Statistical inference

STATISTICS
STATISTICAL INFERENCE
Complementary course
for B.Sc.
MATHEMATICS
III Semester
(2014 Admission)

CU-CBCSS

UNIVERSITY OFCALICUT
SCHOOLOFDISTANCEEDUCATION

Calicut University P.O. Malappuram, Kerala, India 673 635

984
School of Distance Education
2 Statistical Statistical 3
SYLLAB
US

Module 1. Sampling Distributions: Random sample from a population

distribution, Sampling distribution of a statistic, Standard error,
Sampling from a normal population, Sampling distributions of
the sample mean and variance. Chi-square, Student‟s t and F
distributions - derivations, simple properties and inter
relationships. 25 hours

UNIVERSITY OF CALICUT Module 2. Theory of Estimation: Point estimation, Desirable properties

of a good estimator, unbiasedness, consistency, sufficiency,
SCHOOL OF DISTANCE EDUCATION statement of Fisher Neyman factorization criterion,
efficiency. Methods of estimation, method of moments,
Method of maximum likelihood-Properties estimators
STUDY MATERIAL obtained by these methods
25 hours

III Semester Module 3. Interval Estimation: Interval estimates of mean, difference of

means, variance, proportions and difference of proportions,
Large and small sample cases.
Complementary Course 10 hours
for B Sc. Mathematics Module 4. Testing of Hypotheses: Concept of testing hypotheses, simple and
STATISTICS : STATISTICAL INFERENCE composite hypotheses, null and alternative hypotheses, type I
and type II errors, critical region, level of significance and
power of a test. Neymann-Pearson approach-Large sample tests
Prepared and Scrutinised by: concerning mean, equality of means, proportions, equality of
proportions. Small sample tests based on t distribution for
Dr. K.X.Joseph, mean, equality ot means and paired mean for paired data. Tests
Director, based on F distribution for ratio of variances. Test based on chi
Academic Staff College, square- distribution for variance, goodness of fit and for
University of Calicut. independence
Layout & Settings
Computer Section,
©
Reserved
School of Distance School of Distance
2 Statistical Statistical 3
SYLLAB
US

of attributes. 30 hours

Layout & Settings

Computer Section,
©
Reserved
School of Distance School of Distance
2 Statistical Statistical 3
SYLLAB
US

Module 1. Sampling Distributions: Random sample from a population

distribution, Sampling distribution of a statistic, Standard error,
Sampling from a normal population, Sampling distributions of
the sample mean and variance. Chi-square, Student‟s t and F
distributions - derivations, simple properties and inter
relationships. 25 hours

UNIVERSITY OF CALICUT Module 2. Theory of Estimation: Point estimation, Desirable properties

of a good estimator, unbiasedness, consistency, sufficiency,
SCHOOL OF DISTANCE EDUCATION statement of Fisher Neyman factorization criterion,
efficiency. Methods of estimation, method of moments,
Method of maximum likelihood-Properties estimators
STUDY MATERIAL obtained by these methods
25 hours

III Semester Module 3. Interval Estimation: Interval estimates of mean, difference of

means, variance, proportions and difference of proportions,
Large and small sample cases.
Complementary Course 10 hours
for B Sc. Mathematics Module 4. Testing of Hypotheses: Concept of testing hypotheses, simple and
STATISTICS : STATISTICAL INFERENCE composite hypotheses, null and alternative hypotheses, type I
and type II errors, critical region, level of significance and
power of a test. Neymann-Pearson approach-Large sample tests
Prepared and Scrutinised by: concerning mean, equality of means, proportions, equality of
proportions. Small sample tests based on t distribution for
Dr. K.X.Joseph, mean, equality ot means and paired mean for paired data. Tests
Director, based on F distribution for ratio of variances. Test based on chi
Academic Staff College, square- distribution for variance, goodness of fit and for
University of Calicut. independence
Layout & Settings
Computer Section,
©
Reserved
School of Distance School of Distance
2 Statistical Statistical 3
SYLLAB
US

of attributes. 30 hours

Layout & Settings

Computer Section,
©
Reserved
School of Distance School of Distance
4 Statistical inference Statistical inference 5

MODULE I

SAMPLING DISTRIBUTIONS

Here we are interested with the study of population characteristics

based on a sample taken from the population. The process of making
inferences about the population based on samples taken from it is called
statistical inference or inferential statistics. We have already discussed
the sampling theory which deals with the methods of selecting samples
from the given population. The sample selected should be such that it is
capable of exhibiting the characteristics of the population. Here we focus
our attention on characteristics calculated from simple random samples.
If X1, X2,.....Xn are independent and identically distributed r.v.s., we
say that they constitute a random sample from the population given by
their common distribution. According to this definition a random sample
x1, x2,... xn of size „n‟ is a collection of random variables (X 1, X2, Xn)
such that the variables Xi are all independent but identically distributed
as the population random variable X. That means observations of a
random variable from the repetitions of a random experiment can be
treated as i.i.d. random variables. We can justify this interpretation by
means of suitable examples.

Parameter and Statistics

Any measure calculated on the basis of population values is called a
‘parameter’. For example, population mean µ population standard deviation
2
σ, population variance σ , population correlation coefficient p etc. For
example, λ is the parameter of a Poisson distribution, µ and σ are the
parameters of normal distribution. Statistical inferences are usually basedon
„Statistics‟, that is, on random variables X1, X2, X3....Xn constituting a
random sample. In other words, any measure computed on the basis of
sample values is called a statistic. For example, sample mean x , sample
2
standard deviation s, sample variance s , sample correlation coefficient r
etc.

School of Distance School of Distance

6 Statistical Statistical 7
Sampling Distributions A. Sampling distribution of small samples drawn from normal
In the case of random sampling the nature of the sampling distribution population.
of a statistic can be deduced theoretically, provided the nature of the B. Sampling distribution of large samples drawn from any large
population is given, from considerations of probability theory. Let x 1, population.
x2,...xn be a random sample taken from the population under Conventionally by a small sample we mean a sample of size less than
investigation. We can consider the random observations as independent 30 where as a sample of size greater than or equal to 30 is treated as a
random variables X1, X2,.....Xn following the same distribution of the large sample.
population. Let t = g(X1, X2........Xn ) being a function of these r.v.s, is A. Sampling distribution of small samples
also a r.v. That is t is a r.v. The probability distribution of t = g(X 1, X2 ...
The probability distribution of statistics computed from small
Xn) is called sampling distribution of t. In other words, by a sampling samples drawn from a normal population are discussed here. Being a
distribution we mean the distribution of a statistic . If t is a statistic, its small sample we get exact probability distribution of these statistics or
sampling distribution is usually denoted as f(t). The sampling distribution exact sampling distributions.
of one sample differs from the sampling distribution of another even if
both are defined on the same sample. The determination of the sampling Sampling distribution of sample mean
distribution of a statistic depends on the selection procedure of the
sample, the size of the sample, and the distribution of the population. Let x1, x2, xn be a random samples of size n drawn from N( µ σ).
Let x be the sample mean. To find the probability distribution of x we
Standard error shall use the mgf technique. We can consider the random observations as
The standard deviation of the sampling distribution of a statistics is independent and normally distributed r.v.s each having the same normal
called standard error of the statistic. If t is a statistic with sampling law N(µ σ).
distribution f(t)l the standard error (SE) of t is given by µt +1 2 2
M t σ

SE of t 2 2
V (t ) where V(t) = E(t ) – {E(t)} m xi (t) = e 2 ,i = 1, 2, 3......n
= n

Uses of Standard Error Σ xi (t )

mM x (t ) = = Σxi (t / n ) = П x i (t / n )
Standard Error plays a very important role in large sample theory and M
2
n i 1 n

2
forms the basis of testing of hypothesis ( µt + 1 t σ 2  ( µt + 1 t σ
2

1. Since SE is inversely proportional to the sample size n it is very  n2 n2   n 2 n 

П
n

helpful in the determination of the proper size of a sample to be taken = e = e

i 1
to estimate the parameters.    
2. It is used for testing a given hypothesis. ( σ 
2
3. SE gives an idea about the reliability of a sample. The reciprocal of = µt +1 t2 σ = mgf of N µ,
n
SE is a measure of reliability of the sample. e 2n  
4. SE can be used to determine the confidence limits of population
parameters. (σ 
m x →N µ,
School of Distance School of Distance
6 Statistical Statistical 7

Here we discuss sampling distributions under two headings. n
 

School of Distance School of Distance

 8 Statistical Statistical 9
Note:
n
1. When x is the mean of sample of size n drawn from a population ( 1 2 n
which is not normal, the sampling distribution of x can be approximated 2
2
/ 2( y 2 ) –1 ,0 c3 2 c
e
( f(z ) =
σ  –3
as normal N using central limit theorem, provided n is sufficiently n
µ , n 
 = 0 , otherwise
large. Here n is the parameter of the distribution and we write this as z2 → z2
2. In the case of normal population, the distribution of x is normal
(n) df.
N ( µ , σ / n ) for any sample size n.

σ2
3. The above results show that E( x ) = µ, and V( )= n
x
σ
m SE of x = n

The pdf of the random variable x is given by

–n ( x – µ )
2
f( ) = n
,–c x c 
x
e 2σ

σ 2n
Chi square Distribution
Karl Pearson in about 1900 described the well known probability The shape of z2 curve depends on the value of n. For small n, the curve
d i s t r i b u t i o n “ C h i s q u a r e d i s t r i b u t i o n ” o r d i s t r i b u t i o n o f z 2. (Square of greek is positively skewed. As n, the degrees of freedom increases, the curve
letter chi) z2 is a random variable used as a test statistic. Let a random approaches to symmetry rapidly. For large n the z2 is approximately
sample X1, X2,....Xn be taken from a normal population with mean µ and normally distributed. The distribution is unimodal and continuous.
variance σ2.
ie.X →N (µ σ ). We define z statistic as the sum of the squares of Degrees of freedom
i

standard normal variates. The phrase „degrees of freedom‟ can be explained in the following
2 intutive way.
n (X –µ
ie., z = Σ i
Let us take a sample of size n = 3 whose average is 5. Therefore the
i 1  σ sum of the observations must be equal to 15. That means X 1 + X2 + X3 =
15
Definition
We have complete freedom in assigning values to any two of the three
A continuous r.v. z2 assuming values from 0 to  , is said to follow observations. After assigning values to any two, we have no freedom in
a

School of Distance School of Distance

8 Statistical Statistical 9
chi square distribution with n degrees of freedom if its pdf is giving by finding the other value, because the latter is already determined. If we

School of Distance School of Distance

10 Statistical inference assign values to X1 = 3, X2 = 8, then X3 = 4. Statistical inference 11
2
( i –µ
Given the mean, we have only n – 1 degrees of freedom to compute the en 3 (1
mean of a set of n observations. Here we have complete freedom in  σ  →
assigning values to any n – 1 observations. ie., they are independent 2
observations. The
complete freedom is called the degrees of freedom. Usually it is denoted by v nΣ( Xi –µ  →3
2
(n) df, by additive property
. Accordingly the sample variance has n – 1 degrees of freedom. But the
i 1  σ 
sum
of squared deviations from the population
n
mean µ ie., Σ( x i – µ )2 has n degrees of freedom. Thus by degrees of Also we know that x →N(µ,σ/n)
i 1

x –µ
freedom we mean the number of independent observations in a → N (0,1)
distribution or a set. σ/ n
2
Moments of (x –µ 2 2
Mean  → 3 (1)
m
E(z2) = n
σ / n d.f.

