19.

The business
busine. records.
against the physical inventories manager of a la/8 He true proportion of the i n a c c u r a tthe
e
sample
Sampling error should not be moreaby sample
a
sample surve below
survey.
past experiences.
Determine
The than 5% above
size.
proportion of the inaccurate
e 5% above or
records is estimatea 25o J
or

Solution. In the
notations
we a r eg i v e n :
0-35
of $ 16-5
[ (16-60)J
r e c o r d s =

= n
inaccurate
P= of
Population proportion

E =5% =0-05
Confidence
From standard coefficient
= 1 -a1
normal tables, we know that
P(1ZIS 3) =0-9973 1 3. Substituting in (16-60). w
get
Zlies between the limits Hence zal2=
almost surely. 2-0475 819
0-0025
n 2_0.35 x (1 -0-35)x 9 O 5 S
E2 (0-05)
16-6. TESTING OF HYPOTHESIS
Inductive inference is based on
deciding the characteristics of t .
about

POpulation on the hoc or an element Or risk, the risk of takin

Wrong decisions. For sexample, sample study. Such decisions involve,
pharmaceutical concern may re or inducing sleep; we may wantallyto
a
effective for the particular aliment, say, in reducing bl00d pi
decid Y C n toodstuff really is eioht: or which
effectiyve in increasing weight;
which of
of the two brands af.
of a
P a r product, say, foodstuffs, fertilisers, etc., is more effective. It is here that the modern theory of
at th p a y s a very vital role in decision-making and the branch of Staisics wnicn neips us in arivine
i n i t t onfor such decisions is known as testing of hypothesis. The theory of testing of hypothesis was
niated by J. Neyman and E.S. Pearson and emplovs statistical techniques to arrive at decisions in certain
a o n s where there is an element of uncertainty on the basis of a sample whose sizZe is ixed in advance.
i s another technique, known as Sequential Testing. propounded by ADranamwala, in
which sample
S1ze is not fixed but is regarded as a random variable. Here, we shall discuss only the approach
and Pearson. We
of Neyman
explain below certain fundamental concepts associated with testing of hypothesis.
16-6-1. Statistical
Hypothesis Simple and Composite. A statistical hypothesis is some assumption
-

Or siatement, which may or may not be true, about a population or equivalently about the
distribution characterising the probability
given population, which we want to test on the basis of the evidence from a
random sample.
If the hypothesis completely specifies the population, then it is known as simple
hypothesis, otherwise it is known as
composite hypothesis.
In sampling from a normal population N(L, O*), the
hypothesis
H:=Ho and o2=O0?,
is simple hypothesis since it completely specifies the distribution.
a
On the other hand, each of te
following hypothesis is a composite hypothesis:
) =Ho (Nothing known about o)
.

(i) = Oo7. (H not specified)

(ii) u< Ho, O= o02
(iv) >Ho, o = o,?
(v) u= Ho, O> o02 and so on.
 THEOAY O F E S T I
ESTIMATION AND TESTING OF HYPOTHESIS 16-27

statistical hypothesis is a two-action decision after observing a random sample from the
o fa
A 1etation, the two actions being the acceptance or rejection of the hypothesis under consideration
en pop falsity of a statistical hypothesis is thus based on the information contained in the sample
he truthhe
may be
consistent or inconsistent with the hypothesis and accordingly, the hypothesis may be
iected. It should be clearly borne in mind that the acceptance of a statistical hypothesis is due
eceptedt evidence provided by the sample to reject it and does not necessarily imply that it is true.
1 oi n s u f i c e

Tests of ignificance. of the sampling distribution of a statistic, it is

From the knowledge
66-2. value of the
10find the probability that a sample statistic would differ from a given hypothetical
more than a certain amount and hence to answer the question
possir or from another sampie value, by difference between two
paraneance. Accordingly, a procedure to assess the significance of a statistic or
of significance.
We say that
o fs i g n i f i c a n

statisticsis known
as test
ndent statistic and the
e difference
between a corresponding population parameter, or
betrween two independent statistics,
The difference
(i) be attributed to the fluctuations of sampling, otherwise it is said
to be significant.
nificant if it can
Statistics like Binomial, Poisson, Beta, Gamma 1, F, X
i sn o ts i g n i f i c a n t

