of committing type I error. If the cost of committing type I error is high, the
decision makers should specify a small alpha. A small alpha results in a small
rejection region. If the rejection region is small, the sample mean is less
likely to fall there, and the chances of rejecting a true null hypothesis are
small. If the acceptance region is large, the probability of committing a type
II error () will also tend to be large. A decrease in  will increase in .
However, the increase in  will not equal the decrease in . Decision makers
must examine carefully the costs of committing type I and type II errors and,
in light of these costs, attempt to establish acceptable values of  and .

The goodness of the choice of the critical value 9 depends on the attitude of
the decision towards the two types of errors, Type I and Type II errors. The
smaller the two types of errors, the better the test. These errors can be
computed for any particular test. It can be shown that when one of the
probabilities is decreased the other tends to increase. Therefore, the general
agreement is that we fix the probability of type one error at  and try to
minimize type two error .
Hypotheses in general are of two forms
Tests of the form,
Ho:  = o or Ho:  = o
Ha:  N o Ha:   o,
in which the critical region includes either large or small values of the test
statistic are called one- sided or one –tailed tests.
A test statistic is a formula which is used for computing from the data in
testing hypothesis. The value of the test statistic is used in determining
whether or not we may reject the null hypothesis.
And those of the form,
Ho:  = o
Ha:   o
Including both large and small values of the test statistic are called two –
tailed tests.
Steps in testing hypotheses
In general statistical testing of hypothesis follows certain steps that are more
or less taken by most researchers.
1. Formulate the hypothesis. That is, rewrite the problem to a testable
form.
2. Determine alpha ( ), which is usually given
3. Determine the appropriate statistical distribution used.
4. Determine the test statistic.
5. Establish the critical region of the test statistic or give the decision
rule. The decision rule of a statistical hypothesis test is a rule that
specifies the conditions under which the null hypothesis may be
rejected.
6. Gather data.
7. Calculate the value of the test statistic and find the tabulated value.
8. Make a decision, that is, accept or reject the null hypothesis, Ho.
4.3. Tests about a single mean
Hypotheses tests about means, just like confidence interval estimation about
means considered in the previous unit, are dependent on whether the
population is normal or not, samples large or small and population standard
deviation known or unknown. Let us see the different possible combinations
of these cases and make tests about mean.

4.3.1. Normal population, variance known, large or small sample

Suppose we take a random sample of size n (large or small) from an
2 2
N (,σ ) population when σ is known. We want to test
 H o:  = o
Ha:   o, where o is a specific value.

Since the best estimator of  is X , the test statistic must depend on X .

X−μ
2
σ σ
But in this case, X  N (µ, n ) and Z = √ n  N (0, 1)
X− μ o
σ
When Ho is true, Z = √ n  N (0, 1)
X− μ o
σ
Note that √ n is the test statistic for testing single mean.
If  is really greater than o, X would most probably be greater than o and
X− μ o
σ
therefore, Z = √n  C
where, the constant C = Zα , the value of which is found using the level of
significance (α ) from the normal table.

1-

_____________________________________________________________
Z
X−μ o
(
σ
P √ n  C) = 
Critical region for testing the above hypothesis is
 X− μ o
σ
√ n  Z
If Ho:  = o
Ha:   o
In a similar argument as above, the critical region is,
X− μ o
σ
√ n  - Z
To test the two- tailed tests,
Ho:  = o
Ha:   o, there is no difference between the believed population
mean and the specified value.
Critical region can be shown to be,
X− μ o
σ
 √ n   Z/2
Example 1
According to a car manufacturing company, their cars averaged at least 32
miles per gallon in the city. From past records it is known that mileage is
normally distributed with standard deviation of 2.5 miles per gallon. Tests
on 16 cars showed that mean mileage in the city is 31.5 miles per gallon. Do
the data support the claim of the company at 1% level of significance?
Solution
The appropriate hypothesis is,
Ho:  ≥ 32
Ha:  < 32
Here, population is normal, sample size, n = 16, is small and population
standard deviation,  = 2.5, is known, and hence we use the normal
distribution for the test.
Let  = 0.01
Here, we have to reject Ho if
 X− μ o
σ
√ n < - Z
X− μ o 0 .985−1
σ
0 .105
Zcal =  √n  =  5  = 0.728
Z tab = - Z = - Z0.01 = - 2.33
Since Zcal > Z , accept Ho.
tab

Therefore, the claim of the company is correct or the data support the claim
of the company.