Now consider the expression

Variance n
2 n
2 2 2 2 Σ ( x i – 2  Σ ( x i – x + x – µ)
µ)
V(z ) = E(z ) – [E(z )] i 1 i 1
n n n
Moment generating function = Σ(xi x) + 2( – µ ) Σ ( x i – x ) + Σ ( x – µ )
–
i 1 i 1 i 1
2
z (t)= ty = 2 2
= ns + 2( x – µ)  0 + n ( x – µ)
()

E (e ) 2 2
n ( Xi –µ2 ns ( x –µ 
., Σ  2 , by dividing each term term
e
Sampling Distribution of i 1  σ  σ σ / n
2 2
Sample Variance s
y
Let x1, x2,... xn be a random sample drawn from a normal population
School of Distance School of Distance
 2 2 3 2
+ 2 ie., 3 (n) = ns
N(µ σ). Let x be the sample mean and s be its variance. We can consider (1)
the random observations as independent and normally distributed r.v.s. σ 2
z2
with mean µ and variance σ2. Here we have seen that the LHS follows (n). So by additive
ie., Xi → N(µ σ).i = 1, 2,........n property of z2 distribution we have.
X –µ 2 2
m i → N(0,1) ns 3
σ (n – 1)
2 →
σ

School of Distance School of Distance

 Statistical Statistical 1
Therefore the pdf of z2 distribution with (n – 1) df is given Definition
by A continuous random variable t assuming values from –  to + 
n –1
1 2
with the pdf given by
– 2 n –1
ns
 ns 2 2 2 –1 2
 c n +1
2  2
e 2σ ns ,0 c

 f (t ) 1
( t 2  - 2 ,–  c t c
σ n –1 σ σ 1+ n
 n
2  n ,
2 2
2 
ns du n is said to follow a student‟s t distribution with n degrees of freedom.
Put u = , 
σ ds σ The t distribution depends only on „n‟ which is the parameter of the
distribution.
du
2 1
f(s ) = f(u) ds For n = 1, the above pdf reduces to f(t) = , which is known
n(1 + t2
n –1
as the Cauchy pdf. )
(1 2 2 n –1
–ns (ns 2 –1 mn
= 2 e 2σ 2   n (m + n)
n –1 
σ 
σ Note: β(m, n) =
2 
Definition of ‘t’ statistic
n –1 If the random variables Z → N (0, 1) and Y → z2(n) and if Z and Y
n 2 2
n –1
are independent then the statistic defined by
–
2σ 2σ 2 (s 2
= n–
e –1 , 0 c s 2c  Z
)
1
t= Y/n follows the Student‟s „t‟ distribution with n df.
 2
2
This is the sampling distribution of s . It was discovered by a German x- →t df
mathematician. Helmert in 1876. We can determine the mean and t = (n - 1)
variance of s2 similar to the case of z2 distribution. S/ n- 1
Later we can see that the above r.v. will play an important role in
2 n-1 2 building schemes for inference on µ.
Thus E(s ) =
n
s
n - 14 SE of t = n- 1
2
School2.of Distance
n School of Distance

n2
 Statistical Statistical 1
V(s ) = ii. Distribution of t for comparison of two samples means:
2

2 We now introduce one more t statistic of great importance in

SD of s = 2σ (n applications.

School of Distance School of Distance

 1 Statistical Statistical 1
Let x1 and x2 be the means of the samples of size n1 and n2
respectively drawn from the normal populations with means µ1 and
2 2 2
µ2 and with the same unknown variance σ . s1 and s2 be the
variances of the samples.

(x 12 – x )–(µ–µ)
1 2
→ t (n +n – 2)df
1 2
t n 1s1 +n 2 s 2 ( 1 1 
= + 7. For each different
n1 +n 2 –  n1 2 
2
Characteristics of t distribution
1. The curve representing t distribution is symmetrical about
t=0
2. The t curve is unimodal number of degrees of freedom, there is a distribution of t. Accordingly t
3. The mode of t curve is 0 ie, at t = 0, it has the maximum probability distribution is not a single distribution, but a family of distributions.

4. The mean of the t distribution is 0 ie., E(t) = 0 8. The mgf does not exist for t distribution.

5. All odd central moments are 0. 9. The probabilities of the t distribution have been tabulated. (See
tables) The table provides
6. The even central moments are given by

µ  µ  (2 r –1)(2r –3)....3.1 v2r –t 0 

2r 2r (v –2)(v – 4)...(v – 2r ) , v denotes df P{|t| > t0} =  f (t ) dt +  f (t ) dt

v – t0

Putting r = 1, µ2  v –2 , for v > 2

=2


f (t )dt,because of symmetry.

2 v t0
Hence variance σ v –2 , it is slightly greater than 1. Therefore, it Values of t0 have been tabulated by Fisher for different probabilities
has a greater dispersion than normal distribution. The exact shape of the t say 0.10, 0.05. 0.02 and 0.01. This table is prepared for t curves for v=
distribution depends on the sample size or the degrees of freedom v. If v 1, 2, 3,... 60.
is small, the curve is more spread out in the tails and flatter around the
centre. As v increases, t distribution approaches the normal distribution. Snedecor’s F Distribution
Another important distribution is F distribution named in honour of
Sir. Ronald. A. Fisher. The F distribution, which we shall later find to be
of considerable practical interest, is the distribution of the ratio of two
independent chi square random variables divided by their respective

School of Distance School of Distance

 1 Statistical Statistical 1
degrees of freedom.

School of Distance School of Distance

 1 Statistical Statistical 1
If U and V are independently distributed with Chi square distributions
with n1 and n2 degrees of freedom.
F(3, 9)
U / n1
F(9, 1, 2)
then, F = V / n2 is a random variable following an F distribution with
f(F)
(n1, n2) degrees of freedom. F(1, 3)

Definition
A continuous random variable F, assuming values from 0 to  and
having the pdf given by O F
Applications of F distribution arise in problems in which we are
n 1n2 n interested in comparing the variances σ 2 and σ 2 of two normal
n 1 n2
1 –1 1 2
1 and X2 such that X1 →
populations. Let us have two independent r.v.s. X
(n 1 n2 
f (F) = F 2 n 1 +n2 ,0 cFc
β , N (µ σ ) and → N (µ σ ). The random samples of sizes n and n
 X

2  (n 1 F +n2 )
1 2
2 1 2 2
2 2
1 2

is said to follow an F distribution with (n 1, n2) degrees of freedom. are taken from the above population. The sample variances s1 and s2 are
The credit for its invention goes to G.W. Snedecor. He chose the letter F
n sn s
to designate the random variable in honour of R.A. Fisher. 1 1 2 2

The F distributions has two parameters, n

and n corresponding to the computed. Then we can observe that the statistic n1 –1 n2 – has an
1
2
1 2
F distribution with (n1 – 1, n2 – 1) degrees of freedom.
degrees of freedom of two z random variables in the ratio; the degrees of
freedom for the numerator random variable is listed first and the ordering
makes a difference, if n1 s n2. The reciprocal of the F random variable Characteristics of F distribution

1 n2
(
ie.,  V / n2  z
again is the ratio of two independent r.v.s. each
 F U / n1  1. The mean of F distribution is n2 – 2
divided by its degrees of freedom, so it again has the F distribution, now
n2
with n2 and n1 degrees of freedom.
In view of its importance, the F distribution has been tabulated
extensively. See table at the end of this book. This contain values of
ie., E (F) = n2 – 2 , No mean exist for n2 c 2.
F 2. The variance of the F distribution is
2
( ; n 1 ,n 2 ) for α = 0.05 and 0.01, and for various values of n1 and n2, 2n 2 (n 1 +n2 –2)
F V(F) = , No variance exist if n c 4.
where ( ; n 1 , n 2 ) is such that the area to its right under the curve of n 1 (n 2 –2) 2 (n 2 – 4) 2

School of Distance School of Distance

 1 Statistical Statistical 1 2
the F distribution with(n 1, n2) degrees of freedom is equal to α. That is 3. The distribution of F is independent of the population variance σ.
F P(F 4. The shape of the curve depends on n 1 and n2 only. Generally it is non
( ; n 1 , n 2 ) =
( ; n 1 ,n 2 ) is such F symmetric and skewed to the right. However when one or both
that α

School of Distance School of Distance

 Statistical inference 19
18 Statistical inference parameters increase, the F distribution tends to So the square of at variate with n df is F(1, n)
become more and more symmetrical.
2. F is the ratio of two 2

1 c
(n F → F(n , n ) df, then
5. If F
n ) df n s2 n s2
1 2 2, 1 2
Let F(n – 1, n – 1) = 1 1 2

F
1 2
It is called reciprocal property of F distribution
n1 –1 n2 –1
6. Th two important uses of F distribution are (a) to test the equality of n s
two normal population variances and (b) to test the equality of three 1 1 / (n 1 –1)
or more population means. σ
1
2
2
7. Since the applications of F distribution are based on the ratio of sample = 2 2 / (n –1)
variances, the F distribution is also known as variance - ratio distribution. 2
σ 2
2
Inter relationship between t, 2
and F distributions; 2

1. The square of t variate with n df is F(1, n) y (n 1 –1)/ (n 1 –1)

=
Let x1, x2.......xn be a random sample drawn from N( µ, σ). We
2 2
y (n –1)/ (n 2 –1)
Hence the result.
can
consider the random observations as i.i.d. r.v.s.
SOLVED PROBLEMS
1 2 1 2
Then x
n Σx i , s  n Σ( x i – x Example l
 ) 2)
Let X be N (100, 10 distributed. Find the sample size n so as to have
x–µ P ( x  101.645) = 0.05
→ N(0,1)

We know that Z = σ/ Solution

2 Given X → N (100, 102). Take a random sample of size n from

n → 3 (n –1)
the population. Let
10
be its mean,
x
ns n
Y= then → N(100, )
σ 2
Define

We have to find n such that t

X
=  N(0,1)
–1 of3 Distance
y nSchool (n –1) School of Distance
n –1

2
Squareof N (0,1) 3 (1)/ 1
  101.645) = 0.05
( x –100 101.645 – 100
ie P  = 0.05

 10/ n 10/ n 
( 
2 2 P Z 1.645 n
ie, = 0.05
m t2 = 3 (n –1)/ (n –1) = 3 (n –1)/ (n –1) → F (1, n –1)  10 

ie., the square of a variate with n –1 df is F(1, n –1).

School of Distance School of Distance

20 Statistical inference Statistical inference 21

ie,
P ( Z  0.1645 n = 0.05 X4
2 2
 3 (1)
) X1 +X2
3
+ X33 (3)/ 3

ie, P ( 0 c Z c 0.1645 n ) = 0.45

= 2
3 (1)/ 1

From normal table, 0.1645 n = 1.645

mn = 10, ie, n = 100 2

= 3 (3)/ 3 = F(1, 3)
Example 2
Example 3
If X1, X2, X3 and X4 are independent observations from a univariate
normal population with mean zero and unit variance, state giving reasons
If X1, and X2, are independent z 2r.v.s. each with one degree of
the sampling distribution of the following. 1
2 freedom. Find λ such that P(X1 + X2 > λ) =
2 X3 2
2 3X42 2
c z (1)
+ + Solution
22 2
(i) u = (ii) v = X1 X2 X3 Given X1
X1 + X2
Solution X2 → z 2
(1)
2
Given Xi → N (0, 1), i = 1, 2, 3, 4 m Y= + → (2), by additive property
X1 X2

2 We have to find λ such that

m X 2
3 (1) i = 1, 2, 3, 4 1
→i
2
m Xi 2 2 2 2 2 2 P(X1 + X2  λ) =
+ X2 → 3 (2) and Xi + X 2 + X 3 →3 (3) 2
1
2 3
(i) u = ie, P(Y > λ) = 2
2 2  1
X 1 + X2
X3 N(0,1) → t(2) df
ie, λ (y ) dy = 2
= X 1 + X23 (2)
2
22 (1  2
y 2

2
3X 4
School of Distance School of Distance
  2 –2 2 –1 1
ie, λ 2 e y dy = 2
(ii) v = 2 2 2
2 + X3
X1 +X2

School of Distance School of Distance

 2 Statistical Statistical 2
1
1  –y l Student‟s t distribution curve is symmetrical about mean, it means
that
2 λ e 2 dy = 2
a. odd order moments are zero
(e–y / 2  ( λ 
b. even order moments are zero
–
=1, – 2 0 – e  2  = 1 c. both (a) and (b)d. none of (a) and (b)
–1/ 2 2
l If X → N (0, 1) and Y → (n), the distribution of the variate
λ   z
λ λ 1
1, - = x / y / n follows
-=
2e2 a. Cauchy‟s distribution
e 2 2
b. Fisher‟s t distribution
λ c. Student‟s t distribution
λ
m 2 = 2, ie = loge2 d. none of the above
e 2 l The degrees of freedom for student‟s „t‟ based on a random
sample of size n is
mλ = 2 loge2 a. n – 1 b. n c. n – 2 d. (n – 1)/2

EXERCISES l The relation between the mean and variance of z with n df is

a. mean = 2 variance . 2 mean = variance
Multiple Choice Question c. mean = variance d. none of the above
l Simple random sample can be drawn with the help of l Chi square distribution curve is
a. random number tables b. Chit method a. negatively skewed b. symmetrical
c. roulette wheel d. all the above c. positively skewed d. None of the above
l Mgf of chi square distribution with n df is
l Formula for standard error of sample mean x based on sample o a. (1 – 2t) b. (1 – 2it)
–n –n
si z en h avi ng v ar i a nce
s , when population consisting of N items is c. (1 – 2t) d. (1 – 2it)
a. s/n b. s /n –1 F distribution was invented by
c. s / N– d. s / n l

a. R.A. Fisher . G.W. Snedecor

1
l Which of following statement is true
c. W.Z. Gosset d. J. Neymann
a. more the SE, better it is
l The range of F - variate is
b. less the SE, better it is
a. – to +  b. 0 to 1
c. SE in always zero
c. 0 to  d. – to 0
d. SE is always unity
l The relation between student‟s t and F distribution is
l Student‟s „t‟ distribution was discovered by

School of Distance School of Distance

2 Statistical Statistical 2
a. G.W. Snedecor b. R.A. Fisher 2 2
a. F1, 1 = tn b. Fn,1 = t1
c. W.Z. Gosset c. Karl Pearson 2

c. t = F1,n d. none of the above

School of Distance School of Distance

 2 Statistical Statistical 2
l Student‟s t curve is symmetric about lGive an example of an F statistic. l

a. t = 0b. t = µ c. t = 1d. t = n Define sampling error.