mOst of the theoretical distributions in

values of we can use the Normal Test of significance |See
rm to Normal distributionfor large for n,
small n, usually less than 30, the exact
etc. c o n f
Sample Tests] large samples. However, for
Chapter 17 Largextests -
SeeChapters 18 and 19] are used.
tests , P,
sample makes the
samples from the given population
The random selection of the
.6-3. Null Hypothesis. a
applying any test of significance we first set up a hypothesis
significance valid
ofs of statement
for
us. kor Such a statistical hypothesis, which is
under test, 1s
tests
about the population parameter(s).
definite
denoted by Ho. In the
difference and hence is called Null hypothesis. It is usually
of no under
l a bvpothesis
Fisher: "Null hypothesis is the hypothesis which is tested for possible rejection
rds of Prof. A.R.
that it is true".
the assumption first and the foremost problem in any testing
procedure is the
Hypothesis? The
a Null difference.
How to set up taken as a hypothesis of no
hypothesis. As the name suggests, it is always of
the null the outcome
setting up of statistician should always adopt the neutral
or null attitude regarding
the
maker or statistician before drawing
The decision attitude on the part of the
neutral or null (no difference) be borne in mind in setting
the null
the sample..This The following points may
basis of the null hypothesis.
sample is the
hypothesis: form.
hypothesis to be tested in the symbolic
Express the claim
or
1. (a) is not true.
Write the symbolic form if the original claim or hypothesis
(6) as explained below
hypothesis and the alternative hypothesis
(c)Identifythe null the null hypothesis (H%)
above, take the expression involving equality sign as
In steps (a) and (b)
and the other as the alternative hypothesis (H).
can sometimes be regarded
as

of the original claim, the original claim the equality

Thus, depending on the wording can be regarded
as H, (if it does
not contain
and sometimes it
Ho (if it contains equality sign)
Sign). while the alternative
is expressed as an equality e.g.,
Ho: 0= ¬p.
Remark. Usually, the null hypothesis
like
hypotheses are in the form of strict inequality
or H:0#0.
H:0> or H:0<ep and so on.]
could be mean ( 4), proportion (
P), variance (0 )
Ihe parameter
If we want to test, ay,
it suffices to test
Ho:0spagainst H: 0> 6,
Ho:0=0against H : 0>6g, H1:0>0), it obviously
(i.e., accept
because provides sufficient evidence reject Ho: 0=0, to
e data
Provides sufficient evidence to reject 0 < do
16-28
FUNDAMENTALS OF
Hence, null hypothesis is specified as the value of the parameter, nearest tc the one
sided
STATISTIC;
hypothesis.
2. If we want to test the significance ofthe difference between a statistic and the nar.. al
two sample statistics, then we set up the null hypothesis H that the diference is not significn be
ternalve
that the difference is just due to fluctuations of sampling.
mcan
For example. if we want to test if a particular drug is effective, we shalltake a neutral aui.
up the hypothesis that it is not effective. For testing if out of the two foodstuffs, one is better the de and se
we shall set-up the null hypothesis that there is no difference between them. nan the othe
3.1fwe want to test any statement about the population, we set-up the null hypothesis that in
For example, if we want to find if the population mean has specified value Ho. then we set-un thre
nul
hypothesis Ho =HLo
4. To test the significance of any statistic i, we compute the standardised variate [ef (16.38)]

Z-S.E.
For any statistic t, the values of E) and S.E.() are invariably in terms of the population parameters
. (16-58)

For example, for the statistic t , E (T) =4 and S.E. ( X) =d/vn and for the statistic t =p sample
proportion), E(p) = P and S.E. (p) = V PQIn. Hence, if the problem is not of the type described in (1) and
above, sometimes we set up the null hypothesis in such a way that it completely specifies the parameters

neededfor the computation ofZ defined in (16-58).

