Week 9 MIDTERM EXAMINATION

Week 10 Topic: Kendall’s Tau Correlation Between Ranks

Kendall’s Tau, can be applied wherever the Spearman rank-order
coefficient is applicable. As can be seen below, it is somewhat harder to compute
than rs. When there are no ties, the solution is short and simple. Data in the given
table, it is imperative that one of the series of ranks be in its natural order – that is,
starting from high to low on the y variable are counted. For example, 3 persons rank
higher than the first individual on test y and 3 rank lower. In the case of the second
individual, 4 rank higher and 1 ranks lower. The counting continues until the
column is completed. In doing this, one looks only at the Ry’s that lie below the
score of the individual being examined. Each column is then summed, the column
with the number of ranks higher being called P and the other column Q. Tau is then
calculated:

T= P–Q
N(N – 1)/2

Table with No Tied Scores:

Individual Test X Test Y Rx Ry No. of Ranks No. of Ranks

Higher Lower
1 18 24 1 4 3 3
2 17 28 2 2 4 1
3 14 30 3 1 4 0
4 13 26 4 3 3 0
5 12 22 5 5 2 0
6 10 18 6 6 1 0
7 8 15 7 7 0 0
P = 17 Q=4

T= P–Q = 17 – 4 = 13/21 = 0.62

N(N – 1)/2 7(6)/2

When ties appear, certain adjustments have to be made. The number of

individuals ranking higher and lower than each individual on the y variable is again
determined, resulting in P = 33 and Q = 11.
Table with Tied Scores:

Individual Test X Test Y Rx Ry No. of Ranks No. of Ranks

Higher Lower
1 60 60 1 2 8 1
2 54 68 2 1 8 0
3 53 40 3 7 3 4
4 49 52 4.5 3 6 0
5 49 51 4.5 4.5 4 0
6 47 38 6 9 1 3
7 46 51 7 4.5 3 0
8 45 32 9 10 0 2
9 45 39 9 8 0 1
10 45 41 9 6 0 0
P = 33 Q = 11

In handling ties, first take the x distribution and for each test set of ties
determine x (x – 1), where x is the number tied for a rank. These are summed and
divided by 2.

2 (2 – 1) + 3 (3 – 1) = 2 + 6 = _8_ = 4
2 2 2
Repeat the process for the y distribution:

2 ( 2 – 1) = 1
2
Next calculate:
N (N – 1)/2 = 10 (9)/2 = 45
45 – 4 = 41
45 – 1 = 44
Multiplying the two terms (41) (44) = 1804

Take the square root √ 1804 = 42.50

Solve for T = _P – Q_ = ___22___ = 0.52 – moderate correlation

42.50 42.5

Activity 8:
 1. Using Kendall’s Tau coefficient, determine the correlation of Test A
and Test B and interpret the result.
Test A Test B
36 18
34 26
33 24
29 18
29 15
28 12
28 14
28 13
24 12
22 10
19 8
12 9

2. Determine the correlation between variable x and variable y

considering Kendall’s tau coefficient. Interpret the result.
Variable x Variable y
15 4
14 6
10 4
9 8
8 7
8 8
7 10
6 9

Week 11 Topic: Kendall’s Coefficient of Concordance W

To determine the relationship among three or more sets of ranks, one rank could
be selected and a Spearman rs coefficient computed between it and all of the others,
and the process could then be continued until rs coefficient has been obtained
between each set of two ranks. Then these rs’s could be averaged for an overall
measure of relationship. Kendall, though, has developed a technique and a statistic
that make all of this unnecessary.
W= 12 ∑D2
m2 (N) (N2 – 1)

Suppose the five judges (m) rank the projects of ten individuals (N) in a judging
contest, and we wish to determine the overall relationship among the ratings of the
five judges. The rankings of these judges have been set up in a given table. First
the rankings by the five judges of each of the projects are summed. The sums
appear in column 3. Then column 3 is summed to give the total sum of the ranks.
This can be checked for the total sum of the ranks as follows:
Total sum of ranks = m (N) (N + 1)
2
Individual Judges’ Ranks Sum of Difference of
Project 1 2 3 4 5 Ranks the mean and D2
the ranks (D)
1 2 1 2 3 4 12 15.5 240.25
2 1 3 1 2 2 9 18.5 342.25
3 3 4 4 1 3 15 12.5 156.25
4 5 5 5 5 1 21 6.5 42.25
5 4 2 6 7 6 25 2.5 6.25
6 7 8 3 4 7 29 1.5 2.25
7 6 6 8 6 5 31 3.5 12.25
8 8 7 7 8 9 39 11.5 132.25
9 9 10 10 9 8 46 18.5 342.25
10 10 9 9 10 10 48 20.5 420.25
∑ = 275 ∑D2 =
1696.5

