Chapter 7 - Control Techniques in Experimental Research

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 40

Control Techniques in

Experimental Research
OUTLINE

 Control Techniques Carried Out at the Beginning of the Experiment Randomization


 Matching
 Control Techniques Carried Out During the Experiment
 Control of Participant Effects
 Control of Experimenter Effects
 You already have your sample of participants,
so that we can focus on control techniques for
maximizing internal validity (i.e. causation).

 The ideal situation in experimental research


would be to randomly select your sample and
then randomly assign the participants to the
groups to be used in the experiment.

Illustration of the ideal procedure for obtaining participants f


an experiment.
 The primary goal when conducting a psychological experiment is to determine whether the
independent variable causes the changes observed in the dependent variable.

 To make this causal inference, we must control for the influence of extraneous variables.

 The key strategy for eliminating extraneous variables as rival explanations for claims of
causation is to produce an experimental situation that holds the extraneous variables constant
across the different levels of the independent variable.

 The experimental groups (e.g., treatment and control) should have the same levels of each
extraneous variable to eliminate any differential influence.
 The ONLY difference between the experimental groups should be the levels of the independent
variable.

 This phenomenon is called Method of Difference = If groups are equivalent on every variable
except for one, then that one variable is the cause of the difference between the groups
Control Techniques Carried Out at the Beginning
of the Experiment Randomization
 Randomization (also called random assignment) is the most important and basic of all the
control methods.

 It is a probabilistic control technique designed to equate experimental groups at the start of an


experiment on all extraneous variables, both known or unknown.

 It’s randomly assigning a sample of individuals to a specific number of comparison groups.

 Random assignment is the only technique for controlling both known and unknown sources of
extraneous variation.
 Random assignment of participants to the experimental groups assures that each participant
has an equal chance of being assigned to each group.

 In order to achieve equiprobability of events when randomly assigning participants to


treatment conditions, it is necessary to use a randomization procedure.

 Random assignment produces control by virtue of the fact that all variables present in
a group of participants will be distributed in approximately the same manner in all
groups.

 When the distributions of extraneous variables are approximately equal in all groups,
the influence of the extraneous variables is held constant because they cannot exert
any differential influence on the dependent variable.
 For example, gender cannot be the cause of the difference found between two groups if 58% of
the treatment group and 58% of the control group participants are women;.
 For example, Professor X was conducting a study on learning. The extraneous variable
intelligence is correlated with learning, so this factor must be controlled for, or held constant.
 Assume first that no random assignment of participants existed (no control), but that the first 10
participants who showed up for the experiment were assigned to treatment Group A, and the
second 10 participants were assigned to treatment Group B. Assume further that the results of the
experiment revealed that treatment Group B learned significantly faster than treatment Group A.
 Is this difference caused by the different experimental treatments that were administered to the
two groups or by the fact that the participants in Group B may have been more intelligent than
those in Group A?
 Suppose the investigator also considers the intelligence factor to be a possible confounding
variable and gives all participants an intelligence test.
 Intelligence is, therefore, a potentially confounding variable and serves as a rival
hypothesis for explaining the observed performance difference in the two groups.
 To state that the treatment conditions produced the observed effect, researchers must
control for potentially confounding variables such as IQ.
 One means of eliminating such a bias is to randomly assign the 20 participants to the two
treatment groups.

 Random number generator (http://randomizer.org).


 One more example will make this process fully clear! Let’s assume that you have 30
participants, and you want to randomly assign them to three groups, with 10 people
in each group. Here’s how to use the randomizer.org program for random
assignment. Go to the Web site and answer the questions as follows:
Matching

 When random assignment is not possible, matching can be an effective technique to


equate groups.

 If participants in different treatment conditions are matched on intelligence, then the


intelligence level of the research participants will be the same in each treatment group
—intelligence is held constant and therefore controlled.

 The key weakness of matching is that the groups are equated only on the matching
variables.
A- Matching by Holding Variables Constant
• One technique that can be used to control an extraneous variable is to hold the
extraneous variable constant for all groups in the experiment.
• This means that all participants in each treatment group will have the same degree
or type of extraneous ­variable.
• If we are studying conformity, then gender of participants needs to be controlled
because conformity has been shown to vary with the gender of the participant.
• The gender variable can be controlled by using only female (or only male)
participants in the experiment.
 Disadvantages of this technique:
- restricts the size of the participant population. Consequently, in some cases, it might be
difficult to find enough participants to participate in the study.
- results of the study can be generalized only to the type of participant who participated in
the study.
Illustration of matching by holding variables constant.
B- Matching by Building the Extraneous Variable into the Research Design OR blocking
 We are conducting a learning experiment and want to control for the effects of intelligence.