Illustrations
1. The statement by an automobile battery manufacturer
that his battery will last on the average for at
least 48 months is a statistical hypothesis, because the manufacturer does not check the life of each battery

he produces.
claim is tested. The statistical
If there complaints from the consumers, then the manufacturer's
are
hypothesis that is to be tested is called the null hypothesis, and is denoted by Ho
manufacturer's claim we will test_the statistical
example, to prove or disprove the battery
For contains
null hypotheses is (Ho) : 248 months. (Note that this
hypothesis that u248 months. Thus, the
equality sign also).
The alternative hypothesis, (when Ho is not true) is : H: < 48 months.
2. In testing if the die (dice) is (are) unbiased, we set up the null hypothesis that the die (dice) is (are)
17-lj
unbiased, otherwise we ill position to estimate the population parameters. [See Example
not be in a
1s

16-64. Alternative Hypothesis. Any hypothesis which is complementary to the null hypothesis
denoted by H1. It is very important to explicitly state ne
called a n alternative hypothesis. It is usually of Ho
null hypothesis Ho, because the acceptance or rejection
alternative hypothesis in respect of any the
meaningful only if it is being tested against a rival hypothesis. For example, if we want to
test nul
has specified mean Ho. (say), i.e.,
hypothesis that the population
a

Ho =Ho
could be
then the alternative hypothesis
(i) Hi: # Ho» (i.e., u> Ho or u<uHo)
ii) H: u>Ho
(iii) H: H<Ho
The alternative hypothesis in () is known as atwo-tailed alternative and the alternativesin (i) ss of les"

known as right-tailed and left-tauled alernatves respectively. Accordingly,

the corresponding
are
significanceare called two-tailed, rignt-tailed and lett-tailed tests respectively.
sists
16-6-5. Types of Errors in Testing of Hypothesis. As already stated, the inductive inference Co1AS
fron
Or reject n u l nypothesis (H%) after inspecting only a sample
to accept a
in arriving at a decision
T H E O R Y O F

STIMATION AND
E S T I M A T I O

TESTING OF HYPOTHESIS
16-29
of risk the risk
o f risk
taking of
an
e l e m e n t

tually disjoint and exhaustive wrongare

decisions decisions is involved. In any test
such,
procedure, the
rou
tually it is not true,
i.e., when Ho is false.
Accept Ho when it is true.
when it is true.
(i)RejectHo
Accept Ho when it is false.
i)
eisions in (i) and (u) are correct
The decisions

be expresse in the
decisions while the decisions (it) and (iv) are wrong decisions.
These
isioris may following
dichotomous table.
Decision from Sample
Reject Ho Accept Ho
True State Ho True Wrong (Type I Error) Correct
Ho False (H, True) Correct Wrong (Type II Error)
Thats. in testing of hypothesis we are likely to commit two types of errors. The error of rejecting Ho
is known as Type I Error and the error of accepting Ho when H is false (i.e., H is true) 1s
when Ho is trueis
as Type
lI Error.
known
II
Remark. Type I and Type Errors:
We make type I error by rejecting a true null hypothesis.
We make type II error by accepting a wrong null hypothesis.
Ifwe write

P Reject Ho when it is true =P\Type1Error =a

. (16-59)
and
PAccept Howhen it is wrong =P\Type l Error= B
then d and B are also called the sizes of type I error and type II error respectively.
In the terminology of Industrial Quality Control while inspecting the quality of a manufactured lot, the type
I eror amounts to rejecting a good lot and type II error amounts to accepting a bad lot. Accordingly,

a=
PRejecting a goodlot .. (16.59a)

and B=P Accepting abadlot

The sizes of type I and type II errors are also known as producer's risk and consumer's risk

respectively.
these errors. But
An ideal test procedure would be one which is so planned as to safeguard against both
these errors simultaneously. An
practically, in any given problem, it is not possible to minimise both
attempt to decrease a results in an increase in p and
vice-versa. In practice, in most of decision-making
than to reject a
problems in business and social sciences, it is more risky to accept a wrong hypothesis
correct one, i.e., consequences of type II error are likely to be more
serious than the consequences of type I
be reduced simultaneously, a compromise is made by
eror. Since for a given sample, both the errors cannot serious error. Thus, we fix a, the size of type I error
less
nimising more serious errors after fixing up the
and then try to obtain a criterion which minimises B, the size of type lI error. We have:

B P[ Type II Error ]
=P[ Accepting Ho when H, is false or H is true ]
when Hg is truej =1
Now P[Accept Ho when Ho is wrong] + P[Accept Ho
P[Accept Ho when Ho is true]
=1-P[Accept Ho when Ho is wrong
. (16-60)
=1-B
16-30 FUNDAMENTAL: OF STATIS
Obviously, when Ho is true, it ought to be accepted. Hence, minimising ß amon
(1-B), which is called the power of the test. Hence, the usual practice in testing of hun a
the size of type I error and then try to obtain a criterion which minimizes B, the si esisis is
to
maximizes (1-B), the power ofthe test.
e \l error fu t
16-6-6. Level of Significance. The maximum size of the typel eror, which we are
Are prepared t
known as the level ofsignificance. It is usually denoted by and is given by:
a
P[ Rejecting Ho when Ho is true =

Commonly used levels ofsignificance in practice are 5% (0-05) and 1% (0-01). If we ad.
significance, it implies that in 5 samples out of 100, we are likely to reject a correct Ho. In ot
66
lel
evel
implies that we are 95% confident that our decision to reject Hg is correct. Level other wOrds, t
of significance
s alwan
fixed in advance before collecting the sanple information.
Remark. When we reject a null hypothesis Ho we have certain confidence in our deci
depends on the level of significance employed. Thus, at 'a level of significance, the degree ofision
co
in our decision is (1 is also calied the confidence coefficient. However,
-a),which whenwe acc
don't have any confidence in our decision. The acceptance is merely due to the fact that
that the san the
does not provide us sufficient evidence against the null bypothesis, sample dat:
16-67. Critical Region. Suppose we take several samples of the same size ffrom a
given populati
andcompute some statistic t, (say , p etc.), for each ofthese samples. Let h, h be the valuesof
statistic for these samples. Each of these values may be used to test some null hypothesis H Someal
may lead to the rejection of Ho while others may lead to acceptance of Ho. These sample statistics t
(comprising the sample space), may be dividedinto two mutually disjoint groups, one leadine tak
rejection of H, and the other leading to acceptance of Ho. The statistics which lead to the rcjeetionof
give us a region called Critical Region (C) or Rejection Region (R), while those which lead tat
acceptance of Ho give us a region called Acceptance Region (A). Thus, if the statistic 1 e C, Hh is reie
and if t e A, Ho is accepted. jected
The sizes of type I and type Il errors in terms ofthe critical region are defined below:
a =
P[Rejecting Ho when H, is true]
PRejecting H, IHol
= P[te CIHl
and B P[Accepting Ho when Ho is wrong]
=PlAccepting Ho when H, is true]
=P[Accepting HolH]
Plte AlH
where C is the critical (rejection)
region, Ais acceptance region and CnA =0, CUA = S (sample space).
Example 16-20. In order to test whether a coin is perfect, it is tossed 5 times. The null hypothesis f
perfectnes is rejected if and only if more than 4 heads are obtained. Obtain the
(i) critical region, (ü) probability of type I error, and
(ii) probability of type Il error, when the
corresponding probability of getting a head is 0-2.
Solution. Let X denote the number of heads obtained is 5
tosses of a coin.
Ho: The coin is perfect i.e., unbiased i.e., Ho:
p=
Under Ho X-Bn =5.p=
P(X=x IH%)="C,p*q*-i=c G
P(X=x lHo) =C,
zxC,: x=0, 1,...5 ()
 aF ESTIMATION AND TESTING OF HYPOTHESIS 16-31
THEORY OOF

or Rejection Region
Critical Region
Reject Ho if more than 4 heads are obtained.
Critical region {X>4)
= =
{ X=5}
Probability of type I error (a) is given by:
i)
a = P [Reject HolHol P[X =5 IHol =Xx C
=
=
=0-03125 [From ()D

ti The probability of type lI eTror (B) is given by

3 =P [Accept Hol H,l= 1-P[Rcject HoIH]
1- P [X 5 Ip 02] 1- FCp*
= = =
q$-*],x= 5.p -02
= 1 - 5Cs (0-2) = 1 - 0-00032=0:99968.
=

ample 1621. In order to testwhether a coin is perfect, it is tossed 5 times. The null hypothesis of
perfec
o f the coin is accepted if and only if atmost 3 heads are obrained. Then the power of the test
to the alternative hypothesis that probability of head is 0-4, is
c o r r e s p o n d i n g .