To check:
Total sum of ranks = m (N) (N + 1)
2
Total sum of ranks = 5 (10) (10 + 1) = 275
2
Mean: x = 275 ÷ 10 = 27.5
If there were no relationship among the ranks, we should expect the sum of the
ranks for each row to be equal. For this case the sum of each would be the average
sum of ranks or 275/10, which equals 27.5. Then obtain the difference from this
mean of the sum of the ranks of each row and square these differences. Then these
squares are summed.
To compute W, using the formula:
W= 12 ∑D2 = ___12 (1696.5)___ = 0.82
m2 (N) (N2 – 1) 25 (10) (110 – 1)

Interpretation of W:
The size of this coefficient of concordance indicates that there is high
agreement among these five judges in the ranking of the ten projects. Perfect
 agreement is indicated by W = 1 and lack of agreement by W = 0.

Activity 9:
1. Three judges rated eight essays with the following results. Calculate
the coefficient of concordance for these data.

Essay Judges
1 2 3
1 8 7 8
2 6 5 6
3 4 6 5
4 1 2 1
5 3 3 2
6 2 1 3
7 5 4 4
8 7 8 7

2. Four judges rank eight convicts on “parole readiness.” By using the

coefficient of concordance, indicate the degree of consistency of the judges.

Essay Judges
1 2 3 4
1 1 1 1 1
2 2 4 3 2
3 3 3 2 4
4 4 2 4 3
5 5 6 5 5
6 6 5 6 7
7 7 7 8 6
8 8 8 7 8

Week 12 Topic: Distribution-Free Statistical Tests:

The Cochran Q-test:
This test is used to determine whether or not the two related samples are
different. It is suited to data that are dichotomized or in nominal scale. To achieve
matching, N groups of K matched subjects or some N subjects subjected to k
different conditions may be used. This arrangement results in k matched sets of
frequencies or proportions. Are these frequencies or proportions different from one
another? The Cochran test provides an answer to this question.
Cochran test may be used in a testing whether or not N subjects differ in their
responds to k items in a questionnaire.
Equation: Q = (k – 1) [ k ∑G2 – (∑G)2 ]
K ∑T - ∑T2

Example 1. An explanatory study is made by a researcher regarding the opinion of

students of a certain college towards banning of pornographic movies and reading
materials before, during, and after a series of students’ demonstration against
pornography. In this study, a sample of 10 students were selected. The responses of
the same students corresponding to each of the k = 3 conditions (before, during, and
after the series of demonstrations) are then tabulated. The responses are classified
as either “in favor” or “not in favor” of banning pornographic movies and reading
materials.
Students’ responses to the proposal of banning pornographic movies and reading
materials:

Students Before the Demo During the Demo After the Demo

1 Not in Favor In favor In favor

2 In favor In favor Not in Favor
3 Not in Favor In favor In favor
4 Not in Favor Not in Favor Not in Favor
5 Not in Favor In favor Not in Favor
6 In favor In favor In favor
7 In favor In favor In favor
8 In favor In favor In favor
9 Not in Favor In favor In favor
10 Not in Favor In favor In favor

Solution:
Ho: The probability of “in favor” responses is the same for all three conditions
(before, during, and after the demonstrations), that there are no differences in the
student’s attitudes towards the banning of pornographic movies and reading
materials before, during, and after the series of student’s demonstrations.
Ha: The probabilities of “in favor” responses differ according to the condition
under which the interview takes place; that is, there are differences in the student’s
attitudes towards the banning proposal before, during, and after the series of
demonstrations.
Level of Significance is 0.05
Table value: refer to x2 table: df = k – 1 = 3 – 1 = 2; t value = 5.991

Student Before, G1 During, G2 After, G3 T1 T12

1 0 1 1 2 4
2 1 1 0 2 4
3 0 1 1 2 4
4 0 0 0 0 0
5 0 1 0 1 1
6 1 1 1 3 9
7 1 1 1 3 9
8 1 1 1 3 9
9 0 1 1 2 4
10 0 1 1 2 4
Total 4 9 7 ∑ T1 = 20 ∑ T12 = 48

∑G = 4 + 9 + 7 = 20

Q = (k – 1) [ k ∑G2 – (∑G)2 ]
K ∑T - ∑T2

Q = (3 – 1) [ 3(42 + 92 + 72 – (20)2 ] = 6.33

3 (20) - 48

Interpretation: The tabular value (5.991) is less than the computed value (6.33),
therefore the null hypothesis is rejected and the alternative hypothesis is accepted. It
means that the response of the students differs according to the condition under
which the interviewed takes place.

