CONDUCTING AN EXPERIMENT

Science revolves around experiments, and learning the best way of conducting an experiment is crucial to obtaining useful and valid results.
by Martyn Shuttleworth (2008)
When scientists speak of experiments, in the strictest sense of the word, they mean a true experiment, where the scientist controls all of the factors and conditions.
Real world observations, and case studies, should be referred to as observational research, rather than experiments.
For example, observing animals in the wild is not a true experiment, because it does not isolate and manipulate an independent variable.

THE BASIS OF CONDUCTING AN EXPERIMENT

With an experiment, the researcher is trying to learn something new about the world, an explanation of ‘why’ something happens.
The experiment must maintain internal and external validity, or the results will be useless.
When designing an experiment, a researcher must follow all of the steps of the scientific method, from making sure that the hypothesis is valid and testable, to using controls and statistical tests.
Whilst all scientists use reasoning, operationalization and the steps of the scientific process, it is not always a conscious process.
Experience and practice mean that many scientists follow an instinctive process of conducting an experiment, the ‘streamlined’ scientific process. Following the basic steps will usually generate valid results, but where experiments are complex and expensive, it is always advisable to follow the rigorous scientific protocols. Conducting an experiment has a number of stages, where the parameters and structure of the experiment are made clear.
Whilst it is rarely practical to follow each step strictly, any aberrations must be justified, whether they arise because of budget, impracticality or ethics.

STAGE ONE

After deciding upon a hypothesis, and making predictions, the first stage of conducting an experiment is to specify the sample groups. These should be large enough to give a statistically viable study, but small enough to be practical.
Ideally, groups should be selected at random, from a wide selection of the sample population. This allows results to be generalized to the population as a whole.
In the physical sciences, this is fairly easy, but the biological and behavioral sciences are often limited by other factors.
For example, medical trials often cannot find random groups. Such research often relies upon volunteers, so it is difficult to apply any realistic randomization. This is not a problem, as long as the process is justified, and the results are not applied to the population as a whole.
If a psychological researcher used volunteers who were male students, aged between 18 and 24, the findings can only be generalized to that specific demographic group within society.

STAGE TWO

The sample groups should be divided, into a control group and a test group, to reduce the possibility of confounding variables.
This, again, should be random, and the assigning of subjects to groups should be blind or double blind. This will reduce the chances of experimental error, or bias, when conducting an experiment.
Ethics are often a barrier to this process, because deliberately withholding treatment, as with the Tuskegee study, is not permitted.
Again, any deviations from this process must be explained in the conclusion. There is nothing wrong with compromising upon randomness, where necessary, as long as other scientists are aware of how, and why, the researcher selected groups on that basis.

STAGE THREE

This stage of conducting an experiment involves determining the time scale and frequency of sampling, to fit the type of experiment.
For example, researchers studying the effectiveness of a cure for colds would take frequent samples, over a period of days. Researchers testing a cure for Parkinson’s disease would use less frequent tests, over a period of months or years.

STAGE FOUR

The penultimate stage of the experiment involves performing the experiment according to the methods stipulated during the design phase.
The independent variable is manipulated, generating a usable data set for the dependent variable.

STAGE FIVE

The raw data from the results should be gathered, and analyzed, by statistical means. This allows the researcher to establish if there is any relationship between the variables and accept, or reject, the null hypothesis.
These steps are essential to providing excellent results. Whilst many researchers do not want to become involved in the exact processes of inductive reasoning, deductive reasoning and operationalization, they all follow the basic steps of conducting an experiment. This ensures that their results are valid.


Read more: http://www.experiment-resources.com/conducting-an-experiment.html#ixzz1CLJJDaFX


True experimental design
For some of the physical sciences, such as physics, chemistry and geology, they are standard and commonly used. For social sciences, psychology and biology, they can be a little more difficult to set up.
For an experiment to be classed as a true experimental design, it must fit all of the following criteria.
• The sample groups must be assigned randomly.
• There must be a viable control group.
• Only one variable can be manipulated and tested. It is possible to test more than one, but such experiments and their statistical analysis tend to be cumbersome and difficult.
• The tested subjects must be randomly assigned to either control or experimental groups.