l Give four examples of statistics. l Give
Fill in the blanks
four examples of parameters
l If the number of units in a population are limited, it is known as l What is the relationship between t and F.
.................. population.
lWhat are the importance of standard error? l

l Any population constant is called a .................. 2

What are the mean and variance of s
l Another name of population is ..................
Short Essay Questions
l The index of precision of an estimator is indicated by its ..................
2 lExplain the terms (i) parameter (ii) statistic (iii) sampling distribution. l
3 /n What is a sampling distribution? Why does one consider it?
1
1
2 l Explain the meaning of sampling distribution of a statistic T and
l In the above case, the distribution of y2 / n2 is the standard error of T. Illustrate with the sample proportion.
z 2

l The mean of the distribution is of its variance l Explain the terms (i) statistic (ii) standard error and (iii) sampling
distributions giving suitable examples.
l If the df is for Chi square distribution is large, the chi-square
distribution tends to .................. l Define sampling distribution and give an e xample.
l t distribution with 1 df reduces to .................. l Derive the sampling distribution of mean of samples from a
normal population.
l The ratio of two sample variances is distributed as ..................
Long Essay Questions
l The relation between Fisher‟s Z and Snedecor‟s F is ..................
l State the distribution of the sample varience from a normal
l The square of any standard normal variate follows ......... .........
population
distribution. 2
l Define z and obtain its mean and mode.
Very Short Answer Questions 2
l Define z statistic. Write its density and establish the additive
l What is a random sample? l property.
Define the term „statistic‟. l Give the important properties of z 2distribution and examine its
l Define the term „parameter‟. l
relationship with the normal distribution.
What 2
is sampling distribution? l Define
l Define a z variate and give its sampling distribution. Show that
standard error. its variance is twice its mean.
l What is the relationship between SE and sample size.
2 l Define the F statistic, Relate F to the t statistic and Fn,m to Fm,n
lDefine z distribution with n df. l

Define student‟s t distribution.

l Define F distribution.
l Give an example of a t statistic.

School of Distance School of Distance

 2 Statistical Statistical 2

MODULE II 1. Unbiasedness

THEORY OF ESTIMATION An unbiased estimator is a statistic that has an expected value equal
to the unknown true value of the population parameter being estimated.
An estimator not having this property is said to be biased.
The Theory of estimation was expounded by Prof: R.A. Fisher in his
Let X be random variable having the pdf f(x , θ), where θ may be
research research papers round about 1930. Suppose we are given a random
sample from a population, the distribution of which has a known unknown. Let X1, X2....Xn be a random sample taken from the
population represented by X. Let
mathematical form but involves a certain number of unknown parameters.
The technique of coming to conclusion regarding the values of the unknown tn = t(X1, X2.....Xn) be an estimator of the parameter θ.
parameters based on the information provided by a sample is known as If E(tn) = θ for every n, then estimator t n is called unbiased estimator.
the problem of „Estimation‟. This estimation can be made in t wo ways.
i. Point Estimation ii. Interval Estimation 2. Consistency
One of the basic properties of a good estimator is that it provides
Point Estimation increasingly more precise information about the parameter θ with the
If from the observations in a sample, a single value is calculated as an increase of the sample size n. Accordingly we introduce the following
estimate of the unknown parameter, the procedure is referred to as point definition.
estimation and we refer to the value of the statistic as a point estimate. For
Definition
example, if we use a value of x to estimate the mean µ of a population we
The estimator t = t(X X......X ) of parameter θ is called consistent if
are using a point estimate of µ. Correspondingly, we, refer to the statistic 1 2 n

x as point estimator. That is, the term „estimator‟ represents a rule or tn converges to θ in probability. That is, for s > 0
method of estimating the population parameter and the estimate lim P (| t n –8 | c c )  1 or lim P (| t n –8 |  c )  0
represents the value produced by the estimator. n → n →
An estimator is a random variable being a function of random The estimators satisfying the above condition are called weakly
observations which are themselves random variables. An estimate can be consistent estimators.
counted only as one of the possible values of the random variable. So The following theories gives a sufficient set of conditions for the
estimators are statistics and to study properties of estimators, it is consistency of an estimator.
desirable to look at their distributions. Theorem
An estimator t , is such that E(t ) = θ →θ and V(t )→0 as n
Properties of Estimators n n n n

There are four criteria commonly used for finding a good estimator. → , the estimator tn is said to be consistent for θ.
They are:
3. Efficiency
1. Unbiasedness
Let t1 and t2 be two unbiased estimators of a parameter θ. To choose
2. Consistency between different unbiased estimators, one would reasonably consider their
3. Efficiency variances, ie., If V(t1) is less than V(t2) then t1 is said to be more efficient
4. Sufficiency than t2. That is as variance of an estimator decreases its efficiency

School of Distance School of Distance

 2 Statistical Statistical 2
V (t 1
) V (t 2 is called the relative efficiency of with respect Properties
increases. t2 i. Moment estimators are asymptotically unbiased.
)
to t1 and we can use this to compare the efficiencies of estimators.
ii. They are consistent estimators.
4. Sufficiency iii. Under fairly general conditions, the distribution of moment
estimators are asymptotically normal.
An estimator t is said to be sufficient if it provides all information
contained in the sample in respect of estimating the parameter 8. In other
Method of Maximum Likelyhood
words, an estimator t is called sufficient for θ, if the conditional
distribution of any other statistic for given t is independent of θ. The method of moments is one procedure for generating estimators of
unknown parameters, it provides an attractive rationale and is generally
Factorisation Theorem quite easy to employ. In 1921 Sir. R. A. Fisher proposed a different
Let x1, x2 xn be a random sample of size n from a population with rationale for estimating parameters and pointed out a number of reasons
density functions f(x; θ) where θ denotes the parameter, which may be that it might be preferable. The procedure proposed by Fisher is called
unknown. Then a statistic t = t(x1, x2.....xn) is sufficient if and only if the method of Maximum likelyhood and is generally acknowledged to be
joint probability density function of x 1, x2.....xn (known as likelyhood of superior to the method of moments. In order to define maximum
the sample) is capable of being expressed in the form likelyhood estimators, we shall first define the likelyhood function.
L(x1, x2....xn; θ) = L1 (t, θ). L2(x1, x2 xn)
Likelyhood function
where the function L2(x1, x2......xn) is non negative and does not involve The likelyhood function of n random variables X X ....X is defined
the parameter θ and the function L1 (t, θ) is non negative and depending 1, 2

on the parameter θ. is the estimator of θ obtained by the method of moments.

Method of Moments
This is the oldest method of estimation introduced by Karl Pearson.
According to it to estimate k parameters of a population, we equate in
general, the first k moments of the sample to the first k moments of the
population. Solving these k equations we get the k estimators.
Let X be a random variable with the probability density function f(x,
θ). Let µr be the r-th moment about O. µr = E(X ).r In general, µr will
be a known function of θ and we write µr = µr (θ). Let x1, x2...xn be a
random sample of size n drawn from the population with density function
1 n
f(x, θ). Then r-th sample moment will be m= Σ xi . Form the
r n i 1
equation m  = µ  (θ) and solve for θ. Let ˆ
be the solution of θ.
Then 8
r r 8

School of Distance School of Distance

to2 be the joint probability density function of the n random
Statistical
variables, Statistical 2
say f(x x.....x ; θ) which is considerd to be a function of θ. In
particular
1, 2n

suppose that X is a random variable a nd X 1, X......X n is a random sample

of X having the 2

density f(x, θ). Also x x ....x are the observed sample

values. Then the likelyhood
1, 2n

function is defined as

L ( x 1 , x 2 ......x n ; 8 ) = f ( x 1 , x 2....x n ; 8 )

= f( x1;8) f ( x 2 ; 8 )..... f (x n ; 8 )
n
= П f(xi,8)
i 1
The likelyhood function can also be denoted as L(X; θ) or L(θ). The
likelyhood function L ( x 1 , x 2.......x n ; 8 ) give the likelyhood that the
random variables assume a particular value x 1 , x 2.........xn .

School of Distance School of Distance

 3 Statistical Statistical 3
The principle of maximum likelyhood consists in finding an estimator of
the parameter which maximises L for variations in the parameter. Thus the SOLVED PROBLEMS
problem of finding a maximum likelyhood estimator is the problem of
finding the value of θ that maximises L(θ). Thus if there exists a function t Example l
=
t( x 2........x n ) of the sample values which maximises L for variations in
Let x , x , x ...., x be a random sample drawn from a given population
1, 1 2 3 n

x
θ, then t is called Maximum likelyhood Estimator of θ. (MLE). ith mean µ and variance σ . Show that the sample mean is an unbiased
Thus t is a solution if any of estimator of population mean µ.
2
6L 6L Solution
 0 and 0
68 68 2 n

Also L and log L have maximum at the same value of θ we can take We know that x = 1Σ i
log L instead of L which is usually more convenient in computations. n i 1
6 log L Taking expected value, we get
n
Thus MLE is the solution of the equations  0 , provided    
68 n
E
1 
1E 

x


2
6 log L  0 E( x ) = n

Σ i n Σ
1
i 
2 i
i 1
68 1
=

E ( x 1 + x 2 +......+ xn )
The maximum likelyhood estimator can also be used for the
n

simultaneous estimation of several parameters of a given population. In that

case we must find the values of the parameters that maximise the likelyhood =
1 {E ( x 1 ) + E ( x 2 ) +.....+ E ( x n )}
function. n

Properties of Maximum likelyhood estimators. Now E (xi) = µ (given)

Under certain very general conditions (called regularity conditions) 1 nµ
the maximum likelyhood estimators possess several nice properties. {µ + µ +.... + µ}  µ
m E(x) = n
1. Maximum likelyhood estimators are consistent n
2. The distribution of maximum likelyhood estimators tends to normality 5. Maximum likelyhood estimators are not necessarily unbiased.
for large samples. 6. Maximum likelyhood estimators have invariance property, (ie. if t is the
3. Maximum likelyhood estimators are most efficient. m.l.e. of θ , then g(t) is also the MLE of g (θ), g being a single valued
4. Maximum likelyhood estimators are sufficient if sufficient estimators function of θ with a unique inverse).
exists

School of Distance School of Distance

3 Statistical Statistical 3
Therefore sample mean is an unbiased estimator of population mean.

Example 2
Let x1, x2, x3...., xn is a random sample from a normal distribution
N(µ, 1) show that

1 n
t= Σx i is an unbiased estimator of µ +1
n i 1

School of Distance School of Distance

 3 Statistical Statistical 3
Solution Example 4
We are given that x1, x2,...., xn is a random sample from a population following Poisson
E(xi) = µ distribution with parameter λ. Suggest any three unbiased estimators of λ.

V(xi) Solution
= 1 for every i = 1, 2, 3,...n

Now V(xi) = 2 Since xi is a random observation from a Poisson population with parameter λ,
2
2 E(xi) = λ i = 1, 2, ... n
E ( x i ) – [ E ( x i )]
m E(xi )= 2 x +x2
µ +1 mt 1
=x,t = 1 , tn  x 1 + x 2 +...+ xn
2
2 n
1 n 2  1 n 2
are unbiased estimators of λ. It may be noted that
or

E 
E( t) =

x   E (x )
n Σi n Σ i .
E(t1) = E(xi) = λ
L
1 n i 1  i 1 1 1
1 E(t2) = [ E ( x 1 ) + E ( x 2 )]  [λ + λ ]  λ
n Σ( µ +1)  n n ( µ +1) 2 2
i 1

Example 3 2 E(tn) = 1E(x1+x2 +. . .x n )

= μ +1
n
2 = 1 [ E ( x 1 ) + E ( x 2 ) +. ]
If T is an unbiased estimator of 8 , show that and T are the biased
T n
2
estimator of and respectively. n
8
Solution 8 = 1[λ + λ +.... + λ]  λ
Given E(T) = 8 n n
Now var(T) = E[T – E ( T)] s 0 as var > 0 m t1, t2 and tn are unbiased estimators of λ.
2 2 2 2 Example 5
or E{T – 2T 8 +8 } = E(T) –2 8 E (T) + 8 s0
2 2 2 Show that sample variance is a consistent estimator of the population
E{T ) – 2 8 + s 0 variance in the case of normal population N(µ σ).
2 2 2
Solution
or E(T) s 8 , ie., T is biased2 Let x1, x2,...., xn be a random sample from N(µ σ2). Let x be the mean
Alsovar ( T ) = E [ T – E ( T )] s 0
2 2
2
T
School of Distance School of Distance
 3 Statistical Statistical
and s is its variance. From the sampling distribution of s , we have 3
= E (T) – {E ( )} s0
s {E 2
or E(T) = 8 ( n –1 ( 1
T )} 2
E(s ) = σ = 
–
σ2
2
E( T) s 8 . n  n 

Hence the result. ie., T is not an unbiased estimator of 8 .

