Professional Documents
Culture Documents
Designs
Research
4
CHAPTER
Learning Objectives
By the end of the chapter, you should be
able to:
Define an experiment and explain the
concept of causality.
Discuss the necessary conditions for drawing
3. Explain the basic concepts that are causal inferences.
used in experiments.
4. Explain the difference
between internal and external validity of theexperiment.
5. Explain the factors affecting internal
validity of the
6 Describe the factors affecting external validity of theexperiment.
7. Discuss the methods to experiment.
control extraneous variables.
8 Distinguish between laboratory and field experiments.
9. Explain the classification of experimental designs into
four categories--pre-experimental, quasi
experimental, true experimental design and statistical designs.
In 1991 Bajaj Enterprises set up a chain of supermarkets in all the Indian metros. These
supermarkets sell a broad line
of household and kitchen appliances. While the supermarkets in other metros were doing well, the one in
Delhi NCR
was showing a stagnant growth of 2-2.5 per cent per annum. The General Manager (Sales) was concened and was
thinking of ways to boost the sales. A meeting of the senior marketing officials was called to discuss the issue. Many
Suggestions came up including increasing the advertising budget, reducing the prices of slow-moving items, and giving
a discount to loyal customers. One of the suggestions was to offer a discount of 5 per cent in the form of coupons to
customers who opt for a bulk purchase of2,500/- and above. It was decided that these customers would be given 5per
cent discount coupons that they could redeem within a three-month period. It was argued that this would gradually result
n increasing sales and profits of the supermarkets. However, amarket researcher whowas part of the discussion team
argued that the sale increase depended upon a host of factors such as the size of the supermarket, location, the layout,
point-of-purchase (POP) displays, competitor's prices and competitor's advertising expenses besides other variables.
Ihe regulation of many of these was beyond their control. The GM (Sales) also gave athought to designing astudy in
order to examine the impact of the entire idea of discount on the bulk purchase scheme and gradually on the net sales
and profits of thesupermarkets. The members also realized that the extraneous factors would have to be controlled s0
as to infer a
causality.
Research Methodology
70
This chapterdiscussestheissuesinvolvedin inferringa cause and effectrelationship.
Anumber ofconcepts would be discussed which wouldihelpin setting up experiments
limitations of various designs in removing thei
to establish
causality.
extraneous variables
The
will also be covered
under this chapter.
influence of
EXPERIMENT?
WHAT IS AN An experimentis generally used toinfer a causality. In an experiment, a researcher
causal variables and measures their
dependenteffects
LEARNING OBJECTIVE 1 manipulates one or more changes in the on
actively any
Define an experiment the dependent variables
of interest. Since variable
variables, the relationship between cause and
number of other
and expiain the concept
may be caused by a probabilistic in nature. It is virtually impossible to prove a
of caUsality. often tends to be
causality. One can onlyinfer a cause-and-effecttrelationship. It is, therefore, essential
effect
Causality
some ofhis sales personnal
The sales manager of a soft drink bottling company sends
from the training
for a new sales training programme. Three months after they return increases hy
programme, the sales in the territory where this sales force was Working
20 per cent. The sales manager concludes that the training programmne is very
effective and, therefore, the sales force from the other territories should also he
sent for the same. What the sales manager is trying to infer is that the sales training
is a causal variable and increased sales is an effect variable. Do we agree to this
statement? This statement may not be true as the increase in sales may not be due to
the sales training programme alone. It could occur because of a host of factors e.g.,
reduction in the price of the soft drink,a strikeat the competitor's plant,increase in
the price of the competitor's product, reduction in the quality of competing products,
weather conditions and so on. Therefore, it is very important that the sales manager
understands the conditions under which such causal statements can be made. There
are three necessary conditions for making causal inferences.
NECESSARY CONDITIONS FOR MAKING CAUSAL INFERENCES
The following are the necessary conditions for making
causal inferences:
LEARNING OBJECTIVE 2 1. Concomitant variation: Concomitant variation is the extent to which a
Discuss the necesSary and effect Yoccurtogether orvarytogether. This cause A
conditions for drawing means that there hasto bea strong
association between the training programme and increased sales.
