TESTING OF SIGNIFICANCE

13
13.1.Introduction
of sample
In practice we make decision about populations on the basis
and
information. In this chapter an assumption is made on the populaion
the sample informations the
by going through some statistical analysis of
the basis
validity of the assumption is tested. For example, we decide on
of the sample data whether a new drug is really effective in curing
a

disease.

13.2. Statistical Hypothesis

Any assumption taken on a population, regarding its probability
distribution or its is called a Statistical Hypothesis'. Such a
parameters,
Hypothesis may or may not be true.
For example, let we that the mean of a population is 60. Then
assume

"H= 60 is a statistical hypothesis. Drawing samples from this population

we can test the validity of this hypothesis and this is test of hypothesis.
There are two types of hypothesis. These are discussed below

Simple Hypothesis
A Statistical Hypothesis which specifies the probability distribution and
all related parameters of a population is called simple hypothesis.
Illustration: Consider the population of lifetime of electric bulbs
manufactured by a company. Let the lifetime (denoted by X) be normally
distributed with standard deviation 4. We have to test its mean. Let
we

assume the mean = 600 hours and we shall test its validity. Then we
see under this hypothesis the entire character of the population is specified.

Sothis hypothesis
"

= 60" is simple hypothesis.

Composite Hypothesis
A Statistical Hypothesis which does not specify the population

completely is called composite hypothesis.

Illustration Consider the population of weekly wages of the workers
of a big industry. Let the population be normally distributed whose mean
and s.d are unknown. We have to test its mean. Suppose we assume its
mean = 60 and go to test its validity.
 ENGINEERING MATHEMATICS-II
474
the entire character of the
unknown.population
hypothesis
under this nlpopula
we
see
deviation remains
1s
Ther
hen
because the
standard
So th
is not
known
hypothesis. this
"l=60" is a composite
hypothesis

Null Hypothesis

starts with
sis. AA statist
a statistical Hypothesis,
Test of Hypothesis is tested on
acceptance or rejection the
whose possible
hypothesis d asi
a Null Hypothesis. Usually itit is
is
of sample
observation is called denoted by
Ho "the mean of a population
Tllustration. Suppose
we assume that
is 40
population has mean 38 R.
Let a random sample drawn
from this
By going
through some statistical analysis we have to test whether our assumption

may be accepted or rejected. Then the assumption =40 sa Null

Hypothesis; in
notation Ho(u= 40).
Alternative Hypothesis.
A statistical hypothesis which is different from the Null Hypothesisis

called Altermative Hypothesis. It is denoted by H.

Tllustration. Consider the null hypothesis Ho(u = 60). Then H(u 60)

is alternative hypothesis. Again if it is seen from the fact that thereis no

chance of being less than 60 then we may also take Hi(u>60) as
an Altenative Hypothesis to Ho(u=60). Similarly H(u < 60) may
also considered as Altenative Hypothesis.
Both Sided Alternative Hypothesis.
Let Ho(0=0,) be a null hypothesis where 0 is
parameter. Ihan a

thealternative hypothesis H(0+00) is called both sided alternative

hypothesis.
For example, the alternative hypothesis H(60) is both su
alternative hypothesis against the null hypothesis Hou= 60), w
is the mean of the
population.
One Sided Alternative
Let Hypothesis.
Ho0=0o) be a null hypothesis where 0 is Then

the alternative hypothesis

parameter
a

H(0>0,) paramai
is called one sided (Right).
ernative

Hypothesis. The alternative hypothesis (RIg sided(Lell

Alternative Hypothesis. H(0<) is called one (L
TESTING OF SIGNIFIANCE 475

one sided
For example the alternative hypothesis H(u>60) is
alternative hypothesis against the null hypothesis Hou=60).