School of Distance School of Distance

 Statistical Statistical 3
2
But V(s ) = 2. n –1 n
4 → n→ Then L(x , x ,....x ; α β) = 1 1 1 ( 1
σ 0 as
1 2 n . ... =  
n 2 β – β – β –  β – 
Thus the sufficient conditions are satisfied. L is maximum when (β – α) is maximum i.e. when β is minimum and
α is maximum. If the sample observations are arranged in ascending
2
The r ef or e

s
is consistent for σ2 order, we have
 c x 1 c x 2 c x 3.....c xn c β
Example 6
Give an example of estimators which are Here the minimum value of β consistent with the sample is xn and
(a) Unbiased and efficient, maximum value of α is x1. Thus the M.L.E.‟s of α and β are
(b) Unbiased and inefficient,
x1,β
ˆ
ˆ

 xn
(c) Biased and inefficient.
EXERCISES
(a) The sample mean x and modified sample variance Multiple Choice Questions
n 2
l An estimator is a function of
S =n –1 s are two such examples. a. population observations
1 b. sample observations
c. Mean and variance of population
(b) The sample median, and the sample statistic 2 [Q1 +Q3] where Q1 and
d. None of the above
Q3 are the lower and upper sample quartiles, are two such examples.
Both statistics are unbiased estimators of the population mean, since the l Estimate and estimator are
mean of their sampling distribution is the population mean.
a. synonyms b. different
(c) The sample standard deviation s, the modified standard deviation s , c. related to population d. none of the above
the mean deviation and the semi-in-terquartile range are four such l The type of estimates are
examples.
a. point estimate
Example 7 b. interval estimate
For the rectangular distribution oven an interval ( α β); α < β. Find c. estimates of confidence region
the maximum likelihood estimates of α and β. d. all the above
Solution
l The estimator of population mean is
x
For the rectangular distribution over ( α β), the p.d.f. of X is given by 1 , c x c β

School of Distance School of Distance

 Statistical Statistical 3
a. an unbiased estimator b. a consistant estimator
f(x) = β –
c. both (a) and (b) d. neither (a) nor (b)
Take a random sample x1, x2,... xn from (α β)
l Factorasation theorem for sufficiency is known as
a. Rao - Blackwell theorem

School of Distance School of Distance

36 Statistical inference Statistical inference 37
b. Cramer Rao theorem l Consistent estimators are not necessarily ..................
c. Chapman Robins theorem l As estimator with smaller variance than that of another estimator is
d. Fisher - Neymman theorem ..................
l The credit of factorisation theorem for sufficiency goes to
l If t is a consistent estimator for 8 , then
..................
2
a. t is also a consistent estimator for 8 Very Short Answer Questions
2
b. t is also consistent estimator for 8 l Distinguish between an estimate and estimator. l

2 2 What is a point estimate?

c. t is also consistent estimator for 8
d. none of the above l Define unbiasedness of an estimator l

l The credit of inventing the method of moments for estimating Define consistency of an estimator. l Define
parameters goes to efficiency of an estimator.
a. R.A. Fisher b. J. Neymann l Define sufficiency of an estimator.
c. Laplace d. Karl Pearson l State the desirable properties of a good estimator.
l Generally the estimators obtained by the method of moments as l Give one example of an unbiased estimator which is not consistent. l

compared to MLE are Give an example of a consistent estimator which is not unbiased. l Give the
a. Less efficient b. more efficient names of various methods of estimation of a parameter. l What is a
c. equally efficient d. none of these maximum likelyhood estimator?
l Discuss method of moments estimation. l
Fill in the blanks
What are the properties of MLE?
An estimator is itself a ..................
l
l Show that sample mean is more efficient than sample median as
l A sample constant representing a population parameter is known as an estimator of population mean.
..................
l State the necessary and sufficient condition for consistency of
l A value of an estimator is called an .................. an estimator.
A single value of an estimator for a population parameter 8
l
Short Essay Questions
is called its...................estimate
l Distinguish between Point estimation and Interval estimation.
l The difference between the expected value of an estimator and
the value of the corresponding parameter is known as ............ l Define the following terms and give an example for each: (a)
...... Unbiased statistic; (b) Consistent statistic; and (c) Sufficient statistic,
l The joint probability density function of sample variates is called l Describe the desirable properties of a good estimator.
.................. l Explain the properties of a good estimator. Give an example to
show that a consistent estimate need not be unbiased.
l A value of a parameter 8 which maximises the likelyhood function
Define consistency of an estimator. State a set of sufficient conditions
is known as.........................estimate of 8 l

l An unbiased estimator is not necessarily ..................

School of Distance School of Distance
38 Statistical inference for the consistency of an estimate and establish it. Statistical inference 39

l In a N(µ, 1), show that the sample mean is a sufficient estimator of

MODULE III
µ. INTERVAL ESTIMATION
l Describe any one method used in estimation of population parameter.
Thus far we have dealt only with point estimation. A point estimator is
l Explain method of moments and method of maximum likelihood. used to produce a single number, hopefully close to the unknown parameter.
The estimators thus obtained do not, in general, coincide with true value
l Explain the method of moments for estimation and comment on of the parameters. We are therefore interested in finding, for any population
such estimates. parameter, an interval called „confidence interval‟ within which the population
l Explain the maximum likelihood method of estimation. State parameter may be expected to lie with a certain degree of confidence, say
some important properties of maximum likelihood estimate.
 . In other words, given a random sample of n independent values x 1 ,
x2 xn of a random variable X having the probability density f(x ;
l State the properties of a maximum likelihood estimator. Find the
maximum likelihood estimator for 8 based on n observations for
θ), θ being the parameter, we wish to find t 1 and t 2 the function of x 1 ,
the frequency function x 2 x n such that p (t 1 8 t2 )  1 – .

8
This leads to our saying we are 100( 1 – )% confident that our single
f(x, 8 ) = (1 + 8 ) x ; 8 > 0, 0 < x < 8
interval contains the true parameter value. The interval (t 1 , t 2 ) is
= 0 elsewhere. called confidence interval or fiducial interval and 1 – is called
l Given a random sample of size n from „confidence coefficient‟ of the interval (t 1 , t 2 ). The limits t1 and t2 are
–8 x called „confidence limits‟.
f(x ; 8 ) = 8e , x > 0 ; 8 > 0. For instance if we take  = 0.05, the 95% confidence possesses the
find the maximum likelihood estimator of 8 . Obtain the variance meaning that if 100 intervals are constructed based on 100 different
of the estimator. samples (of the same size) from the population, 95 of them will include
the true value of the parameter. By accepting 95% confidence interval for
the parameter the frequency of wrong estimates is approximately equal to
5%. The notion of confidence interval was introduced and developed by
Prof: J. Neyman in a series of papers.
Now we discuss the construction of confidence interval of various
parameters of a population or distribution under different conditions.

Confidence interval for the mean of a Normal population N(

Case (i) when is known.
To estimate µ, let us draw a random sample x 1 , x 2.......xn of size n
from the normal population.

School of Distance School of Distance

 4 Statistical Statistical 4
Let x be the mean of a random sample of size n drawn from the Note:
normal population N( µ σ).
1. If   0.05, z  / 2  1.96 , so the 95% confidence interval for µ
x –µ is
n
Then ); m Z = σ / n → N(0,1)
x cN ( µ , σ /  σ σ 
From the area property of standard normal distribution, we get x –1. 6 ,x +1. 6
L n n 
P | Z | cz =
 / 2  1– 2. If   0.02, z  / 2  2.326 , so the 98% confidence interval for µ

ie. P  –z /2 c Z c z / 2  = 1 – is

 x – µ   σ σ
– 2.326 , + 2.326
ie. P  –z  / 2 c c z / 2  = 1 –  x x
L n 
 σ/ n J n
 σ σ  3. If   0.01, z  / 2  2.58 , so the 99% confidence interval for µ
ie. P –z  / 2 c x – µ c z / 2 n  = 1 –
 n J is
σ
 σ   σ σ 
– – 2.58 , + 2.58
n
ie. P –x /2 c–µc– x + z / 2
σ
 = 1 – L
 x n x n 
 n J
 +z σ n  4. If   0.10, z  / 2  1.645, so the 90% confidence interval for µ
ie. P /  µ  x – z / = 1 – is
x 2
2 J  σ σ 
 n  x –1.645 ,x +1.645 
 n n
σ σ  L 
ie. P x – z  /2 cµcx + z / = 1 –
2
 n n J Case (ii) when is unknown, n is large (n 30)

 σ When the sample is drawn from a normal population or not, by

Here the interval x – z /2 , x + z σ  central limit theorem,
/ 2 is called
 n n J x –µ
100(1 – )% confidence interval for the mean µ of a normal population. Here z / 2 is obtained „area stan normal curve‟ in such a
/n
from the table showing the under a dard way that the area under the

School of Distance School of Distance

 4 Statistical Statistical 4
normal curve to its →N (0,1) as n→
right is equal to  / 2 . Z = s
Here we know that P | Z |
cz /2 = 1 –

Proceeding as above, we get

School of Distance School of Distance

 4 Statistical Statistical 4
 s s  – s +t s 
 = 1 – => t /2 cµc /2 = 1 –
–
P /2 c µ c x + z / 2 x x
nJ P
 n  n –1 n –1 J
is
Thus the 100(1 – )% confidence interval for µ Thus the 100(1 – )% confidence interval for µ is
s 
 s n
– x– s , x +t s 
 t

L
/2

n
, x + z / 2
L /2 n –1  /2 n –1 
Case (iii) when is unknown, n is small (n<30)
where t / 2 is obtained by referring the Student‟s t table for
Let X 2 , X 2 ,.......Xn be a random sample drawn f rom N( µ , σ ) where
(n–1) d.f. and probability  .
2
σ is unknown. Let x be the sample mean and s be its sample variance.
Here we know that the statistic. Interval Estimate of the Difference of two population means

t x – µ →t df Case (i) When

σ
known
s/ n – (n –1) Let x1 be the mean of a sample of size n1 taken from a population
1
Hence 100(1 – )% confidence interval for µ is constructed as with mean µ1 and SD σ 1 .
follows.
Then x 1 →N ( µ 1 , σ 1 n1 )

Let P|t|
= Let x 2 be the mean of a sample of size n 2 taken from a population
ct
 / 2  1 –
with mean µ2 and SD σ 2 .
=> P  –t /2 c t c t / 2 = 1 –
 –µ  Then x 2 c N ( µ 2 , σ 2 n2)
x
=> P –t  / 2  /2 = 1 – Then by additive property,
 s / n –1 J ( σ 2σ 2 
12
 s s 
=> P –t  / 2 c x – µ c t / 2 = 1 –
x1–x2 → N µ1 – µ2, +
 n 1n 2 
 n –1 n –1 J
(x –x )–(µ –µ )
1 2 2
 s s  mZ=
1

→ N(0,1)
=> P –x – / 2 c –µ c x /2 = 1 – σ2 σ
+ of Distance
2

School of Distance School 1

n 1n 2
4 Statistical Statistical 4
 n –1 n –1 J
 + s s  By the area property of ND, we know that
=> /2  µ  x – t / = 1 –
t
x 2
P
 n –1 n –1 J P | Z | c  = 1–α
Z /2 
L

School of Distance School of Distance

44 Statistical inference Statistical inference 45
m P  –Z cZcZ   1–α  
L /2 /2  
s 1s 2 s 1 s 2 
( x –x ) – 1.96 + ,(x –x ) + 1.96
n 1 +,
 


n1 n2 n2

   J
(x –x )–(µ –µ Similarly we can find 98% and 99% confidence intervals replacing 1.96
) /2 respectively by 2.326 and 2.58.
i.e. P –Z  / 2 c
1 2 1 2  1–α
 σ 2 2

 1
+2  Case (iii) When  unknown, small
L n1 n2

Here t = ( σ )–(µ – µ ) / σ /n +σ /n →
On simplification as in the case of one sample, the 100(1 – ) xx1 2 1 2 1 2
%
confidence interval for µ – µ is
1 2 students „t‟ distribution with c = (n 1 +n 2 –2)d.f.
 2 2 2 2 

( ) (n
 σ1 σ2
σ σ  2 2 2
+n –2)
(x –x )–Z

 1 + n 2 ,(x –x )+Z , Where σ = n 1 +n 2
+
 1s 1
s
2
1
 n1 2 n 1 n 2
 J Refer the „t‟ curve for c = (n 1 +n 2 –2)d.f. and probability level P =
here the value of Z / 2 can be determined from the Normal table. α
When α = 0.05, Z / 2 = 1.96. So, 100(1 – )% = 95% The table value of t is t / 2

confidence
interval for µ – µ is Then we have P | t | = α
1
 2
2 2

2 2
t

/2

ie. P | t | ct =
σ1 σ σ σ 
 2  /2 1 –

1 2

ie. P  –t  = 1 – .
(x 1 –x 2 ) – 1.96 + ,(x 1–x 2) + 1.96 + ,
  c+tt c

n1 n2 n1 n2

/ 2 /2
 J
When α = 0.01, Z /2 = 2.58. So, the 99% eq for µ 1 –µ2 Substituting t and simplifying we get the 100(1 – )% confidence
is
+σ
 2 2
2 2  interval for µ –µ as ( 1
–x
)–t (
σ2 /n
1 ) ( /n 2) ,
 σ1 σ2 σ1 σ2  1 2  2 /2
(x 1 –x 2 ) – 2.58 + ,(x 1 –x 2 ) + 2.58 + ,




n1 n2 n1 n2
 J – )+t 2 2
Case (ii) When unkn own, 2 /2 σ
σ large (x 1 x
/n1 +σ /n2
where tα/2 is obtained by referring the t table for n 1 +n 2 –2 df and
probability α.