Causal inferences. of them need to occur
together. However, a strong association betweenMoreover, bon
does not imply causality. The he t
be due to the influence of otherhigh association between these two variables coud
the variables or it may be the ofextraneous factors which may be influencing bou
result of random variations.
2. Time order of
occurrence of
variable must occur prior to variables: This condition means that the
or caus
Concomitant variation is the means that sales training must have simultaneously with the effect
taken place either before or
variable. This
with the increased sales.
extent to which a cause Xand
an increase in sales However, just because sales simultaneousy
place prior
to
effect Yoccur together or vary will not help in training took
together.
a mere inferring causality.
coincidence and thus, cannot help in inferring It might have been dueto
causality.
Experimental Research Designs 71
urthermore, it is quite possible for each of the two events to be both cause and
effect of each other., In the illustrated example,the sales training programme may
cause an increase in sales, and increased sales may result in keeping company
some spare funds for training etc. Therefore, the relationship between the two
variables could be that they alternatively 'feed' each other.
Even if it can be shown that there is a concomitant variation between the
sales training programme and the increased sales and the time occurrence of
all variables, there is still a question left unanswered whether other variables
which could 'cause' increased sales have remained ina constant position. This is
explained in the next point.
3. Absence of other possible causal factors: As mentioned earlier, the increase in
an
Theobjective of measure the sales of soft drink could have been due tomany other factors besides the sales
experimentis to training. There could be a strike at the competitor's plant, resulting in an overall
independent
influenceofthe reduction in supply, weather conditions, the increased price of the competitor's
dependent
variables on a product or a problem at the distribution channel at the competitor's end. The sales
variable while keeping the training programme may be a causal variable if all the other factors mentioned
effect of other extraneous above were kept constant or otherwise controlled.
variables constant.
Asamatter offact, the researcher cannotrule outthe influence ofother causal factors
such as the weather condition. However, it will be seen later that it may be possible to
control some or nmore of the extraneous variables by the use of experimental design. It
may be possible to balance the effect of some uncontrolled factors. This may help in
measuring random variations resulting from uncontrolled measures.
Experiments are used to seekhelp in identifying acause-and-effectrelationship.
The objective of an experiment is to measure the influence of the independent
variables on a dependent variable while keeping the effect of other extraneous
variables constant. Experiments may be used to arrive at conclusive answers in
the following situations:
Can a change in the package design of a product enhance its sales?
purchase to
" Should a supermarket introduce a discount scheme on bulk
increase its sales?
particular
Will an increase in the shelf space allocated to a brand of a
product increase its sales?
restaurant increase
. Will a reduction in the price of the menu items of a
sales?
on their sales?
What will be the impact of POP display of Arrow shirts
increasing the
Which of several promotional techniques is most effective in
sales of a product?
of female counter clerks
What is the impact of increasing the proportion
the store?
from 30 to 60 per cent on the sales of
person to the organizational
Does mentoring help in acclimatizing a
culture? company?
Does organizational climate impact the quality of workinglife of ainvestment
home loan rates on the investor
" What is the impact of change in
in real estate?
understanding of experimentation, it would be useful to
In order to have a good
definition used in experiments.
learn some basic concepts and
Suppose the diference in 'rupee' sales before and 'after' the training
programme is used to measure the effectiveness of the training programme, a
price difference during the time interval could make a substantial difference in
the inference. A'change in price' would be the change of
instrumentation.
Presenting the pre and post-test questionnaire in a different fashion, experience
of the invigilator, and a change in the mood of the investigators are some of the
examples of changing instrumentation.
Statistical regression Ocurs 5. Statistical regression: The effect of statistical regression occurs when the test
when the testunits
with units with extreme scores (either extremely favourable or extremely unfavourable)
sores are hosen for are chosen for exposure to the treatment. The effect is that test units with extreme
etree
e20sure to the treatment. sCores tend to move towards an average score with the passage of time. Suppose
in the example of the sales training programme, the sales people with extremely
poor performance are sent for the training programme. An increase in sales
after the training programme may be attributed to the regression effect. This is
because test units with extreme score have more room for a change, so a variation
0s more likely to be there. Random occurrences (weather, luck, festive seasons),
might have helped good and poor performance of sales people in the pre-test
measurement. These random occurrences will turn some of the poor performers,
into better performers thereby confounding the experiment.
