Welcome to this tutorial on confounding variables and their control.

Confounding variables can be viewed as tricksters who remind us that things may not

be as they seem. In an experiment, a confounding variable is an independent variable

that is conceptually distinct but empirically inseparable from one or more independent

variables. The presence of a confounding variable makes it impossible to distinguish a

variable’s effects in isolation from its effects in conjunction with other variables.

For example, in a study of high school achievement, whether a student attended a

private or a public school prior to high school (the variable of interest) and the student’s

prior academic achievement (the confounding variable) overlap.

A critical step in research design is anticipating possible confounding variables that

might, on closer inspection, cast doubt on the results of the study.

These extraneous variables are not the principal interest of the investigation but may
still affect the outcome and thus influence the results. They must be anticipated and
carefully controlled for during the research study.

Understanding which confounding variables need to be controlled is the first step.

Understanding which methods of control are most appropriate is the second.

By the end of this tutorial, you should be able to

 identify the principal sources of major confounding variables and the different
forms of group equivalence
 examine the effects of sampling error on within-subject design as well as
motivation and role perception on a study
 examine the control method for the confound of communication among
participants
 examine why the placebo effect is a clinically significant response to treatment
 identify the experimenter effect and experimental procedural impact and the
effect on results
 identify the methods of control for apparatus and raters and different nuisance
events
For each of these sources of confounds, we discuss methods for controlling
extraneous variation.

Lesson 2 of 18
Participant Demographic and Personal Characteristics 
The demographic characteristics of participants make up a large category containing
subcategories like age, sex, education, socioeconomic status, marital status, living
arrangements, employment status, and ethnic group membership. Researchers aim to
assign participants to groups so that those groups are comparable on as many
demographic characteristics as possible.
For example, if IQ is a variable that could influence scores, are the
groups comparable in IQ?

If research groups are from clinical sources, are they comparable in

diagnosis, length, and severity of illness; history of treatment; and
presence of concurrent disorders?
In studies involving more than one group, researchers aim to establish equivalence on
all factors that may affect the dependent variable except the independent variable.
Several methods are used to control confounding variables in participant characteristics.
We discuss these methods next.

Lesson 3 of 18
Random Assignment of Participants

Random assignment is the assignment of participants or units to the different conditions

of an experiment entirely at random, so that each participant or unit has an equal
likelihood of being assigned to any particular condition.

Random assignment is the best way of scrambling all possible variables across groups.
However, be aware that some assignment procedures that are labeled “random” are not
random at all.

For example, a procedure that assigns the first 50 participants to Group A and the next

50 to Group B is not random because each participant does not have an equal chance of

being assigned to the treatments. 

Participants who arrive early might differ from those who arrive late in systematic ways

—for example, greater motivation.

Note that random assignment promotes group equivalence but does not guarantee it,
particularly with small samples.
Homogenous Sample 
One way to make groups equivalent is to keep the sample narrow by restricting it to a
homogenous—that is, highly similar—set of participants.
Consider, for example, a researcher conducting a study on the treatment of depression.

The researcher decides to assure equivalence of the experimental and control group

participants by restricting the sample to highly motivated, bright, male, Black, affluent

college graduates, ages 30 to 35 years, with moderate reactive depression of short

duration.

With such a restricted pool, members of the groups resemble each other on these
individual variables, and the researcher succeeds in controlling all of these variables
from potentially being confounded with the treatment.

However, the downside of this strategy is that the generalizability of the study is likely
to be challenged, especially if the author uses the data to make assertions about
treatment of depression in the general population.

Lesson 5 of 18
Matched-Pairs Design 
The matched-pairs design is another method for controlling confounding variables.
Members of one group are paired with similar persons in the other group—that is,
someone who matches them on one or more variables that are not the main focus of the
study but nonetheless could influence its outcome.
Consider the previously described study for depression. The researcher may pair two
25-year-old, male, Black college graduates with moderate reactive depression of short
duration and randomly assign one to the experimental group and the other to the control
group.
 Control Group

Experimental Group
The next pair might be two 60-year-old, female, White high school graduates with
moderately severe depression of 5 years’ duration who are assigned to groups via
similar fully random means.
 Control Group

Experimental Group
This kind of person-for-person matching would continue until a sample of the desired
size and diversity was attained. Because the groups would be nearly identical on the
matched variables, none of the between-group differences at the end of treatment could
be attributed to these particular characteristics.

Unfortunately, most researchers do not have access to the enormous samples that would
be required for matching on numerous variables. They prefer to match on a few
important variables and to find other ways to control for the remaining ones that might
be problematic.

Lesson 6 of 18
Equated Groups 

Group equivalence can be established by first comparing the means, medians, or

percentages of the groups on important participant variables.