Activity1
A producer of an electric bulb claims that the average life length of its
product is more than 1800hrs. Consumer’s protection agency on the other
hand doubts the claim. Therefore, it takes a random sample of 400 bulbs and
found out that the mean life length is 1780hrs with standard deviation of
200hrs. Assuming that life length of bulbs is normal, should the agency
support the producer’s claim at 5% level of significance?

4.3.2. Non-normal population, large sample,  known or unknown

We wish to test,
Ho:  = o
Ha:   o

Here, since n is large by the central limit theorem, X is approximately

normally distributed. Therefore,
X−μ
σ
Z= √ n  N (0, 1)
Thus the test statistic and critical region are the same as the above case,
 X−μ
σ
√ n  Z
The critical regions for Ha:   o and Ha:   o could similarly be shown to
be the same as above.

Note that if σ is unknown it could be estimated by S, the sample standard

deviation given by

S= √ ∑ ( X i −X )2
n−1
Example 2
Workers of a given industry complain that their average monthly salary is at
most birr 600. Workers union wanted to test the claim of the workers and
takes a random sample of 100 workers of the factory that produce a mean
monthly salary of Birr 610 with a standard deviation of Birr 50. What should
the workers union conclude about the claim of the workers at 5% level of
significance?
Solution
The hypothesis is
Ho:   600
Ha:  > 600
Here,  is unknown and n is large. Therefore, by central limit theorem, the
test distribution is Z-distribution.
X−μ
s
Critical region is √ n  Z
X−μ 610−600
s 50
√n = √ 100 = 2
For  = 0.05, Z0.05 = 1.645
Since Zcal > Z , then Ho is rejected. Therefore, the workers union should
tab

reject the claim of the workers.

4.3.3. Normal population, small sample,  unknown
In some cases of testing hypotheses, because of reasons such as time,
money, convenience or availability, is able to gather only a small random
sample of data. We usually say n is small if n  30. In such cases, if the data
are normally distributed in the population and σ is known, the Z- test can
be used.
However, if population standard deviation, σ , is unknown the Z test cannot
be used but rather the t - test based on the t - distribution is applicable.
Here again we replace σ by S, sample standard deviation, where
1
S2 = n−1  (Xi - X ) 2
Suppose we wish to test Ho:  = o
Ha:   o

We have seen earlier that in this case, X has a t- distribution with (n-1)

degree of freedom, X  t (n – 1).

X−μ
S
T= √n  t (n-1)
which gives the critical region as,
X−μ
S
√n  t (n-1)
If Ho:  = o
Ha:   o
 X−μ
S
Critical region is √ n  - t
And also if Ho:  = o
Ha:   o
Then critical region similarly is given by
X−μ
S
 √n   tα /2
Example 3
It is known that the average body temperature of human beings is a normal
variable with mean 370c. A medical person took the temperatures of ten
patients as 39, 37, 41, 35, 37, 37, 34, 38, 40, and 39 in 0 c. Test at the 5%
level of significance whether these sample data are consistent with the
specified human body temperature.
Solution
The hypothesis is
Ho:  = 37
Ha:   37

X =
∑ Xi
n =, 37.7 S= √ ∑ ( Xi−X )2
n−1 = 1.84
X− μ o
S
Critical rejoin is  √n   tα/2
X− μ o 37 . 7−37
S 1. 84
t cal = √n = √10  = 1.2
At  = 0.05, ttab = t (n – 1) = t
/2 0.025 (9) = 2.26
Since t cal < ttab, accept Ho.
Therefore, the sample data are consistent with the specified body
temprature.
Activity2
In a particular school District the I.Q. is known to have a normal distribution
with mean of 110. From one school a sample of 25 students were taken and
found that their average I.Q. is 115 with variance 100. Is the average in this
school different from the district average?