6. Selection bias: This refers to the improper assignments oftest units to treatments.
Test units may be assigned to the treatment groups in such a way that the groups
differ on the dependent variable prior to the presentation of the treatment.
Selection bias can occur if test units self-select their groups or are assigned to the
groups on the basis of the researcher's judgment. The selection of test units to the
treatment group should be random.
7. Test unit mortality: Some of the test units might drop out from the experiment
while it is in progress or some may refuse to continue with the experiment. In the
case of sales training example, some sales people may quit the organization before
completing the training successfully. There is no way of finding outwhether those
who were not improving quit the organization. It is also not possible to measure
whether those who left would have produced the same results as those who
completed the training programme.
The types of extraneous variables discussed above are not mutually exclusive.
They can occur together and interact with each other. These extraneous variables
can provide alternative explanations regarding what is being observed in an
experiment and our objective should be to eliminate the possibility of these effects
confounding the results.
used in experiments?
1 What are the concepts and 'external validity'?
terms 'internalvalidity'
CONCEPT 2 What is meant by the experimental research.
symbols commonly used in experiment.
CHECK 3. Define the set of
affect the internal and external validity of a particular
4. Name the prime factors that In a la
the re
artific
EXTRANEOUS VARIABLES
METHODS TO CONTROL Condu
discussed in the previous sections, extraneous variables pose a threat to the field
LEARNING OBJECTIVE 7 As
Discuss the methods internal and external validity of the experiment. They affect the dependent variable mar
A
CLASSIFICATION OF EXPERIMENTAL DESIGNS
quasi-experimental,
Experimental design can be classified as pre-experimental,
Pre-experimental designs include the one
true experimental and statistical.pre-test-post-test design and the static group
shot case study, the one-group quasi-experimental designs are time series
comparison. Tests included under
True-experimental designs include pre-test-post-test
and multiple time series. The
group, and Solomon four-group design.
control group, post-test-only control
78 Research Methodology
FIGURE 4.1
Experimental
Cassification of Design
ëxerimental design
Factorial
Experimental Research Designs 79
Let us examine another example here. The objective is to study the impact of
an extra ten days' credit period (X) on a credit card
payment time (0) and one
decides to study the relationship/impact by offering this to the customers who
make an average usage of 25,000/- per month. The problem in this case would
be that no measure was taken to establish their
payment behaviour prior to the
extended period. Hence, no valid conclusion can be made from this design. There
1S no pre-treatment observation on performance. The level of'O' might be
by several uncontrolled extraneous factors like history, affected
maturation, selection bias
and test unit mortality. These uncontrolled extraneous variables will confound
the experiment and render the design
One-group pre-test-post-test 2. One-group pre-test-post-test design:internally invalid.
This design is also called before-after
design is also known as before
after without control group design without control group design. This design may be written symbolically as:
and may be symbolically written 0, X 0,
In this design also, test units are not selected at random as the symbol 'R is not
appearing here. The test units are subjected to the treatmentX and both pre
treatment (0,) and post-treatment measurement (0,) are taken. For instance,
in the credit card example, one might take the payment time
before and after
the extended ten-days' period. One may be tempted to compute
treatment
effect as 0, - 0,, which may not be really so, as this difference could be the
result of many uncontrolled extraneous factors like history, maturation, testing,
instrumentation, regression, selection and mortality. This would make the
design invalid for making any causal inferences on account of the following
reasons:
Theeconomic condition might have changed during the two periods (history).
The test units may mature over time (maturation).
The pre-test measurement on the test units may influence the performance
(testing).
" The prices of goods might havechanged over time (instrumentation).
Test units might not have been selected at random (selection bias).
" Some test units might have left before the experiment was complete (mortality).
Test units might be self-selected on the basis of the current poor performance
and may have a better period ahead because of sheer luck (regression).
Static group comparison 3. Static group comparison: This design is symbolically written as:
Uses two treatment groups
in which test units are not Group 1 X
selected at random.This Group 2
design is presented as: This design uses two treatment groups. Test units in both the groups are not
Group 1-X0, selected at random. The first group, called the experimental group, is subjected
Group 2- O, to the treatmentX, whereas the second group, namely, the control group, is not
subjected to any treatmnent. Both groups are measured only after the treatment has
been presented. Thus, it is critical to understand that in this design the exposure
as well as the experimental treatment not under the control of the researcher.