ADVANTAGES

The results of a true experimental design can be statistically analyzed and so there can be little argument about the results.
It is also much easier for other researchers to replicate the experiment and validate the results.
For physical sciences working with mainly numerical data, it is much easier to manipulate one variable, so true experimental design usually gives a yes or no answer.

DISADVANTAGES

Whilst perfect in principle, there are a number of problems with this type of design. Firstly, they can be almost too perfect, with the conditions being under complete control and not being representative of real world conditions.
For psychologists and behavioral biologists, for example, there can never be any guarantee that a human or living organism will exhibit ‘normal’ behavior under experimental conditions.
True experiments can be too accurate and it is very difficult to obtain a complete rejection or acceptance of a hypothesis because the standards of proof required are so difficult to reach.
True experiments are also difficult and expensive to set up. They can also be very impractical.
While for some fields, like physics, there are not as many variables so the design is easy, for social sciences and biological sciences, where variations are not so clearly defined it is much more difficult to exclude other factors that may be affecting the manipulated variable.

SUMMARY

True experimental design is an integral part of science, usually acting as a final test of a hypothesis. Whilst they can be cumbersome and expensive to set up, literature reviews, qualitative research and descriptive research can serve as a good precursor to generate a testable hypothesis, saving time and money.
Whilst they can be a little artificial and restrictive, they are the only type of research that is accepted by all disciplines as statistically provable.


Read more: http://www.experiment-resources.com/true-experimental-design.html#ixzz1CLKCFsAs


MODULE R13
EXPERIMENTAL RESEARCH AND DESIGN

Experimental Research - An attempt by the researcher to maintain control over all factors that may affect the result of an experiment. In doing this, the researcher attempts to determine or predict what may occur.
Experimental Design - A blueprint of the procedure that enables the researcher to test his hypothesis by reaching valid conclusions about relationships between independent and dependent variables. It refers to the conceptual framework within which the experiment is conducted.
Steps involved in conducting an experimental study
Identify and define the problem.
Formulate hypotheses and deduce their consequences.
Construct an experimental design that represents all the elements, conditions, and relations of the consequences.
1. Select sample of subjects.
2. Group or pair subjects.
3. Identify and control non experimental factors.
4. Select or construct, and validate instruments to measure outcomes.
5. Conduct pilot study.
6. Determine place, time, and duration of the experiment.
Conduct the experiment.
Compile raw data and reduce to usable form.
Apply an appropriate test of significance.

Essentials of Experimental Research
Manipulation of an independent variable.
An attempt is made to hold all other variables except the dependent variable constant - control.
Effect is observed of the manipulation of the independent variable on the dependent variable - observation.
Experimental control attempts to predict events that will occur in the experimental setting by neutralizing the effects of other factors.
Methods of Experimental Control
Physical Control
Gives all subjects equal exposure to the independent variable.
Controls non experimental variables that affect the dependent variable.

Selective Control - Manipulate indirectly by selecting in or out variables that cannot be controlled.
Statistical Control - Variables not conducive to physical or selective manipulation may be controlled by statistical techniques (example: covariance).

Validity of Experimental Design
Internal Validity asks did the experimental treatment make the difference in this specific instance rather than other extraneous variables?
External Validity asks to what populations, settings, treatment variables, and measurement variables can this observed effect be generalized?
Factors Jeopardizing Internal Validity
History - The events occurring between the first and second measurements in addition to the experimental variable which might affect the measurement.
Example: Researcher collects gross sales data before and after a 5 day 50% off sale. During the sale a hurricane occurs and results of the study may be affected because of the hurricane, not the sale.
Maturation - The process of maturing which takes place in the individual during the duration of the experiment which is not a result of specific events but of simply growing older, growing more tired, or similar changes.
Example: Subjects become tired after completing a training session, and their responses on the Posttest are affected.
Pre-testing - The effect created on the second measurement by having a measurement before the experiment.
Example: Subjects take a Pretest and think about some of the items. On the Posttest they change to answers they feel are more acceptable. Experimental group learns from the pretest.
Measuring Instruments - Changes in instruments, calibration of instruments, observers, or scorers may cause changes in the measurements.
Example: Interviewers are very careful with their first two or three interviews but on
the 4th, 5th, 6th become fatigued and are less careful and make errors.
Statistical Regression - Groups are chosen because of extreme scores of measurements; those scores or measurements tend to move toward the mean with repeated measurements even without an experimental variable.
Example: Managers who are performing poorly are selected for training. Their average Posttest scores will be higher than their Pretest scores because of statistical regression, even if no training were given.
Differential Selection - Different individuals or groups would have different previous knowledge or ability which would affect the final measurement if not taken into account.

