Lesson – 12

Between subject designs

12.0. Objectives
Having gone through the lesson the learner will be able to understand
1. Two randomized group designs.
2. More than two randomized group design.
3. Factorial design.
4. Matched group designs.

12.1. Introduction

When two or more values of the independent variable are selected for study,

and one value is administered to each group in the experiment the design used is

known as between subjects design. We then calculate the mean of the dependent

variable for each group, compute the mean difference between groups, and assess

the effect of varying the independent variable. In short, for between groups designs

we compare dependent variable values between groups who have been treated

differently.

A basic and important assumption made in this type of design is that the

means of the groups on the dependent variable do not differ at the start of the

experiment. This is achieved by selecting a large sample and randomly allotting the

sample between the groups. Therefore it is assumed that all the groups are equal on

the dependent variable before the start of the experiment and whatever differences

observed are because of the different treatments of the independent variables.

The two randomized group design, the more than two randomized groups

design, the factorial design, and the matched groups design are all examples of

between groups design. These are all parametric tests and are used when the

sample size is large. It is not recommended to use these tests when the sample is

less than 30. When the sample size is small appropriate nonparametric test is used.

137
12.1. Two randomized group designs.

In a two randomized groups design one independent variable is varied in two

ways. A two randomized groups design is so called because here the subjects are

randomly assigned to two groups only. In formulating this type of design the

experimenter, first defines the independent variable and the dependent variables.

Subsequently, he selects two values of the independent variable. These two values

may also be called as “conditions” or “treatment” of the experiment. The main

interest is to examine whether or not these two conditions affect the dependent

variable in a differential way. This is achieved by computing a statistic known as ‘t’

test, which is a ratio between the mean difference of groups and the error variance

in the experiment. The error variance is a direct function of variability of the

dependent variable scores. ‘t’ test is computed by the formula

M1 – M 2
t =

SS1 – SS2
1 + 1
(n1–1) (n2-1)
n1 n2
df = N – 2 , where N= n1+ n2. The significance level is obtained by looking into the

SS1 = ΣX12 – (ΣX1)2 SS2 = ΣX22 – (ΣX2)2

n1 n2
table of siginificance for ‘t’ test.
Example: It is advertised that a certain tranquilzer has a curative effect on

psychotics. A clinical psychologist seeks to determine whether this is true. A well

designed experiment is conducted with the following results on a measure of

psychotic tendencies. Assuming that the criterion for rejecting the null hypothesis is

138
P=0.01 and assuming that the lower the score, the greater the psychotic tendecy,

determine whether the tranquilizer has the advertised effect.

Values of the group that Value of the group that

receives tranquilizer did not receive tranquilizer
2, 3, 5, 7, 7, 8, 8, 8 1, 1, 1, 2, 2, 3, 3

Values of the group that Value of the group that

receives tranquilizer did not receive tranquilizer
X1 X12 X2 X22`
2 4 1 1
3 9 1 1
5 25 1 1
7 49 2 4
7 49 2 4
8 64 3 9
8 64 3 9
8 64
2
ΣX1=48 ΣX1 =328 ΣX2 =13 ΣX1 =29
M1 = 48/8 =6 M2 =13/7=1.86

(ΣX1)2 = 328 – (48)2 = 40

SS1 = ΣX12 –
n1 8

= 29 – (13)2 =5
SS2 = ΣX 2
2
– (ΣX2)2
7
n2

M1 – M 2 6 – 1.86
t = = = 4.5

SS1 – SS2 40 + 5 P < 0.01.

1 + 1 1 + 1
(n1–1) (n2-1) (7) (6)
n1 n2 8 7
N = n1 + n2 = 8 + 7 = 15; df = N – 2 = 15 – 2 = 13

139
Looking into ‘t’ tables: the t value required at 0.01 level of significance is 2.681;

Hence Null hypothesis is rejected at 0.01 level and it is concluded that the drug is

having the advertising effect.