Consider the following example:
Astudy wants to assess the relationship of 'family support' (measured by the
presence of domestic help or spouse/family's help in carrying out domestic
chores)with the work-life balance of BPO women employees. Here, the presence
or absence of help is ascertained and then we can mneasure the work-life balance.
Thusthe design is essentially ex-post facto and any segregation into experimental
or controlgroup is not made by the researcher.
80 Research Methodology
FIGURE4,2 70
60
(%)
Share
Market
50
B
-AC
40
30
20
10
2 3 4 X 5 6 7 8
Source: Adopted with modification from Thomas C. Kinnear &James R.
Taylor,
"Marketing Research: An Applied Approach" McGraw-Hil, Inc., Fifth Edition
Here, the test units in both the experimental and the control group are selected
at random. The experimental group is subjected to the treatment X, and post-test
measurements are taken on both experimental (O,) and control group (0,)at tne
same time. The post-test measurement (0,) on experimental group comprises
treatment effect and all other extraneous variables, whereas O, comprises only
extraneous variables. Therefore, the difference in the post-test measurement of
experimental and control group is taken as a measure of treatment effect. Hente
0, -, = (Treatmenteffect +extraneous factors)- (extraneous factors)
Treatment effect
As pre-test measurement is absent, the effect of instrumentation
andinteracuvo
testing effect is ruled out. As there is a random assignment of test units to both uio
groups, itcan be approximately assumed that both the groups were equal prior o
Experimental Research Designs 83
the difference in
subjected to any treatment,
As the group was not
would only indicate the
effect of extraneous
factors
measuretment
without interactive
esting
effect.
(v) and (vi), one gets:
By taking the average of
0, +0, =(Extraneous factors without testing effect)
2
one obtains:
..vii)
By subtracting (vii) from (viii),
Statistical Designs
Statistical designs allow for statistical control and analysis of external variables. The
main advantages of statistical design are the following:
The effect of more than one level of independent variable on the dependent
variable can be manipulated.
" The effect of more than one independent variable can be examined.
" The effect of specific extraneous variable can be controlled.
Included in this category are the following designs:
Compieteily randomized
design alloWs aresearcher to 1. Completely randomized design: This design is used when a researcher is
nvestigate the effect of one investigating the effect of one independent variable on the dependent variable.
ndependet variable on the The independent variable is required to be measured in nominal scale i.e. it
dependent variable. should have a number of categories. Each of the
variable is considered as the treatment. The basiccategories
of the independent
assumption of this design S
that there are no differences in the test units. All the test units are treated alike and
randomly assigned to the test groups. This means that there are no
variables thatcould influence the outcome. extraneous
Suppose we know that the sales of a product is
In this case, sales are a dependent variable and influenced by the price level.
the price is the independent
variable. Let there be three levels of price, namely, low,
wish to determine the most effective price level, ie.. at medium and high. we
which price level the sale
Erpernental Research Designs 85
is highest. Here the test units are the stores which are randomly assigned to the
three treatment 1levels. The average sales for each price level is computed and
examined to see whether there is anysignificant difference in the sale at various
price levels. The statistical technique to test for such a difference is called analysis
of variance (ANOVA).
mtaion of the This design suffers from the main limitation that it does not take into account
h aan
completely tandomized the effect of extraneous variables on the dependent
does not variable. The possible
design s ihat it extraneous variables in the present example could be the size of the store, the
cffect of
ake nto a«ount the competitor's price and price of the substitute product in question. This design
variables on the
NaneoU assumes that all the extraneous factors have the same influence on all the test
enendent variable units which may not betrue in reality. This design is very simple and inexpensive
to conduct.
TABLE 4.1
Store Size
Packaging
Latinsquare deslqn for
various levels of price
1(Small) X, X, X,
2 (Medium) X, X, X.
3 (Large) X X X
It may be noted that the rOws and columns represent those
extraneous variable.
whose effect is to be controlled and measured. There are three categories of rou
variable (size of store) and three categories of column variable (type of packaging)
This would result in 3 x 3 Latin square.
One point that has to be kept in mnind is that the treatment should be assion ed
randomly tocells in such a way that each treatment occurs once and only once in
each row and in each column. The treatments exhibited in Table 4.1 satisfy this
condition.