Example: A group of subjects who have viewed a TV program is compared with a group which has not. There is no way of knowing that the groups would have been equivalent since they were not randomly assigned to view the TV program.
Experimental Mortality - The loss of subjects from comparison groups could greatly affect the comparisons because of unique characteristics of those subjects. Groups to be compared need to be the same after as before the experiment.

Example: Over a 6 month experiment aimed to change accounting practices, 12 accountants drop out of the experimental group and none drop out of the control group. Not only is there differential loss in the two groups, but the 12 dropouts may be very different from those who remained in the experimental group.
Interaction of Factors, such as Selection Maturation, etc. - Combinations of these factors may interact especially in multiple group comparisons to produce erroneous measurements.
Factors Jeopardizing External Validity or Generalizability
Pre-Testing -Individuals who were pretested might be less or more sensitive to the experimental variable or might have "learned" from the pre-test making them unrepresentative of the population who had not been pre-tested.

Example: Prior to viewing a film on Environmental Effects of Chemical, a group of subjects is given a 60 item antichemical test. Taking the Pretest may increase the effect of the film. The film may not be effective for a nonpretested group.
Differential Selection - The selection of the subjects determines how the findings can be generalized. Subjects selected from a small group or one with particular characteristics would limit generalizability. Randomly chosen subjects from the entire population could be generalized to the entire population.

Example: Researcher, requesting permission to conduct experiment, is turned down by 11 corporations, but the 12th corporation grant permission. The 12th corporation is obviously different then the others because they accepted. Thus subjects in the 12th corporation may be more accepting or sensitive to the treatment.
Experimental Procedures - The experimental procedures and arrangements have a certain amount of effect on the subjects in the experimental settings. Generalization to persons not in the experimental setting may be precluded.

Example: Department heads realize they are being studied, try to guess what the experimenter wants and respond accordingly rather than respond to the treatment.
Multiple Treatment Interference - If the subjects are exposed to more than one treatment then the findings could only be generalized to individuals exposed to the same treatments in the same order of presentation.

Example: A group of CPA’s is given training in working with managers followed by training in working with comptrollers. Since training effects cannot be deleted, the first training will affect the second.
Tools of Experimental Design Used to Control Factors Jeopardizing Validity
Pre-Test - The pre-test, or measurement before the experiment begins, can aid control for differential selection by determining the presence or knowledge of the experimental variable before the experiment begins. It can aid control of experimental mortality because the subjects can be removed from the entire comparison by removing their pre-tests.
However, pre-tests cause problems by their effect on the second measurement and by causing generalizability problems to a population not pre-tested and those with no experimental arrangements.
Control Group -The use of a matched or similar group which is not exposed to the experimental variable can help reduce the effect of History, Maturation, Instrumentation, and Interaction of Factors. The control group is exposed to all conditions of the experiment except the experimental variable.
Randomization - Use of random selection procedures for subjects can aid in control of Statistical Regression, Differential Selection, and the Interaction of Factors. It greatly increases generalizability by helping make the groups representative of the populations.
Additional Groups - The effects of Pre-tests and Experimental Procedures can be partially controlled through the use of groups which were not pre-tested or exposed to experimental arrangements. They would have to be used in conjunction with other pre-tested groups or other factors jeopardizing validity would be present.