13.3. Test Statistie.

we
To test the possible acceptance or rejection of a null hypothesis
have to take the help of
a stanstic whose
sampling distribution is known.
By evaluating the value
or the
statistic we decide whether the Null
Hypothesis is to be rejected not. This statistic is called Test-Statistic.
or

llustration. Let Ho(H=00) be a Null Hypothesis where is the

population mean. Let the population be normally distributed with s.d.
G=3. We consider the statistic , the sample mean. Its sampling

distribution is knowWn to be normal with mean 60 and s.d. Ewhere

15 is the size of the sample. We compute the value of for a sample

decide
drawn from this population and from this computed value we
whether Ho is to be accepted or not. So here is the test statistic.
Note: Selection of test statistic depends on the character of population,
sample and the Null Hypothesis.
and Level of
13.4. Critical Region, Region of Acceptance
Significance.
statistic by
Let Ho be null hypothesis and t be the appropriate test
a
This decision is
which we decide whether Ho would be accepted or not.
0.05
based on probability consideration. Usually a lower probability a (e.g
If
or 0.01 taken under consideration and let P(aStsb)=1-a.
etc) is
decide Ho
a computed value oft
lies within this interval (a,b) then we
of Significance" or 10a% Level of *

is accepted at "a Level

is called Region of Acceptance
Significance". The interval (a,b)
"
Level of Significance".
corresponding to a

the interval (a,b) we decide Ho

If a computed value of t lies beyond
or 100 a % Level of
is rejected at a Level of Significance" "Critical
the interval (a,b) is called
Significance", This region beyond
test at a Level of Significance".
Region" (CR) of the
region is not unique
Note: (1) The region of acceptance/Critical
corresponding to a Level of Significance'
 ENGINEERIN MATILEMATICS-11
476
(2) Thus ifa coniputed alue of the appropriate test statistic t falls in a

then lo 1S rejected.
Critical Region
value of r in CR' is co
comple
lling of the computed
cceptance mentary
7The
(3)
computed value
of t in Region of ary to
the 'alling ofthe
that a computed value of lies within the Critical
t
Thus, Probability
el-Probability that computed a value of t lies withinthe Region
Region of
Acoeptance =1-P{a<t <b) =1-(l-a)
=a.
(4) It would be seen in the latter section of this chapter that
that when the
pdfofthe test statistic tis symmetric about 0 it is
best to take th he
in above) if both sided gon
of acceptanceas(-a,a) (i.e. a Iternative
sided.
hypothesis IS taken both
(5) It would be seen in the latter sections of this chapter that it is
est to
take the region of acceptance as-,a) (i.e. C.R as
s t>a) ifthe
t>a) if the
altermative hypothesis is taken right sided and (a,co) (i.e. C.R. as t<al
<a)
ifthe alternative hypothesis is taken left sided.
Illustrations. () Let be the mean of the population of lifetimes of
clectric bulbs manufactured by a company. Let the population has normal
distribution with s.d 3.30. We want to test the null hypothesis
Holu=17117) by drawing samples of size 400 from the population. We
know the sampling distribution of the sample mean X 1S
normally
distributed with mean 171.17 and s.d = 3.30 0.165. We select
V400 as
thetest statistic.Now z =17117 is standard normal variate.
0.165 We take
the alternative hypothesis as H 171.17), both sided. Since
P-196 <2<1.96)=0.95=1-005 (obtained from statistical
table) so
is best to consider(-196,196) as the
of significance and the region region of acceptance at o level
critical region at 5% level of beyond it. i.e. the region E196 as the
2190 asction
would be discussed significance. (The reason of sucn
later)
Now if it is seen that
a random sample (of size 400) gives its mean
gve
= 17138 then we
compute the value of z/1.38-171.I7127.
0.165
 TESTING OF SIGNIFIANCE 477

So we
We this value lies in the region of acceptance(-1.96, 1.96).
see

may conclude Ho is
accepted at 5% level of significance i.e. the average
1his
lifetime of the bulbs manutactured by the Company may be 171.17,
decision of us is 95% true.
Since the region of acceptance determined by z,

-1.96<2<1.96-196< 1.96 170.85<r<17149,

0.165
so the region of acceptance determined by the test statistic 1s

(17085,171.49) consequently the CR determined by is X<l70.85

together with > 17149.
13.5. Type I Error and Type II Error [WB.U.Tech 2005]
A null hypothesis Ho the basis of sample values only. For
is tested on
this reason it has no gurantee that we always take right decision regarding
error in taking
acceptance or rejection of Ho. There may occur an
decision. Two types of Error may exist:

Type I Error: This error is made when a Null Hypothesis is rejected

though it was really true.
II Error: This is made when a Null Hypothesis is accepted
Type error

though it was really false.

test of hypothesis, the testing should
Note: n order to ensure a good
II error. This is not
be designed so as to minimize the Type I and Type
size an attempt to decrease one
so simple because for any given sample
an increase in other type of error.
In
type of error is accompanied by
be more serious than the other. Thus itis
practice one type of error may
more serious error.
wise to reach a compromise to limit the
Probability of Type I Error.
and be test statistic. Now, the
Let Ho(0=00) be a Null Hypothesis
of Ho(0=0,) on
probability of Type I error=Probability of rejection
is true.
the hypothesis that Ho
=P (computed valueof tlies in C. R)/0 =0,istrue (this 1S

conditional probability)
= Level of Significance of the test.

This is also called Size of Type Error

 ENGINEERING MATHEMATICS-1IA
478

Let P be the proportion of defective itemslarge lot

in a
ustration. of tems in the lot). Let
of defective items in the lot) +(No.
ie Pe(No. items is drawn from the
Ha(P-0.2)be a null hypothesis. A sample of 8
in the sample
number of defective items (U)
lot Ho is accepted if the
test statistic and the CR is [7, 8]
is s6. Here Pis parameter, /is

Now, PU =r)= Probability of r defective items in the sample of 8

( f is (8, P) binomial variate). So the
articles-C, P" (1-P)
of rejection of Ho on the
probability of Type I error=Probability
is true
hypothesis that Ho(P =0.2)
=PU=7,8) on the hypothesis P=02 e rerd 2
=P(f-7)+ P(f =8) assuming P=02
=C(02) (1-02)C(0.2)'(1-02)*
is also the level of significance of the test.
=0.00008448. This
Probability of Type II Error
be test statistic. Now, the
Let Ho(0=00) be a Null Hypothesis and r
of
probability of Type II error Probability acceptance Ho(0=0o)
= of
on the hypothesis that 6#0,.

any value other than . So the probability of

Now 6 may assume

this type of errordepends upon the fact 'what value is taken by e'. Say
assumes the value 0, (which is 0%) Then probability of Type II
.

Error, assuming 0=0, is the Probability of acceptance of Ho on the

hypothesis 6 assumes the value 0

=P (computed value of / falls in Region of Acceptance)/0 0, =

Illustration. Let P be the proportion of defective items in a large lot,

ie. P= No, of defective items+ total number of items in the lot. Let
Ho(P=02) be a null hypothesis. A sample of 8 items is drawn from the
lot; Hy is accepted if the number of defective item (/) in the sanmple is
S6. Here P isparameter, fis test statistic and the CR is [ 7, 8]. Now

PU=r)= probability of r success in 8 trials =C,Pr(1-P) (:/s

(8, P) binomial variate).
 TESTING OF SIGNIFIANCE 479

assumne
Now if we go to find the probability of Type II error we have to
a value of P other than 0.4. Suppose p=0.1

Probability of Type Il error (when P=0.1 is true)

Probability of acceptance ot tHo assuming p = 0.1

= P(0sfs6) assuming P=0.1

=1-{P(f= 7)+ P(f =8)} assuming p=01

-1-C(01) (09)G(01 (09

:fis (8,0.1) Binomial variate]
= 0.99999927.
Note: From the above two illustrations it is easy to understand that if

probability of Type I error (or Type II error) is given, we can find the
critical region. Students are advised to exercise this reverse process 1.e to

find the CR given that the probability of Typer I error is 0.00008448 in

the previous illustration. This will be helpful to understand the notion of
Best Critical Region which is being discussed below.
13.6. Best Critical Region.
Let Ho(0= 0o) be a null hypothesis which is to be tested, by the test
statistic , against the alternative hypothesis H. The test is a procedure
on the choice of a region of acceptance or a
Critical Region. It
depending
select critical region so as to minimize the
would be good if we can a