Activity 10:
1. The data shows the response of the drug experienced individuals before, during and
after imprisoned for at least 1 year. Determine if difference exists on their behavior and
attitudes during these periods were interviewed takes place. Level of significance set at
0.05.

Prison Before During After

1 Fine Fine Fine
2 Fine Not Fine Fine
3 Not Fine Not Fine Fine
4 Not Fine Fine Fine
5 Fine Not Fine Fine
 6 Not Fine Not Fine Fine
7 Not Fine Fine Fine
8 Not Fine Fine Fine
9 Fine Not Fine Fine
10 Fine Not Fine Fine
11 Fine Not Fine Fine
12 Fine Not Fine Fine
13 Not Fine Fine Fine
14 Fine Not Fine Fine
15 Not Fine Not Fine Fine

Week 13 Topic: The Friedman Two-Way Analysis of Variance

When the data are in nominal scale the Cochran test is the proper test to use.
However, when the data are in ordinal scale, the Friedman test is appropriate. The
null hypothesis to be tested is whether the k matched samples have been drawn
from the same population. Unlike the Cochran test, which analyzes the frequencies
of 0 and 1 values, the Friedman test analyzes the ranks of the k values in each row.

X2r = [ 12 ∑C2j ] – 3N (k + 1)
N k (k + 1)
Example: One may consider the following hypothetical case. Suppose it is desired
to analyze the achievement of male students taught by one of the three methods of
instruction, namely: (1) programmed text, (2) lecture, and (3) lecture with discussion.
Suppose further, 12 sets of k = 3 male students per set are chosen in such a way that they
are matched according to age and IQ and that the three subjects in each set are each
randomly assigned to one of the three methods of instruction. The end results in terms of
scores are presented.

Students Programmed Lecture Lecture with

Text Discussion
1 145 134 140
2 130 125 131
3 150 95 120
4 120 132 111
5 121 110 123
6 129 121 119
7 131 125 134
8 151 126 129
9 89 102 110
10 159 60 108
11 125 95 143
 12 126 121 112

Solution:
Ho: The three methods of instruction do not have different effects
Ha: The three methods of instruction have differential effects.
Level of Significance = 0.05
Refer to X2 table: df = k – 1 = 2
Tabular value = 5.991

Students Rank – Prog. Test Rank - Lecture Rank – Lec with Dis.
1 3 1 2
2 2 1 3
3 3 1 2
4 2 3 1
5 2 1 3
6 3 2 1
7 2 1 3
8 3 1 2
9 1 2 3
10 3 1 2
11 2 1 3
12 3 2 1
C1 = 29 C2 = 17 C3 = 26

X2r = [ 12 ∑C2j ] – 3N (k + 1)
N k (k + 1)

X2r = 12[ 292 + 172 + 262] – 3(12) (3 + 1)

12 (3) (3 + 1)

= 150.5 – 144 = 6.50

Interpretation: The Computed X2r is greater that the tabular value of 5.991, which
means that the null hypothesis is rejected and the alternative hypothesis is accepted.
This means that the three methods of instruction have differential effects.

Activity 11:

1. Scores of ten pairs of matched individuals on a test after being

exposed to three different teaching methods, A, B, and C. At 0.05 level of
significance, determine if there is a differential effect on the three methods used.
Pair Method A Method B Method C
1 24 26 28
 2 28 30 42
3 32 37 25
4 18 22 20
5 24 29 25
6 36 40 20
7 40 38 36
8 37 41 45
9 24 29 34
10 20 28 24

2. In an experiment designed to compare the effects of coaching on

scores obtained on an aptitude test used for entrance to a professional school, three
levels of coaching were considered from the six sets of applicants. Determine if
there are differential exists on the three levels of coaching such as none, 4 hours and
12 hours. Use 0.05 level of significance.