The method by which treatments are applied to subjects using these tools to control factors jeopardizing validity is the essence of experimental design.
Tools of Control

Internal Sources Pre-Test/
Post Test Control Group Randomization Additional
Groups
History X
Maturation X
Pre-Testing X
Measuring Instrument X
Statistical Regression X X
Differential Selection X X
Experimental Mortality X
Interaction of Factors X X
External Sources
Pre-Testing X
Differential Selection X X
Procedures X
Multiple Treatment



Experimental Designs
Pre-Experimental Design - loose in structure, could be biased
Aim of the Research Name of the Design Notation Paradigm Comments
To attempt to explain a consequent by an antecedent One-shot experimental case study X » O An approach that prematurely links antecedents and consequences. The least reliable of all experimental approaches.
To evaluate the influence of a variable One group pretest-posttest O » X » O An approach that provides a measure of change but can provide no conclusive results.
To determine the influence of a variable on one group and not on another Static group comparison Group 1: X » O
Group 2: - » O Weakness lies in no examination of pre-experimental equivalence of groups. Conclusion is reached by comparing the performance of each group to determine the effect of a variable on one of them.

True Experimental Design - greater control and refinement, greater control of validity
Aim of the Research Name of the Design Notation Paradigm Comments
To study the effect of an influence on a carefully controlled sample Pretest-posttest control group R - - [ O » X » O
[ O » - » O This design has been called "the old workhorse of traditional experimentation." If effectively carried out, this design controls for eight threats of internal validity. Data are analyzed by analysis of covariance on posttest scores with the pretest the covariate.
To minimize the effect of pretesting Solomon four-group design R - - [ O » X » O
[ O » - » O
[- » X » O
[ - » - » O This is an extension of the pretest-posttest control group design and probably the most powerful experimental approach. Data are analyzed by analysis of variance on posttest scores.
To evaluate a situation that cannot be pretested Posttest only control group R - - [ X » O
[ - » O An adaptation of the last two groups in the Solomon four-group design. Randomness is critical. Probably, the simplest and best test for significance in this design is the t-test.
Quasi-Experimental Design - not randomly selected
Aim of the Research Name of the Design Notation Paradigm Comments
To investigate a situation in which random selection and assignment are not possible Nonrandomized control group pretest-posttest O » X » O
O » - » O One of the strongest and most widely used quasi-experimental designs. Differs from experimental designs because test and control groups are not equivalent. Comparing pretest results will indicate degree of equivalency between experimental and control groups.
To determine the influence of a variable introduced only after a series of initial observations and only where one group is available Time series experiment O » O » X » O » O If substantial change follows introduction of the variable, then the variable can be suspect as to the cause of the change. To increase external validity, repeat the experiment in different places under different conditions.
To bolster the validity of the above design with the addition of a control group Control group time series O » O » X » O » O
O » O » - » O » O A variant of the above design by accompanying it with a parallel set of observations without the introduction of the experimental variable.
To control history in time designs with a variant of the above design Equivalent time-samples [X1 » O1] »[X0 » O2] » [x1 » O3] An on-again, off-again design in which the experimental variable is sometimes present, sometimes absent.
Correlational and Ex Post Facto Design
Aim of the Research Name of the Design Notation Paradigm Comments
To seek for cause-effect relationships between two sets of data Causal-comparative correlational studies -»
Oa ¥ Ob
«- A very deceptive procedure that requires much insight for its use. Causality cannot be inferred merely because a positive and close correlation ratio exists.
To search backward from consequent data for antecedent causes Ex post facto studies This approach is experimentation in reverse. Seldom is proof through data substantiation possible. Logic and inference are the principal tools of this design
Leedy, P.D. (1997). Practical research: Planning and design (6th ed.). Upper Saddle River, NJ: Prentice-Hall, Inc., p. 232-233.

SELF ASSESSMENT
1. Define experimental research.
Define experimental design.
2. List six steps involved in conducting an experimental study.
3. Describe the basis of an experiment.
4. Name three characteristics of experimental research.
5. State the purpose of experimental control.
6. State three broad methods of experimental control.
7. Name two type of validity of experimental design.
8. Define eight factors jeopardizing internal validity of a research design.
9. Define four factors jeopardizing external validity.
10. Describe the tools of experimental design used to control the factors jeopardizing validity of a research design.
11. Define the essence of experimental design.
12. Name and describe the four types of experimental designs.



EXPERIMENTAL
RESEARCH
METHODS
John Davis, Ph.D.
Department of Psychology
Metropolitan State College of Denver
These notes and outlines are part of a site on psychological research methods. They are intended as a brief introduction and overview for undergraduate students in psychological research methods courses.
Click here to go to the HOME PAGE for this site.
Click here to go to the CONTENTS of OUTLINES AND DESCRIPTIONS of research methods.