2853 56 (iii)3125
(iv) none of these.
3125 3125

I.C.W.A. (Intermediate), June 2000]

the number of heads in =5 tosses of a coin and let p =
Probability of a head in
Solution. Let X denote
n

of the coin.
a random toss
Null hypothesis..Ho:p=.
Alternative hypothesis. H1
: p = 0-4 Critical Region: X> 3.
Power test for testing Ho
of the against H1 is given by
1-B =P(Reject Ho when H is truc) =P(Rcject HolH)
5

=
P(X>3 Ip =04) 2 C, (0-4y (0-6)- =
rz4
[:x-B(n 5,p =0-4 under H)] =

=C (O4*(o6)+(04=5x(10x0+(0
=( 3 + ] - s
Hence (i) is the correct answer.
is represented by a portion
and Two-Tailed Tests. In any test, the critical region
16-68. One-Tailed the test statistic.
of the sampling distribution of
of the area under the probability curve or left
where the alternative hypothesis is one-tailed (right-tailed
A test of any statistical hypothesis
test. For example, a test
for testing the mean of a population
tailed) is called a one-tailed
Ho:=Ho
against the alternative hypothesis
H : >Ho (Right-tailed)
or H:4<Ho (Left-tailed).
the critical region lies entirely in the right-tail of
test (H1: > Ho)
S asingle-tailed test. In the right-tailed Ho), the critical region is entirely in the
while for the left-tailed
test (H :u <
ne
Sampling distribution of ,
left tail ofthe distribution of .
two-tailed such as:
where the alternative hypothesis is
A test
of statistical hypothesis
Ho:=Ho
against the alternative hypothesis
and <Ho)
H: * Ho ( >Ho
 FUNDAMENTALS OF STATISTICS
16-32
area lying in
case the critical
region is given by the portion of the
Is known as two-tailed test and in such a statistic.
both the tails of the probability curve
of the test
entirely on the
whether one-tailed or tw0-taicd test is
to
be applied depends
two-tailed test and
In a particular problem, we apply
alternative hypotnesis iswo-tailed,
nature of the altermative hypothesis. If the one-tailed test.
one-tailed, we apply manufactured by standar
the alternative hypothesis is brands or bulbs, one
i there are two population
technique (with m e a n
life ua. If.
For example, suppose that by s o m e new we
the other manufactured and alternat
process (with mean life ) and
then our null hypothesis is Ho :
H H2 =

differ significantly, to test if the hul

However, if we want
want to test if the bulbs two-tailed test.
thus giving us a
produced by
standard process, then we ha
hypothesis will be H: H# H2, life than those
average
produced by new process have higher H:4<
and
the new process is inferior to that
of
.

if the product of
test. Similarly, for testing
thus giving us a left-tailed
standard process, we have: and H:>
Ho: H a two-tailed test or a single-tailed (s .
decision about applying right
right-tailed test. Thus, the
thus giving us a under study.
on the problem value of test statistie
or left) test will depend Critical Region. The
Values and
Significant is called the critical value or sionis
16-6-9. Critical Values
o r

region and the acceptance region

separates the
critical (or rejection)
value. It depends upon :
used and
(i) The level of significance whether it is two-tailed or single-tailed.
(i) The alternative hypothesis, > 30), the
standardised variable corresponding
to
..

out earlier, for large samples, (n

As has been pointed
the statistic t, viz.,
1-EC)NO, 1),
Z S.E.)
.(16-62

(16-62) under the null hypothesis is known as tet

asymptotically as n o. The value of Z given by

statistic.
NORMAL PROBABILITY CURVE
CRITICAL VALUES OF STANDARD
that the total area to the ngh
critical value za is determined so
In case of single-tailed alternative, the the left of -za is a (See diagran
left-tailed test, the total area to
of it (for right-tailed test) is a and for
below), i.e., (lbo
a
For Right-tailed Test: P(Z>Za)
=

(lo%
P(Z<-za) = a
For Left-tailed Test Left-Tailed Test
Right-Tailed Test
't)
(Level of significance 'a) (Level of significance

Rejection region Rejection region

(a) (C)