Set None 4 Hours 12 Hours

1 30 32 35
2 27 30 33
3 26 29 32
4 24 27 30
5 22 25 27
6 20 24 26

Week 14 Topic: The Kruskal -Wallis H-Test

The Kruskal-Wallis H Test is used to test whether or not a group of

independent samples is from the same or different populations. The Kruskal-Wallis
test is a powerful non-parametric test.
The Equation for H:

H= 12 [∑Ri2] – 3(N + 1)
N (N + 1) ni

N = the number in all samples combined

Ri = sum of ranks and ni the numbers in i samples

To illustrate the test a given table with the set of data is presented:
Example:
Variable Rank x Variable Rank y Variable Rank z
x y z
12 13.5 13 16 13 16
16 21 18 22 14 19
14 19 14 19 7 8.5
2 1.5 13 16 8 10.5
12 13.5 8 10.5 4 4.5
7 8.5 3 3
6 7 2 1.5
4 4.5 5 6
9 12
Nx = 5 ∑Rx = 68.5 Ny = 8 ∑Ry = 103.5 Nz = 9 ∑Rz = 81

The scores of the three sets are combined and ranked with the lowest score
receiving a rank value of 1. Ties are treated in the usual fashion for ranking data.
Then the ranks of the three columns are summed. If the sample proportions are
similar, the total sum of the ranks would be divided proportionately among the
various samples on the basis of sample size. The sums of the ranks of the various
samples are then tested against these proportions of the total sum of ranks.

H= 12 [∑Ri2] – 3(N + 1)
N (N + 1) ni

H= 12 [ 68.52 + 103.52 + 812] – 3(23)

22 (23) 5 8 9

H = 2.30

When the samples contain five or more cases, H is interpreted as chi-square with
the number of samples minus 1 degree of freedom. For this problem df is equal to 2
and this is not significant.

Activity 12:

1. The following data from Borislow show the changed response scores
of three groups on the Edwards Personal Preference Schedule. Test the differences
among these three groups at 0.01 and 0.05 level of significance.

Control Group Social Desirability Personal Desirability

Group Group
47 76 87
38 76 76
 34 75 52
32 73 51
32 62 50
30 47 42
29 38
50
35

2. In the data below are three samples of scores obtained on the

arithmetic scale of the Wechsler-Bellevue. Apply Kruskal-Wallis H-test if there are
differences exists among the three samples.

A B C
8 2 12
7 5 4
15 8 11
10 17 5
9 10 14
7 9 11
8 15
2 8

3. One may consider a pilot study involving the socio-economic scores

of households of three places, namely: a rural area, sub-urban area and urban area
believed to be undergoing the process of urbanization. The socio-economic score is
based on the possession of certain items such as radio, telephone, automobile,
television, and so on. One may further wish to find out if there are differences
among the socio-economic scores of the households in the three places. For the
purpose of this study, three independent random samples are drawn: 10 households
from rural area, 12 households from the sub-urban area and 15 households from the
urban area. The scores are shown in the table that follows: Level of significance =
0.05
Ho: There are no differences in the socio-economic scores of households in the
three places considered.
Ha: There are differences in the socio-economic scores of households in the three
places.

Rural Area Sub-Urban Area Urban Area

82 70 80
46 69 84
55 80 77
55 88 65
71 60 54
39 78 64
 45 64 66
59 60 52
67 73 91
44 84 69
80 72
81 89
74
87
80

Week 15 Topic: The Mann-Whitney U-test

The Mann-Whitney U-test - This tool is used when one wishes to test whether
two independent groups have been drawn from the same population. The test is a
useful alternative to the parametric t-test when one wishes to avoid the assumptions
that are made in the use of the t-test. The test is used with independently drawn
random samples, the sizes of which need not be the same. This test requires that the
data be in the ordinal scale. This is a more powerful test than the median test for
uncorrelated data.

When the sample sizes are very small-that is, when both N1 and N2 are made up
of eight or less measures – the reader is referred to Siegel (1956) for a method and
tables that apply to such sample sizes. When the larger of the two samples is 9 or
more, the procedure describe below is used.

Example 1. In the table there are two samples, Nx with seven cases and Ny with
nine cases. In this case we are testing the hypothesis that both samples come from
the same population. First rank all the scores in one composite distribution in an
increasing order size. The score of 3 in column y is the lowest one; hence give it a
rank of 1. The score of 4 in column x is the second lowest, give it a rank of 2.
There is a score of 6 in both x and y distribution, handle the ties by giving them the
average of the next two ranks, which for this distribution is 3.5. Continue the
ranking until all scores receive a rank.