I and Type II This is not so simple because one type of error

Type error.
can be seen
is accompanied by an increase in other type of error. (This
from the two previous illustrations)
the region where
To obtain the best test we find a critical region. (e.g
a such that Probability
>a) corresponding to a level significance
of
P (computed value of t lies in
of Type I error relative to this CR is a i.e.
several CR like this.
the C.R assuming 0=00) =a. There may exist
called Best Critical Region for which
Among these that critical region is
or simply
the Type II error is least. We say this is the Best Critical Region
Critical Region (CR) corresponding to a level of significance.
 ENGINEERING MATHEMATICS-IIA

480 Ho(u=52)
null hypothesis
are to test the
Suppose we is the mean of a
llustration.
H(u= 49)
where j
alternative hypothesis drawing
against the with s.d. 5. wo do this by
normal distribuion has
population, having
mean

population. Since th sarmple

25 from the
sample of size
statistic as
distribution we take the test
normal

F-u vn-u)_v25(H)Fu
5

of significance of the test

variate. Take the level
which is standard normal
z=i-52.
as a=0.01. Assuming u
=52 see we

have
From statistical table
we

P(0<z<03)=0.01
P(-900<2<-806) =3159-3051 =01

=4990-4890 01
P2.29<z<3.1)
P-o<z<-2.32) =0
zITTI
Thus the intervals like (0, -2.32
03). (-900, -806), (2.29,
3.1), -,-2.32) are all
0.01 level of significant. There are more
Critical Region corresponding to
CR like these.
CR(-0,-2.32)
for the
Nowit can be shown that among all these CR,
least (this is not shown
theprobability of Type II error assuming =49 is
1.e.
tolengthy calculation). So we conclude the region-o,-2.32)
-02<-2.32 is the Best Critical Region. This is displayed in the figure
by shading the region.
li
Nowif for a sample its mean =50 then the computed value ol
2=50-52=-2 not lying in the best critical region. Hence we accept
Ho as true at 0.01 level of significance. i.e. we think the mean of the
population may be 52 this decision is 99% correct.
 TESTING OF SIGNIFIANCE 481

Power of a Test.

From the previous discussion we see that the goodness ofa test depends
upon the choice of the critical region. As the probability of Type II error
decreases the goodness Or tne test increases. This notion leads to have

thedefinition:
Power ofa Test=1-Probability of Type II Emor assuming H
So if the Best Critical Region can be selected the Power of the Test
becomes highest.
Ilustrative Example.
Example 1. In order to test whether a coin is perfect the coin is tossed 5
times. The null hypothesis of perfectness is rejected if more than 4 heads
are obtained. What is the probability of Type I Error ? Find the probability of
Type II Error when the corresponding probability of head is 02.
[WB.U.Tech 2007]
Let P be the proportion of the number of obtained head among 5 times

ofthrow. The coin is perfect if P= since probability of head for a

is 1/2. So the null hypothesis is Ho P=

Ho is rejected
perfect coin
if the number of obtained heads, f>4.
Now, P =r)= Probability of r heads in 5 tosses
=C,P(1-P[:f is (5,P) Binomial variate].
So the probability of Typer I Error
=
Probability of rejection of Ho
is true Pf=5) on the hypothesis
on the hypothesis that Ho(P =%
pe
II Error when P= 0.2
Now, Probability of Typer
assuming P=0.2
Probability of acceptance Ho
of
assuming P=02
PUs4) assuming P=02 =1-P(f =5)

1-G(02(1-02-099968.
EM-2A-31
 MATHEMATICS
-

IA
ENGINEERING

482
Exercises 1 3
I shall toss it 6 times.
a coin is perfect,
to test whether head
1. (a) In order perfectness if
and only if I get
no

the null hypothesis of

I shallreject the probability of Type I error for my
test2
What is
6 heads.
thrown six times and is
or
unbiasedness of a die it is
(b) To test the Find
are obtained.
than one sixes
accepted that
the unbiasedness if not more
20051
o W . B . U .Tech
theprobability of Type I Eror
Ho(u=7) is tested against i(=6) where
2. The null hypothesis The test is
distribution. Given o = 2.
and S.d. of a normal
4,G are mean size 25 from the given
drawing a random sample of
performed by 0.16 level of significance.
and using the best cirtical region at
population
II
Find the probability of Type
error.

