Testing of Hypothesis

78
10 9.6
S-o = 0.625
statistic ZZ = SE .64
The value of the test
59% level is 1.96 Thus, the calculated value of
value of Z at the null
But, the critical
than its critical value. This suggests that
significantly hypothesis
differ is
Zis much less
Hence, the S.D. of boys'
marks do not from
quite correct.
the pooled S.D. of marks.
(i) In the 2nd place
Let us have, HÍ:S> = by
of difference between S> and o is given
The S.E.
n1
SEs,-o = 2(n1 + n2) n2
50 92.16
(9.6) 180
=
V 2(50 + 40) 40

460.80 = V.64= 0.8

=
720
S2 + o 10 9.6 = 0.5
Z=
The value of the test statistic S.E. 0.8
5% level is 1.96.
But, the table value of Z at hypothesis holds good
The calculated value being less than this, the null
girls' marks, also, donot significantly differ from the pooled
Hence, the S.D.of the
s.d. of marks.

hypotheses in Sampling of variables of smnall size

15. Test of

Meaning
By the term, 'sampling of variables of small size' we mean selecting samples
which can assume any value within a wider
of 30, or less units from a population income, expenditure etc.
range of a variable like height, weight, diameter,
Characteristics
enumerated as under:
The chief characteristics of asmall sample may be
to 30 at the maximu
(1) The size of the sample is conventionallylimited such sampls
(ii) The distribution of any statistic like, Mean, S.D. etc. of
may, or may not take the shape of anormal distribution.S.D. etc. ofsuct
(ii) It is not possible to substitute any statistic like Mean, for deternuit
asample for the respective parameter of the populationof large samples
the standard error of any difference as it is done in case 'scienca
pihysical
(iv) Such a sampling is preferredto in the fields of biology,costlylaboratori|
etc. where it is very expensive to collect a datathrough eithe
or ficld experiments. symmetrically,
on

(v) The Mcans of such samples vary grcatly, though

ide of the true Mean of the population.
Testing of Hypothesis 79

(vi) The standard deviations of such samples tend to be smaller than tnat o
their universe,and smaller the sample. the smaller would be the standard
deviation.
(Vn) The standard errors that are calculated on the basis of the standard
deviations of such samples tend to be smaller, and under estimated in
compariSon to the actual errors.
(viii) The test statistic which is extensively used in case of large samples is not
applicable in case of such samples. Instead, a special test statistict(Z,
in certain cases) is applied to such samples for testing the signiticance
of differences and various null hypotheses.
(ir) The statistical distribution of such samples takes the form of a
(-distribution, the details of which are discussed here as under.
(i) t-distribution and t-test.

At-distribution is asampling distribution of smallsamples only. The theory

of such adistribution was developed first in 1900 by Sir William S. Gosset
(1876-1937), an employee of the Guinness &Son, a Dublin and Brewery Company
of Ireland to deal with the problems on small samples. This theory was published
in 1905 by Mr. Gosset under his pseudonym, or pen namne, 'Student', as his
employer did not allow him to publish any such research work in his own name.
Hence, this distribution is popularly known after his pen name, as (Student's t
distribution, or simply (-distribution.
According to Gosset'stheory, a t-distribution takes the form of a bell shaped
under :
frequency curve like that of a Normal distribution shown as
Normal distribution

t-distribution for 12 degrees

of freedom

t-distribution for 3 degrees

of freedom

Important Features
important features :
A (-distribution has the following
shaped curve which is symmetrical.
(i) It gives a normal bell i.e. the degree
(Gii) The shape of its curve is determined by the only parameter in such a
of freedom V(nue) or the siZe of the sample (n). A change shown in the
as
parameter brings a change in the Shape of the curve
above diagram.
(i) It is higher than the normal distribution at both the tails of the curve
the Mean (u)
but lower in height at the point of
 Testing of Hypolhe
80
(iv) I has larger arcas at its tails than a normal distribution, and So, il assig
extreme outcomes.
higher probabilities to like that of a normal
zero distribution.
() Its Mean is always unity (). If the sample size
ís greater than
(vi) Its Variance
Variance approaches
then itgives rise to a
unity,
normal
and if the sample
distribution.
sizc
becomes incrvereasyeS,
Yarg
like the distribution of 7 2
(vii) It has a set of critical valuesfor testing the magnitude of the and F
These values are referred to,
value of the test statistic t ;
(viii) Its value ranges from -ve infinity
to + ve infinity.
calculate, .
(x) Its techniques can be used for testing the hypotheses not only in Smal
samples, but also, in large samples, although, techniques of1large sampls:
can not be used in small samples for the purpose.

Assumptions
While dealing with a (-distribution for hypothesis testing, or for fixing te
fiducial limits etc. the following assumptions are made about the distribution:
(i) It is assumed that the parent population is normal.
(iü) It is assumed that the samples have been drawn at random.
(iü) It is assumed that the population standard deviation (o) is not knoy.
Applications
A t-distribution provides us with a powerful instrument called t-test statistic
which can be applied to a number of situations for testing the hypotheses, or the
significance of differences. The following are some such situations to which tht
technique of t-distribution applys advantageously :
(i) When there appears a difference in Means.
(ii) When there appears a difference betweentwo independent sample Means
(ii) When there appears a difference between two dependent sample Mca
(iv) When there appears a difference in an observed correlation coefficien.

t-Test Statistic
The fundamental test statistic of a t-distribution is given by thefollowing mode!