12.3. More than two randomized groups design

The design is so called because the total subjects are randomly assigned to

three or more unbiased groups. In this design there are three or more conditions or

values of the independent variable and accordingly, three or more groups of subjects

participate in the experiment. Such a design is also known as the multi groups

design. In psychological and educational research the use of the more than two

randomized groups design is more common than the two randomized groups

because the multi groups design enables the experimenter to establish an adequate

relationship between the dependent variable and the independent variable but the

two groups designs lacks this trait. The statistics analysis is conducted by

computing a test known as F test.

The steps to compute F test are as follows

(1) Total sum of squares (TSS) =

(ΣX1 + ΣX2 + ΣX3 )2

= ΣX12 + ΣX22 + ΣX32 –
N

(2) Between sum of squares (BSS) =

(ΣX1 + ΣX2 + ΣX3 )2

= (ΣX1)2 + (ΣX2)2 + (ΣX3)2 -
n1 n2 n3 N

(3) With in sum of squares (WSS) = TSS – BSS = 87 – 55 = 32

(4) df (a) Total df = N – 1

140
 (b) Between df = r – 1
(c) Within df = N – r
(r = number of groups)

(5) Mean square

(a) Between mean squares (BMS) = (BSS/Bdf)
(b) Within mean square (WMS) = (WSS/Wdf)

(6) F = (BMS / WMS)

The significance is obtained by looking in the ‘F’ tables between the

intersection of the df associated with the numerator (between df) and the df

associated with the denominator (with in df).

Example: An experimenter was interested in assessing the relative sociability scores

of different students in a college. Random samples of students who were students in

English, art, and science were selected, and they were administered a standardized

test of sociability. Assume a 0.05 criterion for testing the overall null hypothesis that

there is no mean difference between any pair of the three groups.

Sociability scores
English students Art students Science students
0, 1, 3, 5 3, 5, 6, 6 5, 7, 9, 9

Sociability scores
English students Art students Science students
0 0 3 9 5 25
1 1 5 25 7 49
3 9 6 36 9 81
5 25 6 36 9 81
9 35 20 106 30 236
ΣX1 ΣX12 ΣX2 ΣX22 ΣX3 ΣX32

141
(1) Total sum of squares (TSS) =

(ΣX1 + ΣX2 + ΣX3 )2 (9+20+30)2

= ΣX12 + ΣX22 + ΣX32 – = 35+106+236 – = 87
N 4

(2) Between sum of squares (BSS) =

(ΣX1 + ΣX2 + ΣX3 )2

= (ΣX1)2 + (ΣX2)2 + (ΣX3)2 -
n1 n2 n3 N

(3) With in sum of squares (WSS) = TSS – BSS = 87 – 55 = 32

(81+ 400+900 )2
= 55
= (9)2 + (20)2 + (30)
2
-
4 4 4 N

(4) df (a) Total df = N – 1 = 13 – 1 = 12

(b) Between df = r – 1 = 3 – 1 = 2
(c) Within df = N – r = 12 – 3 = 9
(r = number of groups)

(5) Mean square

(c) Between mean squares (BMS) = (BSS/Bdf) = 35/2 = 27.5
(d) Within mean square (WMS) = (WSS/Wdf) = 32/9 = 3.5

(6) F = (BMS / WMS) = 27.5 / 3.5 = 7.85 P < 0.05

The significance is obtained from the ‘F’ tables by observing the intersection

of the df associated with the numerator (between df) 2 and the df associated with the

denominator (with in df) 9 which is 4.26 at 0.05 level. This indicates that the English

students, Arts students, and Science students differ significantly in their sociability

scores.

142
12.4. Matched group designs

Like the randomized groups design, the matched group design may be a two

matched groups design or a more than two matched group design. Whatever the

type, in the matched group design all subjects are first tested on a common task or a

pretest measure (also called the matching variable) and then, they are formed into

groups (as many as needed for the experiment) on the basis of the performance of

the common task or the matching variable. The groups thus formed are said to be

equivalent groups. Subsequently, the different conditions or values of the

independent variable are introduced to each group. If these groups have equivalent

means on the dependent variable before the experimental treatment is given and if a

significant difference occurs after administering the experimental treatment and

controlling the relevant variables, the resulting differences in the dependent variable

may safely be attributed to the experimental treatment.