Use of this design helps to measure statistically the effect of a treatment on
the dependent variable andalso the measurement of an error resulting from two
extraneous variables. This design, indeed has a very complex setup and isquite
expensive to execute.
Afactorial design is
4. Factorial design: A factorial design may be employed to measure the
employed to measure effect of
the effect of two or more two or more independent variables at various levels. The factorial designs allow
independent variables at interaction between the variables. An interaction is said to take place when the
various levels. simultaneous effect of two or more variables is different from the sum of their
individual effects. An individual may have a high preference for mangoes and may
also like ice-cream, which does not mean that he would
like mango ice cream,
leading to an interaction.
The sales of a product may be influenced by two factors,
and store size. There may be three levels of price--low namely, price level
(A,), medium (A) and
high (A,). The store, size could be categorized into small (B,) and big
could be conceptualized as a two-factor design with (B,). This
form of a table. In the table, each level of one information reported in the
and each level of another variable would be factor may be presented as a roW
presented as a
could be summarized in the form of a table having three column. Thisexample
This would require 3 x 2 = 6 cells. Therefore, six rows and twO columns.
different levels of treatment
combinations would be produced, each with a specific
size. The respondents would be randomly selected and level of price and store
six cells. The tabular presentation of 3 × 2 randomly assigned to the
factorial design is given in Table 4.2.
TABLE 4.2
3x2 factorial design for Price Store
price level and store size Small (B,)
Low Level (A,)
Big (B,)
A,B, A,B,
Medium Level (A,) A,B, A,B,
High Level (A,) A,B, A,B,
Respondents in each cell receive a specified treatment
example, respondents in the upper left hand corner cell
combination. ro
price and small store. Similarly, the would face small level o
cell will be respondents in the lower right hand
subjected to both high price level and big store. cornel
Experinental Research Designs 87
SUMMARY
¢ Experiments are used to infer causality where the researcher actively manipulates one or more causal variables
and measure their effects on the dependent variable. There are three necessary conditions for inferring causality: 0)
concomitant variation (i) time order of occurrence of variables, and (ii) the absence of other possible causal factors.
Various concepts like independent variables (treatments), test units, dependent variables, exogenous variables
are used in conducting an experiment. An experiment can be conducted under different environmental conditions,
namely, laboratory and field. The researcher has two goals while conducting an experiment: () to keep the internal
wider popu
validity of the experiment very high and (ii) to make generalization of the results of the experiments to a wh0se influ
the causal factors except the one
lation. Internalvalidity is concerned with examining the absence of all
refers to the generalization
ence is being examined on the dependent variable. External validity, on the other hand, These are
validity of the experiment.
of the results of the experiment. There are various factors affecting the internal
statistical regression, selection bias and test units' mortality. Similarly,
history, maturation, testing, instrumentation,
factors may be common to both the
there are factors influencing the external validity of an experiment. Some of the
effects of extraneous variables
internal and the external validity of the experiment. The methods of controlling the
are also discussed.
true-experimental, and statistical
Experimental designs are classified into pre-experimental, quasi-experimental, (ii) one-group pre-test-post-test
design. Under pre-experimental design are included (0) one-shot case study,
designs do not make use of randomization pro
design and (ii) static group comparison. The pre-experimental
the internal validity of such experiments remains
cedure in order to control the extraneous variables. Therefore,
series design and (ii) multiple time series de
doubtful. Under quasi-experimental design are discussed (i) time
when the measurements are to be taken and on whom
sign. In these designs the researcher has control over ability to
they are taken. However, the design lacks complete control of scheduling of treatment and also lacks
category of true-experimental design are (i) pre
randomize test units exposure to treatments. Included in the
and (ii) Solomon four-group design. In these de
test-post-test control group, (ü) post-test-only control group researcher is
units and treatments to experimental groups. The
Signs, the researcher can randomly assign test statistical de
to eliminate the effect of extraneous variables from both control and experimental groups. The design,
able Latin square
randomized design, (iü) randomized block design, (ii) level of independent
SIgns covered here are (i) completely effect of more than one
designs help to (i) study the
and (iv) factorial design. The statistical study the effect of more than one independent variable and (ii) the effect
Variables on the dependent variable: (ii)
of specific extraneous variables.