 We decided to select several IQ levels (e.g., 90–99, 100–109, and 110–119), and treat them as
we would an independent variable.

 This would allow us to control and detect the influence of the intelligence variable.

 Differential influence of intelligence is eliminated because the treatment groups are compared
within each of the three intelligence levels.
 The technique is recommended if one is interested in the differences produced by the
various levels of the extraneous variable or in the interaction between the levels of the
extraneous variable and other independent variables.

 In the hypothetical learning experiment, one might be interested in the differences


produced by the three levels of intelligence and how these levels interact with the learning
strategies.
Illustration of matching by building the extraneous variable into
the research design.
 Rather than categorizing intelligence into three groups, few psychologists recommend that the
matching variable, such as intelligence, be left in its natural units and then entered into the
study during statistical analysis.

 This form of control is sometimes called statistical control because it is done during data
analysis.

 Statistical control is much more important in quasi-experimental designs than in randomized


experimental designs because quasi-experimental designs lack random assignment, and it is
wise for the researcher to determine the variables the groups might differ on, measure those
variables, and then control for those variables during data analysis
C- Matching by Yoked Control
 A matching technique that matches participants on the basis of the temporal sequence of
administering an event.
 For example, if you want to know the effect of students having the freedom to choose snack
breaks on their classroom productivity, it would be important to know if the effects are due to
having the freedom to choose a snack break or due to having a snack break (determined by
the teacher).
 In a yoked control experiment, each control participant is “yoked” to an experimental
participant. Therefore, when the experimental participant engages in a behavior and receives
an outcome (e.g., taking a self-chosen snack break), the “yoked” participant is given the same
outcome (teacher tells the student to take a snack break).
 If the experimental group scores higher on the dependent variable than the control group, it
can be attributed to the freedom to choose and receive the snack rather than just getting a
snack break.
D- Matching by Equating Participants OR individual matching OR subject matching
• A matching technique in which each participant is matched with another participant on
selected variables.
• You start with a sample of participants to participate in your research study. Then you order
them on the matching variable.
• For example, you could rank all of the participants from lowest to highest IQ scores. Next
you match participants; if you have two treatment groups then the two participants with the
lowest IQ scores would be your first set. You would then randomly assign each of these two
individuals to one of the two treatment groups. You continue this process until you get to the
two participants with the highest IQ scores and randomly assign them to the two groups.
Illustration of individual matching technique.
 This technique is used in strong experimental research where you can randomly
assign each set of matched participants to the treatment groups.

 Also used in quasi-experimental research (where random assignment is not possible)


and in nonexperimental research (where there is no active manipulation).
 Individual matching has three major disadvantages when used without random
assignment as its final step:
- the researcher does not know what extraneous variables the groups differ on, and what to
match.
- difficulty in finding matched participants increases disproportionately as the number of
matching variables increases.
- matching can limit the generalizability of the results of the study. This will occur if you
have to throw out participants for whom you cannot find adequate matches.
Control Techniques Carried Out During the
Experiment
 You must treat the different groups in the same way during the conduct of the
experiment, except for administration of the different levels of the independent
variable.
 We now explain the most important control techniques for employment during the
experimental study.
A- Counterbalancing
Control of Participant Effects
A- Double-Blind Placebo Method
 One of the best techniques for controlling demand characteristics is the double-blind placebo
method.
 Neither the experimenter nor the research participant is aware of the treatment condition
administered to the participant.
 If you were conducting an experiment designed to test the effect of aspartame on disruptive
behavior in young children, you would have to administer this sweetener to one group of children
and a placebo to another group. Expectancies would need to be held constant for both groups. To
make this a double-blind experiment, the experimenter must not know whether a given participant
received the aspartame or the placebo in order to avoid communicating any expectancy of
generating disruptive behavior and participants must have the same perceptions of their condition.
 Use of the double-blind placebo method is a way to eliminate the development of
differential participant perceptions because all participants are told the same thing (that
they might or might not receive the treatment).

 And because the researcher does not know which participants have received the
experimental treatments, he or she cannot communicate this information to the
participants.