At this point make an arithmetic check on the work. The sum of the two
columns of ranks must be equal to N (N + 1)/2.
Table:
X Y Rank (x) Rank (y)
14 18 11.5 15
12 16 9 14
13 15 10 13
10 14 8 11.5
 7 19 5.5 16
6 7 3.5 5.5
4 8 2 7
6 3.5
3 1
Nx = 7 Ny = 9 ∑Rx = 49.5 ∑Ry = 86.5

To check: ∑Rx + ∑Ry = N (N + 1)

2
49.5 + 86.5 = 16 (17)/2
136 = 136
To obtain U as follows:

U1 = N1N2 + N1(N1 + 1) - ∑Rx

2
U2 = N1N2 + N2(N2 + 1) - ∑Ry
2
Or U2 = N1N2 – U1

z= U1 – (N1N2/2)
√[N1N2(N1 + N2 + 1)]/12

For the problem:

U1 = 7(9) + 7 (8) - 49.5 = 41.50

2

U2 = 7(9) + 9(10) – 86.5 = 21.50

2

To check: U2 = N1N2 – U1
= 7(9) – 41.5
= 21.50

Refer to table:
Level of significance = 0.01
Two tailed test - refer critical values of U:
Ni = 7 and N2 = 9, U must be 9 or less to be significant.

Interpretation: Our obtained U of 21.50 is higher than 9, therefore it is not

significant. And the null hypothesis is accepted.

Example 2. (For larger Set of data), Find out if difference exists between the
variable x and y at 0.01 level of significance.
Given:
Variable x Variable y
7 18
10 14
12 15
6 16
8 12
4 10
6 8
14 13
2 13
4 3
6
7
3
5
2
1
13
14
11
10
9
N1 = 21 N2 = 10

Solution:
Ho: There is no significant difference between the scores of variable x and y.
Ha: Significant difference exists between variable x and y.
Level of significance: 0.01
Non-directional: two-tailed test
Table value of z = 2.58

Variable x Rank x Variable y Rank y

7 12.5 18 31
10 18 14 27
12 21.5 15 29
6 10 16 30
8 14.5 12 21.5
4 6.5 10 18
6 10 8 14.5
14 27 13 24
2 2.5 13 24
 4 6.5 3 4.5
6 10
7 12.5
3 4.5
5 8
2 2.5
1 1
13 24
14 27
11 20
10 18
9 16
N1 = 21 ∑Rx = 272.5 N2 = 10 ∑Ry = 223.5

U1 = N1N2 + N1(N1 + 1) - ∑Rx

2

U1 = 21(10) + 21(21 + 1) – 272.50 = 168.5

2

U2 = N1N2 + N2(N2 + 1) - ∑Ry

2

U2 = 21 (10) + 10 (10 + 1) – 223.50 = 41.5

2

Solve for Z:

z= U1 – (N1N2/2)
√[N1N2(N1 + N2 + 1)]/12

Z= 168.5 – 21(10)/2
√ [21(10) (21 + 10 + 1]/12

Z = 2.68

Interpretation: The computed z (2.68) is greater than tabular value (2.58), it means
that the null hypothesis is rejected and the alternative hypothesis is accepted. This
means that difference exists between the scores of variable x and variable y.

Activity 13:

1. Below are the scores of a group of normal and a group of psychotics

on the picture-completion scale of the Wechsler-Bellevue. Test the hypothesis of
no difference using Mann-Whitney U-test.

Normal Psychotics
6 7
6 2
14 12
13 13
15 8
6 6
8 4
7 2
10 2
14 12
10
14

2. Two groups of children made the following scores on the vocabulary

test of the Stanford-Binet. By the use of the Mann-Whitney U test, test for
differences in the two groups. Use 0.05 level of significance.

Group I Group II
10 14
18 18
36 22
22 16
28 38
29 26
32 28
15 12
18 11
36 19
21 16
27
33
36
26
24
31
 13
19
23
25
27
32

3. Suppose one wish to compare the monthly household expenditures of

rural place with that of an urban place. Suppose further that there are 22 samples of
respondents from the urban area, while there are only 15 respondents from the rural
area. Level of significance @ 0.05. Below are the corresponding data of the groups
to be considered.

