1

Hypothesis Testing
One sample (small) -single population.
Consider a population where x is normally distributed with mcan u and standard deviation G.
Let x, x) be a randomLM:
sample obtained from this distribution. We shall denote byx the
sample mean ofN:
and
.2
x = the sample s' =

n-1i=l
There is a difference between sand
In' 2
the divisor is n- 1, which makes it is an unbiased estimator of population variance
or o'. For

Now we consider the following cases:

Case I: Hypothesis test about population mean u: o known. (i..,4 unknown, o known)
Object - Toinvestigate the significance of the difference between an assumed population
mean (say uo) and sample mean x.
Hypotheses:
Here we may be test the hypotheses:
(A) Ho:p= (B) Ho*= o (C) Ho:=o
HË:tAo Hj:p>o Hju<
Method:
Under the null hypothesis we consider the test statistic:

T=
which follows a standard normal distribution under the null
Vn
hypothesis.
Let the level of significance be 100(1- a)%. At this level of significance we shall proceed as
follows:
(A) Incase of alternative hypothesis H:p#po, if for the given sanmple the numerical value of
the observed value (statistic) of t greater than the tabulated value we reject null hypothesis,
i.e., if T > Ta.Otherwise, we shall accept the null hypothesis.
2

(B) In case of alternative hypothesis HË:p> owe shall reject the null hypothesis if the
observed value of t on the basis of the sample observations greater than the tabulated value,
i.e.,r > r, .Otherwise, we shall accept the null hypothesis.
 2

()In case of alternative hypothesis H1:u < Mo we shall reject the null bypothesis if the
Observed value of t on the basis of the sample observations less than the tabulated value, i.e.,
7<Ta Ort <-T,.Otherwise, we shall accept the null hypothesis.
As regards the problem of interval estimation of 4, it has been shown that the limits
x-Talz X and x+Tai2 X , computed for given sample, are the confidence limits to u

with confidence coefficient 1 - .

It follows that -

P -Tal2 STal2=l-a

< (-uwn XSTal2

-TainXs(i-)sta/2X=1-a
Vn

P-* -sx-u-xStal2

P-tan -xs-u[Tal2
=1-a

x-Tal2 X is called the lower limit and x+

t i s the upper limit.
Vn Vn
Sum 1: The breaking strength of pieces of atype of string has mean 8.19 kg. and standard
deviation 0.95 kg. A new method of manufacture of strings is supposed to give a higher mean
breaking strength, but for this the standard deviation is known to be practically the same.
Fifteen pieces manufactured by the new method have breaking strength (in kg.)as follows:
8.05, 7.56, 8.28, 8.10, 9.81,9.45, 8.64, 7.16, 8.91, 8.68,9.23, 7.81, 7.60, 7.92, 9.86.
Do these results indicate that the new method is really a better one?
Solution: Let us denote the breaking strength of a string by x. The method of manufacture
may be said to be better than the existing method if in the population of strings produced by
the new method x has mean (u) greater than 8.19 kg. We have thus to test a null hypothesis in
this case, V1Z.,
Ho:u = 8.19 kg., against the alternative hypothesis H:> 8.19 kg.
For this purpose, we assume (a) that in the population x is distributed in the normal form and
(b) that the given values of x, say
x{i=1,2, 15)are a random sample from the distribution.
The population standard deviation of xis given: G=0.95 kg.
Again, for the given sample x= 8.472 kg., so that
 3

T=V15( x8.19)/0.95 =...=1.150

Since it is smaller than 2.326 as well as 1.645, the hypothesis Ho is to be accepted at both 1%
and 5% levels of significance. In other words, the newmethod of manufacture does not scem
to be superior to the old method.

Case 2:u known, G unknown.

Let x, x2,..x,be a random sample obtained from this distribution. We
shall denote by x the sample mean of x:

1
X X
i=1
One may be interested in testing a hypothesis regarding o or in estimating, o. A sufficient
n

statistic for o is Xx,-u', . It is seen that x, is a normal

variable with mean u and standard deviation, o. Hence

n

i=1

being the sum of square of n independent standard normal variables, is distributed as a z.

with df=n-1
For testing HÍ :0 Oo, We make use of the fact that the statistic
n

i=l i=l

is a y' with df= n- 1under this hypothesis.

(i))HÍ :0 = 00 (ii) HÍ : ¡ = Oo (iii)HÍ :G= Oo
Hj:o~0oH}: >ooH: o<o0
(a) For the alternative HË: g>oo, HÍ is rejected in case for the given sample the value of the
statistic exceeds X´a,n
(b) For the alternative HË: o<oo, HÍ is rejected if for the given sample the value of the
statistic is smaller than x´t-4 n
(c) For the alternative H;: G#00, HÍ is rejected if for the given sample the value of the statistic
smaller than z'u2. n or exceeds z, n In each case a denotes the chosen level of
significance.
 4

Case IIl:u and o both unknown.