V Z=0
Z=0
Fig. 16-3a Fig. 163b

Let z be the critical value of Z at level of significance a. for a two-tailed test so una

P(1ZI>z) =a
P(Z>z)+P(Z<-z) = a
( B ys y m m e t r y )

P(Z>)+ P(Z>z) =a
THEOR JATION AND TESTING OF HYPOTHESIs
OF ESTIMA 16-33

2P (Z> z1) = a

P(Z>z) =
Z o2 [From (16-63)
that the each tail of the curve is (o/2).
a r e a on Two-Tailed Test
implies
(*) the
critical values for a two tailed test at level of (Level of significance a')
Hence
as shown in Figure 16.4.
significance
a, are tzau2 Lower critical Upper critical
critical value value
Thius, the significant o r of Z for a value
Ssingle-tailedrtest (left or right) at level of significance 'a Rejection region
value of 2 for a two-tailed test at Rejection region
as the critical (/2) P(a2)
same
is 2a.
significance
level of
Z=0 Za2
Fig. 16-4

critical values of Z at commonly used levels of significance, for

both two-tailed and
Wegive below the on using the
obtained from equations (16-63) to (16-65) and
tests. These values have been
single-tailed
normal probability Table VI.
standard
TABLE 16-2: CRITICAL VALUES OF Z
Level of Significance (0)
10%
1% 2% 5%
I Z a n = 1-645
za2=2-58 Zau2=2.33 Zau/2= 196
Two-tailed test Za 1 - 2 8
Za2-05
Za1-645
Za2-33
Right-tailed test 1-28
-Za-2-05 - Za-1-645 Za
-Za-2.33
Left-tailed test

1-96 (for a =
0-05) and 2-58 (for a =0-01) for two-tailed
Remarks 1. The significant
values of Z, viz., test should be committed
to memory.
and 2.33 (or a =
0-01) for right -tailed
and 1-645 (for a = 0-05) the alternativve
test
single-tailed test will be determined by
two-tailed test or
The decision about applying a

hypothesis H1.
then from normal probability Table VI, we get,
2. If Z W(O, 1), (By symmetry)
P(1ZIS3) =P(-3Z<3) =2P ( 0 <Z<3) Tables)
2(0-49865) = 09973 (From Normal Probability

P(IZI>3) = 1 - P(IZIS3) =0-0027 lie within the limits + 3.

The
should a standard normal variable to
Hence, in all probability, we expect
critical or significant
value of at all levels of significance. Z
value of 1Zl= 3, is regarded as in that case
distribution of the test statistic
Z will not be normal and
3. If n is small, then the sampling n small, (usually
less than 30),
values of Z given in Table 16-2. In this case viz., out to be Student's
We can't use the significant distribution of Z, which turns
the significant values based on
the exact sampling The significant values
we use distribution [See Chapters 18
and 19].
-distribution, F-distribution or chi-square (2) in the appendix at the end
of
different values of n and a and are given
F'and y2have been tabulated for
or,
the book.

CRITICAL REGION
at level of significance a is given by:
The critical (rejection) region . . (16-66)
() Z>za, for right-tailed test
test
(i)Z<-zu ie., IZI>za for left-tailed left-tailed) test, the
critical (rejection) region at level of
.(16-66a)
Her single-tailed (right-tailed or
e,Jor
SEgnificance a is 1 ZI> Za
 16-34 FUNDAMENTALS OF STATISTICS
The critical region for fwo-tailed test at level of significance a is given by:
Z> a or Z-an i.e., 1Z1> zan
... (16-66h
16-610. P-Value or Probability Value of Test Statistic
P-Value. The probability that the value oftest statistic is at least as extreme as its comu
the basis of the sample data under Ho, is called its P-value. uted value on
1. For the right tailed test, the P-value is the area to the right of the computed val
ed value of
statistic under Ho. IFigure 16:5(a)] the tcst
2. For the left tailed test, the P-value is the area to the left of the computed value of the ...
under Ho. Figure 16-5(6)). statistic
Right-Tailed Test Left-Tailed Test

P-value P-value

W
0
Test Statistic
Test Statistic

Fig. 165(a)
Fig. 165(6)