A matching variable can be selected on its ability to yield a high correlation

with the dependent variable. If possible the dependent variable it self should be used

as a matching variable. Once the matching variable scores are obtained the sample

is divided into pairs of subjects having equal scores on the matching variable. The

subjects in each pair are randomly allotted between the experimental and control

groups.

Example: To test the hypothesis that the higher the induced anxiety, the better the

learning an experimenter formed two groups of participants by matching them on an

initial measure of anxiety. Next, considerable anxiety was induced into the

experimental group but not into the control group. The following scores on a learning

task were then obtained, the higher the score, the better the learning. Assuming

143
 adequate controls were exercised and that a criterion of 0.05 was set, was the

hypothesis confirmed?

Scores of fourteen subjects on anxiety test (matching variable)

Subjects Scores Subjects Scores
1 70 8 90
2 75 9 70
3 69 10 81
4 90 11 85
5 69 12 75
6 81 13 62
7 85 14 62
Matched groups formed on the basis of the scores of the matching variable
Rank on Experimental group Control group

D = D1-D2
matching Subject Score on Score on Subject Score on Score on
variable No. the the No. the the D2
matching dependent matching dependent
variable variable variable variable
(D1) (D2)
1 4 90 8 8 90 6 2 4
2 11 85 8 7 85 7 1 1
3 10 81 7 6 81 4 3 9
4 2 75 6 12 75 5 1 1
5 1 70 5 9 70 3 2 4
6 5 69 3 3 69 1 2 4
7 14 62 1 13 62 2 1 1
38 28 ΣD 10 ΣD2 24
M1 = 38/7=5.4 M2= 28/7=4

df = n – 1 (where n= number of pairs)

M1 – M 2 5.4 – 4
t = = = 2.8 P< 0.05

(ΣD)2 (10)2
ΣD –
2
n 24 – 7
n (n-1) n 7 (7-1) n

144
df= 7 – 1 =6, at 6 df a t value of 2.447 is required to be significant at 0.05 level. This

means that the hypothesis is conformed. The higher the induced anxiety the better is

the learning.

Which design to prefer: Randomized Groups or Matched Groups?

A review of the literature clearly shows that the randomized groups design

has been used more frequently than the matched groups design. There are two

reasons for this.

(i) In the matched groups design the inconvenience of the experimenter is

increased because of two factors. One reason is that it is difficult to obtain a relevant

matching variable and obtain a set of scores upon. Another difficulty arises when the

experimenter is using the initial trail performance scores as the matching variable. If

the problem is to be investigated within a laboratory, the experimenter will have to

bring the subjects twice to the laboratory, the first time for obtaining initial scores and

the second time for obtaining scores after introducing the independent variable. This

requirement of keeping the subjects present twice in the laboratory is troublesome.

(ii) Many statistically advanced techniques can more easily be applied with

the randomized groups design than with the matched groups design.

A part from this, in the matched groups design the experimenter has fewer

degrees of freedom (df) than in the randomized groups design if the total number of

subjects (N) is equal in both the designs.

12.5. Factorial design

So far we have discussed the randomized groups design and the matched

groups design, which are appropriate for studying a single independent variable at a

time. If the single independent variable is varied in two ways the two-randomized

groups or the two matched groups design is recommended and when the single

145
independent variable is varied in more than two ways the more than two randomized

groups design is recommended. It is commonly observed that in many behavioural

researches the investigator is faced with a problem in which he is required to

manipulate two or more than two independent variables simultaneously. The

experimental design suited in such a situation is technically known as the factorial

design. A factorial design, then may be defined as a design in which the selected

values of two or more independent variables are manipulated in all possible

combinations so that their independent as well as interactive effects upon the

dependent variable may be studied.