In this case we replace σ 1 and σ 2 respectively by their estimates

s1 and s2.
So 95% CI for µ – is
1 2
School of Distance Education School of Distance Education
46 Statistical inference Statistical inference 47
Confidence interval for the variance of a Normal population where 3 and 3 are obtained by referring the 3 table for
1 – / 2 / 2
μσ n–1 d.f. and probabilities 1 –  / 2 and  / 2 respectively.
2
Let s be the variance of a sample of size n(n<30) drawn from N( µ , σ Confidence interval for the proportion of success of a
).
We know that the statistic binomial population
2 ns2 2
z → 3 (n –1) d . f . ( x


Let P  bethe proportion ofsuccess ofasample ofsizen drawn

σ
2  n 
2 2 2 from a binomial population with parameters n and p where p is unknown
Now by referring the table we can find a 31 – / 2 and / 2 such
3 and n is assumed to be known. Then we know that
3
that
2 2 p–p
P 3 c32 c3 = Z → N(0,1) for large n
pq n
 1 – / 2 /2 1 –

where 3 and 3 are obtained by referring the table for n–1
1 – / 2 / 2
d.f. and probabilities 1 –/2 and  / 2 respectively. From normal tables we get,
 ns 2  P| Z| cz =

ie. P 3


σ2 2 

= 1 –  /2 1 –
c 3

 –z
 1 – / 2 /2 

 J ie. P / cZc / = 1 –
2
z
2 
 1 σ 2
1 
 
P    

ie. 32 ns 32 = 1 – p –p
 1 – / 2 / 2 J P  –z  / 2 c
z
/2

pq
ie.  ns 2 ie. = 1 –
2 ns J
2
 = 1 –  n get
P  σ  
3 3

School of Distance School of Distance

  1 – / 2 /2 J As in the previous cases, on simplification we
 ns 2 ns 2
  
2 pq
pq 


ie. –z n cp c p  + z  / 2n


 = 1 – ie. P p 

= 1 –
P cσ c  
 J
3 y
 /2 1 – / 2 p is
J

2 So, the 100(1 – )% confidence interval for n
pq

Thus the 100(1 – )% confidence interval for σ is




 ns ns  –z pq , p  + z
, ie. p /2 n /2
32 y2


 J
L /2 1 – / 2 

School of Distance School of Distance

 4 Statistical Statistical 4
But, since p is unknown, we can replace p and q by their unbiased From normal tables we have P(|
estimators p and q. Thus the 100(1 – )% confidence interval for p z| c + 1.96) = 0.95
is ie. P(–1.96 c z c+1.96) = 0.95
  
  From this result we can write the 95% confidence interval for p 1 – p2
p – z/2
pq
,p + z / 
pq
2
L n 
n
where z / 2 can be determined from the normal tables for a given  .
 ( p 1q1 p 2 q 2 
 (  –  )–  n + n ,
as  1 2
p p 1.96 1 2

Note
When  = 0.05, z / 2 = 1.96, so the 95% C.I. for p is ( p 1 q 1 p 2 q 2  
(p  – p  )–1.96  +  
 
  
1 2 n 1 n 2

 Since p1, q1 and p2, q2 are unknown, they are estimated as
p  –1.96 pq
, p +1.96 pq
   
L n p1p1 , q 1 q1,p2 p2 and q  q2 .
n  2

hen  = 0.02, z / 2 = 2.326, so the 98% C.I. for p is The 95% confidence interval for ( p 1 – p2 ) is
 p  q p  q 
     


(p  – p )–1.96 +22
 ,
1 1

p – 2.326 pq
, p + 2.326 pq
1 2
nn
L n  1 2
n
hen  = 0.01, z / 2 = 2.58, so the 99% C.I.
for p is
q  pq 
(p  – p )+1.96
1 2 1 1 +2 2

   nn
p – 2.58 pq
, p + 2.58   J
Note: To construct 98% and 99% confidence 12intervals for
pq

L n p –p , we have to replace 1.96 by 2.326 and 2.58 respectively.

1 2
n 
Interval Estimate of the difference of proportions of
two binomial populations: SOLVED PROBLEMS
From the study of sampling distribution it is known that the difference
Example 1
of proportions obtained from two samples
( Obtain the 95% confidence interval for the mean (when σ known) of
ie. p  – p 2  → N p p 1 q1 + p 2 q 2
 a normal population N( µ , σ ).
1
p
– 1
 when n1, n 2are large

School of Distance School of Distance

4 Statistical Statistical 4
 n1 n2  Solution
( 
m z (p – p  ) – ( –p )   p
+ 2 2

Let x 1 , x 2 , x 2 ..., xn be a random sample of size n drawn from
p q 1 1 p q
L 1 2
2 1  
2
n n


N( µ , σ ). Let be its mean and σ is its variance. Then we know that

 1 2 
x

School of Distance School of Distance

 5 Statistical Statistical 5
x→N(µ,σ/ n) Solution

2
x – → N(0,1) To obtain the CI for σ let us use the result
mZ =
σ /n 2
From normal tables, we will get 2 ns 2

P | Z | c1.96 = 0.95 3  σ (n –1)

2 2
ie. P –1.96 c Z c1.96 = 0.95 2
table, we can find a 3 2 and 3 such that
From 3 0.975 0.025
 
x– μ
2 2 2

ie. P –1 .96 c c1.96 = 0.95 3

P 0.975 c 3 c 30.025 = 0.95



σ/ n J
P –1.96 .  ns 2 
 σ σ c3
 

ie P 3 σ 2 = 0.95
 c
  .75 . 5 

ie.  J
P –x –1.96 c x – µ c1.96 = 0.95
n n J
 ns 2
2 ns 2
σ σ  ie. P   = 0.95
ie. 3
c– µ c –x +1.96 σ  

n 3
= 0.95 2
 nJ  0.975 0.025 J
 σ σ 
µ x –1.96  ns 2
2 ns 2
e. P x +1.96 = .
 n nJ ie. P   = 0.95
32
cσ c 
σ σ 
c µ c x +1.96 3
2
n n  0.025 0.975 J
 J
 2 
ie. P x –1.96 = 0.95 2 ns2 , ns 2
School of Distance School of Distance
 5 Statistical Statistical is 3 2 5 and
 Thus the 95% CI for σ 3 3
 where 0.975

L 0.025 0.975 
2 2
Thus the 95% CI for µ 30 .025 are obtained by referring the 3
L n  probabilities 0.975 and 0.025 respectively.
table for n –1 df and

Example 2
Example 3
Obtain the 95% confidence interval for the variance of a normal Obtain the 99% CI for the difference of means of two normal
population N( µ , σ ) populations N( µ , σ ) and N( µ , σ ) when (i) σ, known (ii)
1 1 2 σ 2
2 1
σ,σ unknown.
1 2

School of Distance School of Distance

 5 Statistical Statistical 5
Solution Example 4
Case (i) when σ , σ Obtain the 99% confidence interval for the difference of means of
1 known.
2
two normal populations N( µ1 , σ1 ) and N( µ2 , σ2 ) when σ1 and σ2
Let X1 and 2 be two independently normally distributed random unknown, by drawing small samples.
variables with means µ and µ and variances σ 2 and σ respectively.
1 2 1 2 Solution
Let x and x be the sample means of n1 and n2 observations. Since σ and σ are unknown, they are estimated from samples, as
1 2

since 2 2
X →
X
N µ,σ
( ,
→N µ,σ
)
( ) 1 2

1 1 1 2 2 2 σ =σ = *2 , where
2 2 1 2 σ
( σ ( σ  2 2
then x 1 → N µ,
1
 , x2 → N µ2, 2
 n s +n s
 1   
2
n n *2 11 2

 σ =
 1 2
n 1 +n2 – 2
( σ2 σ  Then the statistic
therefore x1 – x2 →N µ –µ 2,
1
+ 2


 1 

(x –
n1 n2 ( x12–x ) – (µ – µ )
) – (µ – µ t = ns2 +n
12

s2 ( 11  ct (n 1 + n 2 –2)d
x
) 1 12 2
+
1 21
2
→ ( ) n +n– 2 n n
N 0,1
mZ= σ 2 σ 12 1 2 
+
1 2

n 1 n2 Thus we have
From normal tables we can write
P ( –2.58 c Z c 2.58)  0.99  
Substituting Z and simplifying, we get  
( x12 – x ) – (µ – µ 2 )
 –t  / 2 c
1

p c / 2  1–
 σσ1 + 2 c (µ – µ n21s + n22 s( 1 1
(x x ) – 2.58 )c
P
n n
1 2
+
 1 2 1 2 1 2 n 1 +n 2 – 2  1 2

  J

 where t / 2 is obtained by referring the t table for n 1 +n 2 – 2 df

σσ 

(x 1 – x2 )– 2.58 1

+ 2
 0.99 and probability   0.01.
n
n1 2 
J 2
 2 ( 1

σ 2 σ 2  
(x1 – x 2 ) –t / 2 
1 1 +n 2 s 2 1

+
c
– is (x 1 – x2 ) 2.58 1
+
2 This gives n 1 +n 2 –2

1
Thus the 99% CI for µ n 2 
1 2  1n 2 
5 Statistical Statistical 5
 J
School of Distance Education School of Distance Education
54 Statistical inference Statistical inference 55
 n s 2 +n s 2 ( 11  2
µ –µ c

 From table, 3 = 23.68
(x –x )–  112 2
 +  1 –
1 2  1
t
2 /2 n 1 +n 2 –2  1 2 
14,0.05

 J 3 = 6.571
Thus the 99% CI for µ – µ is 14,0.95
1 2 2 15  4.24 15  4.24 
 Thus 98% CI for σ is ,


(x 1 –x2) 
/2 
n 1 s1 +n 2 s 2 ( 1
+
1   23.68 6.571 J
n 1 +n 2 – 2  1 2  = (2.685, 9.678)
 J
Example 7
Example 5 A medical study showed 57 of 300 persons failed to recover from a
particular disease. Find 95% confidence interval for the mortality rate of
If the mean age at death of 64 men engaged in an occupation is 52.4
the disease.
years with standard deviation of 10.2 years. What are the 98%
confidence limits for the mean age of all men in that occupation? Solution
Solution n = 300, x = 57

Here n = 64, x  52.4, s  10.2 57


 s mp  n  300 0.19
98% CI for the population mean is x  2.326 q   1 – p  1 – 0.19  0.81
 nJ The 95% CI for the mortality rate is
 10.2 
     
ie. 52.4  2.326  p  –1.96 , p  +1.96

pq pq

=
nn
 64 J J
Example 6
 0.19  0.81
+ 1.96


0.19 – 1.96 , 0.19

300
A random sample of size 15 from a normal population gives x  0.19 
3.2
2 2 ie.  0.81
 J


and s  4.24 . Determine the 90% confidence limits for σ .

ie. {0.146, 0.234}
Solution
2
n = 15, s =  2 2 
4.24 n  ns 

The 90% CI for σ is 2 2 ,

3 3
0.05 0.95
School of Distance School of Distance
Example 8
A random sample of
16 values from a normal
population showed a
mean of 41.5 inches and
the sum of squares of
deviations from this
mean equal to 135
square inches. Obtain
the 95% and 99%
confidence interval for
the population mean.

