T-TEST

T Test procedure compares the means of two groups or (one-sample) compares the
means of a group with a constant.

Two Independent Samples

Notation
The following notation is used unless otherwise stated:

X ki

Value for ith case of group k

wki

Weight for ith case of group k

nk

Number of cases in group k

Wk

Sum of weights of cases in group k

Basic Statistics
Means
nk

Xk =

ki wki

i =1

Wk

k = 1, 2

Variances
nk

S k2 =

X
i =1

F
G X
GH
bW 1g
nk

2
ki wki

ki wki

i =1

I
JJ
K

Wk

2 T-TEST

Standard Errors of the Mean

SEM k = S k

Wk

Differences of the Means for Groups 1 and 2

D = X1 X 2

Unpooled (Separate Variance) Standard Error of the Difference

SD =

S12 S22
+
W1 W2

The 95% confidence interval for mean difference is

D t df S D
where t df is the upper 2.5% critical value for the t distribution with df degrees
of freedom.
Pooled Standard Error of the Difference

SD
= Sp

1
1
+
W1 W2

where the pooled estimate of the variance is

S 2p =

bW 1gS + bW 1gS
1

2
1

W1 + W2 2

2
2

T-TEST 3

The 95% confidence interval for mean difference

D t df S D

where df is defined in the following.

The t Statistics for Equality of Means

Separate Variance

t = D SD
df =

1
Z1 + Z 2

where

Zk

F
=G
HS

2
1

S k2 Wk
W1 + S22

I bW 1g
J
W K
2

Pooled Variance

t = D SD

df = W1 + W2 2
The two-tailed significance levels are obtained from the t distribution separately
for each of the computer t values.

4 T-TEST

The Test for Equality of Variances

The Levene statistic is used and defined as
2

(W - 2 )
L=

Wk ( Zk - Z )2

nk

k =1

wki ( Zki - Zk )2
k =1 i =1

where

Z ki = Xki - X k
nk

wki Zki

Z k = i =1

Wk

Wk Zk

Z = k =1
W1 + W2

The t Test for Paired Samples

Notation
The following notation is used unless otherwise stated:

Xi

Value of variable X for case i

Yi

Value of variable Y for case i

wi

Weight for case i

Sum of the weights

Number of cases

T-TEST 5

Basic Statistics
Means
N

X=

w X
i

i =1

Y =

w Y

i i

i =1

Variances
N

S 2X =

wi X i2

i =1

F N I2
G wi X i J
GH i =1 JK

W 1

Similarly for SY2 .

Covariance between X and Y

S XY =

1
W 1

F
GGH X Y w
N

k k k

k =1

Difference of the Means

D = X Y

Standard Error of the Difference

SD =

eS

2
X

+ SY2 2 S XY

jW

F
GG w
H
N

k =1

IF
I
JJ GG w Y JJ
KH
K
N

k Xk

k k

k =1

I
JJ
K

6 T-TEST

t statistic for Equality of Means

t = D SD

with W 1 degrees of freedom. A two-tailed significance level is printed.

95% Confidence Interval for Mean Difference

D tW 1S D

Correlation Coefficient between X and Y

r=

S XY
S X SY

T-TEST 7

The two-tailed significance level is based on

W 2

t=r

1 r2

with W 2 degrees of freedom.

One-Sample t Test
Notation
The following notation is used unless otherwise stated:
N

Number of cases

Xi

Value of variable X for case i ( i = 1,

wi

Weight for case i ( i = 1,

Test value

K, N )

K, N ). The weights must be positive.

Basic Statistics
Mean

X=

1
W

w X

i i

i =1

where W =

is the sum of the weights.

i =1

Variance

S X2

1
=
W 1

w dX X i
i

i =1

8 T-TEST

Standard Deviation

S X = S X2

Standard Error of the Mean

SX = SX / W

Mean Difference

D = X v

The t value

t = D / SX

with W 1 degrees of freedom. A two-tailed significance level is printed.

100p% Confidence Interval for the Mean Difference 0 < p < 1

CI = D t W 1,( p +1) / 2 S X

ca f h

where tW 1,( p +1) / 2 is the 100 p + 1 / 2 % percentile of a Students t distribution

with W 1 degrees of freedom.

References
Blalock, H. M. 1972. Social statistics. New York: McGraw-Hill.