3. For the two-tailed test, P-value is:

(a) twice the area to the left of the computed value of test statistic under Ho, if it is negati
[Figure 16-6a)] or
gative
(b) twice the area to the right of the computed value of test statistic under Ho, if it is posjitive
[Figure 16-66)].
The P-value for two-tailed test is twice the area on either tail (left or right) of the computed value of
test statistic under Ho
Two-Tailed Test Two-Tailed Test

P-value P-value
= Twice this area Twice this area

Test Statistic (Negative) Test Statistic (Positive)

Fig. 16-6@) Fig. 16-6(6)

16-7. PROCEDURE FOR TESTING OF HYPOTHESIS

We can use any one of the following two methods for testing of a statistical hypothesis.
1. Rejection Region Method
2. P-value Testing Method
We summarize below the various steps in testing of a statistical hypothesis in a systematic manner, ln

each of these methods.

Method 1. Rejection Region Method
Step 1: Set up the null hypothesis.
Step 2: Set up thealternative hypothesis.
[For detailed discussion on setting up of null and alternative hypotheses, see a 166:3)
 N AND TESTING OF HYPOTHESIs
O FESTIMATION
16-35
THEORY
tternative hypothesis will enable us to decide whether we have to use a single-tailed (right
Alte
test r two-tailed test.
or left-tailed)
evel of Significance. ChoOse the appropriate level of significance (a) depending on the
Step missible risk. a is fixed in advance, before drawing of the sample.
pern

S t e p4 :
Tdentify the sample statistic to be used and its sampling distribution.
Identi

Test
statistic. Define and compute the
test statistic under
Step 5 Ho
Some of the commonly used distributions in obtaming the test statistic or test criterion are
normal, 1, chi-square, and F, and tests based on them are discussed in chapters 17 to I

Obtain the critical value (or values) and the critical (rejection) region of the test statistic from
Step the appropriate tables, given in the appendix at the end of the book.
value of the test statistic lies in the level of
Giep 7: If thecomputed
significance 'a'.
rejection region, we reject Ho at

If the computed value of test statistic lies outside the rejection region, we fail to reject Ho
It should be clearly understood that whenever we reject a null hypoihesis, we have certain
degree of confidencein our decision which depends on the level of significance used.Thus
at a level of significance, the degree of confidence in our decision to reject Ho is (I - a),
which is also called the confidence coefficient. However, the failure to reject H o is not the
same thing as 'accepting Ho'. By this we simply mean that the sample data does not provide
us sufficient evidence against Ho. which may, therefore;be regarded as true.
Step 8: Write the conclusion of the test in simple language.
Remark. If the calculated value of test statistic is greater than the tabulated value (critical or
sionificant value at level of significance ' ) , then we say that it is significant and the null hypothesis Ho is
gn
rejected at leve of significance o i.e., with confidence coefficient (1 - a).

If calculated value of test statistic is less than the tabulated (critical) value, we say that it is not
Significant. By this we mean that the difference between the sample statistic and the corresponding
narameter value under Ho, is just due to the fluctuations of sampling and the sample data does not provid
us sufficient evidence against the null hypothesis, which may therefore, be accepted at 'a' level of
significance.
Method 2:P-VALUE ESTIMATION METHOD
Steps 1 to 5 and Step 8 are the same as in Method1- The Rejection Region Method'. There is slight
variation in Steps 6 and 7, as given below:
Step 6: Find the P-value of the computed test statistic under Ho in Step 5.
[For details, see § 16-6-10]
Step 7: If P-value < oa, we reject Ho at 'o level of significance.
If P-value>a, we fail to reject Ho at 'a' level of significance.
Step 8: Write conclusion of the test in simple language.
Remark. In chapters 17 to 19, we will be using the method 1 i.e., the Rejection Region Method. In all
nese chapters the reader is advised to strictly follow the order of the steps explained above.

EXERCISE 16-1
. Discuss briefly the importance of estimation theory in decision making in the face of uncertainty.
2.
Explain, with illustrations, the concept of
Point Estimation and (i) Interval Estimation.
. Describe the important properties of a good estimator,
Define the following terms and give one example for each.
i) Unbiased Statistic, (i) Consistent Statistic,
(iii) Efficient Statistic and (iv) Sufficient Statistic.