School of Distance School of Distance

 5 Statistical Statistical 5
Solution
s 2 s2
2 2
Here n = 16, x  41.5, Σ(x – x )
1 2

i.  ns  135 The lower limit = ( x 1 – x2 ) – 1.96 + n

1 2
 s  n (8.83)(8.81)
/2
t n –1 J 580+ 786
The 95% CI for µ is x = (34.45 – 28.02) – 1.96

ie. from table, t15,0.05 = 2.131
= 6.43 – .95 = 5.48.
 3  s s
12
So the required confidence interval is 41.5  2.131
+n
 4J The upper limit = ( x 1 – x2 ) + 1.96 n
ie. {39.902, 43.098}
1 2 (8.83)(8.81)
ii. For 99% CI, t15,0.01 = 2.947 580+ 786
= (34.45 – 28.02) + 1.96
 3
m 99% CI is 41.5  2.947   = {39.29, 43.71} = 6.43 + .95 = 7.38.
 4J
m The 95% confidence interval for ( µ – µ ) is (5.48, 7.38)
1 2
Example 9
A certain psychological test was given to two groups of Army prisoners
(a) first offenders and (b) recidivists. The sample statistics were as follows. EXERCISES
Population Sample size Sample mean Sample S.D. Multiple Choice Questions
a) first offenders 580 34.45 8.83 The notion of confidence interval was introduced and developed by
) recidivists 786 28.02 8.81 a. R.A. Fisher b. J. Neymann
Construct 95% confidence limits of the difference of the means c . Karl Pearson d. Gauss
( µ1 – µ2 ) of the two populations. l The 95% confidence interval for mean of a normal population

Solution N( µ , σ ) is
The 95% confidence interval for ( µ1 – µ2 ) is s σ
a. x 1.96 n b. x 1.96 n
sn

 σ 2 σ 2 σ 22 σ  
σ
+2 x  2.58 x  2.58n
x x –1.96 1 +2 ) + 1.96
(x x 1 c. d.
 1 2 n1 n2
,
1 2 n1 2 
 J and s2.
Since σ and σ are unknown, we shall replace them respectively by s
1 2 1

School of Distance School of Distance

 5 Statistical Statistical 5
l The 100(1 – )% confidence interval for µ of N( µ , σ ) when
σ unknown, using a sample of size less than 30 is

School of Distance School of Distance

 Statistical Statistical 5
Give the 95% CI for the variance of a normal population
t s t s l

a. x /2 n –1 b. x /2 l Give the formula for obtaining confidence limits for the difference
n
between the mean of two normal populations
s s Why interval estimate is preferred to point estimate for
c. x tn –1 d. x tn
l

estimating an unknown parameter.

l A random sample of 16 housewives has an average body weight of
l What do you mean by confidence level?
52kg and a standard deviation of 3.6kg. 99% confidence limits
for body weight in general are Short Essay Questions
a. (54.66, 49.345) b. (52.66, 51.34) l Distinguish between point estimation and interval estimation.

c. 55.28, 48.72) d. none of the above Explain how you will construct 100(1 – )% confidence interval for
normal population mean when population S.D. is (i) known and (ii)
l Formula for the confidence interval for the ratio of variances unknown.
of the two normal population involves
l Explain how you would find interval estimates for the mean
2
a. 3 distribution b. F distribution and variance of a normal population.
c . t distribution d. none of the above l What do you mean by interval estimation? Obtain 99% confidence

Fill in the blanks limits for θ of the normal distribution N( θ, σ 2), with the help of a
random sample of size n.
l The notion of confidence interval was introduced and developed b y
l Explain the idea of interval estimation. Obtain a 100( 1 – )%
l The confidence interval is also called interval confidence interval for the mean of a normal distribution whose
variance is also unknown.
l An interval estimate is determined in terms of
l Obtain 95% confidence interval for the mean of a normal population
l An interval estimate with interval is best
with unknown variance on the basis of a small sample of size n
l Confidence interval is specified by the limits taken from the population. What happens when n is large?
l Confidence interval is always specified with a certain Long Essay Questions
l To determine the confidence interval for the variance of a normal l A random sample of 20 bullets produced by a machine shows an
distribution distribution is used average diameter of 3.5 mm and a s.d. of
0.2 mm. Assuming that the diameter measurement follows N( µ ,
Very Short Answer Questions σ ) obtain a 95% interval estimate for the mean and a 99% interval
l What is an interval estimate? estimate for the true variance.
l Explain interval estimation l The mean and s.d. of a sample of size 60 are found to be 145 and
l State the 95% confidence interval for the mean of a normal 40. Construct 95% confidence interval for the population mean.
distribution N( µ , σ ) when σ is known l Two independent random samples each of size 10 from two

School of Distance School of Distance

6 Statistical Statistical 6
independent normal distributions N( µ , σ ) and N( µ , σ ) yield

School of Distance School of Distance

6 Statistical Statistical 1 6 1 2
2

School of Distance School of Distance

 6 Statistical Statistical 6
MODULE 4
= 8.6,
x

1
= 4.8, s
1
x = 5.6 and s = 7.9. Find 95% confidence
2 TESTING OF HYPOTHESIS
2
interval for –µ.
1 2 Most tests of statistical hypothesis concern the parameters of
distributions, but sometimes they also concern the type, or nature of the
l Two random samples of sizes 10 and 12 from normal populations
distributions, themselves. If a statistical hypothesis completely specifies
having the same variance gave x 1 = 20, the distribution, it is referred to as a simple hypothesis if not, it is referred
s 2 = 25 and x = 24, s = 36. Find 90% confidence limits for to as a composite hypothesis.
1 2 2 The statistical hypothesis that X follows normal with mean 15 is a
( µ – µ ). composite hypothesis since it does not specify the standard deviation of
1 2
the normal population. The statement that X follows a poison distribution
. In a sample of 532 individuals selected at random from a population,
with parameter λ = 2 is a simple hypothesis since it specifies the
l

89 have been found to have Rh-ve blood. Find an interval estimate

population completely.
of the proportion of individuals in the population with Rh-ve
blood with 95% confidence. A statistical hypothesis which refers only to the numerical values of
unknown parameters of a random variable is called a parametric
l Of 250 insects treated with a certain insecticides, 180 were hypothesis. Eg. In a normal population if we test that whether µ = 10 or
killed. Set approximate 95% confidence interval to the value of not is a parametric hypothesis. A hypothesis which refers to the form of
p, the proportion of insects likely to be killed by this insecticides an unknown distribution is called a non parametric hypothesis. eg. The
in future use. form of the density function in a population is normal.
l Suppose a sample of 500 people were interviewed and 200 of Definition of terms
them stated they were in favour of a certain candidate as The following are definitions of some terms which are frequently
president. Obtain the 98% confidence limits for the population used in this module.
proportion in favour of the said candidate.
Test of Hypothesis
l 150 heads and 250 tails resulted from 400 tosses of a coin.
Rules or procedures which enable us to decide whether to accept or
Find 90% confidence interval for the prob: of a head.
reject the hypothesis or to determine whether observed samples differ
l A random sample of 500 apples was taken from a large significantly from expected results are called tests of hypothesis or tests
consignment and of these 65 were bad. Estimate the proportion of significance.
of bad apples by a 90% confidence interval. In our subsequent discussions we are concerned with hypothesis
l A sample poll of 100 voters in a given district indicated that about only one parameter.
55% of them were in favour of a particular candidate. Find 95% Null Hypothesis
and 99% confidence limits for the proportion. The hypothesis to be tested is usually referred to as the „Null hypothesis‟
and is denoted by the symbol H0. Thus a hypothesis which is set up with the
possibility of its being rejected at some defined probability value is called a
null hypothesis. For instance, if we want to show that students of College A
have a higher average IQ than the students of College B, then we might
formulate the hypothesis that there is no difference viz, H0 : µA = µB

School of Distance School of Distance

 6 Statistical Statistical 6
Alternative Hypothesis
Every test of H0 has values for the pair (α β) associated with it. It would
In the testing process H0 is either „rejected‟ or „not rejected‟. If H0 is
seem ideal if we could find the test that simultaneously minimises both α and
not rejected, it means that the data on which the test is based do not
provide β but this is not possible. Since each of α and β is a probability we know
that α  0 and β  0 ; that is 0 is the minimum value for each. No
sufficient evidence to cause rejection. But if H0 is rejected it means matter
that the data on hand are not compatible with some other hypothesis. This
other hypothesis is known as „alternative hypothesis‟, denoted by H1. The whatH0 and H1 state and what observed values occur in the sample, we
rejection or „non rejection‟ of H0 is meaningful when it is being tested could use the test: Accept H0. With this test we would never commit a type
against a rival hypothesis H1 .
I
Type I and Type II errors error, since we would not reject H0 no matter what the sample values were.
Research requires testing of hypothesis. In this process two types of Thus for this test α = 0 implies β = 1. The converse of this test, which would
wrong inferences can be drawn. These are called type I and type II errors. always reject H0 given β = 0, α = 1. Neither of this test is desirable,
Rejecting a null hypothesis H 0 when it is actually true is called type I because they maximise one of the two probabilities of error while
error or error of the first kind . minimising the other. Now our objective is to choose the decision rule that
will lead to probabilities of these errors being as small as possible.
Accepting a null hypothesis H0 when it is false is called type II error
or error of the second kind. Test statistic
These can be schematically shown as below.
The testing of a statistical hypothesis is the application of an explicit set
H0 true H0 false of rules for deciding whether to accept the null hypothesis or to reject it in
Action
favour of the alternative hypothesis, consistent with the results obtained
Reject H0 Type I error No error from the random sample taken from the population. As the sample itself is
set of observations, usually an appropriate function of the sample
Accept H0 No error Type II error
observation
is chosen and the decision either to accept a reject the hypothesis is taken
Any test of H0 will tell us either to accept H 0 or reject H0, based on
based on the value of this function. This function is called „test statistic‟ or
the observed sample values. Thus is not possible to commit both errors
„test criterion‟, in order to distinguish it from an ordinary descriptive statistic
simultaneously.
or s .
We will define
or estimator such as x
α = P(Type I error)
We can note that a test statistic is a random variable, being a measurable
= P(rejecting H0 given H0 is true)
function of random observations which are themselves random variables.
= P(rejecting H0 | H0) The test procedure, therefore, partitions the possible values of the test
β = P(Type II error)
statistic into two subsets: an „acceptance region for H0 and a rejection
= P(accepting H0 given H1 is true) region for H0.
= P(Accepting H0 | H1)
Critical Region
The basis of testing the hypothesis is the partition of the sample space
School of Distance Education into two exclusive regions, namely, the region of acceptance and region of
rejection. If the sample point falls in the region of rejection, H0 is rejected.
The region of rejection is called „critical region‟. Thus critical region is
School of Distance Education
64 Statistical inference the set of those values of the test statistic which leads Statistical inference 65
to the rejection of the null hypothesis. Critical region is denoted byx. The larger the value of 1 – β for fixed α the better is the test in
Acceptance regions is the set of those values of the test statistic for which general. We define sensitiveness of a test as its ability to ascertain the
we are accepting the null hypothesis. Every test is identified with a critical correctness of the alternative hypothesis when it is true for fixed α.
region w and we are facing embarrassing richness of potential tests. Here Thus, power is a measure of the sensitiveness of the test. Therefore if
other things are identical the comparison of two tes ts is the comparison
we want to find best critical region (BCR) w, guided only by the principle
of their respective powers.
of minimising the probabilities of errors of type I and II.
Critical value
The value of test statistic which separates the critical region and
Level of significance acceptance region is called „critical value‟. T he critical value is usually
The validity of H0 against that of H1 can be tested at a certain „level of referred to as Zα or tα depending on the sampling distribution of the test
significance. The level of significance is defined as the probability of rejecting statistic and level of significance used.
We now summarise the steps involved in testing a statistical
the null hypothesis H0 when it is true or probability of type I error. Actually
this is the probability of the test statistic falling in the critical region when hypothesis. Step 1. State the null hypothesis H0 and the alternative
the null hypothesis is true. So significance level is also called „size of the hypothesis H1. Step 2. Choose the level of significance α
critical region‟, „size of the test‟ or producer’s risk. It is denoted by α. α Step 3. Determine the test statistic
is usually expressed as a percentage such a s 1%, 2%, 5% and 10%. Step 4. Determine the probability distribution of the test statistic
ie., α = P(Rejecting H0 |H0) = P(x ϵ w | H0) Step 5. Determine the Best Critical Region
For instance, if the hypothesis is accepted at 5% level, the statistician Step 6. Calculate the value of the test statistic.
in the long run, will be making wrong decisions in 5 out of 100 cases. If Step 7.Decision: If the calculated value of the test statistic falls in the
the hypothesis is rejected at the same level, he runs the risk of rejecting a critical region, reject the null hypothesis H0,
true hypothesis about 5% of the time. otherwise accept it. ie., if the calculated value
The best test for a Simple Hypothesis exceeds the table value, reject H0, otherwise
Often the test statistic is to be determined by controlling α and β. The accept H0.
ideal thing is to minimise α and β simultaneously but in practice when α Neymann Pearson Theory of testing Hypothesis
is minimised, β becomes large and vice versa. Hence the attempt is to The conceptual structure of the theory is as follows. To test the simple
minimise β for a fixed α and if there exists such a test statistic, it is called
the „best H0 : 8 = 88 versus the simple H1 : 8  81 , based on a random sample of
test‟. size n, the solution is given by the celebrated Neymann-Pearson Lemma.
Power of a test This specifies the explicit form of the test (ie., critical region) which has
pre assigned probability of error of type I and a minimal probability of
The probability of rejecting the null hypothesis H 0 when it is actually error of type II. This is same as maximising power subject to the
not true is called power of a test and it is given by 1 - β. Power of a test is
condition that the type I error is a constant. This process is equivalent to
also called power of the critical region.
choosing a critical region of size α which has at least the same power as
ie., Power = P(Rejecting H0 | H1 is true)
any other critical region of the same size. Such a critical region is called
= 1 – P (accepting H0 | H1) the best critical region, abbreviated as BCR. The test based on BCR is
= 1 – P (accepting H0 | H0 is not true) called Most Powerful Test.
= 1 – P (Type II error) = 1 – β
School of Distance School of Distance
66 Statistical inference Statistical inference 67
SOLVED PROBLEMS H1 : 8 = 1 on the basis of a single observation from
Example l –8 x
f(x, 8 ) = 8e , x  0 , obtain the probabilities of type I and type II
Let X follows B(10, p). Consider the following test for testing H0 : p = errors.
1/2 against H1 : p = 1/4: “Reject H0 if Xc2”. Find the significance level
and power of the test. Solution
Solution –8 x
Given f(x, 8 ) = 8e ,x0
Given X → B(10, p)
P(Type I error) = P(Rejecting H0|H0)
Significance level = P(Rejecting
= P(X  1 8 = 2)
H0/H0) = P(X c 2 p = 1/2)