Rural Area Urban Area

186 134
177 890
67 52
78 570
48 153
300 462
244 115
74 405
68 119
300 334
280 157
68 244
236 244
111 362
140 50
344
486
261
338
740
101
117
Week 16 Topic: The Bi-serial and Point Bi-serial Correlation

The Biserial Correlation Coefficient

The bi-serial correlation measures the degree of relationship between two

continuous quantitative variables in which one variable is forced into a dichotomy.

Equation: rb = (y1 – y2) (pq)

SD y
Where y1 is the mean of the measurements falling within the first category, y2 is
the mean of the measurements falling within the second category, SD is the
standard deviation of the measurements, p is the proportion of measurements in the
first category, q is the proportion of measurements in the second category, and y is
the height of the ordinate dividing the area under the standard normal curve into p
and q parts.

Example: In a study to determine the relationship between a person’s intelligence and

socio-economic status, a researcher obtained a random sample of 40 individuals. Each
was administered an intelligence test and his socio-economic status (SES) was evaluated
on a continuous rating scale from 0 to 10.
IQ SES IQ SES IQ SES IQ SES
score rating score rating score rating score rating
(y) (x) (y) (x) (y) (x) (y) (x)
125 9.1 112 6.4 105 4.5 118 8.0
114 7.0 113 6.6 115 5.3 118 8.3
112 6.5 102 3.6 129 9.0 115 7.2
117 7.8 100 3.2 120 8.2 123 8.5
106 4.8 117 7.6 120 5.1 120 8.2
130 9.5 103 4.0 114 7.0 120 8.5
101 3.6 95 2.2 113 6.6 120 7.5
110 5.7 109 5.6 110 5.8 121 8.7
111 6.0 121 8.4 116 7.3 127 8.8
110 5.8 97 2.5 117 7.7 115 6.0

The variable “socio-economic status” is forced into a low-high dichotomy, such

as (0 – 5.5) and (5.6 – 10) respectively. Determine the coefficient of correlation
and interpret the result.

Solution: Computation of the biserial correlation coefficient

SES IQ No. of Sum of Sum of

Dichotomy Measurements Measurements Measurements Squares of
Measurements
 Low 106 101 102
(0 – 5.5) 100 103 95 n1 = 10 ∑y1 = 1044 ∑y12 = 109,534
105 115 120
97
High 125 114 112
(5.6 – 10) 117 130 110
111 110 112
113 117 109
121 129 120 n2 = 30 ∑y2 = 3517 ∑y22 = 413,247
114 113 110
116 117 118
118 115 123
120 120 120
121 127 115
Total N = 40 ∑y = 4,561 ∑y2 = 522,781

Standard Deviation (SD) = N ∑y2 – (∑y)2

√ n (n – 1)

SD = 40 (522,781) – (4,561)2
√ 40 (39)

SD = 8.34

y1 = ∑y1 /n1 y2 = ∑y2/n2

= 1,044/10 = 3,517/30
= 104.40 = 117.23

p = n1/n q = n2/n
= 10/40 = 30/40
= 0.25 = 0.75

From table: pq/y = 0.5900

rb = (y1 – y2) (pq) = (104.40 – 117.23) (0.5900)

SD y 8.34

rb = 0.91

The Point Biserial Correlation Coefficient

The point biserial Correlation coefficient is similar in many respects to the

biserial correlation coefficient, except that one of the variables is genuinely
dichotomous; for example, (male and female), employment status (employed and
unemployed) or locale (urban and rural). Its formula is given by:

Equation: rpb = y1 – y2 √ pq
SD

Where y1 is the mean of the measurements in the first category, y2 is the mean
of the measurements in the second category, SD is the standard deviation of the
measurements, p is the proportion of measurements in the first category, and q is
the proportion of measurements in the second category.

Example 1. Determine the extent of relationship between expenditures and sex, the
average weekly expenditures of 28 males and 22 female respondents were
computed over a one-month period. The data are shown in the table. Unit is in
thousand pesos.