Interaction is one of the most important concepts which is measured in

factorial design. An interaction exits between two independent variables if the

dependent variable value that results from one independent variable is determined

by the specific value assumed by the other independent variable.

Let us illustrate the concept of interaction with an example. Suppose the

problem is: Which of the two is more important for a person’s academic

achievement, Intelligence or the hard work? Obviously, the immediate answer will be

one of the two: Here intelligence is one independent variable which may be varied in

two ways intelligent and not intelligent. Hard work is the second independent

variable which may be varied in two ways students who study for more than two

hours everyday and students who study less than two hours everyday. The

dependent variable academic achievement can be measured by the marks obtained

in the annual examination. By conducting factorial design we will know (i) whether

intelligence has any effect on marks, (ii) whether hours of study has any effect on

the marks. In addition we also study interaction which will tell whether less

intelligence students who work more get better marks than more intelligence

146
students who work less. This is basic notion of the interaction. Thus an interaction

may be said to exist between the two independent variables when changes

produced in the dependent variable score by one independent variable are also

determined by the other independent variable.

When two independent variables are being investigated it is known as AX

B factorial design. If each of the independent variables are varied in 2 ways, the A X

B factorial design is represented as a 2 X 2 factorial design which means that there

are going to be four groups. If A is varied in 2 ways and B in 3 ways it is known as a

2 X 3 factorial design, which means there are going to be 6 groups. Similarly if 3

independent variables are being investigated it is known as A X B X C factorial

design. If each of these independent variables are varied in two ways it is known as

a 2 X 2 X 2 factorial design which means there are going to be 8 groups.

A 2 X 2 factorial design is represented in the following way

CONDITION A
A1 A2

B1
CONDITION B

B2

Example: An experimenter wants to evaluate the effect of a new drug on “curing”

psychotic tendencies. Two independent variables are studied the amount of the drug

administered and the type of psychotic condition. The amount of drug administered

is varied in two ways none and 2cc. The type of psychotic condition is also varied in

two ways schizophrenic and manic-depressive. Diagram the factorial design sued.

147
 In the drug experiment the psychologist used a measure of normality as the

dependent variable. This measure varies between 0 and 10, in which 10 is very

normal and 0 is very abnormal. Seven participants were assigned to each cell. The

resulting scores for the four groups were as follows. Conduct the appropriate

statistical analysis and reach a conclusion about the effect of each variable and the

interaction.

PSYCHOTIC CONDITION
Schizophrenics Manic Depressives
Did not Received drug Did not Received drug
Received Drug Received Drug
2 3 5 4
3 4 6 6
4 5 7 7
4 7 8 9
5 9 8 10
6 10 8 11
7 13 8 14

(1) The first step is to compute ΣX. ΣX2, and n for each condition.

Amount of drug administered

Did not
received drug Received drug
ΣX = 31 ΣX = 51
ΣX2 = 155 ΣX2 = 449
Schizophrenic
Psychotic n=7 n=7
condition
ΣX = 50 ΣX = 61
ΣX2 = 366 ΣX2 = 599
Manic depressive
n=7 n=7

(2) Total SS =

(ΣX1 + ΣX2 + ΣX3 + ΣX4)2

= ΣX12 + ΣX22 + ΣX32 + ΣX42 –
N

(31 + 51 + 50 + 61)2
= 238.68
= 155 + 449 + 336 + 599 –
28
148
(3) Between SS =

(ΣX1 + ΣX2 + ΣX3 + ΣX4)2

(ΣX1)2 + (ΣX2)2 + (ΣX3)2 + (ΣX4)2 –
n1 n2 n3 n3 N

(31 + 51 + 50 + 61)2
=67.25
(31)2 + (51)2 + (50) + (61)
2 2
–
7 7 7 7 28

The between SS consists of between SS for condition A, between SS for condition B