= P(X = 0) + P(X = 1) + P(X = 2) when p = 1/2
=  f ( x )dx when 8 = 2
0 10 1 9 2 8 1
= 10 C 0 ( 1  ( 1  +10 C ( 1  ( 1  +10C 2 ( 1  ( 1 
2  2
 2 2
 –2x ( e –2 x

  1   

2   2  = 2.e dx 2 
10
56 = – 2 
(1
[1 + 10 + 45] 1  1
= 
2 1024
= – (0 – e–2) = e 2 

Power of the test = P(Accepting H0/H0)

= P(Rejecting H0/H1) P(Type I error) = P(Accepting H0|H1)
= P(X c 2 p = 1/4) = P(X c 1 8 = 1)
= P(X = 0) + P(X = 1) + P(X = 2) when p = 1/4
1
0 10 1 9 2 8 –8 x
= 10
C (1 ( 3  +10 C ( 1 (  +10C ( 1 (3 = dx when 8 1
 3

  8e
0   2  0
4 4
 1

  4 4   4   4
9 1
+
8 3 1  1 – dx (e – x
 + 45   e
= (3/ 4)
L16
10
=  –1
16 16  =
84  0

=

8
(3/ 4)  = 5.25 (3/4)
8
– e ( –1
–1 ) –1

Exa 2 L 16 
mple
School of Distance School of Distance
 Example 3 = = 1 – e

If X  1 is the critical region for testing H0 : 8 = 2

Given a binomial distribution
against

School of Distance School of Distance

 68 Statistical inference Statistical inference 69
nC x p x n–
, x  0,1,2,3,4
Example 4
x
q
f(x, p) = 0, otherwise ‟ The hypothesis H0 : 8 = 2 is accepted against H1 : 8 = 5 if
X c 3 when X has an exponential distribution with mean 8 . Find type I
It is desired to test H0 : p = 1/3 against H 1 : p = 1/2 by agreeing to
and type II error probabilities of the test.
accept H0 if x c 2 in four trials and to reject otherwise. What are the
Solution
probabilities of committing.
Given X has exponential distribution with mean 8 , that is X follows
(a) type I error, (b) type II error
an exponential distribution with parameter 8 .
Solution 1 –x/
8 , x  0, 8  0
(a) α = P(type I error) ie., f(x)
= e
8
= P(reject H0 /H0 is true) P(Type I error) = P(Rejecting H0 | H0)
= P(X > 2/p = 1/3)
4 = P(X > 3|8 = 2)
=
 4Cx ( 1
x 4–x 
Σ   ( 2  =  f ( x ) dx when 8 = 2
x 3 3  3 
3
3 4
( 1 (2 4  3 (1 
= 3  + 4C4 
= 1 e–x / 8 when 8 = 2
3 3 3 8
3

2 1 1 1
= 4  +   
3 3 3 9
= 1e– x/2 dx
(b) β = P(type II error)
3 2
= P(accept H0 /when H1 is true) or
= P(X c 2/p = 1/2)  –x / 2
1  
e 

.
2 x 4–x = 2  –1/ 2
= Σ 4Cx (1  (1  J

3

x 0 2  2   0
 


4 1 3 2 2
P(Type II error)
4C ( 1  + 4 C ( 1  (1  + 4C ( 1  ( 1
=

School of Distance School of Distance

 –3/2 3/2 = P(Accept H0 /H1) = P(X c 3 | 8 = 5)
= – [o – e ] = e
2  2  2   2  2 3 1 –x/ 1( e–x / 5
3

 5

(1 4 4
e dx =
= + 4(1 (1
+6
4
=
11 = 05 –1/ 5


4 – 
5 0
2 2  2 2 3/5
= –e –1 = e
1

School of Distance School of Distance

 7 Statistical Statistical 7
Example 5
EXERCISES
Find the probability of type I error of the test which rejects H 0 if X >
1 – α in favour of H1 if X has pdf Multiple Choice Questions
8 –1 l An „hypothesis‟ means
f(x) = 8 x ,0  x 1 with H0 : 8 = 1 and H1 : 8 = 2. Find the a. assumption b. a testable preposition
power of the test.
c. theory d. supposition
Solution
l A hypothesis may be classified as
8 –1 a. simple b. composite
Given f(x) = 8 ,0  x 1 c. null d. all the above
x
P(Type I error)= P(Rejecting H0 | H0 true)
l A wrong decision about H0 leads to
= P(X > 1 –α 8 = 1) a. One kind of error b. Two kinds of error
1 c. Three kinds of error d. Four kinds of error
8 –1
= 8x dx when 8 = 1 l Area of critical region depends on
1– a. Size of type I error b. Size of type II error

1
c. Value of the statistic d. Number of observations

= dx = ( X) l The idea of testing of hypothesis was first set forth by

1– 1– a. R.A. Fisher b. J. Neymann
= 1 – (1 – α) = α c. E.L. Lehman d. A. Wald

Power of the test = P(Accepting H0 | H0 true) l The hypothesis under test is a

= P(Rejecting H0 | H1 true) a. simple hypothesis b. alternative hypothesis
c. null hypothesis d. none of the above.
= P(X > 1 –α 8 = 2)
l Rejecting H0 when H0 is true is
1
a. Type I error b. Standard error
8–
=  1 dx when 8 = 2 c. Sampling error d. Type II error
1– 8x
l H1 is accepted when the test statistic falls in the
1 a. critical region b. probability space
= 2x dx = 2α –α 2
1–

= (x 2 1

)
1 – (1 –
=
α)
2

1– 

School of Distance School of Distance

7 Statistical Statistical 7
c. acceptance region

l Power of a test is
related to
a. Type I error
Type II error
c. both (a) and (b) d.
neither (a) nor (b)

School of Distance School of Distance

 7 Statistical Statistical 7
l Level of significance is the probability of Short Essay Questions
a. type I error b. type II error l What do you mean by a statistical hypothesis? What are the two
c. not committing error d. any of these types of errors? Outline the Neyman-Pearson approach.

Level of significance is also called l Explain the following with reference to testing of hypothesis: (i)
l
Type I and II errors; (ii) Critical region; and (iii) Null and
a. size of the test b. size of the critical region
alternate hypothesis.
c. producer‟s risk d. all the above
l Distinguish between Simple and Composite hypotheses. Give one
Fill in the blanks example each.
l Explain the terms (i) Errors of the first and second kind; (ii)
l A hypothesis is a testable ..................
Critical region; (iii) Power of a test; and (iv) Significance level
The parametric testing of hypothesis was originated by ..................
l
in test of hypothesis.
and ..................
l Explain with illustrative examples the terms : two types or
l The hypothesis which is under test for possible rejection is called
error, critical region and significance level.
..................
l Explain the terms (1) Null hypothesis; (2) Level of significance; and
l .................. error is not severe than...............................error. (3) Critical region.
l Probability of type I error is called .................. l Explain the terms (i) statistical hypothesis; (ii) critical region
l Rejecting H0 when H0 is true is called .................. (iii) power of a test.
l Accepting H0 when H0 is flase is called .................. Long Essay Questions
Very Short Answer Questions l To test the hypothesis H0 : p = 1/2 against H1 : p > 1/2, where p is
l Define the term „test of hypothesis‟ the probability of head turning up when a coin is tossed, the coin
was
l Define simple and composite hypothesis l
tossed 8 times. It was decided to reject H0 in case more than 6 heads
Define null and alternative hypothesis turned up. Find the significance level of the test and its power
l Define type I and type II errors l if H1 : p = .7.
Define level of significance X1 and X2 are independent Bernoulli r.v.s. such that
l

l Define critical region. l

P(X1 = 1) = 8 = P(X2 = 1). To test 8 = 1/3 against 8
Define power of a test l
= 2/3 the test suggested is to reject if X1 + 2X2 > 1. Find the
Define test statistic power of this test.
l State Neymann Pearson lemma l Consider the problem of testing the hypothesis H0 : X has uniform
l Define a parametric test of hypothesis l distribution over (0, 3) against H 1 : X has uniform distribution
What is a statistical hypothesis. over (5, 7). If (5, 7) is the critical region. find the probabilities of
two kinds of errors.
l Define size of the test.
l Which is the first step in testing a statistical hypothesis? l Let X1,... Xn be a r.s. from N( 8 , 4). Obtain Best critical region
for testing H0 : 8 = 10 against H1 : 8 = 15 with a sample of size

School of Distance School of Distance

7 Statistical Statistical 7
n
= 16 and with level of significance 0.05.

School of Distance School of Distance

 7 Statistical Statistical 7
l To test H : 8 = 1 against the alternative 8 = 2 based on X which test statistic. Such a test is referred to as a two-tailed test or two-sided
test.
1 –x/8
has the p.d.f. f(x) = e , x  0 ; = 0. otherwise, the test On the other hand, if we are testing the null hypothesis
8
proposed is to reject if X > log4. Compute probabilities of H0 :8 80 against the one sided alternative H1 : 8 80 , it would seem
committing type I and II errors if this test is used. reasonable to reject H0 only if when 8ˆ is much smaller than 80 . Therefore,

LARGE SAMPLE TESTS in this case it would be logical to let the critical region consist only of the
left hand tail of the sampling distribution of the test statistic. Likewise, in
The statistical tests may be grouped into two. (a) Large sample tests
testing :8 80 against the one sided alternative H1 : 8 80 we reject
(b) Small sample tests. For small sample tests the exact sampling H0
distribution of the test statistic will be known. In large sample tests the H0 only for larger values of 8ˆ and the critical region consists only of the
normal distribution plays the key role and the justification for it is found
right tail of the sampling distribution of the test statistic. Any test where
in the famous central limit theorem. That is when the sample size is large
the critical region consists only one tail of the sampling distribution of
most of the statistics are normally or atleast approximately normally
the test statistic is called a one tailed test, particularly they are called left
distributed. Let Y be a statistic satisfying the conditions of the CLT, then
the statistic given by tailed and right tailed test respectively.
Y –E(Y)
Z → N(0,1), for large n. Best critical regions of z-test
V(Y) To test H0 :8 80 against H1 : 8 80

: 8 8 0
Here V(Y) is called the Standard Error of Y.
Y –E(Y) : 8 s 80
mZ → N(0,1) for the significance level α the best critical regions are respectively.
SEof Y 
c Ξ Z  –Z c Ξ Z  Z and, c Ξ Z  /2
If Z is chosen as the test statistic, the critical region for a given
significance level can be determined from normal tables. The test based For example, when
on normal distribution is called „normal test‟ or „Z test‟.  = 0.05, the best critical regions are respectively
To explain the terminology, let us consider a situation in which we want
c Ξ Z  –1.645 , c Ξ Z 1.645 and c Ξ Z 1.96
to test the null hypothesis H0 :8 80 against the two sided alternative
α a=
hypothesis H1 : 8 s80 . Since it appears reasonable to accept the null α 0 . 0 2 5
= 0 . 0 5 = 0 . 0 2 5

α= 0 . 0 2 5

hypothesis when our point estimate 8ˆ of 8 is close to 80 and to reject it

when 8ˆ is much larger or much smaller than 80 , it would be logical to let ––1 . 6 45 z = 0 +  – z=0 ––1
1.645+ . 9 6 z = 01 . 9 6 + 

the critical region consists of both tails of the sampling distribution of our
School of Distance School of Distance
 7 Statistical Statistical 7
Note:
When  = 0.01, the best critical regions are respectively (i) For  = 0.05,  = 0.02 or  = 0.01 we can fix the critical regions by
determining the critical values using normal area table. (Refer best
critical regions)
α= 0.01 α=0.01α=0.005 α 0.005 (ii) If the population is given to be normal, the test procedure is valid
=
even for small samples, provided σ is known.
(iii) When σ is unknown and n is large, in the statistic we have to replace
––2 . 236 z = 0 +  – z=0 2.236 +– –2.58z=02.58 +
σ by its estimate s.
When  = 0.02, the best critical regions are respectively
Example 1
A sample of 25 items were taken from a population with standard
α=0.02 α=0.02 α 0.01 α=0.01
= deviation 10 and the sample mean is found to be 65. Can it be regarded
as a sample from a normal population with µ = 60.
– –2 . 05 5 z = 0+  – z=0 2.236 +– –2 . 2 36 z =0 2 . 236 + 
Solution
Testing mean of a Population
Given n = 25, σ = 10, x = 65, µ0 = 60
By testing the mean of population we are actually testing the significant
We have to test H0 :µ = 60 against H1 :µ s 60.
difference between population mean and sample mean. In other words we
are deciding whether the given sample is drawn from the population having Let  = 0.05. The best critical region is c Ξ Z 1.96 . Here the test
the mean given by H0 . statistic is

Suppose we want to test the null hypothesis H 0 : µ  µ0 against one of

x – 65 – 60
n
the alternatives H1 : µ  µ0 ; H1 : µ  µ0 or H1 : µ s µ0 on the basis of a
Z σ/  25  2.5
random sample of size n from a normal population with known variance
2 10 25 
 2.5  1.96
σ .
Since Z lies in the critical region, H 0 is rejected.
For the significance level  , the best critical regions are respectively,
That is, the sample cannot be regarded as drawn from a normal
x Ξ Z  –Z , c Ξ Z > Z and c Ξ Z  Z /2 . population with µ = 60
The test statistic is
x– µ Example 2
Z 0
A news stereo needle was introduced into the market claiming that it
n σ/ has an average life of 200 hours with a standard deviation of 21 hours.
Using the sample data, calculate the value of Z. If it lies in the criticl This claim came under severe criticism from dissatisfied customers. A
region, reject H0 otherwise accept. (Usually we will accept H0 if the customer group tested 49 needles and found that they have an average
calculated value is less than the table value) life of 191 hours. Is the claim of the manufacturer justified?