4.4. Test about a single population proportion

As shown in the previous sections, proportion is a value between 0 and 1
that shows the number of items possessing a certain characteristics in the
population or sample. There are many situations in business in which we
must test the validity of statements about population proportions or
percentages. A business man can claim that 90% of population of consumers
like his product. A random sample of consumers could be taken test this
claim in a similar manner as above.

The method here is to compare the sample proportion with the specified
population value or to analyze the qualitative data where we test the
presence or absence of a certain characteristics bases on sample values.

By the central limit theorem, if nP > 5 and nQ > 5, it can be shown that the

sample proportion, ^p , values are approximately normally distributed as

shown in the previous section.
^ is approximately normal with mean P
That is if nP > 5 and nQ > 5, then p

and standard deviation √ PQ

n

^ = sample proportion and Q = 1 – P.

where P = population proportion, p
^p −p

Hence Z= √ PQ
n  N (0, 1)
 ^p −p

Again √ PQ
n is the test statistic for testing single proportion.

The test procedure is the same as that of the mean, shown in the previous
sections. That is the computed value of Z is compared with the table value of
Z, which is also known as the critical value of Z.
If the hypothesis is Ho: P = Po
Ha: P > P o, where Po is a specific value

1-

_____________________________________________________________
Z
^p −p

P( √ PQ
n > Z ) = , where  is probability of type I error
^p −p

Therefore, the critical region is √ PQ

n > Z

Similarly, if Ho: P = Po
Ha: P < Po
^p −p

Critical region is √ PQ
n < - Z

For the two – tailed hypothesis

Ho: P = Po
Ha: P ≠ Po
^
P−P

The critical region is 

PQ
√
n  Z/2
Example 4
A survey on the workers of an industry indicates that 240 workers out of a
randomly selected sample of 500 workers are not professionals. But workers
union believes that less than or equal to half of the workers are
professionals. Test at α = 0.05 the claim of workers union.
Solution
The appropriate hypothesis is
Ho: P ≤ 0.5
Ha: P > 0.5
Here, P = 0.5, Q = 1 – P = 0.05,
500−240
^ = 500
Sample proportion = p = 0.52
^
P−P

Critical region is
PQ
√
n  Z
The calculated value of Z is

^p −p 0 . 52−0 .5

Zcal = √ PQ
√
(0 . 5)( 0. 5 )
n = 500 = 0.89
For  = 0.05, the tabulated value is
Ztab =Z = Z0.05 = 1.645.
Since Zcal < Ztab , accept the Ho.
Therefore, the claim of workers union is correct.

Activity 3
A medical expert claims that at least 60% of diseases of a given community
are related to lack of sanitation. A researcher who wanted to test the claim
takes a random sample of 500 patients from the community and found out
that 350 of the cases are related to sanitation. Test at 1% level of
significance if the claim of the medical expert is correct or not.

4.5. Tests involving finite populations

The same way as in the construction of confidence intervals with finite
population containing N elements and the sample size, n, constitute at least
5% of the population, we need to use the finite population correction factor,

√ N −n
N −1 , in tests of hypotheses too. In this case also the standard error of the
respective estimators must be multiplied by the finite population correction
factor.

In the case of hypothesis tests about the population mean, μ, we incorporate

σ
σ
a finite population correction factor by using the standard error, x = √n

√ N −n
N −1 . When the population standard deviation is unknown, we substitute
the sample standard deviation, s, for it.