Click on the topic you want
Part 1. THE MODEL AND GOALS OF EXPERIMENTAL RESEARCH METHODS

Part 2. VARIABLES AND VALIDITY IN EXPERIMENTAL DESIGNS

Part 3. HYPOTHESES IN EXPERIMENTAL DESIGNS, including alternative hypotheses
Part 4. EXPERIMENTAL DESIGNS

Part 5. SINGLE-SUBJECT DESIGNS

THE MODEL UNDERLYING EXPERIMENTAL RESEARCH METHODS

Experimental research designs are founded on the assumption that the world works according to causal laws. These laws are essentially linear, though complicated and interactive. The goal of experimental research is to establish these cause-and-effect laws by isolating causal variables.
A softer view of the philosophical assumptions behind experimental designs is that SOMETIMES and IN SOME WAYS, the world works according to causal laws. Such cause-and-effect relationships may not be a final view of reality, but demonstrating cause and effect is useful in some circumstances.
Both of these views agree that some (if not all) important psychological questions are questions about what causes what. Experimental research designs are the tools to use for these questions.
The GOAL OF EXPERIMENTAL RESEARCH METHODS is to establish cause-and-effect relationships between variables.
We hypothesize that the Independent Variable caused the changes in the Dependent Variable. However, these changes or effects may have been caused by many other factors or Alternative Hypotheses.
The PURPOSE, therefore, of experimental designs is to eliminate alterntive hypotheses. If we can successfully eliminate all alternative hypotheses, we can argue--by a process of elimination--that the Independent Variable is the cause.
Good experimental designs are those which eliminate more alternative hypotheses.

FOR EXAMPLE: Say I am testing whether a new form of psychotherapy is successful at improving mental health. I hypothesize that this psychotherapy is the cause of improved mental health in the research participants.
I will use an experimental design to eliminate all (or as many as possible) alternative hypotheses. If I can eliminate alternative explanations, I will be able to make the case that the psychotherapy was the cause of the improvements in the research participants.


EXPERIMENTAL RESEARCH METHODS
John Davis, Ph.D.
Part Two: Types of Variables and Validity
Contents
TYPES OF VARIABLES
RELIABILITY AND VALIDITY
Types of Variables
1. Independent Variable (IV): IV has levels, conditions, or treatments. Experimenter may manipulate conditions or measure and assign subjects to conditions; supposed to be the Cause
In the example, it is the psychotherapy.

2. Dependent Variable (DV): measured by the experimenter; the Effect or result.
In the example, it is the mental health of the participants.

3. Control Variables: held constant by the experimenter to eliminate them as potential causes.
For instance, if I use only research participants who have been problems with anxiety or depression, this diagnosis would be a control variable.

4. Random Variables: allowed to vary freely to eliminate them as potential causes.
Many other characteristics of the research participants, as long as they really do vary freely. Examples might include age, personality type, or career goals.

5.






Confounding Variables: vary systematically with the independent variable; may also be a cause. Good experimental designs eliminate them.
Say I divide the research participants into two groups, one of which gets the new psychotherapy (the experimental group) and one of which does not (the control group). If there is some systematic difference between these two groups, it will not be a fair test.
If those in the psychotherapy group know they are getting a new treatment and therefore expect to get better while those in the control group know they are not getting any treatment and expect to get worse, the expectations will be a confounding variable. If the experimental group does improve, we will not know whether it was because of the psychotherapy itself (the Independent Variable) or because of the participants' expectations (a Confounding Variable).
Go to TOP of this page
RELIABILITY AND VALIDITY
1. RELIABILITY
• Are the results of the experiment repeatable?
• If the experiment were done the same way again, would it produce the same results?
• Reliability is a requirement before the validity of the experiment can be established.


2.
INTERNAL VALIDITY
• Accuracy or truth-value
• Does the research design lead to true statements?
• Did the independent variable cause the effects in the dependent variable?
• In experimental research, this usually means eliminating alternative hypotheses.
• In the example evaluating a new psychotherapy, the issue of internal validity is whether the psychotherapy really was the causal factor in improving participants' mental health.

3.

EXTERNAL VALIDITY
• Generalizability
• Can the results can be applied in another setting or to another population of research participants?