and interaction SS for A X B.
(i) Between SS for amount of drug administered

(ΣX1 + ΣX3)2 (ΣX2 + ΣX4)2 (ΣX1 + ΣX2 + ΣX3 + ΣX4)2

= + –
n1 + n3 n2 + n4 N

(31 + 50)2 (51 + 61)2 (31 + 51 + 50 + 61)2

= 34.32
= + –
7+ 7 7+ 7 28

(ii) Between SS for psychotic condition

(ΣX1 + ΣX2)2 (ΣX3 + ΣX4)2 (ΣX1 + ΣX2 + ΣX3 + ΣX4)2

= + –
n1 + n2 N3 + n4 N

(31 + 51)2 (50+ 61)2 (31 + 51 + 50 + 61)2

= 30.04
= + –
7+ 7 7+ 7 28

(iii) The sum of squares for the interaction component (A X B) =

Between SS – Between SS for A – Between SS for B =
67.25 – 34.32 – 30.04 = 2.89

149
(4) With in sum of squares = Total SS – Overall among SS
= 238.68 – 67.25 = 171.43

(5) Degrees of freedom (df)

(a) Total df = N – 1 = 28 – 1 = 27
(b) Between df = r – 1 = 4 – 1 = 3 (Where r = number of groups)
The components of the between df are
(i) Between df for A = r1 – 1 (where r1 = number of ways the firsts
independent variable varied) = 2 – 1 = 1
(ii) Between df for B = r2 – 1 (where r2 = number of ways the second
independent variable varied) = 2 – 1 = 1
(iii) Interaction df = between df for the first independent variable X between df
for the second independent variable = 1 X 1 = 1
(c) Within df = N – r = 28 – 4 = 24

(7) Compute the various mean squares. This is accomplished by dividing the several
sums of squares by the corresponding degrees of freedom. For our example, these
operations, as well as the results of the preceding ones, are summarized.
Source of variation Sum of df Mean F
squares square
Between A (Amount of drug 34.32 1 34.32 4.81 (P<0.05)
administered)
Between B (Psychotic 30.04 1 30.04 4.21Not significant
condition)
A X B (Interaction) 2.89 1 2.89 0.40 Not significant
With in groups 171.43 24
Total 238.68 27

(8) Compute an F for each “between” source of variation. In a 2 X 2 factorial design

there are three F tests to run.
F for A = Between mean square for A / Within mean square

150
 F for B = Between mean square for B / Within mean square
F for A X B = Interaction mean square / Within mean square
These F’s have been computed and entered in the preceding table.

(9) Looking into F tables to determine the probability associate with each F. To do

this find the column for the number of degrees of freedom associated with the

numerator and the row for the number of degrees of freedom associated with the

denominator. In our example they are 1 and 24, respectively. The F of 4.81 for

between the amounts of drug administered would thus be reliable beyond the 0.05

level, and accordingly we would reject the null hypothesis for this condition, which

means that there is difference in normality between the psychotic patients who

receive drug and who do not receive drug. The higher the score the more the

normality, and since the group which received the drug has the higher mean it

indicates that drug has curative effect on psychotic condition. The F between

psychotic condition (4.21) and for the interaction (0.40), however, are not significant

at the 0.05 level; hence, we would fail to reject the null hypothesis for these two

sources of variation. The lack of significant F’s for the between psychosis (B) and

interaction (A X B) indicates that there is no difference in normality as a function of

type of psychosis, nor is there an interaction between the variables.

12.6. Source
1. Garrett, H.E. (1966) Statistics in Psychology and Education. Bombay :
Vakils Feefer & Simon Pvt. Ltd.
2. Guilford J.P. (1965). Fundamental Statistics in Psychology and Education
(4th Edn.). New Delhi: Subject Publications.
3. Mc.Guigan, F.J. (1990) Experimental Psychology New Delhi: Prentice
Hall of India Limited.
4. Singh A.K. (1997) Tests measurements and Research Methods in
Behavioural science Patna: Bharati Bhavan Publishers and Distributors.

151