Similarly, for testing hypotheses about population proportion, P, with finite

population and the sample constitutes at least 5% of the population, we
incorporate the finite population correction factor to get a standard error of

√ √
^
P(1− ^ N−n
P)
^ , as σ ^p =
the sample proportion, p n N−1
Example 5
Suppose that a random sample of 200 electric bulbs from a total of 2000
bulbs produces average life length of 12,000 hrs with a standard deviation of
Birr 1500 hrs. Test at 5% level of significance that the average life length of
the 2000 electric bulbs is at most 11,800 hrs.
Solution
For N = 2000 and n = 200, the sampling fraction, n/N = 200/2000 = 0.1,
which is more than 0.05. Therefore, the finite population correction factor
must be used in the test.
The hypothesis is
Ho: μ ≤ 11,800
Ha: μ > 11,800
x−μ o

The test statistic is

s
√
N−n
√ n N−1
x−μ o

The critical region is

s
√
N−n
√ n N−1 > Z

x−μ o 12 , 000−11, 800

s
√ N−n
√ n N−1 =
1500 2000−200
√
√ 200 2000−1 = 1.99
Z = Z0.05 = 1.645
Reject Ho. Therefore, the average life length of the electric bulbs is more
than 11,800 hrs.

4.6. Tests about difference of two means

The comparison of two population means is another important and
frequently used statistical technique in business. In testing hypotheses about
difference of two means the assumptions we saw in testing hypotheses on
single mean are also applicable.
 4.6.1. Normal populations, variances known
Suppose that we have a random sample of size n 1 from first population and
another independent sample of size of n 2 from the second population. The
problem here is to see whether there is significant difference between the two

population means based on the sample mean difference, X 1 - X 2 . Where X 1

and X 2 are sample means from the first and the second population
respectively.
Here, it is obvious that the difference in the sample means is normally
σ σ
12 22
+
distributed with mean, μ1- μ2 and variance n1 n2
2 2
σ 1 σ 2
+
That is, ( X 1 −X 2 )  N (1 -2, n1 n2 )

( X 1 −X 2 )−( μ1 −μ2 )

√
σ σ
12 22
+
Thus, Z = n1 n2  N (0, 1)
Suppose we wish to test,
Ho: 1 = 2 or Ho: 1 - 2 = 0
Ha: 1  2 Ha: 1 - 2 > 0
X 1− X 2

√
σ 2 σ 2
1 2
+
Here under Ho, the test statistic becomes n1 n2  N (0, 1)

This gives the critical region as

X 1− X 2

√
σ 2 σ 2
1 2
+
n1 n2  Zα

If the test is of the form,

Ho: 1 = 2 or Ho: 1 - 2 = 0
Ha: 1  2 Ha: 1 - 2 < 0
Critical region becomes
X 1− X 2

√
σ 2 σ 2
1 2
+
n1 n2  - Zα

Similarly if the test is a two- tailed test

Ho: 1 = 2 or Ho: 1 - 2 = 0
Ha: 1  2 Ha: 1 - 2 ≠ 0
The critical region is
X 1− X 2

√
σ 2 σ 2
1 2
+
 n1 n2   Zα /2
Example 6
A sample of 200 students graduated from college in 1990 produce an average
age of 25 years and another sample of 250 students graduated from college
in 2005 produce an average age of 23.5. Test at 5% level of significance that
the average age of graduation decreases in 2005, assuming normal
population and standard deviation of ages of graduation of both years are
equal to be 5 years.
Solution
Let 1 = the average age of all students graduated in 1990
2 = the average age of all students graduated in 2005
The required test is
Ho: 1 = 2
Ha: 1 > 2
Here, populations are normal, 1 = 2 = 5, then the test distribution is normal.
Therefore, critical region is
X 1− X 2

√
σ 2 σ 2
1 2
+
n1 n2  Zα
 X 1− X 2 25−23 .5

√ √
σ 2 σ 2 25 25
1
+
2
+
Zcal = n1 n2 = 200 250 = 3.16
Ztab = Zα = Z0.05 = 1.645
Therefore, we reject Ho, which means that the average age of graduation
decreases from that of 1990 significantly.