Expendi Sex Expendi Sex Expendi Sex Expendi Sex

tures tures tures tures
2.01 F 1.78 F 2.30 F 4.74 F
0.59 F 4.11 M 8.50 M 1.60 F
4.09 M 8.34 M 4.10 M 9.14 M
5.07 M 1.74 F 6.95 F 6.37 M
1.59 F 4.65 F 8.20 F 1.03 F
8.63 M 4.20 M 7.93 M 4.79 F
6.10 M 8.11 M 4.20 M 4.60 M
0.55 F 6.82 M 1.40 F 5.08 F
4.14 M 2.07 F 5.96 F 8.91 M
2.87 M 5.82 M 9.31 M 2.35 F
5.95 F 2.88 M 1.57 M 8.70 M
8.6 F 7.23 F 6.12 M
3.77 M 1.69 M 6.20 M

Solution: Computation of the point biserial correlation coefficient

IQ No. of Sum of Sum of
SEX Measurements Measurements Measurements Squares of
Measurements
4.09 5.07 8.63
6.10 4.14 2.87
4.11 8.34 4.20
8.11 6.82 5.82
8.50 4.10 7.93 n1 = 28 ∑y1 = 162.29 ∑y12 = 1090. 08
Male
4.20 9.31 9.14
6.37 4.60 8.91
3.77 2.88 1.69
1.57 6.12 6.20
 8.70
2.01 0.59 1.59
0.55 1.78 1.74
4.65 2.07 2.30
Female 6.95 8.20 1.40 n2 = 22 ∑y2 = 81.16 ∑y22 = 439.63
5.96 4.74 1.60
1.03 4.79 5.08
2.35 5.95 8.60
7.23
Total N = 50 ∑y = 243.45 ∑y2 = 1,529.71

Standard Deviation (SD) = N ∑y2 – (∑y)2

√ n (n – 1)

SD = 50 (1,529.71) – (243.45)2
√ 50 (49)

SD = 2.65

y1 = ∑y1 /n1 y2 = ∑y2/n2

= 162.29/28 = 81.16/22
= 5.80 = 3.69

p = n1/n q = n2/n
= 28/50 = 22/50
= 0.56 = 0.44

√ pq = 0.4964

rpb = y1 – y2 √ pq
SD

rpb = 5.80 – 3.69 (0.4964)

2.65
rpb = 0.40

Activity 14:
You are a principal of a particular High School. You conducted a follow-up study to
see the correlation of external productivity of your school. See if the following variables
are significantly correlated.
A B C D E F
Successful Grade English Mathematic Involvement Exposure to
School in Extra Community
Graduates Grade s Grade
Average Curricular Programs
Activities
 + 85 74 73 + -
- 83 85 88 - -
+ 78 86 87 + +
+ 84 71 86 - -
+ 86 89 85 - +
- 79 74 90 + -
- 85 86 91 + +
- 84 85 93 - +
+ 93 92 89 - +
+ 91 90 87 + -
- 87 91 89 + -
+ 86 97 88 - -
- 85 98 87 - -
+ 94 73 88 + +
+ 95 85 80 + +
+ 98 88 84 - +
- 96 87 86 - -
+ 90 88 85 - -
- 89 91 84 + +
- 87 93 83 - +
+ 89 90 82 + +
+ 83 92 80 + -
- 84 90 79 - -
+ 85 89 88 - +
+ 87 88 85 + +
- 80 87 80 - +
- 80 86 82 + +

Find the correlation of the following:

1. A and E
2. A and D
3. A and F
4. B and F
5. C and F

Week 17 Topic: Methods of Computing Reliability Coefficients

There are number of techniques used in the computation of reliability

coefficients. For basic sample techniques there are methods of test-retest, parallel
forms, split halves, and Kuder-Richardson Formula 20.
Test-Retest Method:

With the test-retest method a test is administered and then at a later date the
same test is re-administered to the same individuals. A Pearson product-moment
correlation coefficient is computed between the two sets of scores. The length of
the time between the two administrations of the test is important in determining the
size of the reliability coefficient. In general, the longer the time between the two
administrations of the test, the lower the correlation. If the period between the
administrations is very short, such as the second test immediately following the
first, individuals may remember their answers and put the same responses down the
second time without making a new effort to react to the test item. Such behavior
tends to make reliability coefficients artificially high. When the period between
testing is short, memory may be an important factor affecting the results. As the
period increases in length, learning, maturation, senescence, and many other
variables may enter the situation to lower the correlation coefficient. At the present
time, the test-retest method is frequently used when determining the reliability of
paper-and -pencil tests. Coefficients computed by this method are frequently called
coefficients of stability.

Parallel Forms Method:

This technique is also referred to as the method of equivalent forms. With this
technique, we administer form A to a group of individuals and follow this
immediately or fairly soon with form B on the same test. These two forms of the
same test are said to be parallel or equivalent, because they are made up of the same
types of item covering the same materials; they have the same means and variances;
and if one form correlates to a certain extent with some other measure, then the
other form correlates to the same degree.