Here, x and s are jointly sufficient for uand o.
(1) u Test: Here to test H: ! =o or to have confidence limits to u, one cannOt use n
statistic n(r-u)lo, since o is unknown. gis in this case replaced by its sample estimate,

The resulting expression will be

Vn(r-ls
Now, it has been shown that

(n-1)/s'2

is a x´ with df =n - 1and is distributed independently of x. Thus,

Vn (x - )
(n- 1)s?
/n - 1

being of the form , where y has df=n-1and is independent of t, is

ly'/n -1
distributed as t with df= n1.
To test Ho: o we may, therefore, use the statistic

Vn(r-)/s,
Which is distributed t with df =n -1, under HÍ. We shall have to compare the value of the
Ia ,n - I
statistic computed from the given sample with or- 0 r t with al2,n -1 ,
according as the alternative of interest isH;:u > o or HËp < o or H1:u # L. This
procedure is called Student's - test
Sum 3: The following are 12 determinations of melting point of a compound (in degrees
Celsius) made by an analyst, the true melting point being 165°C. Would you conclude from
these data that his determinations are free from bias?
164.4, 169.7, 163.9, 162.1, 160.9, 160.8, 161.4, 162.2, 168.5, 163.4, 162.9, 167.7

Solution: The determinations made by the analyst may be said to be unbiased if the mean
determination in the population, that could be obtained if he took an infinite number of
readings, can be supposed to the 165 degrees. We have, therefore, to test the null hypothesis
Ho :å = 165, against all alternatives
HË :u 165
 5

Itwill be assumed (a) that the population distribution of determinations is of the normal type
and (b) that the sample observations are random and mutually independent.
Under the assumptions, a test for Ho is provided by the statistic
1= Vn x( X- 165)/s'.
Which have degrees of freedom =n-1
For the given observations,
n= 12
X= 163.992

and s´ = Vn-j2r-x)°=3.0:9
1=..... =-1.149
(ii) s Test:
Wemay have the problem of testing Ho:0 Go or the problem of obtaining confidence limits
to o.
From what has been said above. it is clear that

(n-1)s'

is, under the hypothesis Ho ay' with df= n-1. This provides us with tests for Ho. The value
of this statistic, computed from the given sample, is compared with
2
%an-1 or %´1-a.n- 1, according as the alternative is
H): G>o0 or H;: g<oo. For the alternative, H:0÷00, on the other hand, the computed value is
to be compared with both z'-u2,n-jand 'u2,n-|.Ho being rejected if the computed value is
smaller than the former or exceeds the latter vale.
Sum 2:A firm manufacturing rivets wants to limit variation in their length as much as
possible. The lengths (in cm.) of 10rivets manufactured by a new process are:
2.15, 2.01, 1.99, 1.98, 2.05, 2.03, 2.12, 2.25, 2.17, 1.93
Inthe past, the standard deviation of length of rivets manufactured by the firm has been
seems to be superior 0. 145 cm. Examine whether the new process seems to be superior to the
old.
Solution: Ifo be the standard deviation of length for allrivets manufactured by the new
process, then this nmay be considered superior if g< 0.145
The nullhypothesis is then
Ho:0 =0.145, which is tobe tested against the alternative, H,:0<0.145.
Under the usual assumptions, thetest willbe given by the statistic,

X X

2
(0.1 4 5
Which is, under Ho, a y´ with degrees of freedom (df) = n-1. For the present data,
 6

X (1.98
=(2.15 - 2.068)' + (2.01-2.068)' + (1.99- 2.068) +
2.068)' + (2.05- 2.068)' +
(2.03-2.068 )' + (2.12- 2.068)y' + (2.25- 2.068)' + (2.17- 2.068)' + (1.93- 2.068)*
= 0.006724 + 0.003364 + 0.006084 + 0.007744 + 0.000324 + 0.001444 + 0.002704 +
0.033124 +0.010404 + 0.019044 = 0.09096
Where,

2.15+2.01+1.99+1.98+2.05+2.03+2.12+2.25+2.17+1.93 20.68
X= =2.068
10 10
Hence the observed value of the statistic is

i X

(0.1 4 5 )
0.09096 0.09096 0.09096
(0.145) 0.145x 0.145 0.021025 =4.326
Now, the critical values at 1% and 5% levels of
significance are X1-a), n-1 Or X1-0.01), 10-1 Or
Z099,9= 2.088 or X1-a), n-! or X1-0.05), 10-1 Or Xo 95.9 3.325
thus insignificant, the null hypothesis is to be respectively. The observed value is
be superior to the old. accepted; i.e., the new process does not seem to