4.6.2. Non- normal population, large samples

Here, since samples are large, by central limit theorem the sample mean
difference is approximately normal, and hence the test statistic and critical
regions for the various tests in this case is the same as above. It must be
σ σ
noted that in this case results are only approximate. If 12 and 22 are unknown
they can be replaced with sample variances S 12 and S22 respectively, where
1 1
n −1 ∑ ( X 1 −X 1 )2
n −1 ∑ ( X 2 i −X 2 )2
S12 = 1 i
, S22 = 2 and X 1 and X 2 are sample
means of the first and the second populations respectively.

4.6.3. Normal population, variances unknown, small samples

The objective again is to test hypotheses about difference of means of two
populations in this case. We have two cases here.

σ σ
Case1: When population variances are equal, which is, 12 = σ 2.
22 =

2
Because the population variances are equal, σ is logically estimated by the
pooled sample variances, Sp2, given as
( n1 −1) S 2 +( n 2−1 ) S 2
1 2

Sp2 = n1 +n 2−1
2 2
σ 1 σ 2
+
And ( X 1 −X 2 )  t (1 -2, n1 n 2 ).

Then,
 ( X 1 −X 2 )−( μ1 −μ2 )

Sp
√ 1 1
+
n1 n2  t (n1 + n2 – 2)
Suppose we wish to test
Ho: 1 = 2
Ha: 1  2
X 1 −X 2

Under Ho,
Sp
√ 1 1
+
n1 n2  t (n +n – 2)
1 2

Thus the critical region for the above test is

X 1 −X 2

Sp
√ 1 1
+
n1 n2  t (n +n – 2)
 1 2

If Ho: 1 = 2
Ha: 1  2
Critical region is
X 1 −X 2

Sp
√ 1 1
+
n1 n2  - t (n +n – 2)
 1 2

Similarly for the test

Ho: 1 = 2
Ha: 1  2
Critical region is
X 1 −X 2


Sp
√ 1 1
+
n1 n2   - tα 2 (n +n – 2)
1 2

Example 7
A transport company wants to compare the fuel efficiency of two types of
lorries it operates. It obtains data from samples of the two types of lorries, A
and B with the following result.
Lorry type
A B
Average mileage per liter 10.9 10.0
Standard deviation 2.3 2.0
Sample size 20 25

The population variances, although unknown, are assumed to be equal.

Assuming normal population, test at 1% level of significance if there is
significance difference in the fuel efficiency between the two lorries.
Solution
The test could be put as
Ho: A = B
Ha: A ≠ B
X A −X B

The critical region is │

Sp
√ 1 1
+
n A n B │  t (n +n – 2)
α/2 1 2

(19 )(5 . 29)+(24 )(4 )

Sp2 = 20+25−2 = 4.69
X A −X B

tcal = │
Sp
√ 1 1
+
n A n B │ = 1.39

At  = 0.05, ttab = t (43) = 2.01

0.025

Accept Ho. Therefore, there is no significance difference in fuel efficiency

between the two lorries.
 σ σ
Case2: When population variances are not equal, which is, 12  22 .

( X 1 −X 2 )−( μ1 −μ2 )

√
s s
12 22
+
Here, it can be shown that, n1 n2 (approx.) t ()

[ ]
2
S 2 S 2
1 2
+
n1 n2

( ) ( )
2 2
S 2 S 2
1 2
n1 n2
+
where, the degree of freedom =  = n1 −1 n 2−1

Suppose that Ho: 1 = 2

Ha: 1  2
X 1− X 2

√
s 2 s 2
1 2
+
Under Ho, n1 n2 (Approx.) t ()

X 1− X 2

√
s 2 s 2
1 2
+
Critical region is, n1 n2  t ()

Similarly the critical region for the tests with alternative hypotheses
Ha: 1  2 and Ha: 1  2 respectively are
X 1− X 2

√
s 2 s 2
1 2
+
n1 n2  - t ()


X 1− X 2

√
s 2 s 2
1 2
+
And  n1 n2   tα 2 ()