Data on equated groups for categorical variables are usually reported in percentages or
numbers of participants (or both) to demonstrate that differences were not significant.
Differences between the groups should be of no practical or psychological import. If
groups are nonequivalent, the researcher can drop, add, or switch group participants for
the sole purpose of equating groups, provided that it is done before any research data
have been collected.
For example, if the experimental group members are older than the control group
members, the investigator can switch two or three of the oldest members to the control
group and move two or three younger members from the control to the experimental
group.
Move these participants to the experimental group
And move these participants to the control group
When dropping or exchanging individuals, the researcher should ensure that the means
of other variables are not changed, as this would throw the group equivalence out of
balance in other ways.
Researchers can also use statistical methods to deal with nonequivalence between
groups. The researcher can examine the breakdowns of variables that might be
confounds and apply statistical controls to those that are not balanced.

For example, in a study in which people have been randomly assigned to treatment and

the mean ages of the groups are different, age can be treated as a covariate in an analysis
of covariance. Doing so adjusts for the influence of a variable that is not being

investigated (age), but nonetheless is related to the dependent variable and could

influence the study results.

Lesson 7 of 18
Blocking Variables 
A block design is a type of research study in which participants are divided into
relatively homogeneous subsets (blocks) from which they are assigned to the
experimental or treatment conditions. The purpose of a block design is to ensure that a
characteristic of the study participants that is related to the target outcome is distributed
equally across treatment conditions.

In a block design, a blocking variable is an attribute or variable used as a basis for

subdividing study participants or other sampling units into homogenous subsets.

Instead of ruling out the effect of a confounded variable such as sex, the researcher may
be interested in studying its effect and trying to determine whether it interacts with the
treatment variable.

In this case, the researcher uses the extraneous variable in the analysis and treats it like
another independent variable (although it remains confounded with the experimental
condition).

The researcher may divide participants into two blocks by sex and then randomly assign
an equal number of male and female participants to each of the treatments.

For example, in a study of the effects of drug X for

headache pain relief, the researcher has the choice of
using only men 
or only women and not being able to generalize to the omitted sex
 using an equal number of men and women in treated
and  untreated control groups, thus neutralizing the
effects of sex as a source of confound

making sex into a blocking variable, determining both

whether a sex difference in pain exists and whether
men and women react differently to the drug

Lesson 8 of 18
Within-Subjects Design 
Sampling error is the predictable error that occurs in studies that draw samples of cases
or observations from a larger population. It indicates the possible variance between the
true value of a parameter in the population and the estimate of that value made from the
sample data.
Sampling error means that different people in two groups might perform
differently from each other even if they receive the same treatment.
The larger the sample, the smaller the sampling error (for example, if the entire
population were sampled, there would be no error in the sample estimate).

Method of Control for Sampling Error

A researcher may choose to remove sampling error by using the same participants on
two or more occasions under two or more different conditions. This can be termed
within-subjects, a repeated-measures design in which each participant is exposed to an
experimental and a control condition. Within-subjects designs can be adopted to control
some of the confounding variables that materialize when different groups are used.
For example, if members of one group have a higher IQ than members of
the other group, this difference could be confounded with a learning
experience. Such problems are ruled out when participants serve as their
own controls (although other issues of internal validity do arise in this
design).
Group 1: Average IQ 120
 Group 2: Average IQ 100
Questions About Within-Subjects Designs
 Could something else have happened between the first and second condition that
might have caused any differences?  
 Although the same participants are involved on multiple occasions, are they
truly the same on the second occasion as they were on the first, or have they changed
over time?
 Has exposure to initial experience changed them in a way that would alter their
performance on a second occasion, even if the conditions were identical with the first?
 Could one group have done better than the other because they had received some
treatment before? Has the researcher taken the precaution of counterbalancing the order
of treatments?
 Extraexperimental Changes in Participants
 Sometimes, in a study in which considerable time elapses between the pretest
and posttest measures, participants may undergo changes that have little or
nothing to do with the experimental treatment. Extraexperimental changes are
changes that occur as a result of factors extraneous to the study design.


 Participants may grow or otherwise change in physical or mental

state (for example, become more depressed or happier) for reasons
unrelated to the treatment.


 Changes may also occur because of external events outside the

experimental treatment. These are extraexperimental changes in
participants.
 Extraexperimental change cannot be prevented over the long run, but it can be
controlled by random assignment. If the assignment has been truly random, the
odds are greatly against anything happening systematically to one group that
does not happen to the other.

Lesson 10 of 18
 Motivation and Role Perception 
 In an experiment, the personal motivation of participants affects the results.


