d.

f = n-1 = 12-1 = 11
x = 0.65
s2 = 1.568 g2

1.568
sx 0.36
12

t = 1.81

Tabulated t 0.05(2), 11 = 2.201

Since the tabulated value is larger than the calculated one, we do not reject the Ho.

10.2.1 Two-sample hypothesis: Equal and unequal variances

In the two sample testing, we are interested in whether the two sample means are
different or not. To carry out the test, we have to obtain the difference between the two
sample means and divide this by the standard error of the difference between them. For
two-sample test the t is calculated as:

x x2
t
S x1 x2

Where:
x1 x2 = Difference between the two means.

S x1 x2 = Standard error for the difference between the 2 means

When x1 x2 is too large, then x1 x2

Calculation of S x1 x2 depends on whether

1. The two populations have a common variance, s2

2
2. Value of of the common variance are known or estimated.
3. The 2 samples are of the same size.

85
 4. The observations are paired.

Chance of rejection region depends on:

1. Level of significance chosen

2. Sample size
3. Test required, whether one or 2-tailed.

When testing two samples from two populations, we have to know whether the two
populations have equal variances or not, before testing.

For two samples, the variances of the samples are calculated and then compared. The
outcome of the variance test is the one that determines whether we use the test procedure
for equal or unequal variances.

Test of equality of variances:

H 0 : S12 S22

H A : S12 S 22

= 0.05
S12 S22
F or F , the larger variance is always the numerator.
S22 S12
Working example

T1 41, 34, 33, 36, 40, 25, 31, 37, 34, 30, 38
T2 52, 57, 62, 55, 64, 57, 56, 55

n1= 11
d.f.1 = 10
S12 = 21.87

86
 n=8
d.f.2 = 7
S22 = 15.36

S12 21.87
F 1.42
S22 15.36
Tabulated F0.05, 10,7 = 3.64, its value is greater than the calculated value of 1.42. We
therefore do not reject H0 and conclude that the two variances are not different.

Use of F tables: Tables Annex 4& 5, pages147-148 Table A.6:

The F table has the first column on the denoted as denominator d.f. This
represents the degrees of freedom of the variance denominator F value. The
first row on top of the table represents the numerator degrees of freedom. The
second column represents the probabilities associated with various critical
values of F that are given within the table body text. As in the t-tables, we are
only interested in the 0.05 probability. To obtain the value of Fin the tables,
we select the numerator and denominator degrees of freedom, select the 0.05
probability value and move on the F value where the numerator d.f and
denominator d.f intersect, this value is the one selected as the critical F value.
In the example above, this 3.64 for numerator d.f of 10 and denominator value
of 7.

If the two variances are not different, then they be pooled together in calculating t. A
variance common to the two samples is calculated and this is referred to as the pooled
variance. This is calculated as shown below:

(n1 1) S12 (n1 1) S22

Pooled variance, S P2
(n1 n2 ) 2

(df1 ) S12 ( df 2 ) S22

Or simply, S P2
df1 df 2

2 S2P S2P 2 S 2P S P2
S x x2 + S x x2 +
n1 n2 n1 n2

87
 x1 x2
t
S P2 S P2
n1 n2

Working example
Two samples with the following values:
A B
N 10 8
x 26.5 32.5
2
S 3.621 3.88

32.9 31.04
S p2 3.97
10 8 2
Sp 3.97 1.93

26.5 32.51
t 6.55
1.93 1.93
10 8

d.f = n1+n2-2 = df1 + df2

d.f = 10 + 18 – 2 = 9 + 7 = 16

Tabulated t0.05 (2), 16 = 2.12;

W reject the H0, since the calculated value is larger than the tabulated value and conclude
the two sample means are different.

10.2.2 Unequal variances

If variances are unequal, we do not calculate the pooled variance. We use a different
version of the t-test that gives a correct, though approximate result.

88
The t is calculated in the same way as previously, but instead of calculating a pooled
variance, the individual variances are used as shown below.

x x2 x1 x2
t
Se( x1 x1 ) S12 S22
n1 n2

The difference in the t-test calculation involves obtaining the degrees of freedom. Instead
of adding up the two degrees of freedom, we calculate the d.f as follows:
1
df 2
u (1 u ) 2
n1 1 n2 1

S12
n1
where u
S12 S22
n1 n2
The calculated d.f value is not always an integer but it has to be truncated and value to
the right of the decimal point discarded. If we obtained a d.f of 19.456, we use only the
integer part, 19 and discard the 0.456.

Summary Procedure

H A : x1 x1 , if (t ) t (2), df reject H 0

H A : x1 x1 , if (t ) t (1) df reject H 0

H A : x1 x1 , if (t ) t (1) df reject H 0

10.3 Violations of 2-sample t-test

The 2-sample t-test assumes an underlying theory that both samples came at random from
normal populations with equal variances. The t-test has been shown to be robust enough
to withstand considerable deviations from its theoretical assumptions. This is especially

89
so if the sample sizes are equal or nearly equal; and where 2-tailed hypothesis are
considered.

With large the sample size, the t-test is more robust. The power of the test is little
affected by skewness. The power of the test is high when the sampled populations are
leptokurtic and the power is lower when the sampled populations are platykurtic (Figure
10.3 iii).

Figure 10.3: (i), Mesokurtic (normal) with few observations in the tail, (ii), leptokurtic,
where there are relatively more observations in the tails compared to (i) ,and (iii):
Platykurtic distribution with a high concentration of values in the middle and very
few in the tails.

10.4 Paired t-test

The paired t-test is performed if 2 samples are not independent but consist of natural
pairs. This is under situations where two members of each pair fall into different
categories. The situation is common especially in before and after treatments, where an

90