School of Distance School of Distance

 7 Statistical Statistical 7
Solution
Calculate the value of Z using the sample information, and if it lies in
Given µ0 = 200, σ = 21, n = 49, x =191
the critical region reject H0 , otherwise accept it.
We have to test
Note:
H0 : µ = 200 against H1 : µ < 200
(i) When we deal with independent random samples from populations with
Let  = 0.05. The BCR is c Ξ Z  –1.645 unknown variances which may not even be normal we can still use the
The test statistics is test which we have just described with s1 substituted for σ1 and s2

x – µ0 191 – 200 63 substituted for σ 2 so long as both samples are large enough to
Z    21 3 invoke the central limit theorem.
σ/n 21 49
(ii) To test H0 : µ1 – µ2  δ against
Since Z lies in the critical region. H 0 is rejected.
That is, the claim of the manufacturer is not justified. H1 : µ 1 – µ 2  δ , H1 : µ 1 – µ 2  δ , H1 : µ 1 – µ 2 s δ
the procedure is exactly the same as in the case of equality of two
Testing the Equality of two population Means
population means. In this case the test statistic is given by
By testing the equality of two population means we are actually –
testing the significant difference between two sample means. In other
words we are deciding whether the two samples have come from Z (x 1
σσ x2
)– δ
1+

populations having the same mean.

In applied research, there are many problems in which we are interested n 1 n2
in hypothesis concerning difference between the means of two populations. Example 1
Suppose we want to test the null hypothesis.
Suppose that 64 senior girls from College A and 81 senior girls from
H0 : µ1 – µ2  0 (or H0 : µ1 = µ2) against one of the alternatives. College B had mean statures of 68.2” and 67.3” respectively. If the
standard deviation for statures of all senior girls is 2.43, is the difference
H1 : µ1 – µ2  0 , H1 : µ1 – µ2  0 or H1 : µ1 – µ2 s 0 , based between the two groups significant?
on independent random samples of sizes n1 and n2 from two populations Solution
2 2
having the means µ1 and µ2 and the known variances σ1 and σ2 .
Given n1 = 64, n2 = 81, x1 = 68.2, x2 = 67.3
For the significance level  , the critical regions are respectively
σ1  σ2  σ = 2.43
c Ξ Z  – Z , c Ξ Z  Z and c Ξ Z /2
We have to test

x1– x2 H0 : µ1 – µ2 = 0 against H1 : µ1 – µ2 s 0
The test statistic is Z
σσ 2
12 + Let  = 0.05. The BCR is c Ξ Z 1.96
12

School of Distance School of Distance

 8 Statistical Statistical 8

The test statistic is Z  x1– x2 47 – 49 –2

σ 2σ 2
23 + 30
2
=529 + 900
12 + =
10001500 1000 1500
n1 n2
–2 –2
68.2 – 67.3 = .529 + 0.6 = 1.129 =
=2.43+2.43 1.882
6481 = 2.21
m Z  1.882  2.326

Z  2.21  1.96 Since Z lies in the critical region, H 0 is accepted. That is, there is no
m
significant difference between the samples.
Since Z lies in the critical region, we reject H0 .
Testing the proportion of success of a population
That is, the two groups are significantly different with reference to By testing population proportion of success we mean the testing of
their mean statures. the significant difference between population proportion of success and
the sample proportion of success.
Example 2
Now let us familiarise the following notations.
A random sample of 1000 workers from factory A shows that the
mean wages were Rs. 47 per week with a standard deviation of Rs. 23. A
p : population proportion of success (unknown) p0
random sample of 1500 workers from factory B gives a mean wage of
: the assumed value of p (given)
Rs. 49 per week with a standard deviation of Rs. 30. Is there any
significant difference between their mean level of wages?
 x
Solution
p : n ; the proportion of success of a sample
Given n1 = 1000, n2 = 1500, x1 = 47, x2 = 49 x : the number of successes

s1 = 23 s2 = 30 n : sample size
We have to test
Suppose we want to test the null hypothesis
H0 : µ1 – µ2 = 0 against H1 : µ1 – µ2 s 0
H 0 : p = p0 against one of the alternatives
Let  = 0.02. The BCR is c Ξ Z  2.326
H 1 : p  p0 or H 1 : p  p0 or H 1 : p s p0 based on a large sample

x1– x 2 of size n whose proportion of success is p .

The test statistic is Z 
s s
For the significance level  , critical regions are respectively.
1
+ 2

n1 n2 c Ξ Z  –Z , c Ξ Z  Z and c Ξ Z  Z / 2

School of Distance School of Distance

 Statistical Statistical 8

The test statistic is Z  Let us consider the following notations.

p  – p0
p0q0 p1 : proportion of success of the first population
n p2 : porportion of success of the second population.
x1 : number of successes in the first sample
Calculate the value of Z and if it lies in the critical region reject H0, x2 : number of successes in the second sample
otherwise accept it.
n1 : first sample size
Example 1
In a survey of 70 business firms it was found that 45 are planning to n2 : second sample size
expand their capacities next year. Does the sample information contradict

1 : porportion of success of the first sample = x 1 / n1
the hypothesis that 70% the firms are planning to expand next year. 
Solution 2
: porportion of success of the second sample = x2/ 2

x 45 Suppose we want to test the null hypothesis

Here we have p = n = 70 = 0.643 H 0 : p1 – p2  0 against one of the alternatives

p0 =70% = 0.70, n = 70 H 1 : p1 – p2  0 or H 1 : p1 – p2  0 or H 1 : p1 – p2 s 0 based
Here we are testing on two independent large samples of sizes n 1 and n2 with proportions of

H 0 : p = 0.70 against H1 : p < 0.70 success p1and p2 respectively.

For the significance level  , the critical regions are respectively.
Let  = 0.05. The BCR is c Ξ Z  –1.645
p  – p0 p c Ξ Z  –Z , c Ξ Z  Z and c Ξ Z  Z
The test statistic is Z  0  – p
q p1 2

The test statistic is Z

0
n  1
p*q* 1
+
n n
0.643 – 0.70  1 2 
= 0.7  0.3 = 1.04 n 1 p1
 
+n 2 p 2

70 n1
where p* = +n2
Since Z lies in the acceptance region, H0 is accepted. That is, 70% of the proportions of success.

firms are planning to expand their capacities next year.

Testing difference of two population proportions
By testing the difference of two population proportions we are testing
the equality of two population proportions or the significance difference
between two sample proportions. In other words we are deciding whether
the two samples have come from populations having the same

School of Distance School of Distance

 Statistical Statistical 8
and q* = 1 p*
Calculate Z and if it lies in the critical region, reject H0, otherwise accept it.

Example 1
Before an increase in excise duty on tea 800 persons out of a sample
1000 perons were found to be tea drinkers. After an increase in duty 800
people were tea drinkers in a sample of 1200 people. Test whether there
is significant decrease in the consumption of tea after the increase in duty.

School of Distance School of Distance

 8 Statistical Statistical 8
Solution We are testing

H 0 : p1  p2 against H1 : p1 s p2
 800  800
We have = 1000 = 0.8 .
p1 2 1200

Here we have to test Let  = 0.01. The BCR is c Ξ | Z|  2.58

H 0 : p1 – p2  0 against H1 : p1 – p2   
0 The test statistic is Z
p
1

– 2

 1
Let  = 0.05. The BCR is c Ξ Z 1.645 p*q* + 1
 
n
The test statistic is Z p1

(
– p2

 1 1  n1 2 
p*q* + + 
n p
n p
450 + 450 900
n1 2  1 1 2 2
=  = .
 
n p
800 + 800 1600 Now p* =n 1 600 + 900 1500

+n2
+n 2 p 2
1 1
 q* = 1 p* = 1 0.6 = 0.4
Now p* = n 1 +n2 1000 + 1200 2200 = 0.727
q* = 1 p* = 1 0.727 = 0.273 Z = 0.75 – 0.50 = 9.68
11
0.80 – 0.67 0.6  0.4 +
Z= = 6.816  600900 
1 1
0.727  0.273  + 
Z  9.68  2.58
10001200  m

m Since Z lies in the critical region, H0 is rejected.

Z  6.816  1.645
That is, the two cities are significantly different w.r.t. prevalence of
Since Z lies in the critical region, H0 is rejected. smoking.
That is, there is a significant decrease in the consumption of tea after
the increase in duty.
EXERCISES
Example 2
In a sample of 600 men from city A, 450 are found to be smokers.
Out of 900 from city B, 450 are smokers. Do the data indicate that the 450 450
cities are significantly different with respect to prevalence of s moking.  
Solution
School of Distance School of Distance
 8 Statistical Statistical 8
Multiple Choice Questions 100
l Large sample tests are conventionally meant for a sample l A parametric test is performed as a large sample test using
a. central limit theorem b. Techebysheff inequality
size a. n = 20 b. n < 30 c. n  30 d. n =
= = c. Weak law of large numbers d. none of these
Here 600 = 0.75 p2 900 = 0.50
p1

School of Distance School of Distance

 8 Statistical Statistical 8
l For a two failed test with  = 0.05, the best critical region of a l The critical value of one sided left tailed Z test, for α = 0.05 is
Z test is .....................
a. Z < 2.58 b. Z > 2.58 l A two sided test is used for testing a null hypothesis
H 0 : µ = µ 0 against .................
c. Z 1.96 d. Z  2.58
Very Short Answer Questions
l To test H0 : µ  µ0 against H0 : µ  µ0 when σ is known, the
l Distinguish between large sample and small sample tests. l How
appropriate test is
will you decide the best critical regions of a Z test?
a. t-test b. Z-test
l Give the Z statistic to test the mean of a population when σ is
c. F-test d. none of these known.
l To test H0 : µ  500 against H0 : µ  500, we use l State the test statistic for testing H 0 : p1  p2 against H1 : p1 s p2
a. one sided left tailed test .
b. one sided right tailed test Short Essay Questions
c. two-tailed test d. all the above l Distinguish between large sample and small sample tests
illustrating with suitable examples.
l Testing H0 : µ  200 against H0 : µ s 500 leads to
l Explain the importance of normal distribution in large sample tests.
a. left tailed test b. right tailed test
l Discuss the use of standard error in large sample tests. Illustrate
c. two-tailed test d. none of these with an example.
l To test an hypothesis about proportions of success in a class, the l Explain the method of testing the significance of the difference
usual test is between a large sample mean and population mean.
a. t-test b. F-test c. Z-test d. None of these Long Essay Questions

Fill in the blanks l An electrical firm manufactures light bulbs that have a length of
life that is aporoximatey normally distributed with a mean of 800
l A test based on the outcome of tosing of a coin is a.....................test.
hours and a standard deviation of 40 hours. Test H0 : µ = 800
l When σ is known, the hypothesis about population mean is
tested by ................ hours,

l If the smple drawn from a population is large, then the hypothesis against the alternative H1: µ s 800 hours if a random sample of 30
about µ can be tested by ................ bulbs has an average life of 788 hours.
l A large population of heights of person is distributed with mean l A random sample of 36 drinks from a soft-drink machine has an
66 inches and SD = 10 inches. A sample of 400 persons had the average content of 7.4 ounces with a standard deviation of 0.48
mean
ounces. Test the hypothesis H0 : µ =7.5 ounces against H1: µ <
height = 62 inches. The data..........the hypothesis H0 : µ = 66 inches.
7.5 at  = 0.5
School of Distance Education School of Distance Education