 Unfortunately, many types of experiments cannot use such a technique because all
conditions cannot be made to appear identical in all respects. In such cases, other
techniques must be employed.
B- Deception
 Deception involves providing all research participants with a hypothesis that is unrelated to or
different from the real research hypothesis.
 Deception can range from minor deceit (an omission or a slight alteration of the truth) to more
elaborate schemes.
 Is it better to use such deception or simply to refrain from giving any rationale for the
tasks to be completed in the experiment?
 It seems as though providing participants with a false, but plausible, hypothesis is the
preferred procedure because the participants’ curiosity might be satisfied so that they do
not try to devise their own hypotheses.
 If different participants perceive the study to be investigating different hypotheses, their
responses can create a source of bias.
C- Control of Participant Interpretation
Techniques that can be used to gain insight into participants’ perceptions:
1- A retrospective verbal report consists of a technique such as the postexperimental inquiry,
which is exactly what it says it is: questioning the participant regarding the essential aspects of the
experiment after completion of the study. What did the participant think the experiment was
about? What did he or she think the experimenter expected to find?

- Such information will help to expose the factors underlying the participant’s perception and
ways that perception might have influenced behavior.
- The primary disadvantage of retrospective reporting is that participants might fail to recall and
report perceptions they had earlier in the experiment.
2- Concurrent verbal reports include techniques such as:
In Solomon’s sacrifice groups, each group of participants is “sacrificed” by being stopped at a
different point in the experiment and probed regarding the participants’ perceptions of the
experiment.
You will not have data on the dependent variable from the participants that are “sacrificed,” but
you will gain insight into how the experiment is understood by the participants.
“Sacrificing” can be done at different points in the procedure, not just at the end of the
experiment, as in retrospective reporting.
- Concurrent probing requires participants to report at the end of each trial the perceptions they
have regarding the experiment.
- The think-aloud technique requires participants to verbalize any thoughts or perceptions they
have regarding the experiment while they are performing the experimental task.

- The obvious disadvantage to concurrent probing and the think-aloud technique is that
verbalizing one’s thoughts during the experiment might affect participants’ behavior and,
therefore, the dependent variable
Control of Experimenter Effects

 Experimenter effects = The biasing influence that can be exerted by the experimenter.

 Therefore, potential experimenter effects must be eliminated or minimized.


A- Control of Recording Errors
 Errors resulting from the misrecording of data can be minimized if the person recording
the data remains aware of the necessity of making careful observations to ensure the
accuracy of data transcription.

 An even better approach is to use multiple observers or data recorders.


B- Control of Experimenter Attribute Errors
 The influence of experimenter attributes should be held constant across all treatment
conditions.
 As long as experimenter 1 administers condition A and condition B, any observed difference
between conditions A and B cannot be due to the experimenter (because the experimenter was
constant across groups A and B).
 The key point here is to never use one experimenter in one condition and a different
experimenter in the other condition; this would make the groups different not just on the
independent variable but also on the experimenter.
 In a weight reduction experiment, the weight of the therapist might be correlated with the
success of the program. Therefore, to identify the relative effectiveness of different weight
reduction techniques, it would be necessary, at the very least, to make sure the therapists were
of approximately the same weight. Such an attribute consideration might not, however, have
an influence in a verbal learning study.
C- Control of Experimenter Expectancy Error
A number of techniques can be used for eliminating or, at least, minimizing expectancy
effects.

1- The Blind Technique


 In the blind technique, the experimenter knows the hypothesis but is blind as to which
treatment condition the research participant is in.
 Consequently, the experimenter cannot unintentionally treat groups differently.
2- The Partial Blind Technique
 The experimenter is kept ignorant of the condition the research participant is in for a portion of
the study.
 The experimenter could remain blind while initial contact was made with the participant and
during all conditions prior to the actual presentation of the independent variable.
 When the treatment condition was to be administered to the participant, the experimenter could
use random assignment to designate which condition the participant was in.
 Therefore, all instructions and conditions prior to the manipulations would be standardized and
expectancy minimized.
 If the experimenter can leave the room immediately following administration of the
independent variable and allow another person (who is ignorant of the experimental
manipulations administered to the participant) to measure the dependent variable, the
solution would come closer to approaching completeness.
3- Automation
 Many human studies can also be completely automated by having instructions written, tape-
recorded, filmed, televised, or presented by means of a computer, and by recording responses
via timers, counters, pen recorders, computers, or similar devices.
 These procedures are easily justified to the participant on the basis of control and
standardization, and they minimize the participant–experimenter interaction.

You might also like