in a large lot of items is p.To

3. proportion of defective items
The
we take a random sample
of 8 items and accept
testthe hypothesis p 0.2, =

number of defectives in the sample

is 6 or less. Find
thehypothesis if the
II error if
probability of Type I error of the test. What is the Type
the
p-03? [W.B.U.Tech.2008]
head when given coin
a
4.Let p denote the probability of gettinga
is rejected in favour
istossed once. Suppose that the hypothesis Ho:p=0.5
Calculate the probability
of H:p=0.6 if 10 trials result in 7 or more heads.
of Type I and Type II error.

5. Given the density function S(x,0)= 0sr0

=0, elsewhere
and thatyou are testing the null hypothesis Ho:0=1 against H:0-2 by
means of a single observed value x. Determine the size (i.e. probabiniy
oT 1ype I and Type II error if you chose the interval 05 sx as the critica
Tegion.
Answer
1.(a) 1/32 (b) 0.263 3. 0.00008448; 0.99870967
4 11 ,1-0.382=0.618 5. 0.5, 0.25
 TESTING O F SIGNIFIANCE

483

Multiple Choice Questions

1. Consider the population of life times of

norm
tyres manufactured by
whose mean and
s.d are
unknown. Then the
a company
kms 1s
assumption averag
life time3600
(a) simple hypothesis
(6) composite hypothesis
(c) alternative hypothesis (d) none of these.
2 Consider the normal population of the body weight all
B.Tech students
in West Bengal whoSe standard deviation is 5.
Then the assumption "the
average bodyweight= 50kg" is a

(a) simple hypothesis (b) composite hypothesis

(c) alternative hypothesis (d) none of these.
3. Simple hypothesis does
(a) specity the population completely

(b) not specify the population completely

(c) simplify the calculation of parameters of the population

(d) none of these.

4. If m be the mean of a population having t distribution. Then the

assumption m=20 is to be tested. Then this is a

(a) alternative hypothesis (b) simple hypothesis

(c) Null hypothesis (d) none of these.

5. An alternative hypothesis
(a) is same as the null hypothesis

(b) different from the null hypothesis

not same as the null hypothesis
(c) may or may
(d) none of these.
which one of the following is
6. If (o3) be the null hypothesis then
alternative hypothesis
(c) (a=0) (d) (o+3)
(a) (a-4) (b) (G=1)
 484
ENGINEERING MATHEMATICS-1IA
7. If u is a parameter and
H(u 5) is null hypothesis, then which
=

of the following is Left sided one

alternative hypothesis
(a) H(u5) (b) H(u<5) (c) H(u5) (d)
8. If is a parameter and
H(u=7) is null
H(u=4)
of the following is Both sided alternaive hypothesis, then which
one
hypothesis
(a) H(u 7)
(b) H(=8)
(c) H(u<7) (d) none of these.
9.If H(> 60) be an alternative hypothesis then
the Null hypothesis is
(a) Ho(u<60)
(b) Ho(u2 60)
(c)HuS 60) (d) Hou= 60)
10. To test the
acceptance of the null hypothesis Ho
=10) for a population having normal distribution with s.d4(population
the test
mean
statistic
(a) sample s.d (b) sample proportion
(c) sample mean (d) none of these.
11. If x be test statistic and
(a,b) is region of acceptence correspending
to 3% level of significance than
P(aSxsb)=
(a) 0.9 (b) 0.97 (c) 0.99 (d) 0.03.
12. Ift be test statistic and (a,b) is the critical
significance then Pla <t <b) = region at 4% level of
(a) 0.04
(b) 0.4
(c) 0.96
(d) none of these.
13,
If-3.9 <t<39 be a region of acceptance in test of
then the critical
region is
a
hypothesis
(a) 1<-3.9
14. In a test of
(b)139()0<t<39 (d) none of these.
hypothesis
hypothesis is rejected if theif(-4.6,6.8)
Null is
region of acceptance then the
computed value of the test statistic 1s
(a) 5 (b) 6
(c)-3 (d)-6
 E S T I N G O F S I G N I F I A N C E