 Are groups equally motivated to participate? Do they gain the same personal

reward? Unequal benefits can affect motivation, as in one group of individuals

who have high hopes of obtaining symptom relief and another group who do not

share this hope.

 Participants’ perception of their role in research can also differ systematically in

study groups. They may define themselves, for example, as “good participants,”

“resistant participants,” or “negative participants." 



 Ego, motivation, benefits received from participation, and the need to be good

participants can all affect the study and act as possible confounds, if participants

differ in them across conditions.

  Methods for Control
 To control these confounds, the researcher strives to keep motivation constant by
ensuring that each group of participants receives the same benefits and rewards.
Tests and measures constructed to control role selection and response sets are
also useful.
 To learn more, see the tutorial Criteria and Criteria Measures.

Lesson 11 of 18
Communication Among Participants
The possibility that participants who have been seen by the researcher may
communicate their experiences with those who are waiting their turn is another potential
confound.
 1
 2
 3

Any study in which participants are drawn from within the same
institution, school, club, church, or social group and perform a task on
which performance could be affected by foreknowledge is vulnerable to
contamination by communication among participants.
Sometimes the setting of the study can lend itself to communication
among participants.
Communication can also become a confounding variable when treatment
groups are seen over a period of time and participants get to know each
other and start to exchange information.
Methods for Control
 1
 2
 3
 4

One method for control that works fairly well, at least with adults, is to
explain to each participant the reason for not talking about the session
and then ask for cooperation.
To test this, ask new participants about their expectations for the study,
and carefully analyze their responses for any evidence of foreknowledge.
Obtain participants from different places in the hope that they will not
know each other or have an opportunity to talk it over.
Another method is to have a team of data collectors work quickly and
finish before communication can become a factor.

Lesson 12 of 18
Placebo Effects and Control
A placebo effect is a clinically significant response to a therapeutically inert substance
or nonspecific treatment, deriving from the recipients’ expectations or beliefs regarding
the intervention.
For instance, a person who takes a sugar pill but believes it is medicine may feel better,

even though the pill is only sugar. This response is due to the recipient’s expectations or

beliefs about the intervention, which can directly affect what they experience and report

after receiving it.

 A person in the control group thinks, “I didn’t get the treatment, so I
won’t feel better,” 

while a person in the active intervention group thinks, “This treatment

will help me feel better”
If participants in the control and active intervention groups have different expectations
and beliefs about the intervention they receive, this introduces a serious confound. It
will be impossible to determine whether any differences between the groups are due to
the different interventions they receive or due to their different expectations and beliefs
about this intervention.
A placebo is used so that participants in the control group think they are receiving the

active intervention, and therefore have the same expectations and beliefs as participants

in the active intervention group.

Both the person in the placebo group and the person in the active
intervention group think, 

“This treatment will help me feel better”

 “This treatment will help me feel better”
If both groups have the same expectations and beliefs, this is no longer a confound and
the only difference between the groups is the intervention itself.
If both groups have the same expectations and beliefs, this is no longer a confound and
the only difference between the groups is the intervention itself.

For participants in the placebo group to think they are receiving the active intervention,
the placebo should resemble the active intervention as much as possible, but not include
its active ingredients—that is, the specific elements that are thought to cause change. In
drug outcome studies, the placebo is often a sugar pill that looks identical to the drug
pill but does not contain the active ingredient of medication.

The following are some characteristics of psychotherapy outcome studies:

 The active ingredients are the interventions specific to the

psychotherapy of interest—for instance, in cognitive behavioral
therapy, identifying and challenging irrational beliefs.
 The inactive ingredients are the nonspecific elements that could
be part of any psychotherapy, such as interacting with a
therapist.
 In psychotherapy outcome studies, these inactive ingredients
are often used for a placebo condition.
For example, in a study examining the effectiveness of
emotion-focused therapy (or EFT) for depression,

the active intervention group would meet with a therapist to receive the
EFT intervention, and
 the placebo group would meet with a therapist and receive equal
attention for an equivalent number of sessions, but the therapist would
not provide any specific treatment.
Because nothing intentionally therapeutic is done with the placebo group, the
intervention is considered to be inert.
Methods for Control

In situations in which expectancies of participants can affect results, such as drug

studies, the accepted procedure is to keep the participants unaware of which treatment

they have received by administering an inert placebo to a random half of the sample.

Lesson 13 of 18
Experimenter Effects 
Experimenters can have unintentional effects on the outcome of a study. Experimenter
effects can be divided into two broad categories: noninteractional
effects and interactional effects.
Noninteractional Effects
Noninteractional effects rest on how experimenters observe and interpret data. The

influence of experimenters first came to prominence when it was realized that

astronomers differed from each other in their recordings and observations.