S/yOr
Where, VS/n
= the test statistic of a
= the Mean of a
t-distribution to be calculated.
small sample i.e. EX
= the actual, or
S = the hypothetical
standard deviation Mean of the
of the sample i.e. population.
Testing of Hypothesix 81

VE(X-N?
(X-X)
n+1
=size of thesample i.e. s 30 and d deviation from the assumed mean.
If thevaluc of the sample s.d. is given without using n -I as thedenominator,
then the value of t will be calculated by
X-4 Or X-4
S/Vn-1 VS²/n-1
The above fundamental model of the (-test statistic is, howevcr, subject to
necessary modifications in diferent spccial situations which will be shown at
their appropriate places a little later.
Fiducial limits

When it is desired to estimate the fiducial limits of the population Mean at

aparticular probability level for a given degree of freedom (v), the same can be
estimated rcadily by the following model :
Lmts =X+la (v)
Where, X =sample Mean
la (v) = critical value of t at the desired probability level (a) for the
given degrees of frecdom (v).
S
And = Standard error of the diference of Mean.

t-Table
The table showing the critical, or the expected values of the test statistic t
at different levels of significance, or confidence for different degrees of freedom
(v) is known as at-table. A specimen of such a table is given at the end of this
book for reference, while comparing the calculated value of twith its corresponding
critical value given in the table. As a rule, if the calculated value of t happens
to be greater than its critical value, the null hypothesis is taken as rejccted, and
In the reverse ca[e, the same is takén as accepted.
The (-tables referred to, were compiled by Prof. Fisher and Prot. Yaes
tails
They are given in two different forms viz : (i) one exclusive of both the
arca, and (ii) the other exclusive of cither of the tails urea.
Liven at various
However, the crit ical values of t in both the lorms are
significance level for different degrees of freedom. can be
one tailed test tailed
If we refer to the two tailed table, the value of tor aavalue of thc (wo
located by looking at the (-value for 2 (i.e. l0 per cent
a given degrece
lest is equal to 5 percent value of a one tailed test)). Thus, t for the two tailed
for
of freedom at 1% level for a one tailed test = I for 2 percent
!est.
82 Testing of Hypothesis
We can casily check up the following critical values of t at the 95% level
from the t-table given at the end of this book :
Degree of freedom (v) One tailed t.0s two tailed ,025
5 2.015 2.571
10 1.812 2.228
15 1.753 2.131
20 1.725 2.086
25 1.708 2.060
30 1.699 2.045
1.645 1.960

The following illustrations will show how (-test of hypotheses are carried out
under different situations pertaining to small samples, where there appears a
difference in the Means and the population s.d. is not known.
(0) To test the Significance of the Mean of a Random Sample
ILLUSTRATION 31.
The Thomson T.V. Company claims that its T.V.
800 days. A random sample of 10 such sets give the sets have a mean lite of
following data:
Life in days : 540 600 700 780 720 750 810 840 870 930
In the light of the above facts, state if the
5% significance level.
claim of the company is valid a.

SOLUTION
Computation of the Mean, and S.D. of the
Samples
Life in days Deviations Square of Deviations
X
540 X-X(1s4)
-214
(X-X)'
45796
600
-154 23716
700
-54 2916
780
26 676
720
-34 1156
750
-4 16
810
56 3136
840
86 7396
870
116 13456
930 30976
176
Total 7540 n = 10 129240
 83

Testing of Hypothesis
2AW = 754
sample is given by V= 10
the
The mean of 129240
V20®-X)
by s= 10-1
The s.d. of the sample is given n-1
119.83
= V14360 =

Thus we have, = 9, and a = .05

800, n = 10, v = n-1
X=754, S = 119.83, u
Let us have, HÍ: X = u
(> two tailed test)
And
size of the sample is < 30, the relevant test statistic is given by
Since the
754-800 46 = 1.2131
t= 119,83 37.893
V10
significance
But, the criticalvalue of t for 9 degrees of freedom (i.e. v) at 5%
level (i.e. a = 05) for the two tailed test or tos (9) = 2.62.

Conclusion
The calculated value of t being less than its critical value, the differen ce
between Xand is not significant. This suggests that the null hypothesis
(Ho) is accepted, and thus the claim of the company that its T:V. sets have a
mean life of 800 days is quite valid.
Aliter

If the s.d. of the sample is calculated with the denominator n rather than
n-1, the value of the factors, S andt will be as follows:
S= V E(X-X 129240
=113.68
n 10
754 800 46
And = 1.2131
113.68 37.893
Vn-1 V10-1
Thus, in this case also, the conclusion will remain the same as above.
ILLUSTRATION 32.
Arandom sample of 25 spinsters gives a Mean of 2 issues with a sum of the
squares of deviations from the Mean of 16. From this, can it be said at 9se
and 99% probability levels that the sample is from a prostitute population having
a Mean of 3 issues ? Also, determine the fiducial limits of the population Mean
at the above twO confidence levels.
SOLUTION
Given, X =2,u = 3, 2 (X-X) = 16, a = .05, and .01, n = 25, and hence,
the degree of freedom or v = 24