485

the
colnputed
ited val
value of the test
15
Falling of
Falling statisticcritical region'
ofthe
ven Falling of
even the computed value the test statistic
conplement
acceptance.

of
the reg
in the
region
pothesis (-o0,1.02) is critical regionatat 01
hypothesis
a test of level
level of
16. Ifin statisti
test 01 ot
significance
then, ifr is tic,
=0.0
(a) P0<t< 102) (b) P(1.02<t)=0.99
(c) P 0 < t ) = 0 . 9 6
(d) none of these.
TCthe alternative pocsts is taken right sided with test
statistic t
which one
which one of the
followings
then indicate
may be a
possible region of
acceptance

(a) r<a (b) r>a (c) t2a (d) r=a-

19 If the alternative nypotnesIS taken left sided then indicate which
1S
ne of the followings may be a possible critical region
one
(a) r<a (b) 1>a (c) 12a (d)=a
19. In a test of hypothesis lype I Error is committed when

(a) Null hypothesis is rejected though it was really false.

(b) Null hypothesis is rejected though it was really true.

(c) Null hypothesis is accepted though it was really false.

(d) Null hypothesis is rejected though it was really false.

20. In a test of hypothesis Type II error is committed when

(a) Null hypothesis is rejected when it was really false

(b) Null hypothesis is rejected when it was really true

(c) Null hypothesis is accepted when it was really false

(d) none of these. [W.B.U.Tech20071

21. If in a test of hypothesis the null hypothesis Ho is accepted at 2%
levelofsignificance then the probability that the statement in H is false is
(a).2 (b) .02 ().98 (d) none of these.
 ENGINEERING MATHEMATICS-1IA
486

22, If Ho(f=5)
where t is test statistic is null hypothesis then probability

of Type I
error=

g
(a) Probability of acceptance of Ho assuming 11=5
(b) Probability of rejection of H, assuming 1 5 l l pne
of H assuming (=5
(c) Probability of rejection
(45.
(d) Probability of acceptance of Ho assuming
of rejection of
23. If Ho(0= 2) is null hypothesis then the probability
0=2 is true is probability of
Ho though
(b) Type II error
(a) Type I error

of these.
(
(c) Probability of 0 2 (d) none

is same as level of
24. In a test of hypothesis Probability of Type I error

significance of the test.

(b) No. l
(a) Yes
be null hypothesis then the probability of Type II
25. If (0= 10)
error
0= 10
(a) Probability of rejection of H, assuming
010
(b) Probability of rejection of H, assuming
of Ho assuming 0 = 10
(c) Probability of acceptance
0 10
(d)Probability of acceptance of Ho assuming
to a particular level of
26. In a test of hypoi sis corresponding
critical region' has
significance, among all ciical region 'Best
(a) least Type I error (b) least Type II error

(d) greatest Type lI error.

(c) greatest Type I error
27. In a test of hypothesis the critical region corresponding to a parteu

level of significance is unique.

(a) True (b) False.

 TESTING OF SIGNIFIANCE 487

28. If a ull hypothesis is accepted at .05 level of significance then this

decision 1s

(a) 5% correct (b) .05% correct

(b) .95%% correct (d) 95% correct.
29. As the probability ot lype II error decreases the goodness of the

test
(a) decreases (b) increases
(c) does not chang (d) none of these.

Answers

1.b 2.a 3.a 4.c 5.b 6.d 7.b

9.d 10.c 11.b 12.a 13.d 14.d

8.a

17.a 18.a 19.b 20.c 21.b

15.b 16.b

25.d 26.b 27.b 28.d

22.c 23.a 24.a

29.b