Psychologists are also subject to systematic differences in observations and vary in the

recording of time and other measurements.

Methods for Control of Noninteractional Effects
To control for these errors, experimenters consider how their assumptions might
influence their work. They pay attention to slight but consistent errors of measurement
or ways of entering, transcribing, or analyzing data that could affect the results to
support their hypothesis.
Interactional Effects
Interactional effects have to do with the ways that researchers and participants
influence each other’s behavior selectively during the research.

Interactional experimenter effects can be classified into biosocial

effects and psychosocial effects.
Biosocial effects are experimenter characteristics such as age, sex, ethnic group,

and so forth, which could exert selective influence on how the parties respond.  For

example, in a study on sexuality, male participants may respond differently to female

than to male interviewers.

Methods for Control of Biosocial Effects
To control for biosocial effects, the researcher can anticipate them, rule them out or
minimize them by design, analyze them, and/or report them to readers.
Psychosocial effects are experimenter characteristics such as habits, behavior,

appearance, or manner that might have a selective bearing on how participants

respond. Included in this group are modeling effects—how researchers frown, smile,

look anxious or relieved, or distance themselves in ways that can influence participant

responses. These are called demand characteristics because the researcher is perceived

to make demands on the participant to respond in a particular way.

Methods for Control of Psychosocial Effects

Psychosocial effects can be controlled in similar ways as biosocial effects. Minimize the
researcher’s behavioral influence by doing pilot studies in which the researcher has
opportunities for practice and rehearsal. Also keep the researcher unaware of
participants’ assigned conditions (e.g., whether they were given an active or inert drug).

Lesson 14 of 18
Experimental Procedural Impact 
Sometimes variables that need to be controlled include the nature, amount, and duration
of the experimental treatment and the conditions of its application.
Did a poor translation of instructions result in lack of comprehension for
one of the groups? 
 Did delays cause another group to receive treatment just before lunch,
when participants were distracted and hungry?
The results of the experiment are affected by the
●    procedure,
●    wording of instructions to participants,
●    unplanned events,
●    order of experimental treatments,
●    timing of procedures,
●    magnitude of stimulus situations, and
●    experimental setting.
Methods for Control
Researchers strive to hold experimental procedures constant or at least keep them as
invariant as possible. As an extra safeguard, they can perform a manipulation check, or
a postsession interview, to determine how the procedure was perceived and to ensure
the integrity of the delivery.

Lesson 15 of 18
Methods for Control of Apparatus and Raters

Changes in apparatus during the course of a study can be a confound if instruments are
poorly calibrated or malfunction altogether more often in one condition than another. 

Also, judges, raters, and scorers can differ in their observations, interpretations, and
recording of events. These differences are confounding sources.
Methods for Control of Apparatus
To control for apparatus as a study confound, experimenters should 
 check their equipment regularly, 
 recalibrate their equipment if necessary, and
 attest that the apparatus was functioning correctly, working reliably, and
checked and recalibrated frequently throughout the research.

Methods for Control of Raters

To control for individual differences between judges and raters, experimenters should
 enlist competent and qualified people,
 train them for their role,
 and regularly evaluate their performance.

Lesson 16 of 18
 Nuisance Events 
 Any kind of accidental, unplanned event during the course of the experiment can
affect the results.


 For example, during a task requiring great concentration, an

airplane flying low overhead can distract all of the participants.
 

 Or on the way to treatment, one participant might be shaken up by

a fender-bender accident.
 Because these nuisance events typically happen randomly to only a
portion of the sample, they could skew the results.
 Method for Control
 The best a researcher can do is to anticipate as many things that could go wrong
as possible and to be prepared for them should they occur.

Lesson 17 of 18
Review
Let’s review our key points.
 An extraneous variable is a factor that is not under investigation in an
experiment but may potentially affect the outcome or dependent variable and
thus influence the results. Such potential influence often requires that an
extraneous variable be controlled during research.
 In an experiment, a  variable that is conceptually distinct but empirically
inseparable from one or more independent variables is known as a confound.
 In experimental design using random assignment, participants or units are
randomly assigned to the different conditions of an experiment, so that each
unit or participant has an equal likelihood of being assigned to any particular
condition.
 Group equivalence is aimed for in all experiments as a means to control for
confounds.
  Sampling error, which is the predictable error that occurs in studies that draw
samples of cases or observations from a larger population, is an important
source of error that is introduced into any research design that has different
people in each group.
 Experimenters can have unwitting effects on the outcome of a study.
 Researchers should make  a conscious effort to control for extraneous influences
and to hold conditions as constant as possible for all participants.

Lesson 18 of 18
 Related Resources
 To learn more about confounding variables and
their control, please consult the following
publications shown below and in the
downloadable Word document.