As in the previously discussed method, A Pearson product-moment correlation

coefficient is computed between the two sets of scores. This coefficient is
sometimes referred to as coefficient of equivalence. This method is widely used at
the present time. It may be called the technique with the most universal
applicability. When other method cannot be used, we can usually fall back on this
one. The memory factor, which was important with the test-retest method, is ruled
out here. But such factors as learning, growth, and change are still present, and
here again, the longer the period between the two test administrations, the lower the
reliability coefficient tends to be.

The Split-Half Method:

An advantage of this method is that only one test is needed for the computation
of the reliability coefficient. The test papers are scored so that from every single
paper we have two scores. This is usually done by counting the number of odd-
numbered items answered correctly and the number of even-numbered items
answered correctly. Sometimes other splits are made, such as items 1 and 2 go into
the first score, 3 and 4 into the second score and 5 and 6 into the first score and so
on. Almost any split will be acceptable, except one taking the first half of the items
against the second half, because tests are usually made with the easier items first
and because students tend to complete the first half of a test and not the second, a
division of this sort will result in two tests of a different nature.

Each paper now has an even and an odd score on it, or two other types of
scores, depending on the type of split made. Again a Pearson product-moment
correlation coefficient is computed between the two sets of scores. A reliability
coefficient of this type is often called a coefficient of internal consistency. It so
happens that the reliability of a test is directly related to the length of the test.
When we scored our test on an odd-even basis, we actually cut the length of our
original test in half. The reliability coefficient we have computed is then the
equivalent of one for a test of half of the size of our original test. We make a
correction for this effect by using what is known as the Spearman-Brown formula
as follows:

rtt = __2roe___
1 + roe
Where:
rtt = the reliability of the original test
roe = the reliability coefficient obtained by correlating the scores on the
odd items with the scores of the even items

Kuder-Richardson Formula 20:

The Kuder-Richardson formula 20 also yields a coefficient of internal

consistency, and as such has some of the limitations of the split-half method. If an
item analysis has been made for a test, the Kuder-Richardson formula is easily
applied to the data to obtain the reliability coefficient. One of the outcomes of an
item analysis is a difficulty measure for each item on the test. Difficulty is defined
as the proportion or percentage of those responding to an item who answered it
correctly. The symbol p is used to represent difficulty. An item with p = 0.89 was
answered correctly by 89 percent of those who responded to it.

In working a Kuder-Richardson solution, we first set up a worksheet like that shown

below. Column 1 consists of the number of the item. In column 2 the difficulty value (p)
of the item is recorded from the item analysis work. Column 3 is labeled q, which always
 means 1 – p. In all statistical work p + q = 1. Column 4 is labelled pq, which is the
product of columns 2 and 3. This column should be carried to four decimal places. Then
this last column is summed.
(1) (2) (3) (4)
item p q = (1 – p) pq
1 0.60 0.40 0.2400
2 0.30 0.70 0.2100
3 0.71 0.29 0.2059
etc ∑pq = 0.6559

The equation for Kuder-Richardson Formula 20 is:

rtt = k [1 - ∑pq ]
k–1 s2
where:
k = number of items on the test
s2= the variance of the test
pq = the quantity obtained from the worksheet

Questions/Activity 15:
1. In your own understanding, discuss the techniques used in the computation
of reliability coefficients.

2. Give specific example in each technique, specify a data and present

computation. Interpret the result.

Week 18 FINAL EXAMINATION

References:

: Webster, Allen L., (2005); Applied Statistics for Business and Economics; 2nd
Edition; Von Hoffman Press Inc. USA.

: Downie, N.M. and Robert W. Heath; Basic Statistical Methods; Harper and Row
Publishers, New York, USA

: Freund, John E. and Frank J. Williams; Business Statistics: The Modern Approach;
Sixth Edition; Prentice-Hall, Englewood Cliffs, New Jersey, USA

: Walpole, Ronald E. and Raymond H. Myers, (2003); Probability and Statistics; Fifth
Edition; Macmillian Publishing Company; New York

: Montaña, Rizalina A. and Cristobal M. Pagoso, (2000); Introductory Statistics; Rex

Printing Company, Inc. Quezon City, Philippines
Prepared by:

ENGR. JOEL C. VILLARUZ, Ph.D.

Professor